{"id": "2512.03199", "title": "Does Head Pose Correction Improve Biometric Facial Recognition?", "abstract": "Biometric facial recognition models often demonstrate significant decreases in accuracy when processing real-world images, often characterized by poor quality, non-frontal subject poses, and subject occlusions. We investigate whether targeted, AI-driven, head-pose correction and image restoration can improve recognition accuracy. Using a model-agnostic, large-scale, forensic-evaluation pipeline, we assess the impact of three restoration approaches: 3D reconstruction (NextFace), 2D frontalization (CFR-GAN), and feature enhancement (CodeFormer). We find that naive application of these techniques substantially degrades facial recognition accuracy. However, we also find that selective application of CFR-GAN combined with CodeFormer yields meaningful improvements.", "date": "2025-12-02", "categories": []}
{"id": "2606.11615", "title": "Adv-TGD: Adversarial Text-Guided Diffusion for Face Recognition Impersonation Attacks", "abstract": "The widespread adoption of face recognition (FR) technologies raises serious privacy concerns, as facial data can be exploited without consent. To address this challenge, we propose Adv-TGD, a generative adversarial attack framework that synthesizes photorealistic faces capable of impersonating target identities and deceiving face recognition systems. Built upon Stable Diffusion v2.1, Adv-TGD performs per-sample LoRA fine-tuning conditioned on concise textual prompts to generate natural yet adversarially manipulated identities. Unlike conventional identity attack approaches, our method optimizes lightweight cross-attention adapters for each source-target pair within a fixed-timestep denoising process. Latent blending is constrained by a face-local heatmap mask to ensure spatially precise identity manipulation while preserving non-sensitive regions. We introduce a composite objective that integrates masked epsilon-MSE reconstruction, thresholded identity divergence in FR embedding space, directional feature alignment, and source-similarity suppression to balance adversarial attack and visual realism. Optionally, LLaVA-generated attribute prompts enhance fine-grained semantic details without reintroducing identity cues. Under the black-box evaluation protocol, Adv-TGD attains an average attack success rate (ASR) of 85.90% across IR152, IRSE50, MobileFace, and FaceNet, surpassing the semantic SOTA baseline Adv-CPG by 6.25 points, the diffusion-based makeup method DiffAIM by 3 points, and the noise-based P3-Mask by 16 points. Despite its strong attack efficacy, Adv-TGD preserves high visual fidelity (PSNR = 28.18 dB, SSIM = 0.981). Furthermore, we demonstrate the flexibility of our framework by successfully extending it to in-the-wild datasets (LADN), general object classification (ImageNet), and transformer-based diffusion models (FLUX.1).", "date": "2026-06-10", "categories": []}
{"id": "2606.18510", "title": "Architectural Bias in Face Presentation Attack Detection: A Comparative Study of Vision Transformers and Convolutional Neural Networks", "abstract": "Face Presentation Attack Detection (PAD) systems constitute a critical security layer in biometric authentication; however, existing approaches exhibit systematic performance disparities across demographic groups, disproportionately affecting individuals with darker skin tones. This paper presents a comparative empirical investigation of whether Vision Transformer architectures reduce demographic bias in face PAD systems relative to convolutional baselines. Experiments are conducted on the CASIA-SURF Cross-Ethnicity Face Anti-Spoofing (CeFA) dataset. Three architectures are evaluated: a Multimodal ViT-Tiny trained from scratch, a ResNet18 CNN baseline, and a pretrained DeiT-S fine-tuned on CeFA across African, East Asian, and zero-shot Central Asian demographic groups. DeiT-S achieves the highest overall accuracy of 97.27% and the lowest EER of 0.86%, outperforming ResNet18 at 90.15% accuracy. In terms of fairness, DeiT-S reduces the inter-ethnic ACER gap between African and East Asian subjects to 0.13%, compared to 0.75% reported in an LBP-based work [6], representing an 83% reduction. Most notably, while ResNet18 records a BPCER of 10.44% on zero-shot Central Asian subjects, DeiT-S maintains 2.89% on the same unseen group, demonstrating a 3.6x generalization advantage. These results suggest that pretrained Vision Transformers achieve superior PAD accuracy, produce smaller demographic performance gaps, and generalize more equitably across unseen demographic groups, indicating that cross-demographic fairness in PAD may partly be influenced by architectural design.", "date": "2026-06-16", "categories": []}
{"id": "2606.18528", "title": "A Prototypical Signature Approach for Writer-Independent Offline Signature Verification", "abstract": "Offline handwritten signature verification aims to distinguish genuine from forged signatures using static images. Since real forgeries are rarely available, negative samples are usually randomly drawn from genuine signatures of other users to create training data. However, this random selection often lacks diversity, increases redundancy, and escalates computational cost, leading to inefficient training. We propose a data-driven strategy to generate diverse, informative negative samples using prototypical signatures, which are compact, non-identifiable summaries of genuine signature features. Based on the experiments results, we conclude that (i) prototypical signatures yield more informative negative samples, improving the detection of skilled forgeries; (ii) the proposed approach is backbone-agnostic, showing robustness across architectures; and (iii) when combined with a primal-form linear SVM, it serves as an alternative to RBF-based models while significantly improving scalability and computational efficiency. Implementation of the method is available at https://github.com/kdmoura/proto_hsv.", "date": "2026-06-16", "categories": []}
{"id": "2606.17436", "title": "UoU: A Universal Fingerprint Foundation Model Based on Large-Scale Unsupervised Learning", "abstract": "Fingerprint recognition is still dominated by task-specific pipelines, where enhancement, structural parsing, alignment, and matching are optimized in isolation. Although effective in narrow settings, this design limits representation reuse across sensors, qualities, and downstream applications. We therefore present UoU, short for ``a \\textbf{U}niversal fingerprint foundation model based \\textbf{o}n large-scale \\textbf{U}nsupervised learning,'' which reframes fingerprint feature extraction as a domain-specific foundation-model problem. UoU is organized around a multi-level representation hierarchy spanning image restoration, structural fields, semantic tokens, point-level biometric entities, and compact global descriptors. Its training recipe combines a supervised cold start on precise annotations, large-scale weakly supervised refinement, and large-scale unsupervised consolidation, with the latter two stages iterated during large-scale training so that weak supervision broadens semantic coverage while unsupervised learning stabilizes correspondences, invariances, and representation geometry. Rather than treating fingerprint imagery as generic texture, UoU exploits domain-specific symmetries and intermediate structure, including orientation flow, periodic ridge patterns, sparse biometric entities, and spatial equivariance. The framework is intentionally architecture-agnostic: while the present study includes an initial transformer-based structured-prediction instantiation, the broader design supports multi-task learning, scalable model configurations, and downstream specialization for matching, alignment, enhancement, registration, and related fingerprint applications. This paper presents the technical motivation, system design, and validation protocol of UoU, and part of the baseline implementation is publicly available at https://github.com/XiongjunGuan/UoU.", "date": "2026-06-16", "categories": []}
{"id": "2605.15796", "title": "Cross-Modal Registration Between 3D and 2D Fingerprints via Pose-Aware Unwrapping and Point-Cloud Fusion", "abstract": "Three-dimensional (3D) fingerprints preserve global finger geometry and local ridge structure while avoiding contact-induced deformation, but they remain difficult to integrate with legacy two-dimensional (2D) fingerprint systems. This paper addresses the intermediate stage between 3D acquisition and cross-modal matching, and presents a unified framework for 3D fingerprint preprocessing and registration across contactless and contact-based 2D modalities. The framework combines four components: 1) a nonparametric visualization and unwrapping method that converts a 3D fingerprint point cloud into a rolled-equivalent 2D representation without relying on a global finger-shape model; 2) a point-cloud fusion pipeline that registers and mosaics multiple partial 3D captures into a more complete fingerprint model; 3) an ellipse-based pose normalization method for canonical finger alignment; and 4) a pose-aware cross-modal registration strategy that improves compatibility between 3D fingerprints and both contactless and contact-based 2D fingerprints. Experiments on a self-collected multimodal fingerprint database containing 150 fingers show that the proposed framework achieves ridge-level 3D registration accuracy, robust pose estimation, and consistent gains in 2D compatibility. In particular, the 3D fusion error is concentrated around 0.09 mm, contactless 2D--3D registration reaches ridge-scale projection accuracy, and pose-aware unwrapping improves genuine matching scores relative to generic 3D unwrapping. These results support the use of 3D fingerprints as an effective geometric bridge across heterogeneous fingerprint modalities. The baseline implementation has been publicly released at https://github.com/XiongjunGuan/3DFpVisual.", "date": "2026-05-15", "categories": []}
{"id": "2606.15355", "title": "Sustainable Face Recognition on Low-Power Devices with VQ-VAE Embeddings", "abstract": "Face recognition has become a cornerstone of modern AI applications, yet conventional approaches often rely on computationally intensive models deployed in cloud environments, leading to increased network traffic, high energy consumption, and a heavy carbon footprint. This work introduces a sustainable, edge-deployable face recognition framework based on Vector-Quantized Variational Autoencoders (VQ-VAE), which generates compact and semantically rich latent representations of facial images. By leveraging the compression capacity and reconstruction quality of VQ-VAE embeddings on the edge and combining them with the power of pre-trained face embeddings in a knowledge distillation setup, our system achieves comparable accuracy to state-of-the-art face embedding models while significantly reducing memory and computation requirements on the edge, making it suitable for low-power edge devices. The integration of VQ-VAE compression minimizes network overhead while keeping the matching accuracy high by retaining only the most informative facial features in the latent space. As a result, the reconstructed images preserve the key identity characteristics, improving the robustness and overall performance of the face embeddings.", "date": "2026-06-13", "categories": []}
{"id": "2606.15574", "title": "Toward the Whole Picture: Accumulative Fingerprint Mapping and Reconstruction for Small-Area Mobile Sensors", "abstract": "Small-area fingerprint sensing on mobile devices creates a fundamental mismatch between acquisition and recognition: each touch captures only a tiny, pose-varying local patch, while reliable biometric matching ultimately requires a stable and sufficiently complete fingerprint representation. Existing pipelines largely cope with this mismatch by treating repeated touches as independent partial templates, which leads to repeated registration, repeated matching, and no guarantee of adequate global coverage. In this paper, we advocate a different formulation, namely \\emph{accumulative fingerprint mapping and reconstruction} for small-area mobile sensing. Rather than matching every partial patch separately, the proposed perspective converts a sequence of local observations into a unified fingerprint state that is progressively refined as new touches arrive and can be matched only once after consolidation. As a concrete baseline, we present a classical pipeline that performs patch-wise structural feature extraction, feature-level registration and fusion, fingerprint map construction, and phase-based ridge reconstruction. More importantly, we position this baseline within a broader mobile fingerprint framework that integrates structured token learning, two-stage pose reasoning, and diffusion-based generative reconstruction. This viewpoint reframes mobile fingerprint recognition from multi-capture multi-match processing to accumulative map building, state refinement, and one-shot matching, offering a principled route toward efficient, pose-robust, and deployment-friendly biometrics for small-area mobile platforms. The baseline implementation has been publicly released at https://github.com/XiongjunGuan/FpReconstruction.", "date": "2026-06-14", "categories": []}
{"id": "2606.16161", "title": "Multimodal LLM-Empowered Re-Ranking for Generalizable Person Re-Identification", "abstract": "Domain Generalizable (DG) person re-identification (Re-ID) has attracted growing research interest due to its potential for deployment in unseen real-world scenarios. Most existing approaches address DG Re-ID by focusing on training domain-generalizable encoders but ignore the possible refinements in inference stage. In contrast, this work explores an alternative direction which improves inference re-ranking to enhance DG Re-ID. Conventional re-ranking methods typically rely on neighborhood-based distances to refine the initial ranking list, inherently depending on features produced by the Re-ID encoder. However, they deteriorate on target domains since the encoder lacks sufficient generalizability to produce reliable feature distances on unseen scenarios. Inspired by the remarkable generalization capabilities of recent Multimodal Large Language Models (MLLMs), we propose an MLLM-empowered distance metric to improve re-ranking in DG Re-ID. Specifically, we first adapt an MLLM to Re-ID data through supervised fine-tuning, which incorporates a domain-agnostic prompt and a query-candidate hard mining scheme. Then, the adapted MLLM is employed to compute a $\\mu$-distance during inference, which is robust to domain gap and significantly enhances subsequent re-ranking performance. Our approach is model-agnostic and can be seamlessly integrated into previous re-ranking frameworks. Extensive experiments demonstrate that our approach consistently yields substantial performance improvements across multiple DG Re-ID benchmarks. The code of this work will be released at https://github.com/RikoLi/MUSE soon.", "date": "2026-06-15", "categories": []}
{"id": "2601.02177", "title": "Why Commodity WiFi Sensors Fail at Multi-Person Gait Identification: A Systematic Analysis Using ESP32", "abstract": "WiFi Channel State Information (CSI) has shown promise for single-person gait identification, raising interest in its use for contactless biometrics, continuous authentication, and passive identification. However, the feasibility of multi-person identification on low-cost commodity devices remains unclear. A critical question is whether weak multi-person performance is primarily an algorithmic limitation, or whether it reflects a more fundamental sensing ceiling on commodity WiFi hardware. We address this question through a systematic empirical study using commodity ESP32 WiFi sensors. We evaluated six different signal separation methods--FastICA, SOBI, PCA-ICA, NMF, Wavelet, and Tensor decomposition--across seven scenarios spanning 1-10 people in both controlled and realistic indoor environments. To investigate beyond classification accuracy, we introduce three diagnostic metrics: intra-subject variability (ISV), inter-subject distinguishability (ISD), and performance degradation rate (PDR). In all methods, performance remains moderate (39%-56% accuracy), with limited evidence that algorithmic choice alone solves the problem. The best-performing method, NMF, reaches 56% accuracy, while all methods exhibit extremely high feature-space overlap (97%-99%), unstable within-subject representations, and marked environmental sensitivity. These findings suggest that, under commodity ESP32 CSI constraints, dense multi-person gait identification is limited more by sensing quality and spatial diversity than by the chosen separation algorithm. Our results have direct implications for security and privacy: they call into question the practicality of commodity WiFi CSI as a robust multi-user biometric primitive for authentication, while also placing important bounds on the passive identification capabilities achievable with low-cost off-the-shelf WiFi hardware.", "date": "2026-01-05", "categories": []}
{"id": "2606.13410", "title": "Person Identification from Contextual Motion", "abstract": "We consider the problem of identifying people based on their motion styles. We present a generative model describing the action instance creation process and derive a probabilistic identity inference scheme for two common person identification scenarios motivated by the surveillance and authentication applications. We introduce a novel, \\emph{interactive}, scenario for person identification from motion patterns. To this end, we formalize the identification process in the context of a sequential message exchange session between the subject and the system. The subject's behavior is modeled using a probabilistic generative model inspired by the Human Information Processing (HIP) paradigm. At each stage, the system presents a visual stimulus (a cue) to the subject and records their motion response. The cue is selected so as to maximize the mutual information of the expected response and the subject's identity. Once recorded, the response is used to update the a posteriori probability over possible subjects' identities. The process terminates once a sufficient classification confidence level is reached. To the best of our knowledge, this is the first time person identification is addressed in such interactive setting. We report high recognition rates on five publicly available datasets and our own novel dataset consisting of 4,476 recordings of 22 test subjects responding to 15 cues.", "date": "2026-06-11", "categories": []}
{"id": "2606.13528", "title": "What's Old is New Again: Classical Dimensionality Reduction for Efficient Saliency-Guided Biometric Attack Detection", "abstract": "Saliency-guided training is a paradigm in visual recognition that encourages models to focus on the most relevant image regions during learning. While its application in biometric presentation attack detection (PAD) has shown strong benefits in robustness and generalization, adoption is often limited by the high cost, domain specificity, and limited scalability of existing saliency acquisition methods, such as human annotations over a limited dataset. We present a novel, cost-efficient, and highly-scalable approach to saliency acquisition using maps inspired by classical dimensionality reduction techniques: PCA and LDA. Our proposed methods generate saliency maps directly from raw training data, requiring no human annotation nor domain knowledge. We contextualize the effectiveness of these saliency sources in three saliency-explored domains (iris PAD, synthetic face detection, fingerprint PAD) and demonstrate its scalability in two saliency-novel domains (fingerprint vein PAD and ID card PAD). Across all domains tested, models trained using dimensionality reduction-sourced saliency maps exceed baseline and sometimes SOTA saliency methods without any resource investment or domain-specific tooling. Our findings overcome an important yet unaddressed barrier to saliency-guided training for biometric attack detection and beyond.", "date": "2026-06-11", "categories": []}
{"id": "2606.11505", "title": "On the Study of Biometric Spoofing Detection using Deep Learning", "abstract": "Biometric systems are increasingly deployed in security applications; however, they remain vulnerable to spoofing attacks, in which attackers exploit counterfeit biometric data to gain unauthorized access. This research evaluates the effectiveness of state-of-the-art machine learning models, MobileNetV2, DenseNet-121, Inception-v3, and Spoof Trace Disentanglement (STD) in detecting spoofing attacks within facial recognition systems. Using the CelebA-Spoof dataset, the study evaluates model effectiveness using metrics such as accuracy, precision, recall, and F1 Score. Cross-dataset validation is carried out on the MSU-MFSD dataset to assess generalizability. The results show MobileNetV2 as the most efficient model, achieving 92% accuracy while balancing computational effectiveness, making it appropriate for real-life applications. Inception-v3 shows moderate robustness, while DenseNet-121 and STD struggle with generalization. The findings highlight the need for advances in domain adaptation and hybrid architectures to enhance biometric security systems.", "date": "2026-06-09", "categories": []}
{"id": "2606.11661", "title": "Learning Instance-Adaptive Low-Rank Orthogonal Subspaces for Clothes-Changing Person Re-Identification", "abstract": "Clothes-changing person re-identification (CC-ReID) aims to recognize individuals despite drastic appearance changes caused by clothing variation. While existing methods rely on adversarial learning to disentangle clothing features, we propose Ortho-ReID, which explicitly models a low-rank clothing subspace from VLM text descriptions and extracts clothing-invariant representations via direct geometric constraints. A critical component is our transformer-based Basis Maker, which refines a shared, low-dimensional clothing prior into an instance-adaptive low-rank subspace through cross-attention with image patches, enabling robust clothing feature extraction even under varying visibility conditions. This instance-adaptive subspace is supervised via alignment with clothing text embeddings, while identity features are extracted via a learnable projection head and geometrically constrained to be strictly orthogonal to it. Extensive experiments demonstrate state-of-the-art performance on PRCC (+5.9% top-1), Celeb-reID-light (+3.5%), and LaST (+5.3%), with competitive results on LTCC.", "date": "2026-06-10", "categories": []}
{"id": "2606.12023", "title": "ViT-FREE: Efficient Face Recognition via Early Exiting and Synthetic Adaptation", "abstract": "Vision Transformers (ViTs) have gained significant attention in computer vision and shown strong potential for face recognition (FR). However, their high computational cost makes deployment on resource-constrained devices challenging, motivating the need for methods that balance efficiency and accuracy. In this work, we investigate early exiting in pretrained ViTs as a simple yet effective training-free strategy for efficient FR inference. Leveraging the uniform feature dimensionality across transformer encoder blocks, we introduce ViT-FREE, a multi-exit framework that enables face verification directly from intermediate representations without modifying or retraining the backbone model, and thus, reducing inference cost. Empirically, we show that patch embeddings and attention maps evolve progressively across depth, exhibiting high similarity between consecutive ViT blocks and increasing alignment with the final representation. This indicates gradual feature refinement and attention convergence, suggesting that intermediate layers already provide stable and discriminative representations suitable for early exiting. Through extensive experiments on multiple FR benchmarks, we systematically analyze the accuracy-efficiency trade-off across exit depths. Our results demonstrate that later exits achieve a highly favorable balance, with exiting at layer 10 yielding up to a 20% speedup while incurring only a 1.5 drop in verification performance on benchmarks such as IJB-C. Also, we propose ViT-FREE_FT, a lightweight exit-specific fine-tuning strategy that adapts only the projection layers using a small synthetic dataset while keeping the transformer backbone frozen. This approach improves the performance of shallow exits while preserving the efficiency benefits and leaving deeper exits largely unaffected.", "date": "2026-06-10", "categories": []}
{"id": "2606.12036", "title": "Vision Transformers for Face Recognition Need More Registers", "abstract": "Recent advances in Vision Transformers (ViTs) for face recognition (FR) have moved beyond the standard CLS-token paradigm. In this paradigm, a special classification token (CLS) is prepended to the patch embeddings and used as a representation of the input for downstream tasks. An alternative approach, Concatenated Patch Embeddings (CPE), instead leverages all patch tokens by concatenating them into a single vector, which is then projected into a compact face representation. CPE has been shown to improve recognition performance in comparison to CLS-based ones, but our qualitative analysis of attention maps showed the presence of artifacts that limit their interpretability. To address this issue, we incorporate register tokens, learnable tokens concatenated to the initial patch embeddings, and processed jointly through the ViT encoder blocks. This mechanism has been shown to produce more structured and interpretable attention maps compared to baseline ViT. We empirically demonstrate that these artifacts consistently appear across various ViT backbones, including small and large models, and that introducing register tokens effectively mitigates them. Adding four or eight registers significantly enhances interpretability, with eight registers providing the highest verification accuracies and smoothest attention structures. Our resulting model, ViT-8R, corresponds to a CPE-based ViT-B architecture augmented with eight register tokens achieves state-of-the-art performance among ViT-based FR models on large-scale IJB-B and IJB-C benchmarks. Also, ViT-8R produces substantially clearer attention maps compared with the baseline model, which offer deeper insight into the model's attention behavior (https://github.com/TaharChettaoui/ViT-FR-Registers)", "date": "2026-06-10", "categories": []}
{"id": "2606.12051", "title": "MFEN:Multi-Frequency Expert Network for Visible-Infrared Person Re-ID", "abstract": "Visible-infrared person re-identification (VI-ReID) is challenging due to the large modality discrepancy between visible and infrared images. We contend that this discrepancy is largely related to differing lighting conditions, including differences in light wavelength and light source type. Recently, frequency-based VI-ReID approaches have achieved notable success because frequency information can better extract identity-relevant contours and details while excluding irrelevant lighting and color. However, existing methods either do not distinguish different frequency bands or focus on only one band, which is insufficient under diverse lighting conditions. To perform comprehensive frequency domain learning, we propose a Multi-Frequency Expert Network (MFEN) that enables multi-frequency modulation and adaptively combines different bands through a mixture-of-experts design. We further introduce Random Frequency Augmentation (RFA) and Frequency Auxiliary Optimization (FAO) to better train MFEN. The three modules are complementary and jointly capture critical frequency-domain details for robust representation learning. Extensive experiments on three VI-ReID datasets demonstrate the effectiveness of our approach.", "date": "2026-06-10", "categories": []}
{"id": "2606.12074", "title": "Non-frontal face recognition using GANs and memristor-based classifiers", "abstract": "Face recognition systems have advanced significantly through deep learning techniques, delivering high performance and robustness in complex scenarios. However, these approaches incur substantial computational overhead, limiting their in situ applicability in resource-constrained platforms such as drones, where they can address challenges including non-frontal facial imagery. Memristor-based neuromorphic systems have emerged as a compelling approach for edge AI applications, combining biologically inspired processing with efficient and scalable computation. In this work, we propose a facial recognition framework that addresses non-frontal pose variations by integrating lightweight generative adversarial network (GAN)-based pose frontalisation with memristor-based neuromorphic recognition. The experimental results on two datasets demonstrate the effectiveness of combining adversarial learning with memristive technology, achieving up to 96% identification accuracy. The proposed approach alleviates the computational bottlenecks of conventional AI and offers a scalable, efficient solution for face recognition in dynamic real-world environments.", "date": "2026-06-10", "categories": []}
{"id": "2606.12258", "title": "Bridging Day and Night: Unsupervised Cross-Domain Re-Identification with Synergistic Prompt and Prototype Learning", "abstract": "Cross-domain day-night re-identification (ReID) is fundamentally challenged by the substantial visual appearance discrepancies between daytime and nighttime scenes. Existing fully supervised methods rely heavily on labor-intensive annotations, which are costly and exhibit limited generalization across domains. In this work, we investigate unsupervised day-night ReID and propose a novel framework that synergistically combines prompt learning and prototype-based representation learning to associate identities across domains without requiring manual labels. Our approach follows a progressive two-stage training strategy. In the first stage, we exploit the vision-language model to generate instance-specific textual prompts in an annotation-free manner. We employ an instance-level alignment mechanism to embed visual features and textual prompts into a unified semantic space, aligning unlabeled day/night images with learnable prompts via instance-aware dynamic-bias adaptation. In the second stage, we construct domain-specific prototype memory banks and introduce two complementary modules: i) an intra-domain identity association module to enhance feature discriminability within each domain, and ii) a cross-domain prototype matching module to reliably identify positive and negative prototype pairs, thereby establishing robust identity correspondences across day and night. Extensive experiments on public benchmarks validate the effectiveness of our method. Under the unsupervised setting, our framework attains Rank-1 accuracy comparable to state-of-the-art fully supervised methods.", "date": "2026-06-10", "categories": []}
{"id": "2605.20735", "title": "Lowering the Barrier to IREX Participation: Open-Source Algorithms, Toolkit, and Benchmarking for Iris Recognition", "abstract": "NIST Iris Exchange (IREX) offers an appealing solution to evaluating new open-source iris recognition algorithms, but it presents high barriers to entry because these algorithms must be written in C++, using a specific API, and adapted to meet strict IREX speed and memory constraints. The main goal of this paper is to lower these barriers and advance open-source iris recognition large-scale evaluations by offering: (a) two new modern deep learning-based open-source iris matchers (ArcIris and TripletIris), along with their C++ IREX X-compliant implementations, which are the first open-source iris recognition methods included into the IREX X leaderboard (and thus IREX-vetted), as well as new segmentation and iris circular approximation models that can be incorporated into any new iris recognition method, and (b) a performance assessment (according to IREX X testing protocols) of all major and currently available open-source iris recognition solutions. The paper also provides Python implementations of the new ArcIris and TripletIris methods and discusses the differences one may encounter between C++ and Python implementations of the same conceptually equivalent approaches. Finally, the paper offers open-source, IREX X-compliant C++ implementations of two existing methods: (a) an iris image filtering-based algorithm utilizing human saliency-driven kernels (HDBIF), and (b) a human-interpretable algorithm for detecting and comparing Fuchs' crypts (CRYPTS). In addition to IREX X evaluation results, the paper reports the performance of all methods on major academic benchmarks: Quality-Face/Iris Research Ensemble (Q-FIRE), Warsaw-Biobase Post-Mortem Iris, CASIA-Iris-Thousand-V4, CASIA-Iris-Lamp-V4, IIT Delhi Iris Database, IIITD Contact Lens Iris Database, NDIris3D, and Notre Dame Variable Iris Image Quality Release 2 (VII-Q-R2).", "date": "2026-05-20", "categories": []}
{"id": "2606.08437", "title": "X-Palm: Paired Multispectral-to-Smartphone Dataset for Cross-Domain Palmprint Authentication", "abstract": "Palmprint modality offers a privacy-preserving biometric solution, yet its deployment is hindered by the domain gap between controlled enrollment and unconstrained authentication. Existing datasets are largely restricted to controlled setups and fail to capture the compound variability of real-world environments. In this paper, we introduce X-Palm, a cross-domain dataset comprising 6,006 palm images from 103 individuals (206 hands). To the best of our knowledge, X-Palm is the first palmprint dataset providing novel paired-identity acquisition specifically designed to bridge the gap between reliably controlled multispectral enrollment and unconstrained mobile authentication while encompassing a broad spectrum of in-the-wild variability. Unlike existing datasets that focus on single to a few variations, X-Palm addresses the massive modality and environmental shifts encountered in practical deployments by capturing paired data for identities across two distinct domains: (1) a controlled Multispectral Palmprint setting using our custom-developed scanner, and (2) an unconstrained smartphone palmprint setting that is participant-driven, incorporating simultaneous variations in hardware, hand pose, illumination, background, camera-to-hand distance, perspective, and palm surface conditions (e.g., moisture and occlusions). Our extensive benchmarks of 12 SOTA models reveal that while existing methods achieve high performance on controlled data, they experience severe performance collapse on X-Palm. Conversely, models trained on X-Palm demonstrate consistent robustness across domains, positioning X-Palm as a valuable resource for training a model towards real-world, cross-domain generalization. Data access instructions and the related benchmarking codes are publicly available at: https://github.com/X-Palm/X-Palm-2026", "date": "2026-06-07", "categories": []}
{"id": "2606.09353", "title": "Beyond Humans: Multispecies Animal Face Recognition Using Transfer Learning", "abstract": "Individual animal recognition can be useful in the search for lost or stolen pets, the tracking of individuals of endangered species, and the recognition of animals in crowded farms. Present recognition techniques mostly use physical devices, e.g., microchips, often impractical and difficult to apply. These could be replaced by remote recognition via the animal's face; if accurate enough, it provides several advantages: it is non-invasive, can work at a distance, and is difficult to counterfeit, as, for instance, in the case of substituting sick animals for healthy ones in the food industry. The few existing datasets with sufficient per-subject images annotated with a single animal identity are not large enough to train current deep learning architectures. We rather investigate the possibility of transfer learning, exploiting pre-trained network models as backbones. Our experiments compared FaceNet, which is specifically trained on large databases of human faces, with the Vision Transformer (ViT) pre-trained on ImageNet, i.e., on object categories. We used three face datasets of very different animals: dogs, primates (lemurs, golden monkeys, and chimpanzees), and cattle. We report the results and, for each dataset, compare them with the state of the art (SOTA) ad hoc-trained deep networks. The capture conditions differ among the three datasets. Image quality (resolution, motion blur, diverse poses, etc.) decreases from dogs to cattle to primates. The best performance was achieved with dogs, where ViT reached a mean verification accuracy of 96.85% and a Rank-1 Identification Rate of 84.34%. The results for endangered primates are still encouraging, but performance varies across animal classes and tasks (verification or identification), and does not always outperform SOTA. For cattle, the ViT results outperform SOTA, while FaceNet is still competitive.", "date": "2026-06-08", "categories": []}
{"id": "2606.06755", "title": "PromptPrint: Behavioral Biometrics Through Natural Language Prompting in LLMs", "abstract": "Authorship attribution research has traditionally focused on long-form, expressive texts; however, interactions with large language models (LLMs) are typically brief and task-driven prompts. This raises a fundamental question: do such prompts contain a stable, author-identifiable, and distinctive signal? We introduce PromptPrint, a systematic study of prompt-based identity, the hypothesis that a user's habitual vocabulary, syntax, and discourse patterns form a learnable behavioral biometric. Using 20,680 real prompts from 1,034 users, we establish three key findings. First, lexical representations significantly outperform semantic encoders, supporting the \"lexical stability hypothesis\": identity is primarily encoded in surface-level word choice rather than abstract intent. Second, stylometric features exhibit a \"uniqueness-consistency paradox\": users are highly distinctive across the population, yet behaviorally inconsistent across contexts. Third, adversarial analysis reveals a clear vulnerability spectrum: identity signals are robust to minor lexical perturbations but degrade substantially under semantic paraphrasing. Overall, our results demonstrate strong identification performance at scale, establishing prompt-based identity as a viable behavioral biometric. This work introduces a new perspective on user modeling in LLM interactions, with important implications for security and privacy. Data and code will be released upon the acceptance of our work.", "date": "2026-06-04", "categories": []}
{"id": "2503.22929", "title": "Self-supervised Feature Disentanglement and Augmentation Network for One-class Face Anti-spoofing", "abstract": "Face anti-spoofing (FAS) techniques aim to enhance the security of facial identity authentication by distinguishing authentic live faces from deceptive attempts. While two-class FAS methods risk overfitting to training attacks to achieve better performance, one-class FAS approaches handle unseen attacks well but are less robust to domain information entangled within the liveness features. To address this, we propose an Unsupervised Feature Disentanglement and Augmentation Network (\\textbf{UFDANet}), a one-class FAS technique that enhances generalizability by augmenting face images via disentangled features. The \\textbf{UFDANet} employs a novel unsupervised feature disentangling method to separate the liveness and domain features, facilitating discriminative feature learning. It integrates an out-of-distribution liveness feature augmentation scheme to synthesize new liveness features of unseen spoof classes, which deviate from the live class, thus enhancing the representability and discriminability of liveness features. Additionally, \\textbf{UFDANet} incorporates a domain feature augmentation routine to synthesize unseen domain features, thereby achieving better generalizability. Extensive experiments demonstrate that the proposed \\textbf{UFDANet} outperforms previous one-class FAS methods and achieves comparable performance to state-of-the-art two-class FAS methods.", "date": "2025-03-29", "categories": []}
{"id": "2407.13922", "title": "CounterFace: A Synthetic Face Dataset for Fine-Grained Counterfactual Evaluation of Face Recognition Systems", "abstract": "Face recognition (FR) systems are widely deployed in critical applications, making their reliability and robustness across diverse populations and conditions essential. Standard evaluation of FR systems typically relies on datasets such as LFW to estimate average recognition accuracy. Some benchmarks also capture coarse-grained intra-identity variations such as aging, pose, and lighting. However, human faces undergo more fine-grained changes, including appearance changes such as hairstyles and makeup, that are underrepresented in existing benchmarks. Counterfactual evaluation provides a method to assess FR robustness under such fine-grained variations. Existing counterfactual face datasets synthesized with image generators, however, are limited in attribute coverage due to the use of humans for verification in the pipeline.\n  We propose CounterFace, a new counterfactual evaluation dataset comprising 20 facial attributes and 8 demographic factors, exceeding prior synthetic face datasets by 14 attributes and 2 demographics. The dataset is generated using a fully automated pipeline based on off-the-shelf image generators with custom verifiers, removing human need for verification. CounterFace contains 11,821 counterfactual face pairs, and a post-hoc user study confirms the faithfulness of the generated counterfactuals. We evaluate two commercial and four open-source FR systems (AWS Rekognition, Face++, AdaFace, MagFace, ArcFace, FaceNet) across 160 attribute-demographic combinations. Our dataset helps in the isolation of precise failure modes for individual systems unlike standard evaluation benchmarks. Results indicate that the performance degradation varies across attributes and demographics for all six systems and occluding attributes (e.g., facemask and facial hair) universally degrade performance.", "date": "2024-07-18", "categories": []}
{"id": "2605.06637", "title": "DPM++: Dynamic Masked Metric Learning for Occluded Person Re-identification", "abstract": "Although person re-identification has made impressive progress, occlusion caused by obstacles remains an unsettled issue in real applications. The difficulty lies in the mismatch between incomplete occluded samples and holistic identity representations. Severe occlusion removes discriminative body cues and introduces interference from background clutter and occluders, making global metric learning unreliable. Existing methods mainly rely on extra pre-trained models to estimate visible parts for alignment or construct occluded samples via data augmentation, but still lack a unified framework that learns robust visibility-consistent matching under realistic occlusion patterns. In this paper, we propose DPM++, a Dynamic Masked Metric Learning framework for occluded person re-identification. DPM++ learns an input-adaptive masked metric that dynamically selects reliable identity subspaces for each occluded instance, enabling matching to emphasize visibility-consistent evidence while suppressing unreliable components. Built upon the classifier-prototype space, DPM++ introduces a CLIP-based two-stage supervision scheme, where ID-level semantic priors are learned from the text branch and transferred into the classifier-prototype space for dynamic masked matching. To strengthen the masked metric, we introduce a saliency-guided patch transfer strategy to synthesize controllable and photo-realistic occluded samples during training. Exploiting real scene priors, this strategy exposes the model to realistic partial observations and provides richer supervision than random erasing. In addition, occlusion-aware sample pairing and mask-guided optimization improve the stability and effectiveness of the framework. Experiments on occluded and holistic person re-identification benchmarks show that DPM++ consistently outperforms previous state-of-the-art methods in both holistic and occlusion scenarios.", "date": "2026-05-07", "categories": []}
{"id": "2606.04469", "title": "Adaptive Calibration for Fair and Performant Facial Recognition", "abstract": "We introduce Adaptive Calibration (AC), a novel calibration strategy for facial recognition that maps cosine similarity between normalized embeddings to well-calibrated probabilities. By incorporating local context into calibration, Adaptive Calibration corrects for a fundamental mismatch in cosine similarity, whereby the same distance can correspond to different match probabilities in different embedding regions. Our approach improves both overall performance and results in a fairer calibration without requiring demographic metadata.\n  Our approach consistently dominates existing methods both on accuracy and fairness metrics across a variety of pretrained models and standard benchmarks. AC provides a practical solution for equitable facial recognition, without requiring demographic group annotations, and while improving overall performance. Unlike existing approaches, our method provides continuous, region-specific calibration that avoids \"leveling down\" where fairness comes at the cost of degraded performance for some groups.", "date": "2026-06-03", "categories": []}
{"id": "2208.00967", "title": "Counterfactual Intervention Feature Transfer for Visible-Infrared Person Re-identification", "abstract": "Graph-based models have achieved great success in person re-identification tasks recently, which compute the graph topology structure (affinities) among different people first and then pass the information across them to achieve stronger features. But we find existing graph-based methods in the visible-infrared person re-identification task (VI-ReID) suffer from bad generalization because of two issues: 1) train-test modality balance gap, which is a property of VI-ReID task. The number of two modalities data are balanced in the training stage, but extremely unbalanced in inference, causing the low generalization of graph-based VI-ReID methods. 2) sub-optimal topology structure caused by the end-to-end learning manner to the graph module. We analyze that the well-trained input features weaken the learning of graph topology, making it not generalized enough during the inference process. In this paper, we propose a Counterfactual Intervention Feature Transfer (CIFT) method to tackle these problems. Specifically, a Homogeneous and Heterogeneous Feature Transfer (H2FT) is designed to reduce the train-test modality balance gap by two independent types of well-designed graph modules and an unbalanced scenario simulation. Besides, a Counterfactual Relation Intervention (CRI) is proposed to utilize the counterfactual intervention and causal effect tools to highlight the role of topology structure in the whole training process, which makes the graph topology structure more reliable. Extensive experiments on standard VI-ReID benchmarks demonstrate that CIFT outperforms the state-of-the-art methods under various settings.", "date": "2022-08-01", "categories": []}
{"id": "2509.16635", "title": "Towards Anytime Retrieval: A Benchmark for Anytime Person Re-Identification", "abstract": "In real applications, person re-identification (ReID) is expected to retrieve the target person at any time, including both daytime and nighttime, ranging from short-term to long-term. However, existing ReID tasks and datasets can not meet this requirement, as they are constrained by available time and only provide training and evaluation for specific scenarios. Therefore, we investigate a new task called Anytime Person Re-identification (AT-ReID), which aims to achieve effective retrieval in multiple scenarios based on variations in time. To address the AT-ReID problem, we collect the first large-scale dataset, AT-USTC, which contains 403k images of individuals wearing multiple clothes captured by RGB and IR cameras. Our data collection spans 21 months, and 270 volunteers were photographed on average 29.1 times across different dates or scenes, 4-15 times more than current datasets, providing conditions for follow-up investigations in AT-ReID. Further, to tackle the new challenge of multi-scenario retrieval, we propose a unified model named Uni-AT, which comprises a multi-scenario ReID (MS-ReID) framework for scenario-specific features learning, a Mixture-of-Attribute-Experts (MoAE) module to alleviate inter-scenario interference, and a Hierarchical Dynamic Weighting (HDW) strategy to ensure balanced training across all scenarios. Extensive experiments show that our model leads to satisfactory results and exhibits excellent generalization to all scenarios.", "date": "2025-09-20", "categories": []}
{"id": "2605.30364", "title": "Hamiltonian-Inspired Attention Mechanism for Scalable RF Transmitter Fingerprinting", "abstract": "Radio-frequency (RF) fingerprinting identifies wire-less transmitters using hardware-induced imperfections present in baseband I/Q signals. However, deep learning models often degrade under receiver and channel distribution shifts, particularly as transmitter populations grow. This work proposes the Hamiltonian Transformer, a physics-informed attention architecture that enforces norm preserving value dynamics within each attention head using a learned skew-symmetric generator and a St\\\"ormer-Verlet leapfrog integration step. An additional phase-increment embedding exposes oscillator dynamics at the input layer. All experiments use non-equalized raw I/Q signals from the WiSig dataset under four protocols: same-day classification, cross-receiver generalisation, cross-day generalisation, and transmitter scaling up to 150 devices. The Hamiltonian Transformer achieves 99.12% accuracy under same-day conditions and 61.64% at 150 transmitters, consistently outperforming CNN and Transformer baselines across all scale points. A controlled ablation study identifies norm-preservation in the value update as the primary inductive bias driving the scaling advantage, with the phase increment embedding providing the single largest per-component improvement. These results indicate that embedding physics-informed structural priors into attention mechanisms is an effective approach to large-scale transmitter identification on raw wireless signals.", "date": "2026-05-17", "categories": []}
{"id": "2605.30784", "title": "Text-guided Feature Disentanglement for Cross-modal Gait Recognition", "abstract": "Gait recognition is a biometric technique that identifies individuals based on their walking patterns, offering advantages in long-range, non-intrusive scenarios. However, real-world scenarios often involve heterogeneous sensing modalities such as LiDAR and RGB cameras, making LiDAR-Camera Cross-modal Gait recognition (LCCGR) a critical yet challenging task due to the substantial modality gap between 2D videos and 3D point cloud sequences. To address this challenge, we propose TCFDNet, a Text-guided Cross-modal Feature Disentanglement Network, which leverages modality-aware textual priors as semantic anchors to guide the learning of disentangled modality-shared representations. Specifically, we construct a Gait Modality Text Dictionary (GMTD) using large language models to generate rich semantic descriptions of gait across modalities and viewpoints. A CLIP-based Multi-grained Feature Encoder then aligns visual and textual features within a unified vision-language space. Furthermore, the Text-guided Feature Disentanglement (TFD) module selects the topk matched textual descriptions to reconstruct modality-specific representations and derive modality-shared features via residual decomposition and orthogonality constraints. To mitigate the fragility of the disentangled shared features, we propose a Feature Stability Enhancement (FSE) module, which models spatial and channel-wise correlations to improve feature robustness. In addition, a cross-modal patch exchange strategy is introduced to further improve generalization. Extensive experiments on SUSTech1K and FreeGait datasets demonstrate that TCFDNet achieves new state-of-the-art results and validate the effectiveness of the proposed modules.", "date": "2026-05-29", "categories": []}
{"id": "2605.31001", "title": "Iterative Framework For Data Augmentation Of Segmented Fingerprints", "abstract": "Infant biometrics presents unique challenges due to the physiological differences between infants and adults, compounded by the scarcity of available data for research that limits the development of robust matching systems. This paper proposes a novel data augmentation method that uses iterative techniques to generate diverse variants of segmented fingerprints by inducing errors in a convolutional neural network trained to extract fingerprint ridges and valleys. Experiments on real infant fingerprints demonstrate the method's effectiveness in expanding fingerprint variability, with augmentations exhibiting significant fluctuations in minutiae counts while still retaining visual similarity to the originals. The study also highlights the method's customizable nature for applying varying levels of changes to fingerprint segmentations. Future research includes training segmentation and matching neural networks using datasets augmented by the proposed framework.", "date": "2026-05-29", "categories": []}
{"id": "2510.00936", "title": "Resolution as a Direction: Vector-Panning Feature Alignment for Cross-Resolution Re-Identification", "abstract": "Cross-resolution person re-identification (CR-ReID) remains challenging in practical surveillance, where camera quality and capture distance lead to substantial resolution gaps between low-resolution (LR) queries and high-resolution (HR) gallery images. Prior approaches commonly rely on super-resolution (SR) or resolution-invariant representation learning, which often increases system complexity and may not directly address the feature mismatch induced by resolution degradation. In this work, we report a new empirical finding from a dedicated analysis in which identity-specific variation is averaged out: the HR--LR feature discrepancy produced by standard ReID backbones exhibits a consistent, resolution-related semantic direction in the embedding space. We further support this observation with statistical analyses based on Canonical Correlation Analysis (CCA) and Pearson correlation analysis. Motivated by this finding, we propose Vector Panning Feature Alignment (VPFA), a lightweight post-hoc module that learns to pan LR features along the learned resolution direction to obtain pseudo-HR representations. VPFA operates after feature extraction and can be integrated into existing ReID systems with negligible overhead. Extensive experiments on multiple CR-ReID benchmarks show that VPFA achieves state-of-the-art performance while improving efficiency compared to SR-based or jointly trained alternatives.", "date": "2025-10-01", "categories": []}
{"id": "2605.29720", "title": "Efficient, Validation-Free Intrinsic Quality Estimation for Large-Scale Face Recognition Datasets", "abstract": "We propose Intrinsic Quality (IQ), a validation-free metric designed to estimate the inherent potential of face recognition (FR) datasets to produce high-performance models without the need for full-scale training. IQ integrates two components: (i) a Neighbor-Consistency Score that quantifies local identity label agreement via nearest neighbors, and (ii) Global Representation Subspace Complexity (Effective Rank, ER), which captures the underlying embedding geometry and dataset diversity. IQ allows for rapid evaluation using lightweight proxy models or data subsets, facilitating dataset diagnosis and curation prior to resource-intensive full-scale training. We describe an experimental protocol tailored to clean, noisy, and mixed-quality FR datasets, and outline evaluation methodologies to validate IQ's predictive power for downstream performance.", "date": "2026-05-28", "categories": []}
{"id": "2605.29868", "title": "Ciphera: A Decentralised Biometric Identity Framework", "abstract": "Centralised biometric identity systems expose users to single points of failure, opaque verification processes, and irreversible biometric compromise. Decentralised Identifiers (DIDs) and Verifiable Credentials (VCs) offer stronger privacy guarantees, yet their integration with biometric authentication and distributed verification remains insufficiently explored. This paper presents Ciphera, a decentralised biometric identity framework combining privacy-preserving facial recognition, multi-node verification, IPFS-based credential metadata storage, and blockchain-anchored revocation. Evaluated across functional, performance, security, and distributed consistency dimensions, Ciphera achieved an 81% functional success rate, with stable enrolment and authentication but measurable revocation propagation delays and occasional audit-log inconsistencies. Performance testing demonstrated sub-second p95 verification latency of approximately 820ms under concurrent multi-node conditions. Security analysis confirmed strong confidentiality and integrity guarantees, though incomplete liveness detection leaves susceptibility to deepfake and replay attacks. The results demonstrate the feasibility of decentralised biometric identity while identifying key engineering challenges for production-grade deployment.", "date": "2026-05-28", "categories": []}
{"id": "2507.14109", "title": "An Adversarial-Driven Experimental Study on Deep Learning for RF Fingerprinting", "abstract": "Radio frequency (RF) fingerprinting, which extracts unique hardware imperfections of radio devices, has emerged as a promising physical-layer device identification mechanism in zero trust architectures and beyond 5G networks. In particular, deep learning (DL) methods have demonstrated state-of-the-art performance in this domain. However, existing approaches have primarily focused on enhancing system robustness against temporal and spatial variations in wireless environments, while the security vulnerabilities of these DL-based approaches have often been overlooked. In this work, we systematically investigate the security risks of DL-based RF fingerprinting systems through an adversarial-driven experimental analysis. We observe a consistent misclassification behavior for DL models under domain shifts, where a device is frequently misclassified as another specific one. Our analysis based on extensive real-world experiments demonstrates that this behavior can be exploited as an effective backdoor to enable external attackers to intrude into the system. Furthermore, we show that training DL models on raw received signals causes the models to entangle RF fingerprints with environmental and signal-pattern features, creating additional attack vectors that cannot be mitigated solely through post-processing security methods such as confidence thresholds.", "date": "2025-07-18", "categories": []}
{"id": "2508.00748", "title": "Is It Really You? Exploring Biometric Verification Scenarios in Photorealistic Talking-Head Avatar Videos", "abstract": "Photorealistic talking-head avatars are becoming increasingly common in virtual meetings, gaming, and social platforms. These avatars allow for more immersive communication, but they also introduce serious security risks. One emerging threat is impersonation: an attacker can steal a user's avatar, preserving his appearance and voice, making it nearly impossible to detect its fraudulent usage by sight or sound alone. In this paper, we explore the challenge of biometric verification in such avatar-mediated scenarios. Our main question is whether an individual's facial motion patterns can serve as reliable behavioral biometrics to verify their identity when the avatar's visual appearance is a facsimile of its owner. To answer this question, we introduce a new dataset of realistic avatar videos created using a state-of-the-art one-shot avatar generation model, GAGAvatar, with genuine and impostor avatar videos. We also propose a lightweight, explainable spatio-temporal Graph Convolutional Network architecture with temporal attention pooling, that uses only facial landmarks to model dynamic facial gestures. Experimental results demonstrate that facial motion cues enable meaningful identity verification with AUC values approaching 80%. The proposed benchmark and biometric system are available for the research community in order to bring attention to the urgent need for more advanced behavioral biometric defenses in avatar-based communication systems.", "date": "2025-08-01", "categories": []}
{"id": "2603.00177", "title": "Detecting Cognitive Signatures in Typing Behavior for Non-Intrusive Authorship Verification", "abstract": "The proliferation of AI-generated text has intensified the need for reliable authorship verification, yet current output-based methods are increasingly unreliable. We observe that the ordinary typing interface captures rich cognitive signatures, measurable patterns in keystroke timing that reflect the planning, translating, and revising stages of genuine composition. Drawing on large-scale keystroke datasets comprising over 136 million events, we define the Cognitive Load Correlation (CLC) and show it distinguishes genuine composition from mechanical transcription. We present a non-intrusive verification framework that operates within existing writing interfaces, collecting only timing metadata to preserve privacy. Our analytical evaluation estimates 85 to 95 percent discrimination accuracy under stated assumptions, while limiting biometric leakage via evidence quantization. We analyze the adversarial robustness of cognitive signatures, showing they resist timing-forgery attacks that defeat motor-level authentication because the cognitive channel is entangled with semantic content. We conclude that reframing authorship verification as a human-computer interaction problem provides a privacy-preserving alternative to invasive surveillance.", "date": "2026-02-26", "categories": []}
{"id": "2605.25307", "title": "Recursive Class Connectivity Classification (R3C) Applied to Binary Image Segmentation for Improved Infant Fingerprint Enhancement", "abstract": "Image enhancement plays a crucial role in infant fingerprint matching, as child-specific characteristics such as smaller finger dimensions and thinner ridge structures often degrade image quality during acquisition. To address these limitations, enrollment typically depends on specialized highresolution scanners, which most existing enhancement methods are not designed to support. Consequently, identification rates for children remain significantly lower than those achieved with adult fingerprints. This study introduces Recursive Class Connectivity Classification (R3C), a novel framework that iteratively refines binary segmentation outputs from existing enhancement methods by extending ridge structures. R3C does not require modifications to the underlying classifier and operates without training data, which is not currently available for infant fingerprints. Instead, the method improves segmentation by repeatedly feeding the classified image back into the classification process, while combining each intermediate segmentation with the original input image. Experiments conducted on three fingerprint datasets using four different enhancement classifiers show that R3C can increase the True Acceptance Rate (TAR) by up to 4% for children and over 40% for newborns, compared to using the enhancement methods alone. A qualitative analysis further demonstrates that R3C reconnects fragmented ridge patterns, improving the visual quality of segmentation. Because it functions independently of the enhancement method used, R3C provides a flexible and broadly applicable solution for improving binary segmentation.", "date": "2026-05-25", "categories": []}
{"id": "2507.17640", "title": "Not All Starting Points Are Equal: Pre-trained Priors and Their Outsized Impact on Person Identification", "abstract": "Recent years have seen an explosion of diverse general purpose pre-training methodologies for computer vision. However, the impact that these pre-training methodologies have on person identification tasks (re-id) remains under-explored. We show that under equated domain adaptation pipelines, there is dramatic variance in person identification outcomes using different starting models (architectures and pre-trained weights). We show that a range of intuitive explanations for differing downstream performance on a range of re-id tests are insufficient and propose that pre-trained weights serve as a strong prior to the weights learned during domain adaptation. This framework allows for domain adapted solutions to be viewed as a maximum probability point estimate of the Gibbs posterior with the pre-trained weights acting as a prior. Under this framework, we show that large, pre-trained foundation models with simple domain adaptation achieve SOTA solutions on a range of re-id datasets (Market, PRCC, DeepChange, BTS) with solutions that are very close in the parameter space to the starting parameters. Moreover, we perform ablations on these solutions and show that they can be reached with small transfer sets and with varying transfer datasets but are sensitive to choice of optimizer, weight-decay, and loss function. Ultimately, we propose that the simple approach of direct fine-tuning using large vision foundation models (CLIP, Dino, EVA, AIM, etc.) needs to serve as an important baseline for future work in re-id.", "date": "2025-07-23", "categories": []}
{"id": "2603.27747", "title": "AI-Powered Facial Mask Removal Is Not Suitable For Identification", "abstract": "Recently, crowd-sourced online criminal investigations have used generative-AI to enhance low-quality visual evidence. In one high-profile case, social-media users circulated an \"AI-unmasked\" image of a federal agent involved in a fatal shooting, fueling a wide-spread misidentification. In response to this and similar incidents, we conducted a large-scale analysis evaluating the efficacy and risks of commercial AI-powered facial unmasking, specifically assessing whether the resulting faces can be reliably matched to true identities.", "date": "2026-03-29", "categories": []}
{"id": "2603.28675", "title": "Why Aggregate Accuracy is Inadequate for Evaluating Fairness in Law Enforcement Facial Recognition Systems", "abstract": "Facial recognition systems are increasingly deployed in law enforcement and security contexts, where algorithmic decisions can carry significant societal consequences. Despite high reported accuracy, growing evidence demonstrates that such systems often exhibit uneven performance across demographic groups, leading to disproportionate error rates and potential harm. This paper argues that aggregate accuracy is an insufficient metric for evaluating the fairness and reliability of facial recognition systems in high-stakes environments. Through analysis of subgroup-level error distribution, including false positive rate (FPR) and false negative rate (FNR), the paper demonstrates how aggregate performance metrics can obscure critical disparities across demographic groups. Empirical observations show that systems with similar overall accuracy can exhibit substantially different fairness profiles, with subgroup error rates varying significantly despite a single aggregate metric. The paper further examines the operational risks associated with accuracy-centric evaluation practices in law enforcement applications, where misclassification may result in wrongful suspicion or missed identification. It highlights the importance of fairness-aware evaluation approaches and model-agnostic auditing strategies that enable post-deployment assessment of real-world systems. The findings emphasise the need to move beyond accuracy as a primary metric and adopt more comprehensive evaluation frameworks for responsible AI deployment.", "date": "2026-03-30", "categories": []}
{"id": "2605.07816", "title": "ICDAR 2026 Competition on Writer Identification and Pen Classification from Hand-Drawn Circles", "abstract": "This paper presents CircleID, a large-scale ICDAR 2026 competition on writer identification and pen classification from scanned hand-drawn circles. The primary objective is to investigate how biometric writer characteristics and physical pen features naturally entangle within minimal, static traces. CircleID comprises two distinct tasks: (1) open-set writer identification, requiring models to recognize known writers while explicitly rejecting unknown ones, and (2) cross-writer pen classification, evaluated across both seen and unseen writers. Participants were provided with a new, controlled dataset of 46,155 tightly cropped circle images, digitized at 400 DPI and annotated for writer identity and pen type. The dataset comprises samples from 44 known and 22 unknown writers using eight different pens. Hosted on Kaggle as two separate tracks with public and private leaderboards, the competition provided participants with a ResNet baseline. In total, 389 teams (436 participants) made 3,185 submissions for the pen classification task, and 113 teams (141 participants) made 1,737 submissions for the writer identification track. The best-performing private leaderboard submissions achieved a Top-1 accuracy of 64.801% for writer identification and 92.726% for pen classification. This paper details the dataset, evaluates the winning methodologies, and analyzes the impact of out-of-distribution writers on model generalization and feature disentanglement. In this large-scale competition, CircleID establishes a new baseline for minimal-trace analysis.", "date": "2026-05-08", "categories": []}
{"id": "2605.19020", "title": "A Systematic Failure Analysis of Vision Foundation Models for Open Set Iris Presentation Attack Detection", "abstract": "Vision foundation models have demonstrated strong transferability across diverse visual recognition tasks and are increasingly considered for biometric applications. Their suitability for iris Presentation Attack Detection (PAD), particularly under realistic open-set operating conditions, remains insufficiently examined. This work presents a systematic failure analysis of general-purpose vision foundation models for open-set iris PAD using periocular imagery. Five representative foundation models are evaluated under three open-set protocols that explicitly separate different sources of distribution shift: unseen Presentation Attack Instruments (PAIs), unseen datasets captured with different sensors and cross-spectral transfer from near-infrared (NIR) to visible spectrum (VIS) imagery. Both frozen feature representations and parameter-efficient task adaptation using Low-Rank Adaptation (LoRA) are assessed within a unified experimental framework. The results indicate that foundation models can transfer across datasets with similar sensing characteristics, but fail to generalise reliably to unseen attack instruments and degrade sharply under cross-spectral evaluation. While LoRA improves performance in certain cross-dataset settings, it frequently amplifies failure under attack-level and spectral shifts. Additional validation experiments using segmented iris inputs, full backbone fine-tuning, joint cross-dataset and cross-PAI shifts, and reverse VIS to NIR transfer further confirm that these failures are not simply artefacts of periocular input, weak adaptation, or one-directional spectral evaluation. These findings show that strong closed-set or cross-dataset performance should not be treated as evidence of robust open-set security, and highlight the need for PAD representations that maintain sensitivity to presentation artefacts while remaining stable under realistic deployment variation.", "date": "2026-05-18", "categories": []}
{"id": "2605.19527", "title": "Dual-Prompt CLIP with Hybrid Visual Encoders for Occluded Person Re-Identification", "abstract": "Occluded person re-identification focuses on matching partially visible pedestrians across multiple camera views. However, occlusions disrupt body-region cues, thereby complicating cross-view matching. Most person ReID methods built on pretrained vision-language models only focus on enhancing prompt-based feature learning while ignoring the semantic information of occluders. Based on the success of CLIP-ReID, we propose a novel Dual Prompt Learning ReID (DPL-ReID) model for occluded person ReID. It incorporates a Dual Prompt Learning (Dual-PL) strategy, which can utilize textual cues to capture complete pedestrian semantics and keep robustness against occlusion, and a Real-World Occlusion Augmentation (RWOA) method that realistically simulates occlusion scenarios encountered in real word to enrich occluded samples. In addition, we also design a Weighted Gated Feature Fusion (WGFF) method, which in corporates LSNet to capture global information and act as a feature-gating mechanism. This mechanism can effectively guide the CLIP visual encoder toward generating more comprehensive feature representations. Extensive experiments on several benchmark occluded ReID datasets show that our proposed DPL-ReID achieves the state-of-the art performance. The occlusion instance library are available at https://github.com/stone-qiao/DPL-ReID.", "date": "2026-05-19", "categories": []}
{"id": "2508.17431", "title": "FedKLPR: KL-Guided Pruning-Aware Federated Learning for Person Re-Identification", "abstract": "Person re-identification (re-ID) is a fundamental task in intelligent surveillance and public safety. Federated learning (FL) provides a privacy-preserving paradigm for collaborative model training without centralized data collection. However, deploying FL in real-world re-ID systems remains challenging due to statistical heterogeneity caused by non-IID client data and the substantial communication overhead incurred by frequent transmission of large-scale models. To address these challenges, we propose FedKLPR, a lightweight and communication-efficient federated learning framework for person re-ID. FedKLPR consists of three key components. First, KL-Divergence-Guided training, including the KL-Divergence Regularization Loss (KLL) and KL-Divergence-aggregation Weight (KLAW), is introduced to mitigate statistical heterogeneity and improve convergence stability under non-IID settings. Second, unstructured pruning is incorporated to reduce communication overhead, and the Pruning-ratio-aggregation Weight (PRAW) is proposed to measure the relative importance of client parameters after pruning. Together with KLAW, PRAW forms KL-Divergence-Prune Weighted Aggregation (KLPWA), enabling effective aggregation of pruned local models under heterogeneous data distributions. Third, Cross-Round Recovery (CRR) adaptively controls pruning across communication rounds to prevent excessive compression and preserve model accuracy. Experiments on eight benchmark datasets demonstrate that FedKLPR achieves substantial communication savings while maintaining competitive accuracy. Compared with state-of-the-art methods, FedKLPR reduces communication cost by 40\\%--42\\% on ResNet-50 while achieving better overall performance.", "date": "2025-08-24", "categories": []}
{"id": "2605.16908", "title": "BIDO: A Biometric Identity Online Authentication Framework", "abstract": "Security systems demand continuous, cryptograph- ically robust identity verification without requiring subjects to carry physical tokens, smart cards, or dedicated hardware authenticators. This paper presents BIDO (Biometric Identity Online), a device-free authentication standard that achieves Au- thenticator Assurance Level 2 (AAL2) per NIST SP 800-63B with- out storing long-lived biometric templates, facial images, or any other form of Personally Identifiable Information (PII). BIDO derives Elliptic Curve Digital Signature Algorithm (ECDSA) key material deterministically from a live biometric measurement salted with a user-defined memorized secret at every authen- tication event, eliminating persistent private-key storage while enabling verification from any commodity sensor terminal. The generated credentials are non-discoverable (non-resident) Web Authentication (WebAuthn) credentials, fully compatible with all FIDO2-enabled websites and services without modification on the server side. A multi-stage pipeline, comprising capture of 200 valid biometric samples, feature extraction using the Dlib 68- point facial landmark predictor, affine face alignment, frontality gating, Euclidean distance computation from the inter-eye mid- point, floor-division quantization with divisor q = 8, inter-session drift stabilization, and majority-voting SHA-256 hash binding, produces a Verification Seed (Vseed) from which the WebAuthn credential is transiently derived and immediately zeroized after signing. Evaluated against three prominent face benchmarks (VGGFace2, LFW, and MegaFace), achieving 99.51% verification accuracy on LFW and 92.14% Rank-1 identification accuracy on MegaFace Challenge 1 at 10^6 distractors, with a cryptographic False Accept Rate (FAR) of 0.03%, a False Reject Rate (FRR) of 0.90%.", "date": "2026-05-16", "categories": []}
{"id": "2605.17367", "title": "Bridging Data Trials and Task Barriers: A Unified Framework for Sketch Biometric Identification", "abstract": "Different from existing cross-modality identification tasks (e.g., heterogeneous face recognition, sketch re-identification, etc.), we introduce a novel yet practical setting for these related identification tasks, named \\textbf{sketch biometric identification}, which aims to continually train a unified model across different data domains, even diverse identification tasks. Sketch biometric identification faces challenges, including scarce real sketch data, high annotation costs, privacy risks, and insufficient generalization ability of cross-task models. Existing methods usually rely on limited real data or single-task optimization, making it difficult to effectively address the joint challenges of cross-modality and cross-task. This paper proposes a unified framework that integrates efficient synthetic sketch generation and task-sequential continual learning. First, we design an efficient pipeline to generate a large-scale and high-quality synthetic person and face sketch data, which significantly reduces costs and avoids privacy risks. Meanwhile, we enhance the model's robustness by fusing real data. Second, we construct a universal unified framework for sketch biometric identification, which adopts a task-sequential training strategy: the model first completes sketch person re-identification learning on the person dataset; subsequently, it maintains the acquired person recognition capability through a trusted sample replay technique and seamlessly performs incremental training on the face dataset. This enables a single model to simultaneously handle the cross-task capabilities of multiple sketch biometric identification tasks. To support the study of the mentioned sketch biometric identification, we built a new large-scale benchmark, SketchUnified-BioID, with several practical evaluation protocols.", "date": "2026-05-17", "categories": []}
{"id": "2605.17673", "title": "A simple approach for biometrics: Finger-knuckle prints recognition based on a Sobel filter and similarity measures", "abstract": "The objective of this work is to propose a novel methodology for the finger knuckle print recognition, which is essentially a digital photo of the finger-knuckle region. We have employed very simple concepts of visual computing such as a filter based on the Sobel operator for finding edges and a simple noise reduction algorithm. These operations are exceptionally fast and produce binary images, which are very efficient to process and to store. Furthermore, alongside this preprocessing, some similarity measures were also regarded and evaluated for the task. After preprocessing an input finger it is compared to all the images of fingers in the dataset, one by one. We have obtained up to 17.02% of successful recognitions (true positive rate) with a large dataset.", "date": "2026-05-17", "categories": []}
{"id": "2605.17685", "title": "Attention-Guided Fusion of 1D and 2D CNNs for Robust ECG-Based Biometric Recognition", "abstract": "Electrocardiogram (ECG)-based biometric recognition has emerged as a promising solution for secure authentication and liveness detection. However, most existing methods rely on unimodal deep learning architectures that independently process either one-dimensional (1D) temporal signals or two-dimensional (2D) time-frequency representations, limiting robustness and generalization. To address this issue, this paper proposes a hybrid framework integrating 1D and 2D convolutional neural networks (CNNs) within a unified end-to-end architecture. The 1D branch extracts temporal and morphological features from raw ECG signals, while the 2D branch captures discriminative spectral information from time-frequency representations. An attention-guided fusion mechanism dynamically weights both modalities according to input characteristics, overcoming the limitations of conventional static fusion strategies.\n  The framework was evaluated on three benchmark datasets (ECG-ID, MIT-BIH, and PTB), including healthy subjects and patients with cardiac pathologies, achieving identification accuracies of 99.56%, 100.00%, and 99.89%, respectively. To assess long-term biometric permanence, experiments were also conducted on the multi-session Heartprint dataset spanning ten years. The proposed approach achieved same-session accuracies of 98.54% (S1), 99.09% (S2), 94.93% (S3R), and 96.08% (S3L), while cross-session evaluations reached 56.33% (S1-S2) and 53.27% (S2-S3R), demonstrating the ability to capture stable biometric signatures over time. The optimal configuration combines InceptionTime for 1D processing, ResNet-34 for 2D analysis, and attention-based fusion. Ablation studies confirm that the proposed attention mechanism consistently outperforms conventional fusion approaches. Overall, the proposed framework provides a robust, scalable, and high-performance solution for ECG biometric recognition.", "date": "2026-05-17", "categories": []}
{"id": "2605.18192", "title": "View-Aware Semantic Alignment for Aerial-Ground Person Re-Identification", "abstract": "Aerial-Ground Person Re-Identification (AGPReID) remains highly challenging due to drastic viewpoint variations between drones and fixed cameras. Existing methods typically follow a view-invariant paradigm, aligning shared features across views to achieve robustness. However, view-invariant inherently enforces part-level alignment, which ignores view-specific cues and discriminative identity information. To this end, this work proposes ViSA (View-aware Semantic Alignment), a view-aware framework that achieves cross-view semantic consistency containing an Expert-driven Token Generation Module (ETGM) and a Dual-branch Local Fusion Module (DLFM). Technically, the former constructs a set of view-aware experts to generate adaptive semantic queries that perceive viewpoint-specific patterns, while the latter leverages graph reasoning to extract and align local regions responsive to different experts. Extensive experiments on three AGPReID benchmarks including AG-ReID.v2, CARGO and LAGPeR demonstrate that ViSA consistently achieves superior performance, with a notable 10.06\\% mAP improvement on the challenging CARGO cross-view protocol. The code is available at \\href{https://github.com/Cat-Zero/ViSA}{https://github.com/Cat-Zero/ViSA}.", "date": "2026-05-18", "categories": []}
{"id": "2605.18238", "title": "Non-Colliding Biometric Identities for Digital Entities: Geometry, Capacity, and Million-Scale Virtual Identity Provisioning", "abstract": "Digital entities such as AI agents and humanoid robots increasingly operate alongside real humans, yet their identity infrastructure is based on credentials rather than embodied biometric identity. We introduce Biometric Identity Provisioning (BIP), a new problem and solution framework that addresses: given an enrollment gallery of real human identities, provision virtual identities that are non-colliding with every enrolled identity, maintain sufficient inter-class separability, and are realizable as high-fidelity face images. The key geometric insight is that real face identities occupy a low-dimensional subspace of the embedding hypersphere, leaving no residual subspace for virtual identities. Hence, virtual identities must instead be allocated as unclaimed gaps within the real face manifold itself. BIP is therefore a constrained packing problem: available gaps vastly exceed any foreseeable enrollment scale, and provisioned identities remain non-colliding even as new real identities are subsequently enrolled. Grounded in this geometry, our repulsion-based allocation is not bounded by any fixed provisioning count; we demonstrate 10M non-colliding virtual identity embeddings against a gallery of 360K real identities. Realizing these embeddings as face images requires a generator that operates outside the training distribution of real face images; we introduce GapGen, a gap-aware generator trained with a curriculum that progressively extends synthesis into non-colliding regions, validated at 1M photorealistic virtual face images. We further construct v-LFW, a virtual counterpart to LFW face dataset, with protocols for virtual face verification, cross-reality matching, real-vs-virtual detection, and unified recognition and detection.", "date": "2026-05-18", "categories": []}
{"id": "2605.10867", "title": "BEACON: A Multimodal Dataset for Learning Behavioral Fingerprints from Gameplay Data", "abstract": "Continuous authentication in high-stakes digital environments requires datasets with fine-grained behavioral signals under realistic cognitive and motor demands. But current benchmarks are often limited by small scale, unimodal sensing or lack of synchronised environmental context. To address this gap, this paper introduces BEACON (Behavioral Engine for Authentication & Continuous Monitoring), a large-scale multimodal dataset that captures diverse skill tiers in competitive Valorant gameplay. BEACON contains approximately 430 GB of synchronised modality data (461 GB total on-disk including auxiliary Valorant configuration captures) from 79 sessions across 28 distinct players, estimated at 102.51 hours of active gameplay, including high-frequency mouse dynamics, keystroke events, network packet captures, screen recordings, hardware metadata, and in-game configuration context. BEACON leverages the high precision motor skills and high cognitive load that are inherent to tactical shooters, making it a rigorous stress test for the robustness of behavioral biometrics. The dataset allows for the study of continuous authentication, behavioral profiling, user drift and multimodal representation learning in a high-fidelity esports setting. The authors release the dataset and code on Hugging Face and GitHub to create a reproducible benchmark for evaluating next-generation behavioral fingerprinting and security models.", "date": "2026-05-11", "categories": []}
{"id": "2504.09549", "title": "SD-ReID: View-aware Stable Diffusion for Aerial-Ground Person Re-Identification", "abstract": "Aerial-Ground Person Re-IDentification (AG-ReID) aims to retrieve specific persons across cameras with different viewpoints. Previous works focus on designing discriminative models to maintain the identity consistency despite drastic changes in camera viewpoints. The core idea behind these methods is quite natural, but designing a view-robust model is a very challenging task. Moreover, they overlook the contribution of view-specific features in enhancing the model's ability to represent persons. To address these issues, we propose a novel generative framework named SD-ReID for AG-ReID, which leverages generative models to mimic the feature distribution of different views while extracting robust identity representations. More specifically, we first train a ViT-based model to extract person representations along with controllable conditions, including identity and view conditions. We then fine-tune the Stable Diffusion (SD) model to enhance person representations guided by these controllable conditions. Furthermore, we introduce the View-Refined Decoder (VRD) to bridge the gap between instance-level and global-level features. Finally, both person representations and all-view features are employed to retrieve target persons. Extensive experiments on five AG-ReID benchmarks (i.e., CARGO, AG-ReIDv1, AG-ReIDv2, LAGPeR and G2APS-ReID) demonstrate the effectiveness of our proposed method. The source code and pre-trained models are available at https://github.com/924973292/SD-ReID.", "date": "2025-04-13", "categories": []}
{"id": "2605.14548", "title": "Local Spatiotemporal Convolutional Network for Robust Gait Recognition", "abstract": "Gait recognition, as a promising biometric technology, identifies individuals through their unique walking patterns and offers distinctive advantages including non-invasiveness, long-range applicability, and resistance to deliberate disguise. Despite these merits, capturing the intrinsic motion patterns concealed within consecutive video frames remains challenging due to the complexity of video data and the interference of external covariates such as viewpoint changes, clothing variations, and carrying conditions. Existing approaches predominantly rely on either static appearance features extracted from individual silhouette frames or employ complex sequential models (\\eg, LSTM, 3D convolutions) that demand substantial computational resources and sophisticated training strategies. To address these limitations, we propose a Local Spatiotemporal Convolutional Network (LSTCN), a structurally simple yet highly effective dual-branch architecture that endows standard two-dimensional convolutional networks with the capacity to extract temporal information. Specifically, we introduce a Global Bidirectional Spatial Pooling (GBSP) mechanism that reduces the dimensionality of gait tensors by decomposing spatial features into horizontal and vertical strip-based local representations, enabling the temporal dimension to participate in standard 2D convolution operations. Building upon this, we design a Local Spatiotemporal Convolutional (LSTC) layer that jointly processes temporal and spatial dimensions, allowing the network to adaptively learn strip-based gait motion patterns. We further extend this formulation with asymmetric convolution kernels that independently attend to the temporal, spatial, and joint spatiotemporal domains, thereby enriching the extracted feature representations.", "date": "2026-05-14", "categories": []}
{"id": "2605.14845", "title": "Exploring Vision-Language Models for Online Signature Verification: A Zero-Shot Capability Study", "abstract": "Recent advancements in Vision-Language Models (VLMs) have demonstrated strong capabilities in general visual reasoning, yet their applicability to rigorous biometric tasks remains unexplored. This work presents an exploratory study evaluating the zero-shot performance of state-of-the-art VLMs (GPT-5.2 and Gemini 2.5 Pro) on the Signature Verification Challenge (SVC) benchmark. To enable visual processing, raw kinematic time-series are converted into static images, encoding pressure information into stroke opacity whenever available in the source data. Furthermore, we introduce a scoring protocol that extracts latent token probabilities to compute robust biometric scores. Experimental results reveal a significant performance dichotomy dependent on signal quality and forgery type. In random forgery scenarios, the zero-shot VLM achieves exceptional discrimination, with GPT-5.2 reaching an Equal Error Rate of 0.32% in mobile tasks, outperforming supervised state-of-the-art systems. Conversely, in skilled forgery scenarios, where the task is more challenging because both signatures are almost identical, the results are significantly worse, and a critical \"Rationalization Trap\" emerges: chain-of-thought (CoT) reasoning degrades performance as the model produces kinematic hallucinations to justify forgery artifacts as natural variability.", "date": "2026-05-14", "categories": []}
{"id": "2605.11901", "title": "AccLock: Unlocking Identity with Heartbeat Using In-Ear Accelerometers", "abstract": "The widespread use of earphones has enabled various sensing applications, including activity recognition, health monitoring, and context-aware computing. Among these, earphone-based user authentication has become a key technique by leveraging unique biometric features. However, existing earphone-based authentication systems face key limitations: they either require explicit user interaction or active speaker output, or suffer from poor accessibility and vulnerability to environmental noise, which hinders large-scale deployment. In this paper, we propose a passive authentication system, called AccLock, which leverages distinctive features extracted from in-ear BCG signals to enable secure and unobtrusive user verification. Our system offers several advantages over previous systems, including zero-involvement for both the device and the user, ubiquitous, and resilient to environmental noise. To realize this, we first design a two-stage denoising scheme to suppress both inherent and sporadic interference. To extract user-specific features, we then propose a disentanglement-based deep learning model, HIDNet, which explicitly separates user-specific features from shared nuisance components. Lastly, we develop a scalable authentication framework based on a Siamese network that eliminates the need for per-user classifier training. We conduct extensive experiments with 33 participants, achieving an average FAR of 3.13% and FRR of 2.99%, which demonstrates the practical feasibility of AccLock.", "date": "2026-05-12", "categories": []}
{"id": "2605.12431", "title": "GaitProtector: Impersonation-Driven Gait De-Identification via Training-Free Diffusion Latent Optimization", "abstract": "Conventional gait de-identification methods often encounter an inherent trade-off: they either provide insufficient identity suppression or introduce spatiotemporal distortions that impede structure-sensitive downstream applications. We propose GaitProtector, an impersonation-driven gait de-identification framework that formulates privacy protection as a unified objective with two tightly coupled components: (i) obfuscation, which repels the protected gait from the source identity, and (ii) impersonation, which attracts it toward a selected target identity. The target identity serves as a semantic anchor that biases optimization toward structurally plausible gait patterns under the pretrained diffusion prior, helping preserve dominant body shape and motion dynamics. We instantiate this idea through a training-free diffusion latent optimization pipeline. Instead of retraining a generator for each dataset, we invert each input silhouette sequence into the latent trajectory of a pretrained 3D video diffusion model and iteratively optimize latent codes with a differentiable adversarial objective to synthesize protected gaits. Experiments on the CASIA-B dataset show that GaitProtector achieves a 56.7% impersonation success rate under black-box gait recognition and reduces Rank-1 identification accuracy from 89.6% to 15.0%, while maintaining favorable visual and temporal quality. We further evaluate downstream utility on the Scoliosis1K dataset, where diagnostic accuracy decreases only from 91.4% to 74.2%. To the best of our knowledge, this work is the first to leverage pretrained 3D diffusion priors in a training-free manner for silhouette-based gait de-identification.", "date": "2026-05-12", "categories": []}
{"id": "2411.18111", "title": "When Large Vision-Language Models Meet Person Re-Identification", "abstract": "Large Vision-Language Models (LVLMs) that incorporate visual models and large language models have achieved impressive results across cross-modal understanding and reasoning tasks. In recent years, person re-identification (ReID) has also started to explore cross-modal semantics to improve the accuracy of identity recognition. However, effectively utilizing LVLMs for ReID remains an open challenge. While LVLMs operate under a generative paradigm by predicting the next output word, ReID requires the extraction of discriminative identity features to match pedestrians across cameras. In this paper, we propose LVLM-ReID, a novel framework that harnesses the strengths of LVLMs to promote ReID. Specifically, we employ instructions to guide the LVLM in generating one semantic token that encapsulates key appearance semantics from the person image. This token is further refined through our Semantic-Guided Interaction (SGI) module, establishing a reciprocal interaction between the semantic token and visual tokens. Ultimately, the reinforced semantic token serves as the representation of pedestrian identity. Our framework integrates the semantic understanding and generation capabilities of LVLM into end-to-end ReID training, allowing LVLM to capture rich semantic cues during both training and inference. LVLM-ReID achieves competitive results on multiple benchmarks without additional image-text annotations, demonstrating the potential of LVLM-generated semantics to advance person ReID.", "date": "2024-11-27", "categories": []}
{"id": "2605.08169", "title": "Optimized Culprit Identification Using Mobilenet and Attention Mechanisms", "abstract": "Automated culprit identification in surveillance systems is a critical task that requires high accuracy along with computational efficiency for real-time deployment. In this paper, an optimized deep learning framework is proposed using a lightweight MobileNet architecture integrated with channel and spatial attention mechanisms. The proposed model enhances feature representation by selectively focusing on the most discriminative regions while suppressing irrelevant background information, thereby improving identification performance. The framework incorporates efficient preprocessing, attention based feature refinement, and a robust classification strategy optimized using the Adam Optimizer. Experiments were conducted on benchmark face recognition datasets, including Labelled Faces in the Wild (LFW), CASIA-WebFace, and a subset of VGGFace2, under realistic conditions with variations in illumination, pose, and occlusion. The results demonstrate that the proposed model achieves a high classification accuracy of 97.8%, outperforming conventional models such as baseline CNN, ResNet, and standard MobileNet. The confusion matrix analysis indicates strong class-wise discrimination with minimal misclassification, while ROC-AUC evaluation confirms robust performance across all classes. Additionally, the proposed approach maintains low computational complexity and reduced inference time, making it suitable for real-time surveillance and edge-based applications.", "date": "2026-05-04", "categories": []}
{"id": "2605.06810", "title": "Enhancing Eye Movement Biometrics for User Authentication via Continuous Gaze Offset Score Fusion", "abstract": "Eye movement biometrics (EMB) use subject-specific gaze dynamics for user authentication and identification. Recent deep learning-based EMB systems achieve strong performance by modeling temporal eye movement behavior. However, these systems typically overlook continuous gaze offset, despite prior evidence that it contains user-discriminative information. This work examines whether continuous gaze offset can improve biometric performance when combined with existing biometric features. We evaluate linear and nonlinear fusion methods on two publicly available datasets, collected via the lab-grade eye tracker and virtual reality headset across multiple tasks and observation durations. Results indicate that fusion offers performance benefits on both datasets, particularly when using nonlinear fusion. Additionally, fusing biometric information across multiple tasks further improves authentication performance. These findings support the hypothesis that continuous gaze offset may serve as useful auxiliary information under conditions of degraded or noisy eye tracking.", "date": "2026-05-07", "categories": []}
{"id": "2605.07655", "title": "Towards Billion-scale Multi-modal Biometric Search", "abstract": "Searching a multi-biometric database of a billion records for a country-level identity system requires pushing the limits of all aspects of a biometric system, including acquisition, preprocessing, feature extraction, accuracy, matching speed, presentation attack detection, and handling of special cases (e.g., missing finger digits). This is the first paper that gives insights into such a large-scale multimodal biometric search system, called Bharat ABIS, based on open-source architectures. The end-to-end pipeline of Bharat ABIS processes fingerprint, face and iris modalities through modality-specific stages of preprocessing (segmentation), quality assessment, presentation attack detection, and learning an embedding (feature extraction), producing a concatenated template of 13.5KB per person. We present a detailed analysis of the modalities and how they are integrated to create an efficient and effective solution for 1:N search (de-duplication). Evaluations on a demographically stratified gallery of 220 million identities, randomly sampled from 1.55 billion records in India's Aadhaar database, yield an FNIR of 0.3% at an FPIR of 0.5%, for adult probes (over 18 years). We also compare the performance of Bharat ABIS against three state-of-the-art COTS systems on a 20M gallery. Our system achieves a throughput of 100 searches per second on a gallery of 40M on a single server (8xNvidia H100 GPUs, 2TB RAM).", "date": "2026-05-08", "categories": []}
{"id": "2603.17859", "title": "VISER: Visually-Informed System for Enhanced Robustness in Open-Set Iris Presentation Attack Detection", "abstract": "Human perceptual priors have shown promise in saliency-guided deep learning training, particularly in the domain of iris presentation attack detection (PAD). Common saliency approaches include hand annotations obtained via mouse clicks and eye gaze heatmaps derived from eye tracking data. However, the most effective form of human saliency for open-set iris PAD remains under-explored. In this paper, we conduct a series of experiments comparing hand annotations, eye tracking heatmaps, segmentation masks, and foundation model embeddings to a state-of-the-art deep learning-based baseline on the task of open-set iris PAD. Results for open-set PAD in a leave-one-attack-type out paradigm indicate that denoised eye tracking heatmaps show the best generalization improvement over cross entropy in Attack Presentation Classification Error Rate (APCER) at Bona Fide Presentation Classification Error Rate (BPCER) of 1%. Along with this paper, we offer trained models, code, and saliency maps for reproducibility and to facilitate follow-up research efforts.", "date": "2026-03-18", "categories": []}
{"id": "2505.03597", "title": "Fixed-Length Dense Fingerprint Representation with Alignment and Robust Enhancement", "abstract": "Fixed-length fingerprint representations, which map each fingerprint to a compact and fixed-size feature vector, are computationally efficient and well-suited for large-scale matching. However, designing a robust representation that effectively handles diverse fingerprint modalities, pose variations, and noise interference remains a significant challenge. In this work, we propose a fixed-length dense descriptor of fingerprints, and introduce FLARE-a fingerprint matching framework that integrates the Fixed-Length dense descriptor with pose-based Alignment and Robust Enhancement. This fixed-length representation employs a three-dimensional dense descriptor to effectively capture spatial relationships among fingerprint ridge structures, enabling robust and locally discriminative representations. To ensure consistency within this dense feature space, FLARE incorporates pose-based alignment using complementary estimation methods, along with dual enhancement strategies that refine ridge clarity while preserving the original fingerprint modality. The proposed dense descriptor supports fixed-length representation while maintaining spatial correspondence, enabling fast and accurate similarity computation. Extensive experiments demonstrate that FLARE achieves superior performance across rolled, plain, latent, and contactless fingerprints, significantly outperforming existing methods in cross-modality and low-quality scenarios. Further analysis validates the effectiveness of the dense descriptor design, as well as the impact of alignment and enhancement modules on the accuracy of dense descriptor matching. Experimental results highlight the effectiveness and generalizability of FLARE as a unified and scalable solution for robust fingerprint representation and matching. The implementation and code will be publicly available at https://github.com/Yu-Yy/FLARE.", "date": "2025-05-06", "categories": []}
{"id": "2604.09127", "title": "FaceLiVTv2: An Improved Hybrid Architecture for Efficient Mobile Face Recognition", "abstract": "Lightweight face recognition is increasingly important for deployment on edge and mobile devices, where strict constraints on latency, memory, and energy consumption must be met alongside reliable accuracy. Although recent hybrid CNN-Transformer architectures have advanced global context modeling, striking an effective balance between recognition performance and computational efficiency remains an open challenge. In this work, we present FaceLiVTv2, an improved version of our FaceLiVT hybrid architecture designed for efficient global--local feature interaction in mobile face recognition. At its core is Lite MHLA, a lightweight global token interaction module that replaces the original multi-layer attention design with multi-head linear token projections and affine rescale transformations, reducing redundancy while preserving representational diversity across heads. We further integrate Lite MHLA into a unified RepMix block that coordinates local and global feature interactions and adopts global depthwise convolution for adaptive spatial aggregation in the embedding stage. Under our experimental setup, results on LFW, CA-LFW, CP-LFW, CFP-FP, AgeDB-30, and IJB show that FaceLiVTv2 consistently improves the accuracy-efficiency trade-off over existing lightweight methods. Notably, FaceLiVTv2 reduces mobile inference latency by 22% relative to FaceLiVTv1, achieves speedups of up to 30.8% over GhostFaceNets on mobile devices, and delivers 20-41% latency improvements over EdgeFace and KANFace across platforms while maintaining higher recognition accuracy. These results demonstrate that FaceLiVTv2 offers a practical and deployable solution for real-time face recognition. Code is available at https://github.com/novendrastywn/FaceLiVT.", "date": "2026-04-10", "categories": []}
{"id": "2605.04769", "title": "Lightweight Cross-Spectral Face Recognition via Contrastive Alignment and Distillation", "abstract": "Heterogeneous Face Recognition (HFR) aims at matching face images captured across different sensing modalities, such as thermal-to-visible or near-infrared-to-visible, enhancing the usability of face recognition systems in challenging real-world conditions. Although recent HFR methods have achieved significant improvements in performance, many rely on computationally expensive models, making them impractical for deployment on resource-limited edge devices. In this work, we introduce a lightweight yet effective HFR framework by adapting a hybrid CNN-Transformer model originally developed for RGB homogeneous face recognition. Our approach enables efficient end-to-end training with only a small amount of paired heterogeneous data, while still maintaining strong performance on standard RGB face recognition benchmarks. This makes it suitable for both homogeneous and heterogeneous settings. Comprehensive experiments on several challenging HFR and face recognition benchmarks show that our method achieves state-of-the-art or competitive performance while keeping computational requirements low.", "date": "2026-05-06", "categories": []}
{"id": "2605.04904", "title": "Exploring Clustering Capability of Inpainting Model Embeddings for Pattern-based Individual Identification", "abstract": "In this paper, we explore deep learning techniques for individual identification of animals based on their skin patterns. Individual identification is crucial in biodiversity monitoring, since it enables analysis of decline or growth of populations, or intra-species interactions within populations. Models trained for the task of individual identification often do not focus on the skin pattern of animals, but on background details or body shape details. These characteristics are not individually specific, or can change drastically through time. We focus on techniques that will make machine learning models more responsive to skin pattern structure when extracting individual visual embeddings from images. For this, we explore image inpainting of task-specific masks as an auxiliary task to enhance ML-based individual identification from animal skin patterns. We propose a comparative analysis among four models as an encoder backbone for the individual identification task. We focus on the case study of zebrafish, which is a widely recognized biological model organism, and which exhibits individually identifying skin patterns. To evaluate encoder backbone performance, we present standard metrics for classification accuracy, embedding clustering metrics, and GradCAM visualizations.", "date": "2026-05-06", "categories": []}
{"id": "2605.04977", "title": "ICPR 2026 Competition on Privacy-Preserving Person Re-Identification from Top-View RGB-Depth Camera (TVRID)", "abstract": "This companion paper reports the ICPR 2026 TVRID competition on privacy-aware top-view person re-identification. We present the competition setting, the released RGB-Depth dataset, and a summary of final results with descriptions of the top entries. TVRID contains 86 identities captured by four synchronized overhead Intel RealSense D455 cameras, with paired RGB/Depth streams and structured geometric variation across flat, ascent, descent, and oblique viewpoints. The evaluation protocol includes three tracks: RGB Re-ID, Depth Re-ID, and RGB$\\leftrightarrow$Depth cross-modal retrieval. Submissions are ranked using mAP and CMC-1 under a unified server-side evaluation. The final results show a clear difficulty ordering (RGB $>$ Depth $>$ Cross-Modal), highlighting both the challenge of modality-constrained retrieval and the feasibility of strong performance with modality-invariant learning. By releasing the dataset at https://zenodo.org/records/17909410, the evaluation scripts at https://github.com/RaphaelDel/ICPR-TVRID, and the accompanying documentation, TVRID establishes a reproducible benchmark for top-view, depth-based, and cross-modal person re-id.", "date": "2026-05-06", "categories": []}
{"id": "2605.05027", "title": "Prompt-Anchored Vision-Text Distillation for Lifelong Person Re-identification", "abstract": "Lifelong person re-identification (LReID) aims to train a generalizable model with sequentially collected data. However, such models often suffer from semantic drift, limited adaptability, and catastrophic forgetting as new domains emerge. Existing exemplar-free approaches largely rely on visual-only distillation or parameter regularization, while overlooking the potential of auxiliary modalities, such as text, to preserve semantic stability and enable incremental plasticity. We observe that the frozen text encoder in pretrained vision-language models can serve as a stable semantic anchor across domains. To decouple the roles of vision and text, we propose Prompt-Anchored vision-text Distillation (PAD), an asymmetric vision-text framework for semantic alignment and cross-domain generalization. On the textual side, we distill prompts to preserve vision-text alignment under a fixed semantic space, acting as a global semantic reference rather than a dominant learning signal. On the visual side, an EMA-based teacher with an adaptive prompt pool enables domain-wise adaptation by allocating new slots while freezing past ones. Extensive experiments show that PAD substantially outperforms state-of-the-art methods across seen and unseen domains, achieving a strong balance between stability and plasticity. Project page is available at https://github.com/zu-zi/PAD.", "date": "2026-05-06", "categories": []}
{"id": "2603.17679", "title": "Illumination-Aware Contactless Fingerprint Spoof Detection via Paired Flash-Non-Flash Imaging", "abstract": "Contactless fingerprint recognition enables hygienic and convenient biometric authentication but poses new challenges for spoof detection due to the absence of physical contact and traditional liveness cues. Most existing methods rely on single-image acquisition and appearance-based features, which often generalize poorly across devices, capture conditions, and spoof materials. In this work, we study paired flash-non-flash contactless fingerprint acquisition as a lightweight active sensing mechanism for spoof detection. Through a preliminary empirical analysis, we show that flash illumination accentuates material- and structure-dependent properties, including ridge visibility, subsurface scattering, micro-geometry, and surface oils, while non-flash images provide a baseline appearance context. We analyze lighting-induced differences using interpretable metrics such as inter-channel correlation, specular reflection characteristics, texture realism, and differential imaging. These complementary features help discriminate genuine fingerprints from printed, digital, and molded presentation attacks. We further examine the limitations of paired acquisition, including sensitivity to imaging settings, dataset scale, and emerging high-fidelity spoofs. Our findings demonstrate the potential of illumination-aware analysis to improve robustness and interpretability in contactless fingerprint presentation attack detection, motivating future work on paired acquisition and physics-informed feature design. Code is available in the repository.", "date": "2026-03-18", "categories": []}
{"id": "2604.22515", "title": "Different Strokes for Different Folks: Writer Identification for Historical Arabic Manuscripts", "abstract": "Handwritten Arabic manuscripts preserve the Arab world's intellectual and cultural heritage, and writer identification supports provenance, authenticity verification, and historical analysis. Using the Muharaf dataset of historical Arabic manuscripts, we evaluate writer identification from individual line images and, to the best of our knowledge, provide the first baselines reported under both line-level and page-disjoint evaluation protocols. Since the dataset is only partially labeled for writer identification, we manually verified and expanded writer labels in the public portion from 6,858 (28.00%) to 21,249 lines (86.75%) out of 24,495 line images, correcting inconsistencies and removing non-handwritten text. After further filtering, we retained 18,987 lines (77.51%). We propose a Convolutional Neural Network (CNN)-based model with attention mechanisms for closed-set writer identification, including rare two-writer lines modeled as composite writer-pair classes. We benchmark fourteen configurations and conduct ablations across different feature extractors and training regimes. To assess generalization to unseen pages, the page-disjoint protocol assigns all lines from each page to a single split. Under the line-level protocol, a fine-tuned DenseNet201 with attention achieves 99.05% Top-1 accuracy, 99.73% Top-5 accuracy, and 97.44% F1-score. Under the more challenging page-disjoint protocol, the best observed results are 78.61% Top-1 accuracy, 87.79% Top-5 accuracy, and 66.55% F1-score, thus quantifying the impact of page-level cues. By expanding the Muharaf dataset's labeled subset and reporting both protocols, we provide a clearer benchmark and a practical resource for historians and linguists engaged with culturally and historically significant documents. The code and implementation details are available on GitHub.", "date": "2026-04-24", "categories": []}
{"id": "2605.03830", "title": "Identity-Consistent Multi-Pose Generation of Contactless Fingerprints", "abstract": "Contactless fingerprint recognition has gained increasing attention due to its advantages in hygiene and acquisition flexibility. However, the absence of physical contact constraints introduces severe nonlinear geometric distortions caused by free finger poses in 3D space, resulting in a substantial cross-modal domain gap between contactless and conventional contact-based fingerprints. Existing solutions largely rely on explicit geometric correction or image enhancement, which are fragile under extreme pose variations. In this paper, we propose Identity-Consistent Multi-Pose Generation of Contactless Fingerprints (IMPOSE), a physics-inspired framework that synthesizes identity-preserving, multi-pose contactless fingerprint samples to empower recognition models. IMPOSE consists of three stages: (1) rolled fingerprint identity generation via latent diffusion with discrete codebook representations, (2) cross-modal translation from rolled to contactless modality guided by Sauvola-based local adaptive binarization as an identity anchor, and (3) physics-based multi-pose simulation through 3D finger model texture mapping and projection. The generated samples maintain strict identity consistency at the ridge topology level and spatial alignment with standard fingerprint coordinate space. Extensive experiments on the UWA and PolyU CL2CB databases demonstrate that fine-tuning fixed-length dense descriptors (FDD) with IMPOSE-synthesized data achieves state-of-the-art cross-modal matching, reducing EER to 8.74% on UWA and 2.26% on PolyU CL2CB. Synthetic data also yields consistent gains across mainstream representations including DeepPrint and AFRNet, and the hybrid strategy combining synthetic and real data achieves the best overall results. The code and generated samples are available at https://github.com/Yu-Yy/IMPOSE.", "date": "2026-05-05", "categories": []}
{"id": "2605.03857", "title": "A Deeper Dive into the Irreversibility of PolyProtect: Making Protected Face Templates Harder to Invert", "abstract": "This work presents a deeper analysis of the \"irreversibility\" property of PolyProtect, a biometric template protection method initially proposed for securing face embeddings. PolyProtect transforms embeddings into protected templates via multivariate polynomials, whose coefficients and exponents are distinct for each subject enrolled in the face recognition system. A polynomial is applied to consecutive sets of elements from a given embedding, where the amount of overlap between the sets is a tunable parameter. We begin our irreversibility analysis by demonstrating that PolyProtected templates are easier to invert using a numerical solver based on cosine distance, as opposed to Euclidean distance (used in the earlier PolyProtect work). To make this inversion more difficult, we then propose a \"key selection algorithm\", which tries to choose \"keys\" (coefficients and exponents of the PolyProtect polynomial) that enhance the irreversibility of PolyProtected templates, compared to when the keys are purely random. Our experiments show that this algorithm is effective at generating PolyProtected templates that are significantly more difficult to invert, and that it approximately equalises the irreversibility of PolyProtected templates generated using different \"overlap\" parameters. This allows for better control of the irreversibility versus accuracy trade-off, known to exist across different overlaps. We also show that accuracy in the PolyProtected domain can be affected by the range in which the embedding elements lie, but that this can be improved by normalizing the embeddings prior to applying PolyProtect. This work is reproducible using our open-source code.", "date": "2026-05-05", "categories": []}
{"id": "2603.14632", "title": "Continual Few-shot Adaptation for Synthetic Fingerprint Detection", "abstract": "The quality and realism of synthetically generated fingerprint images have increased significantly over the past decade fueled by advancements in generative artificial intelligence (GenAI). This has exacerbated the vulnerability of fingerprint recognition systems to data injection attacks, where synthetic fingerprints are maliciously inserted during enrollment or authentication. Hence, there is an urgent need for methods to detect if a fingerprint image is real or synthetic. While it is straightforward to train deep neural network (DNN) models to classify images as real or synthetic, often such DNN models overfit the training data and fail to generalize well when applied to synthetic fingerprints generated using unseen GenAI models. In this work, we formulate synthetic fingerprint detection as a continual few-shot adaptation problem, where the objective is to rapidly evolve a base detector to identify new types of synthetic data. To enable continual few-shot adaptation, we employ a combination of binary cross-entropy and supervised contrastive (applied to the feature representation) losses and replay a few samples from previously known styles during fine-tuning to mitigate catastrophic forgetting. Experiments based on several DNN backbones (as feature extractors) and a variety of real and synthetic fingerprint datasets indicate that the proposed approach achieves a good trade-off between fast adaptation for detecting unseen synthetic styles and forgetting of known styles.", "date": "2026-03-15", "categories": []}
{"id": "2605.01548", "title": "ECG-biometrics-bench: A Unified Framework for Reproducible Benchmarking of ECG Biometrics", "abstract": "Electrocardiogram (ECG) biometrics have emerged as a promising modality for continuous, liveness-aware authentication in wearable systems. However, many prior studies report overly optimistic results due to data leakage (e.g., random splits within the same session). To address this issue, we introduce ECG-biometrics-bench, a modular, reproducible benchmarking framework that standardizes preprocessing, segmentation, and evaluation across seven widely used public ECG datasets spanning clinical, ambulatory, and large-scale cohort settings. The framework supports both closed-set and open-set (i.e., subject-disjoint generalization in this work) evaluation, as well as progressively realistic protocols including cross-session and long-term temporal separation. To facilitate reproducible research in the community, the ECG-biometrics-bench repository will be made publicly accessible on GitHub upon the acceptance of this manuscript. Through a comprehensive multi-dataset analysis, we expose the Random Split Fallacy, demonstrating that intra-session evaluation protocols artificially inflate performance while masking severe degradation caused by temporal drift and unseen identities. Furthermore, by evaluating multiple architectures, including DeepECG, ResNet1D, and CNN-LSTM, we show that these failures are not model-specific but are likely inherent to current supervised feature-learning paradigms. Finally, we demonstrate that performance degradation due to temporal aging can be partially mitigated through a heavy enrollment, lightweight authentication strategy based on dynamic multi-session template fusion. These findings establish a more realistic baseline for ECG biometrics and highlight critical challenges that must be addressed for reliable real-world deployment.", "date": "2026-05-02", "categories": []}
{"id": "2605.00111", "title": "AIDA-ReID: Adaptive Intermediate Domain Adaptation for Generalizable and Source-Free Person Re-Identification", "abstract": "Person re-identification (Re-ID) aims to match images of the same individual across non-overlapping camera views and remains challenging due to domain shifts caused by variations in illumination, background, camera characteristics, and population distributions. Although supervised models perform well under matched training and testing conditions, their performance degrades significantly when deployed in unseen environments. Existing intermediate domain approaches such as IDM and IDM++ alleviate this gap by constructing bridge feature distributions between domains; however, they rely on fixed mixing strategies and joint source-target access, limiting their applicability to multi-source and source-free settings. To address these limitations, this paper proposes Adaptive Intermediate Domain Adaptation (AIDA), also referred to as Source-Free Multi-Source Intermediate Domain Adaptation (SF-MIDA). The proposed framework treats intermediate-domain learning as a dynamically regulated process, where feature mixing and regularization strength are adaptively controlled using feedback signals derived from model uncertainty and training stability. A multi-source intermediate domain generator synthesizes diverse intermediate representations, while a pseudo-mirror regularization strategy preserves identity consistency under domain perturbations. Extensive experiments across domain generalization and source-free settings demonstrate the effectiveness of the proposed framework.", "date": "2026-04-30", "categories": []}
{"id": "2605.00120", "title": "GAFSV-Net: A Vision Framework for Online Signature Verification", "abstract": "Online signature verification (OSV) requires distinguishing skilled forgeries from genuine samples under high intra-class variability and with very few enrollment samples. Existing deep learning methods operate directly on raw temporal sequences, restricting them to 1D architectures and preventing the use of pretrained 2D vision backbones. We bridge this gap with GAFSV-Net, which represents each signature as a six-channel asymmetric Gramian Angular Field image: three kinematic channels (pen speed, pressure derivative, direction angle) are each encoded into complementary GASF and GADF matrices that capture pairwise temporal co-occurrence and directional transition structure respectively. A dual-branch ConvNeXt-Tiny encoder processes GASF and GADF independently, with bidirectional cross-attention enabling each branch to query discriminative patterns from the other before metric-space projection. Training uses semi-hard triplet loss with skilled-forgery hard-negative injection; verification is performed via cosine similarity against a small enrollment prototype. We evaluate on DeepSignDB and BiosecurID, outperforming all sequence-based baselines trained under identical objectives, demonstrating that the representational gain of 2D temporal encoding is consistent and independent of training procedure, with ablations characterising each design choice's contribution.", "date": "2026-04-30", "categories": []}
{"id": "2605.00218", "title": "Selfie-Capture Dynamics as an Auxiliary Signal Against Deepfakes and Injection Attacks for Mobile Identity Verification", "abstract": "Mobile remote identity verification (RIdV) systems are exposed to attacks that manipulate or replace the facial video stream, including presentation attacks, real-time deepfakes, and video injection. Recent European requirements, including ETSI TS 119 461 and CEN/TS 18099, motivate complementary evidence channels beyond camera-based presentation-attack detection. This paper investigates whether passive motion traces recorded during selfie capture provide auxiliary evidence for spoof screening and user verification. We introduce CanSelfie, a dataset of 375 bona fide multi-sensor sequences collected at 50\\,Hz from 30 participants using a commercial mobile RIdV application, together with stationary, handheld, and temporally shifted attack-proxy scenarios. We benchmark 7 multivariate time-series classifiers and 8 whole-series anomaly detectors across sensor configurations and temporal windows. For spoof screening, accelerometer-only ROCKAD obtains 0.00\\% false rejection rate (FRR) and 43.8\\% false acceptance rate (FAR), while QUANT+3-NN obtains the lowest overall FAR of 32.0\\% at 2.37\\% FRR; both reject all stationary attack proxies. For same-device and same-session user verification, WEASEL+MUSE reaches 1.07\\% equal error rate (EER) using 9 sensor channels. The analysis shows that raw accelerometer data, preserving gravity and orientation cues, is the most informative modality, and that closed-set classification accuracy alone does not imply good verification performance because threshold calibration depends on score distributions. The findings suggest that short selfie-capture motion traces contain measurable spoof-related and identity-related information, supporting their use as a low-friction auxiliary signal while also identifying the need for cross-device, cross-session, and real injection-attack evaluation.", "date": "2026-04-30", "categories": []}
{"id": "2604.27122", "title": "InterPartAbility: Text-Guided Part Matching for Interpretable Person Re-Identification", "abstract": "Text-to-image person re-identification (TI-ReID) relies on natural-language text description to retrieve top matching individuals from a large gallery of images. While recent large vision-language models (VLMs) achieve strong retrieval performance, their decisions remain largely uninterpretable. Existing interpretability approaches in TI-ReID rely solely on slot-attention to highlight attended regions, but fail to reliably bind visual regions to semantically meaningful concepts, limiting explanations to qualitative visualizations over a restricted vocabulary. This paper introduces InterPartAbility, an interpretable TI-ReID method that performs explicit part-wise matching and enables phrase-region grounding. A new open-vocabulary, lightweight supervision, patch-phrase interaction module (PPIM) is proposed to train a standard TI-ReID model with concept-level guidance. Concept-based part phrases provide evidence that encourages the model to attend to corresponding image regions. InterPartAbility further constrains CLIP ViT self-attention to produce spatially concentrated patch activations aligned with each part-level phrase, yielding grounded explanation maps. A quantitative interpretability protocol for TI-ReID is introduced by adapting perturbation-based evaluation metrics, including counterfactual region masking that measures retrieval degradation when top-ranked explanatory regions are removed. Empirical results\\footnote{Our code is included in the supplementary materials and will be made public.} on challenging benchmarks like CUHK-PEDES and ICFG-PEDES show that InterPartAbility achieves state-of-the-art (SOTA) interpretability performance under these metrics, while sustaining competitive retrieval accuracy.", "date": "2026-04-29", "categories": []}
{"id": "2604.27218", "title": "AttriBE: Quantifying Attribute Expressivity in Body Embeddings for Recognition and Identification", "abstract": "Person re-identification (ReID) systems that match individuals across images or video frames are essential in many real-world applications. However, existing methods are often influenced by attributes such as gender, pose, and body mass index (BMI), which vary in unconstrained settings and raise concerns related to fairness and generalization. To address this, we extend the notion of expressivity, defined as the mutual information between learned features and specific attributes, using a secondary neural network to quantify how strongly attributes are encoded. Applying this framework to three transformer-based ReID models on a large-scale visible-spectrum dataset, we find that BMI consistently shows the highest expressivity in deeper layers. Attributes in the final representation are ranked as BMI > Pitch > Gender > Yaw, and expressivity evolves across layers and training epochs, with pose peaking in intermediate layers and BMI strengthening with depth. We further extend the analysis to cross-spectral person identification across infrared modalities including short-wave, medium-wave, and long-wave infrared. In this setting, pitch becomes comparable to BMI and attribute trends increase monotonically across depth, suggesting increased reliance on structural cues when bridging modality gaps. Overall, the results show that transformer-based ReID embeddings encode a hierarchy of implicit attributes, with morphometric information persistently embedded and pose contributing more strongly under cross-spectral conditions.", "date": "2026-04-29", "categories": []}
{"id": "2604.27564", "title": "Learning from a single labeled face and a stream of unlabeled data", "abstract": "Face recognition from a single image per person is a challenging problem because the training sample is extremely small. We consider a variation of this problem. In our problem, we recognize only one person, and there are no labeled data for any other person. This setting naturally arises in authentication on personal computers and mobile devices, and poses additional challenges because it lacks negative examples. We formalize our problem as one-class classification, and propose and analyze an algorithm that learns a non-parametric model of the face from a single labeled image and a stream of unlabeled data. In many domains, for instance when a person interacts with a computer with a camera, unlabeled data are abundant and easy to utilize. This is the first paper that investigates how these data can help in learning better models in the single-image-per-person setting. Our method is evaluated on a dataset of 43 people and we show that these people can be recognized 90% of time at nearly zero false positives. This recall is 25+% higher than the recall of our best performing baseline. Finally, we conduct a comprehensive sensitivity analysis of our algorithm and provide a guideline for setting its parameters in practice.", "date": "2026-04-30", "categories": []}
{"id": "2604.26285", "title": "Event-based Liveness Detection using Temporal Ocular Dynamics: An Exploratory Approach", "abstract": "Face liveness detection has been extensively studied using RGB cameras, achieving strong performance under controlled conditions but often failing to generalize across sensors and attack scenarios. In this work, we explore event cameras as an alternative sensing modality for liveness detection based on temporal ocular dynamics. Event cameras capture sparse, asynchronous changes in brightness with microsecond resolution, enabling precise analysis of fast eye movements such as saccades. Replay attacks cannot faithfully reproduce these dynamics due to temporal resampling and display artifacts, leading to distinctive spatio-temporal patterns in the event domain. We design a data collection protocol to extend RGBE-Gaze with replay-attack recordings, yielding an event-based fake counterpart for liveness detection. We analyze event-driven temporal features from eye regions and evaluate their effectiveness for ocular motion segmentation and liveness classification. Our results show that event-based representations enable reliable discrimination between genuine and replayed sequences, achieving up to 95.37% top-1 accuracy with a spiking convolutional neural network. These preliminary findings highlight the potential of event-based sensing for robust and low-latency liveness detection.", "date": "2026-04-29", "categories": []}
{"id": "2604.26363", "title": "CO-EVO: Co-evolving Semantic Anchoring and Style Diversification for Federated DG-ReID", "abstract": "Federated domain generalization for person re-identification (FedDG-ReID) aims to collaboratively train a pedestrian retrieval model across multiple decentralized source domains such that it can generalize to unseen target environments without compromising raw data privacy. However, this task is significantly challenged by the inherent stylistic gaps across decentralized clients. Without global supervision, models easily succumb to shortcut learning where representations overfit to domain specific camera biases rather than universal identity features. We propose CO-EVO, a novel federated framework that resolves this semantic-style conflict through a co-evolutionary mechanism. On the semantic side, Camera-Invariant Semantic Anchoring (CSA) learns identity prompts with cross-camera consistency to establish purified and domain-agnostic anchors that filter out local imaging noise. On the visual side, Global Style Diversification (GSD), powered by a Global Camera-Style Bank (GCSB), synthesizes realistic perturbations to expand the visual boundaries of training data. The core of CO-EVO is its co-evolutionary loop where purified anchors act as gravitational centers to guide the image encoder toward robust anatomical attributes amidst diverse style variations. Extensive experiments demonstrate that CO-EVO achieves state-of-the-art (SOTA) performance, proving that the synergy between semantic purification and style expansion is essential for robust cross-domain generalization. Our code is available at: https://github.com/NanYiyuzurn/ACL-LGPS-2026.", "date": "2026-04-29", "categories": []}
{"id": "2604.26598", "title": "FunFace: Feature Utility and Norm Estimation for Face Recognition", "abstract": "Face Recognition (FR) is used in a variety of application domains, from entertainment and banking to security and surveillance. Such applications rely on the FR model to be robust and perform well in a variety of settings. To achieve this, state-of-the-art FR models typically use expressive adaptive margin loss functions, which tie the feature norm to concepts related to sample quality, such as recognizability and perceptual image quality. Recently, through the development of Face Image Quality Assessment (FIQA) techniques, biometric utility has become the preferred measure of face-image quality and has been shown to be a better predictor of the usefulness of samples for face recognition compared to more human-centric aspects, such as resolution, blur, and lighting, tied to general image quality. While image quality expressed through feature norms exhibits a certain level of correlation with biometric utility, it does not fully encapsulate all aspects of utility. To address this point, we propose a new adaptive margin loss, FunFace (Face Recognition Through Utility and Norm Estimation), which incorporates biometric utility, estimated by the Certainty Ratio, into the adaptive margin, taking inspiration from AdaFace. We show that FunFace (when used to train a face recognition model) achieves competitive results to other state-of-the-art FR models on benchmarks containing high-quality samples, while surpassing them on low quality benchmarks.", "date": "2026-04-29", "categories": []}
{"id": "2512.06757", "title": "XM-ALIGN: Unified Cross-Modal Embedding Alignment for Face-Voice Association", "abstract": "This paper introduces our solution, XM-ALIGN (Unified Cross-Modal Embedding Alignment Framework), proposed for the FAME challenge at ICASSP 2026. Our framework combines explicit and implicit alignment mechanisms, significantly improving cross-modal verification performance in both \"heard\" and \"unheard\" languages. By extracting feature embeddings from both face and voice encoders and jointly optimizing them using a shared classifier, we employ mean squared error (MSE) as the embedding alignment loss to ensure tight alignment between modalities. Additionally, data augmentation strategies are applied during model training to enhance generalization. Experimental results show that our approach demonstrates superior performance on the MAV-Celeb dataset. The code will be released at https://github.com/PunkMale/XM-ALIGN.", "date": "2025-12-07", "categories": []}
{"id": "2604.25071", "title": "Scalable Secure Biometric Authentication without Auxiliary Identifiers", "abstract": "The prevalence of biometric authentication has been on the rise due to its ease of use and elimination of weak passwords. To date, most biometric authentication systems have been designed for on-device authentication of the device owner (e.g., smartphones and laptops). Recently, biometric authentication systems have started to emerge that are designed to authenticate users against cloud databases storing representations of biometrics for large numbers of users (potentially millions), such as those facilitating biometric payments. However, the use of a large cloud database introduces a significant attack vector, as a breach of the database could lead to the compromise of all enrolled users' sensitive biometric data. Indeed, all such existing systems either do not adequately protect against such a breach, or are impractical to deploy and use due to their high computational overhead. In this work, we present a new biometric authentication system that provides provable security guarantees against data breaches, while remaining scalable and performant. To do so, we marry artificial intelligence with advanced cryptographic techniques in a novel fashion, providing several optimizations along the way. Our work is the first to show that real-world scalable privacy-preserving biometric authentication without auxiliary identifiers is feasible, and we believe that it will spur widespread industrial adoption and further research in this area.", "date": "2026-04-27", "categories": []}
{"id": "2510.10113", "title": "ImmerIris: A Large-Scale Dataset and Benchmark for Off-Axis and Unconstrained Iris Recognition in Immersive Applications", "abstract": "Recently, iris recognition is regaining prominence in immersive applications such as extended reality as a means of seamless user identification. This application scenario introduces unique challenges compared to traditional iris recognition under controlled setups, as the ocular images are primarily captured off-axis and less constrained, causing perspective distortion, intra-subject variation, and quality degradation in iris textures. Datasets capturing these challenges remain limited. This paper fills this gap by presenting a large-scale iris dataset collected via head-mounted displays, termed ImmerIris. It contains 499,791 ocular images from 546 subjects, and is, to our knowledge, the largest public iris dataset to date and among the first dedicated to immersive applications. It is accompanied by a comprehensive set of evaluation protocols that benchmark recognition systems under various challenging conditions. This paper also draws attention to a shared obstacle of current recognition methods, the reliance on a pre-processing, normalization stage, which is fallible in off-axis and unconstrained setups. To this end, this paper further proposes a normalization-free paradigm that directly learns from minimally adjusted ocular images. Despite its simplicity, it outperforms normalization-based prior arts, indicating a promising direction for robust iris recognition.", "date": "2025-10-11", "categories": []}
{"id": "2604.23247", "title": "Micro-Expression-Aware Avatar Fingerprinting via Inter-Frame Feature Differencing", "abstract": "Avatar fingerprinting, i.e., verifying who drives a synthetic talking-head video rather than whether it is real, is a critical safeguard for authorized use of face-reenactment technology. Existing methods rely on a fixed, non-differentiable landmark extraction stage that prevents the fingerprinting model from being optimized end-to-end from raw pixels. We propose a preprocessing-free system built on a micro-expression-aware backbone operating on raw video frames, with inter-frame feature differencing as the core design principle: consecutive feature maps are subtracted in the learned deep feature space, so that temporally stable appearance dimensions contribute zero to the output while driver-specific motion dynamics are preserved. A controlled ablation on NVFAIR confirms that temporal motion accounts for the large majority of discriminative performance, and that raw appearance features actively degrade identity separation. Both the choice of backbone and the differencing principle are essential: differencing alone is insufficient when applied to a generic encoder, as appearance-dominated features collapse to near-identical representations across adjacent frames, while the micro-expression-aware F5C backbone retains measurable motion variation that the differencing operation can exploit. Without any external preprocessing, our model achieves an overall AUC of 0.877 on NVFAIR and matches or exceeds the landmark-based baseline on the majority of cross-generator pairs.", "date": "2026-04-25", "categories": []}
{"id": "2604.22190", "title": "From Global to Local: Rethinking CLIP Feature Aggregation for Person Re-Identification", "abstract": "CLIP-based person re-identification (ReID) methods aggregate spatial features into a single global \\texttt{[CLS]} token optimized for image-text alignment rather than spatial selectivity, making representations fragile under occlusion and cross-camera variation. We propose SAGA-ReID, which reconstructs identity representations by aligning intermediate patch tokens with anchor vectors parameterized in CLIP's text embedding space -- emphasizing spatially stable evidence while suppressing corrupted or absent regions, without requiring textual descriptions of individual images. Controlled experiments isolate the aggregation mechanism under two qualitatively distinct conditions -- synthetic masking, where identity signal is absent, and realistic human distractors, where an overlapping person introduces semantically confusing signal -- with SAGA's advantage over global pooling growing substantially as occlusion increases across both conditions. Benchmark evaluations confirm consistent gains over CLIP-ReID across standard and occluded settings, with the largest improvements where global pooling is most unreliable: up to +10.6 Rank-1 on occluded benchmarks. SAGA's aggregation outperforms dedicated sequential patch aggregation on a stronger backbone, confirming that structured reconstruction addresses a bottleneck that backbone quality and architectural complexity alone cannot resolve. Code available at https://github.com/ipl-uw/Structured-Anchor-Guided-Aggregation-for-ReID.", "date": "2026-04-24", "categories": []}
{"id": "2604.14734", "title": "Find the Differences: Differential Morphing Attack Detection vs Face Recognition", "abstract": "Morphing is a challenge to face recognition (FR) for which several morphing attack detection solutions have been proposed. We argue that face recognition and differential morphing attack detection (D-MAD) in principle perform very similar tasks, which we support by comparing an FR system with two existing D-MAD approaches. We also show that currently used decision thresholds inherently lead to FR systems being vulnerable to morphing attacks and that this explains the tradeoff between performance on normal images and vulnerability to morphing attacks. We propose using FR systems that are already in place for morphing detection and introduce a new evaluation threshold that guarantees an upper limit to the vulnerability to morphing attacks - even of unknown types.", "date": "2026-04-16", "categories": []}
{"id": "2604.21324", "title": "Temporal Prototyping and Hierarchical Alignment for Unsupervised Video-based Visible-Infrared Person Re-Identification", "abstract": "Visible-infrared person re-identification (VI-ReID) enables cross-modality identity matching for all-day surveillance, yet existing methods predominantly focus on the image level or rely heavily on costly identity annotations. While video-based VI-ReID has recently emerged to exploit temporal dynamics for improved robustness, existing studies remain limited to supervised settings. Crucially, the unsupervised video VI-ReID problem, where models must learn from RGB and infrared tracklets without identity labels, remains largely unexplored despite its practical importance in real-world deployment. To bridge this gap, we propose HiTPro (Hierarchical Temporal Prototyping), a prototype-driven framework without explicit hard pseudo-label assignment for unsupervised video-based VI-ReID. HiTPro begins with an efficient Temporal-aware Feature Encoder that first extracts discriminative frame-level features and then aggregates them into a robust tracklet-level representation. Building upon these features, HiTPro first constructs reliable intra-camera prototypes via Intra-Camera Tracklet Prototyping by aggregating features from temporally partitioned sub-tracklets. Through Hierarchical Cross-Prototype Alignment, we perform a two-stage positive mining process: progressing from within-modality associations to cross-modality matching, enhanced by Dynamic Threshold Strategy and Soft Weight Assignment. Finally, {Hierarchical Contrastive Learning} progressively optimizes feature-prototype alignment across three levels: intra-camera discrimination, cross-camera same-modality consistency, and cross-modality invariance. Extensive experiments on HITSZ-VCM and BUPTCampus demonstrate that HiTPro achieves state-of-the-art performance under fully unsupervised settings, significantly outperforming adapted baselines and establishes a strong baseline for future research.", "date": "2026-04-23", "categories": []}
{"id": "2604.21689", "title": "StyleID: A Perception-Aware Dataset and Metric for Stylization-Agnostic Facial Identity Recognition", "abstract": "Creative face stylization aims to render portraits in diverse visual idioms such as cartoons, sketches, and paintings while retaining recognizable identity. However, current identity encoders, which are typically trained and calibrated on natural photographs, exhibit severe brittleness under stylization. They often mistake changes in texture or color palette for identity drift or fail to detect geometric exaggerations. This reveals the lack of a style-agnostic framework to evaluate and supervise identity consistency across varying styles and strengths. To address this gap, we introduce StyleID, a human perception-aware dataset and evaluation framework for facial identity under stylization. StyleID comprises two datasets: (i) StyleBench-H, a benchmark that captures human same-different verification judgments across diffusion- and flow-matching-based stylization at multiple style strengths, and (ii) StyleBench-S, a supervision set derived from psychometric recognition-strength curves obtained through controlled two-alternative forced-choice (2AFC) experiments. Leveraging StyleBench-S, we fine-tune existing semantic encoders to align their similarity orderings with human perception across styles and strengths. Experiments demonstrate that our calibrated models yield significantly higher correlation with human judgments and enhanced robustness for out-of-domain, artist drawn portraits. All of our datasets, code, and pretrained models are publicly available at https://kwanyun.github.io/StyleID_page/", "date": "2026-04-23", "categories": []}
{"id": "2509.08539", "title": "Motion-Based User Identification across XR and Metaverse Applications by Deep Classification and Similarity Learning", "abstract": "This paper examines the generalization capacity of two state-of-the-art classification and similarity learning models in reliably identifying users based on their motions in various Extended Reality (XR) applications. We developed a novel dataset containing a wide range of motion data from 49 users in five different XR applications: four XR games with distinct tasks and action patterns, and an additional social XR application with no predefined task sets. The dataset is used to evaluate the performance and, in particular, the generalization capacity of the two models across applications. Our results indicate that while the models can accurately identify individuals within the same application, their ability to identify users across different XR applications remains limited. Overall, our results provide insight into current models generalization capabilities and suitability as biometric methods for user verification and identification. The results also serve as a much-needed risk assessment of hazardous and unwanted user identification in XR and Metaverse applications. Our cross-application XR motion dataset and code are made available to the public to encourage similar research on the generalization of motion-based user identification in typical Metaverse application use cases.", "date": "2025-09-10", "categories": []}
{"id": "2604.20585", "title": "On the Impact of Face Segmentation-Based Background Removal on Recognition and Morphing Attack Detection", "abstract": "This study investigates the impact of face image background correction through segmentation on face recognition and morphing attack detection performance in realistic, unconstrained image capture scenarios. The motivation is driven by operational biometric systems such as the European Entry/Exit System (EES), which require facial enrolment at airports and other border crossing points where controlled backgrounds usually required for such captures cannot always be guaranteed, as well as by accessibility needs that may necessitate image capture outside traditional office environments. By analyzing how such preprocessing steps influence both recognition accuracy and security mechanisms, this work addresses a critical gap between usability-driven image normalization and the reliability requirements of large-scale biometric identification systems. Our study evaluates a comprehensive range of segmentation techniques, three families of morphing attack detection methods, and four distinct face recognition models, using databases that include both controlled and in-the-wild image captures. The results reveal consistent patterns linking segmentation to both recognition performance and face image quality. Additionally, segmentation is shown to systematically influence morphing attack detection performance. These findings highlight the need for careful consideration when deploying such preprocessing techniques in operational biometric systems.", "date": "2026-04-22", "categories": []}
{"id": "2604.19218", "title": "Thinking Before Matching: A Reinforcement Reasoning Paradigm Towards General Person Re-Identification", "abstract": "Learning identity-discriminative representations with multi-scene generality has become a critical objective in person re-identification (ReID). However, mainstream perception-driven paradigms tend to identify fitting from massive annotated data rather than identity-causal cues understanding, which presents a fragile representation against multiple disruptions. In this work, ReID-R is proposed as a novel reasoning-driven paradigm that achieves explicit identity understanding and reasoning by incorporating chain-of-thought into the ReID pipeline. Specifically, ReID-R consists of a two-stage contribution: (i) Discriminative reasoning warm-up, where a model is trained in a CoT label-free manner to acquire identity-aware feature understanding; and (ii) Efficient reinforcement learning, which proposes a non-trivial sampling to construct scene-generalizable data. On this basis, ReID-R leverages high-quality reward signals to guide the model toward focusing on ID-related cues, achieving accurate reasoning and correct responses. Extensive experiments on multiple ReID benchmarks demonstrate that ReID-R achieves competitive identity discrimination as superior methods using only 14.3K non-trivial data (20.9% of the existing data scale). Furthermore, benefit from inherent reasoning, ReID-R can provide high-quality interpretation for results.", "date": "2026-04-21", "categories": []}
{"id": "2604.19365", "title": "Detection of T-shirt Presentation Attacks in Face Recognition Systems", "abstract": "Face recognition systems are often used for biometric authentication. Nevertheless, it is known that without any protective measures, face recognition systems are vulnerable to presentation attacks. To tackle this security problem, methods for detecting presentation attacks have been developed and shown good detection performance on several benchmark datasets. However, generalising presentation attack detection methods to new and novel types of attacks is an ongoing challenge. In this work, we employ 1,608 T-shirt attacks of the T-shirt Face Presentation Attack (TFPA) database using 100 unique presentation attack instruments together with 152 bona fide presentations. In a comprehensive evaluation, we show that this type of attack can compromise the security of face recognition systems. Furthermore, we propose a detection method based on spatial consistency checks in order to detect said T-shirt attacks. Precisely, state-of-the-art face and person detectors are combined to analyse the spatial positions of detected faces and persons based on which T-shirt attacks can be reliably detected.", "date": "2026-04-21", "categories": []}
{"id": "2604.16341", "title": "Deep Learning for Virtual Reality User Identification: A Benchmark", "abstract": "Virtual Reality (VR) applications require robust user identification systems to ensure secure access to equipment and protect worker identities. Motion tracking data from VR headsets and controllers has emerged as a powerful behavioral biometric, with recent studies demonstrating identification accuracies exceeding 94% across a large user base. However, the application of modern deep learning architectures, particularly State Space Models (SSM), to VR scenarios remains largely unexplored. In this work, we benchmark user identification performance across the large-scale Who is Alyx VR dataset, gathering data from 71 users playing the popular Half-Life:Alyx game. We evaluate both established architectures (Long Short-Term Memory (LSTM), Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN), Temporal Convolutional Network (TCN), Transformer) and the emerging SSMs on time series motion data. Our results provide the first comprehensive benchmark of state-of-the-art and novel architectures for VR user identification, establishing baseline performance metrics for future privacy preserving authentication systems in manufacturing environments.", "date": "2026-03-14", "categories": []}
{"id": "2604.16546", "title": "A B-Spline Function Based 3D Point Cloud Unwrapping Scheme for 3D Fingerprint Recognition and Identification", "abstract": "Three-dimensional (3D) fingerprint recognition and identification offer several advantages over traditional two-dimensional (2D) recognition systems. The contactless nature of 3D fingerprints enhances hygiene and security, reducing the risk of contamination and spoofing. In addition to surface ridge and valley patterns, 3D fingerprints capture depth, curvature, and shape information, enabling the development of more precise and robust authentication systems. Despite recent advancements, significant challenges remain. The topological height of fingerprint pixels complicates the extraction of ridge and valley patterns. Furthermore, registration issues limit the acquisition process, requiring consistent direction and orientation across all samples. To address these challenges, this paper introduces a method that unwraps 3D fingerprints, represented as 3D point clouds, using B-spline curve fitting to mitigate height variation and reduce registration limitations. The unwrapped point cloud is then converted into a grayscale image by mapping the relative heights of the points. This grayscale image is subsequently used for recognition through conventional 2D fingerprint identification methods. The proposed approach demonstrated superior performance in 3D fingerprint recognition, achieving Equal Error Rates (EERs) of 0.2072%, 0.26%, and 0.22% across three experiments, outperforming existing methods. Additionally, the method surpassed 3D fingerprint flattening technique in both recognition and identification during cross-session experiments, achieving an EER of 1.50% when fingerprints with varying registrations were included.", "date": "2026-04-17", "categories": []}
{"id": "2604.17321", "title": "R-FLoRA: Residual-Statistic-Gated Low-Rank Adaptation for Single-Image Face Morphing Attack Detection", "abstract": "Face morphing attacks pose a substantial risk to the reliability of face recognition systems used in passport issuance, border control, and digital identity verification. Detecting morphing attacks from a single facial image remains challenging owing to the lack of a trusted reference and the diversity of attack generation methods. This paper presents a new Single-Image Face Morphing Attack Detection (S-MAD) framework that integrates high-frequency Laplacian residual statistics with representations from a frozen, foundation-scale vision transformer. The approach employs residual-statistic-gated low-rank adapters (R-FLoRA) and feature-wise residual fusion (Res-FiLM) to enhance sensitivity to local morphing artefacts while preserving the semantic context of the backbone. A novel residual-contrastive alignment loss further regularises the fused token space, improving discrimination under unseen morphing conditions. Comprehensive experiments on four ICAO-compliant datasets, encompassing seven morph generation techniques, demonstrate that the proposed method consistently surpasses nine recent state-of-the-art S-MAD algorithms in detection accuracy and cross-domain (or dataset) generalisation. With a frozen backbone and minimal trainable parameters, the model achieves real-time efficiency and interpretability, making it suitable for real-life scenarios in biometric verification systems.", "date": "2026-04-19", "categories": []}
{"id": "2604.18358", "title": "LBFTI: Layer-Based Facial Template Inversion for Identity-Preserving Fine-Grained Face Reconstruction", "abstract": "In face recognition systems, facial templates are widely adopted for identity authentication due to their compliance with the data minimization principle. However, facial template inversion technologies have posed a severe privacy leakage risk by enabling face reconstruction from templates. This paper proposes a Layer-Based Facial Template Inversion (LBFTI) method to reconstruct identity-preserving fine-grained face images. Our scheme decomposes face images into three layers: foreground layers (including eyebrows, eyes, nose, and mouth), midground layers (skin), and background layers (other parts). LBFTI leverages dedicated generators to produce these layers, adopting a rigorous three-stage training strategy: (1) independent refined generation of foreground and midground layers, (2) fusion of foreground and midground layers with template secondary injection to produce complete panoramic face images with background layers, and (3) joint fine-tuning of all modules to optimize inter-layer coordination and identity consistency. Experiments demonstrate that our LBFTI not only outperforms state-of-the-art methods in machine authentication performance, with a 25.3% improvement in TAR, but also achieves better similarity in human perception, as validated by both quantitative metrics and a questionnaire survey.", "date": "2026-04-20", "categories": []}
{"id": "2604.15979", "title": "MMGait: Towards Multi-Modal Gait Recognition", "abstract": "Gait recognition has emerged as a powerful biometric technique for identifying individuals at a distance without requiring user cooperation. Most existing methods focus primarily on RGB-derived modalities, which fall short in real-world scenarios requiring multi-modal collaboration and cross-modal retrieval. To overcome these challenges, we present MMGait, a comprehensive multi-modal gait benchmark integrating data from five heterogeneous sensors, including an RGB camera, a depth camera, an infrared camera, a LiDAR scanner, and a 4D Radar system. MMGait contains twelve modalities and 334,060 sequences from 725 subjects, enabling systematic exploration across geometric, photometric, and motion domains. Based on MMGait, we conduct extensive evaluations on single-modal, cross-modal, and multi-modal paradigms to analyze modality robustness and complementarity. Furthermore, we introduce a new task, Omni Multi-Modal Gait Recognition, which aims to unify the above three gait recognition paradigms within a single model. We also propose a simple yet powerful baseline, OmniGait, which learns a shared embedding space across diverse modalities and achieves promising recognition performance. The MMGait benchmark, codebase, and pretrained checkpoints are publicly available at https://github.com/BNU-IVC/MMGait.", "date": "2026-04-17", "categories": []}
{"id": "2603.24470", "title": "Counting Without Numbers and Finding Without Words", "abstract": "Every year, 10 million pets enter shelters, separated from their families. Despite desperate searches by both guardians and lost animals, 70% never reunite, not because matches do not exist, but because current systems look only at appearance, while animals recognize each other through sound. We ask, why does computer vision treat vocalizing species as silent visual objects? Drawing on five decades of cognitive science showing that animals perceive quantity approximately and communicate identity acoustically, we present the first multimodal reunification system integrating visual and acoustic biometrics. Our species-adaptive architecture processes vocalizations from 10Hz elephant rumbles to 4kHz puppy whines, paired with probabilistic visual matching that tolerates stress-induced appearance changes. This work demonstrates that AI grounded in biological communication principles can serve vulnerable populations that lack human language.", "date": "2026-03-25", "categories": []}
{"id": "2604.14874", "title": "Open-Set Vein Biometric Recognition with Deep Metric Learning", "abstract": "Most state-of-the-art vein recognition methods rely on closed-set classification, which inherently limits their scalability and prevents the adaptive enrollment of new users without complete model retraining. We rigorously evaluate the computational boundaries of Deep Metric Learning (DML) under strict open-set constraints. Unlike standard closed-set approaches, we analyze the impact of data scarcity and domain shift on recognition performance. Our approach learns discriminative L2-normalised embeddings and employs prototype-based matching with a calibrated similarity threshold to effectively distinguish between enrolled users and unseen impostors. We evaluate the framework under a strict subject-disjoint protocol across four diverse datasets covering finger, wrist, and dorsal hand veins (MMCBNU 6000, UTFVP, FYO, and a dorsal hand-vein dataset). On the large-scale MMCBNU 6000 benchmark, our best model (ResNet50-CBAM) achieves an OSCR of 0.9945, AUROC of 0.9974, and EER of 1.57%, maintaining high identification accuracy (99.6% Rank-1) while robustly rejecting unknown subjects. Cross-dataset experiments evaluate the framework's generalisation across different acquisition setups, confirming that while the model handles large-scale data robustly, performance remains sensitive to domain shifts in low-data regimes. Ablation studies demonstrate that triplet-based objectives combined with a simple 1-NN classifier offer an optimal trade-off between accuracy and efficiency, enabling real-time deployment on commodity hardware.", "date": "2026-04-16", "categories": []}
{"id": "2604.15090", "title": "Beyond Visual Cues: Semantic-Driven Token Filtering and Expert Routing for Anytime Person ReID", "abstract": "Any-Time Person Re-identification (AT-ReID) necessitates the robust retrieval of target individuals under arbitrary conditions, encompassing both modality shifts (daytime and nighttime) and extensive clothing-change scenarios, ranging from short-term to long-term intervals. However, existing methods are highly relying on pure visual features, which are prone to change due to environmental and time factors, resulting in significantly performance deterioration under scenarios involving illumination caused modality shifts or cloth-change. In this paper, we propose Semantic-driven Token Filtering and Expert Routing (STFER), a novel framework that leverages the ability of Large Vision-Language Models (LVLMs) to generate identity consistency text, which provides identity-discriminative features that are robust to both clothing variations and cross-modality shifts between RGB and IR. Specifically, we employ instructions to guide the LVLM in generating identity-intrinsic semantic text that captures biometric constants for the semantic model driven. The text token is further used for Semantic-driven Visual Token Filtering (SVTF), which enhances informative visual regions and suppresses redundant background noise. Meanwhile, the text token is also used for Semantic-driven Expert Routing (SER), which integrates the semantic text into expert routing, resulting in more robust multi-scenario gating. Extensive experiments on the Any-Time ReID dataset (AT-USTC) demonstrate that our model achieves state-of-the-art results. Moreover, the model trained on AT-USTC was evaluated across 5 widely-used ReID benchmarks demonstrating superior generalization capabilities with highly competitive results. Our code will be available soon.", "date": "2026-04-16", "categories": []}
{"id": "2403.08462", "title": "Grammar as a Behavioral Biometric: Using Cognitively Motivated Grammar Models for Authorship Verification", "abstract": "Authorship Verification (AV) is a key area of research in digital text forensics, which addresses the fundamental question of whether two texts were written by the same person. Numerous computational approaches have been proposed over the last two decades in an attempt to address this challenge. However, existing AV methods often suffer from high complexity, low explainability and especially from a lack of clear scientific justification. We propose a simpler method based on modeling the grammar of an author following Cognitive Linguistics principles. These models are used to calculate $\\lambda_G$ (LambdaG): the ratio of the likelihoods of a document given the candidate's grammar versus given a reference population's grammar. Our empirical evaluation, conducted on twelve datasets and compared against seven baseline methods, demonstrates that LambdaG achieves superior performance, including against several neural network-based AV methods. LambdaG is also robust to small variations in the composition of the reference population and provides interpretable visualizations, enhancing its explainability. We argue that its effectiveness is due to the method's compatibility with Cognitive Linguistics theories predicting that a person's grammar is a behavioral biometric.", "date": "2024-03-13", "categories": []}
{"id": "2510.17043", "title": "Person Re-Identification via Generalized Class Prototypes", "abstract": "Advanced feature extraction methods have significantly contributed to enhancing the task of person re-identification. In addition, modifications to objective functions have been developed to further improve performance. Nonetheless, selecting better class representatives is an underexplored area of research that can also lead to advancements in re-identification performance. Although past works have experimented with using the centroid of a gallery image class during training, only a few have investigated alternative representations during the retrieval stage. In this paper, we demonstrate that these prior techniques yield suboptimal results in terms of re-identification metrics. To address the re-identification problem, we propose a generalized selection method that involves choosing representations that are not limited to class centroids. Our approach strikes a balance between accuracy and mean average precision, leading to improvements beyond the state of the art. For example, the actual number of representations per class can be adjusted to meet specific application requirements. We apply our methodology on top of multiple re-identification embeddings, and in all cases it substantially improves upon contemporary results.", "date": "2025-10-19", "categories": []}
{"id": "2604.12221", "title": "BarbieGait: An Identity-Consistent Synthetic Human Dataset with Versatile Cloth-Changing for Gait Recognition", "abstract": "Gait recognition, as a reliable biometric technology, has seen rapid development in recent years while it faces significant challenges caused by diverse clothing styles in the real world. This paper introduces BarbieGait, a synthetic gait dataset where real-world subjects are uniquely mapped into a virtual engine to simulate extensive clothing changes while preserving their gait identity information. As a pioneering work, BarbieGait provides a controllable gait data generation method, enabling the production of large datasets to validate cross-clothing issues that are difficult to verify with real-world data. However, the diversity of clothing increases intra-class variance and makes one of the biggest challenges to learning cloth-invariant features under varying clothing conditions. Therefore, we propose GaitCLIF (Gait-oriented CLoth-Invariant Feature) as a robust baseline model for cross-clothing gait recognition. Through extensive experiments, we validate that our method significantly improves cross-clothing performance on BarbieGait and the existing popular gait benchmarks. We believe that BarbieGait, with its extensive cross-clothing gait data, will further advance the capabilities of gait recognition in cross-clothing scenarios and promote progress in related research.", "date": "2026-04-14", "categories": []}
{"id": "2604.12834", "title": "Rapid LoRA Aggregation for Wireless Channel Adaptation in Open-Set Radio Frequency Fingerprinting", "abstract": "Radio frequency fingerprints (RFFs) enable secure wireless authentication but struggle in open-set scenarios with unknown devices and varying channels. Existing methods face challenges in generalization and incur high computational costs. We propose a lightweight, self-adaptive RFF extraction framework using Low-Rank Adaptation (LoRA). By pretraining LoRA modules per environment, our method enables fast adaptation to unseen channel conditions without full retraining. During inference, a weighted combination of LoRAs dynamically enhances feature extraction. Experimental results demonstrate a 15% reduction in equal error rate (EER) compared to non-finetuned baselines and an 83% decrease in training time relative to full fine-tuning, using the same training dataset. This approach provides a scalable and efficient solution for open-set RFF authentication in dynamic wireless vehicular networks.", "date": "2026-04-14", "categories": []}
{"id": "2602.18047", "title": "CityGuard: Graph-Aware Private Descriptors for Bias-Resilient Identity Search Across Urban Cameras", "abstract": "City-scale person re-identification across distributed cameras must handle severe appearance changes from viewpoint, occlusion, and domain shift while complying with data protection rules that prevent sharing raw imagery. We introduce CityGuard, a topology-aware transformer for privacy-preserving identity retrieval in decentralized surveillance. The framework integrates three components. A dispersion-adaptive metric learner adjusts instance-level margins according to feature spread, increasing intra-class compactness. Spatially conditioned attention injects coarse geometry, such as GPS or deployment floor plans, into graph-based self-attention to enable projectively consistent cross-view alignment using only coarse geometric priors without requiring survey-grade calibration. Differentially private embedding maps are coupled with compact approximate indexes to support secure and cost-efficient deployment. Together these designs produce descriptors robust to viewpoint variation, occlusion, and domain shifts, and they enable a tunable balance between privacy and utility under rigorous differential-privacy accounting. Experiments on Market-1501 and additional public benchmarks, complemented by database-scale retrieval studies, show consistent gains in retrieval precision and query throughput over strong baselines, confirming the practicality of the framework for privacy-critical urban identity matching.", "date": "2026-02-20", "categories": []}
{"id": "2604.09989", "title": "FlowPalm: Optical Flow Driven Non-Rigid Deformation for Geometrically Diverse Palmprint Generation", "abstract": "Recently, synthetic palmprints have been increasingly used as substitutes for real data to train recognition models. To be effective, such synthetic data must reflect the diversity of real palmprints, including both style variation and geometric variation. However, existing palmprint generation methods mainly focus on style translation, while geometric variation is either ignored or approximated by simple handcrafted augmentations. In this work, we propose FlowPalm, an optical-flow-driven palmprint generation framework capable of simulating the complex non-rigid deformations observed in real palms. Specifically, FlowPalm estimates optical flows between real palmprint pairs to capture the statistical patterns of geometric deformations. Building on these priors, we design a progressive sampling process that gradually introduces the geometric deformations during diffusion while maintaining identity consistency. Extensive experiments on six benchmark datasets demonstrate that FlowPalm significantly outperforms state-of-the-art palmprint generation approaches in downstream recognition tasks. Project page: https://yuchenzou.github.io/FlowPalm/", "date": "2026-04-11", "categories": []}
{"id": "2604.09990", "title": "Gait Recognition with Temporal Kolmogorov-Arnold Networks", "abstract": "Gait recognition is a biometric modality that identifies individuals from their characteristic walking patterns. Unlike conventional biometric traits, gait can be acquired at a distance and without active subject cooperation, making it suitable for surveillance and public safety applications. Nevertheless, silhouette-based temporal models remain sensitive to long sequences, observation noise, and appearance-related covariates. Recurrent architectures often struggle to preserve information from earlier frames and are inherently sequential to optimize, whereas transformer-based models typically require greater computational resources and larger training sets and may be sensitive to irregular sequence lengths and noisy inputs. These limitations reduce robustness under clothing variation, carrying conditions, and view changes, while also hindering the joint modeling of local gait cycles and longer-term motion trends. To address these challenges, we introduce a Temporal Kolmogorov-Arnold Network (TKAN) for gait recognition. The proposed model replaces fixed edge weights with learnable one-dimensional functions and incorporates a two-level memory mechanism consisting of short-term RKAN sublayers and a gated long-term pathway. This design enables efficient modeling of both cycle-level dynamics and broader temporal context while maintaining a compact backbone. Experiments on the CASIA-B dataset indicate that the proposed CNN+TKAN framework achieves strong recognition performance under the reported evaluation setting.", "date": "2026-04-11", "categories": []}
{"id": "2604.10040", "title": "Intra-finger Variability of Diffusion-based Latent Fingerprint Generation", "abstract": "The primary goal of this work is to systematically evaluate the intra-finger variability of synthetic fingerprints (particularly latent prints) generated using a state-of-the-art diffusion model. Specifically, we focus on enhancing the latent style diversity of the generative model by constructing a comprehensive \\textit{latent style bank} curated from seven diverse datasets, which enables the precise synthesis of latent prints with over 40 distinct styles encapsulating different surfaces and processing techniques. We also implement a semi-automated framework to understand the integrity of fingerprint ridges and minutiae in the generated impressions. Our analysis indicates that though the generation process largely preserves the identity, a small number of local inconsistencies (addition and removal of minutiae) are introduced, especially when there are poor quality regions in the reference image. Furthermore, mismatch between the reference image and the chosen style embedding that guides the generation process introduces global inconsistencies in the form of hallucinated ridge patterns. These insights highlight the limitations of existing synthetic fingerprint generators and the need to further improve these models to simultaneously enhance both diversity and identity consistency.", "date": "2026-04-11", "categories": []}
{"id": "2604.11250", "title": "Variational Latent Entropy Estimation Disentanglement: Controlled Attribute Leakage for Face Recognition", "abstract": "Face recognition embeddings encode identity, but they also encode other factors such as gender and ethnicity. Depending on how these factors are used by a downstream system, separating them from the information needed for verification is important for both privacy and fairness. We propose Variational Latent Entropy Estimation Disentanglement (VLEED), a post-hoc method that transforms pretrained embeddings with a variational autoencoder and encourages a distilled representation where the categorical variable of interest is separated from identity-relevant information. VLEED uses a mutual information-based objective realised through the estimation of the entropy of the categorical attribute in the latent space, and provides stable training with fine-grained control over information removal. We evaluate our method on IJB-C, RFW, and VGGFace2 for gender and ethnicity disentanglement, and compare it to various state-of-the-art methods. We report verification utility, predictability of the disentangled variable under linear and nonlinear classifiers, and group disparity metrics based on false match rates. Our results show that VLEED offers a wide range of privacy-utility tradeoffs over existing methods and can also reduce recognition bias across demographic groups.", "date": "2026-04-13", "categories": []}
{"id": "2402.02540", "title": "Embedding Non-Distortive Cancelable Face Template Generation", "abstract": "Biometric authentication systems are crucial for security, but developing them involves various complexities, including privacy, security, and achieving high accuracy without directly storing pure biometric data in storage. We introduce an innovative image distortion technique that makes facial images unrecognizable to the eye but still identifiable by any custom embedding neural network model. Using the proposed approach, we test the reliability of biometric recognition networks by determining the maximum image distortion that does not change the predicted identity. Through experiments on MNIST and LFW datasets, we assess its effectiveness and compare it based on the traditional comparison metrics.", "date": "2024-02-04", "categories": []}
{"id": "2404.02696", "title": "Deep Privacy Funnel Model: From a Discriminative to a Generative Approach with an Application to Face Recognition", "abstract": "In this study, we apply the information-theoretic Privacy Funnel (PF) model to face recognition and develop a method for privacy-preserving representation learning within an end-to-end trainable framework. Our approach addresses the trade-off between utility and obfuscation of sensitive information under logarithmic loss. We study the integration of information-theoretic privacy principles with representation learning, with a particular focus on face recognition systems. We also highlight the compatibility of the proposed framework with modern face recognition networks such as AdaFace and ArcFace. In addition, we introduce the Generative Privacy Funnel ($\\mathsf{GenPF}$) model, which extends the traditional discriminative PF formulation, referred to here as the Discriminative Privacy Funnel ($\\mathsf{DisPF}$). The proposed $\\mathsf{GenPF}$ model extends the privacy-funnel framework to generative formulations under information-theoretic and estimation-theoretic criteria. Complementing these developments, we present the deep variational PF (DVPF) model, which yields a tractable variational bound for measuring information leakage and enables optimization in deep representation-learning settings. The DVPF framework, associated with both the $\\mathsf{DisPF}$ and $\\mathsf{GenPF}$ models, also clarifies connections with generative models such as variational autoencoders (VAEs), generative adversarial networks (GANs), and diffusion models. Finally, we validate the framework on modern face recognition systems and show that it provides a controllable privacy--utility trade-off while substantially reducing leakage about sensitive attributes. To support reproducibility, we also release a PyTorch implementation of the proposed framework.", "date": "2024-04-03", "categories": []}
{"id": "2512.03745", "title": "Dual-level Modality Debiasing Learning for Unsupervised Visible-Infrared Person Re-Identification", "abstract": "Two-stage learning pipeline has achieved promising results in unsupervised visible-infrared person re-identification (USL-VI-ReID). It first performs single-modality learning and then operates cross-modality learning to tackle the modality discrepancy. Although promising, this pipeline inevitably introduces modality bias: modality-specific cues learned in the single-modality training naturally propagate into the following cross-modality learning, impairing identity discrimination and generalization. To address this issue, we propose a Dual-level Modality Debiasing Learning (DMDL) framework that implements debiasing at both the model and optimization levels. At the model level, we propose a Causality-inspired Adjustment Intervention (CAI) module that replaces likelihood-based modeling with causal modeling, preventing modality-induced spurious patterns from being introduced, leading to a low-biased model. At the optimization level, a Collaborative Bias-free Training (CBT) strategy is introduced to interrupt the propagation of modality bias across data, labels, and features by integrating modality-specific augmentation, label refinement, and feature alignment. Extensive experiments on benchmark datasets demonstrate that DMDL could enable modality-invariant feature learning and a more generalized model. The code is available at https://github.com/priester3/DMDL.", "date": "2025-12-03", "categories": []}
{"id": "2604.07740", "title": "Beyond Pedestrians: Caption-Guided CLIP Framework for High-Difficulty Video-based Person Re-Identification", "abstract": "In recent years, video-based person Re-Identification (ReID) has gained attention for its ability to leverage spatiotemporal cues to match individuals across non-overlapping cameras. However, current methods struggle with high-difficulty scenarios, such as sports and dance performances, where multiple individuals wear similar clothing while performing dynamic movements. To overcome these challenges, we propose CG-CLIP, a novel caption-guided CLIP framework that leverages explicit textual descriptions and learnable tokens. Our method introduces two key components: Caption-guided Memory Refinement (CMR) and Token-based Feature Extraction (TFE). CMR utilizes captions generated by Multi-modal Large Language Models (MLLMs) to refine identity-specific features, capturing fine-grained details. TFE employs a cross-attention mechanism with fixed-length learnable tokens to efficiently aggregate spatiotemporal features, reducing computational overhead. We evaluate our approach on two standard datasets (MARS and iLIDS-VID) and two newly constructed high-difficulty datasets (SportsVReID and DanceVReID). Experimental results demonstrate that our method outperforms current state-of-the-art approaches, achieving significant improvements across all benchmarks.", "date": "2026-04-09", "categories": []}
{"id": "2604.06782", "title": "EventFace: Event-Based Face Recognition via Structure-Driven Spatiotemporal Modeling", "abstract": "Event cameras offer a promising sensing modality for face recognition due to their inherent advantages in illumination robustness and privacy-friendliness. However, because event streams lack the stable photometric appearance relied upon by conventional RGB-based face recognition systems, we argue that event-based face recognition should model structure-driven spatiotemporal identity representations shaped by rigid facial motion and individual facial geometry. Since dedicated datasets for event-based face recognition remain lacking, we construct EFace, a small-scale event-based face dataset captured under rigid facial motion. To learn effectively from this limited event data, we further propose EventFace, a framework for event-based face recognition that integrates spatial structure and temporal dynamics for identity modeling. Specifically, we employ Low-Rank Adaptation (LoRA) to transfer structural facial priors from pretrained RGB face models to the event domain, thereby establishing a reliable spatial basis for identity modeling. Building on this foundation, we further introduce a Motion Prompt Encoder (MPE) to explicitly encode temporal features and a Spatiotemporal Modulator (STM) to fuse them with spatial features, thereby enhancing the representation of identity-relevant event patterns. Extensive experiments demonstrate that EventFace achieves the best performance among the evaluated baselines, with a Rank-1 identification rate of 94.19% and an equal error rate (EER) of 5.35%. Results further indicate that EventFace exhibits stronger robustness under degraded illumination than the competing methods. In addition, the learned representations exhibit reduced template reconstructability.", "date": "2026-04-08", "categories": []}
{"id": "2604.06987", "title": "CAAP: Capture-Aware Adversarial Patch Attacks on Palmprint Recognition Models", "abstract": "Palmprint recognition is deployed in security-critical applications, including access control and palm-based payment, due to its contactless acquisition and highly discriminative ridge-and-crease textures. However, the robustness of deep palmprint recognition systems against physically realizable attacks remains insufficiently understood. Existing studies are largely confined to the digital setting and do not adequately account for the texture-dominant nature of palmprint recognition or the distortions introduced during physical acquisition. To address this gap, we propose CAAP, a capture-aware adversarial patch framework for palmprint recognition. CAAP learns a universal patch that can be reused across inputs while remaining effective under realistic acquisition variation. To match the structural characteristics of palmprints, the framework adopts a cross-shaped patch topology, which enlarges spatial coverage under a fixed pixel budget and more effectively disrupts long-range texture continuity. CAAP further integrates three modules: ASIT for input-conditioned patch rendering, RaS for stochastic capture-aware simulation, and MS-DIFE for feature-level identity-disruptive guidance. We evaluate CAAP on the Tongji, IITD, and AISEC datasets against generic CNN backbones and palmprint-specific recognition models. Experiments show that CAAP achieves strong untargeted and targeted attack performance with favorable cross-model and cross-dataset transferability. The results further show that, although adversarial training can partially reduce the attack success rate, substantial residual vulnerability remains. These findings indicate that deep palmprint recognition systems remain vulnerable to physically realizable, capture-aware adversarial patch attacks, underscoring the need for more effective defenses in practice. Code available at https://github.com/ryliu68/CAAP.", "date": "2026-04-08", "categories": []}
{"id": "2604.07282", "title": "Are Face Embeddings Compatible Across Deep Neural Network Models?", "abstract": "Automated face recognition has made rapid strides over the past decade due to the unprecedented rise of deep neural network (DNN) models that can be trained for domain-specific tasks. At the same time, foundation models that are pretrained on broad vision or vision-language tasks have shown impressive generalization across diverse domains, including biometrics. This raises an important question: Do different DNN models--both domain-specific and foundation models--encode facial identity in similar ways, despite being trained on different datasets, loss functions, and architectures? In this regard, we directly analyze the geometric structure of embedding spaces imputed by different DNN models. Treating embeddings of face images as point clouds, we study whether simple affine transformations can align face representations of one model with another. Our findings reveal surprising cross-model compatibility: low-capacity linear mappings substantially improve cross-model face recognition over unaligned baselines for both face identification and verification tasks. Alignment patterns generalize across datasets and vary systematically across model families, indicating representational convergence in facial identity encoding. These findings have implications for model interoperability, ensemble design, and biometric template security.", "date": "2026-04-08", "categories": []}
{"id": "2604.05877", "title": "Automatic dental superimposition of 3D intraorals and 2D photographs for human identification", "abstract": "Dental comparison is considered a primary identification method, at the level of fingerprints and DNA profiling. One crucial but time-consuming step of this method is the morphological comparison. One of the main challenges to apply this method is the lack of ante-mortem medical records, specially on scenarios such as migrant death at the border and/or in countries where there is no universal healthcare. The availability of photos on social media where teeth are visible has led many odontologists to consider morphological comparison using them. However, state-of-the-art proposals have significant limitations, including the lack of proper modeling of perspective distortion and the absence of objective approaches that quantify morphological differences.\n  Our proposal involves a 3D (post-mortem scan) - 2D (ante-mortem photos) approach. Using computer vision and optimization techniques, we replicate the ante-mortem image with the 3D model to perform the morphological comparison. Two automatic approaches have been developed: i) using paired landmarks and ii) using a segmentation of the teeth region to estimate camera parameters. Both are capable of obtaining very promising results over 20,164 cross comparisons from 142 samples, obtaining mean ranking values of 1.6 and 1.5, respectively. These results clearly outperform filtering capabilities of automatic dental chart comparison approaches, while providing an automatic, objective and quantitative score of the morphological correspondence, easily to interpret and analyze by visualizing superimposed images.", "date": "2026-04-07", "categories": []}
{"id": "2406.16042", "title": "Pose-dIVE: Pose-Diversified Augmentation with Diffusion Model for Person Re-Identification", "abstract": "Person re-identification (Re-ID) often faces challenges due to variations in human poses and camera viewpoints, which significantly affect the appearance of individuals across images. Existing datasets frequently lack diversity and scalability in these aspects, hindering the generalization of Re-ID models to new camera systems or environments. To overcome this, we propose Pose-dIVE, a novel data augmentation approach that incorporates sparse and underrepresented human pose and camera viewpoint examples into the training data, addressing the limited diversity in the original training data distribution. Our objective is to augment the training dataset to enable existing Re-ID models to learn features unbiased by human pose and camera viewpoint variations. By conditioning the diffusion model on both the human pose and camera viewpoint through the SMPL model, our framework generates augmented training data with diverse human poses and camera viewpoints. Experimental results demonstrate the effectiveness of our method in addressing human pose bias and enhancing the generalizability of Re-ID models compared to other data augmentation-based Re-ID approaches.", "date": "2024-06-23", "categories": []}
{"id": "2601.02318", "title": "Fusion2Print: Deep Flash-Non-Flash Fusion for Contactless Fingerprint Matching", "abstract": "Contactless fingerprint recognition offers a hygienic and convenient alternative to contact-based systems, enabling rapid acquisition without latent prints, pressure artifacts, or hygiene risks. However, contactless images often show degraded ridge clarity due to illumination variation, subcutaneous skin discoloration, and specular reflections. Flash captures preserve ridge detail but introduce noise, whereas non-flash captures reduce noise but lower ridge contrast. We propose Fusion2Print (F2P), the first framework to systematically capture and fuse paired flash-non-flash contactless fingerprints. We construct a custom paired dataset, FNF Database, and perform manual flash-non-flash subtraction to isolate ridge-preserving signals. A lightweight attention-based fusion network also integrates both modalities, emphasizing informative channels and suppressing noise, and then a U-Net enhancement module produces an optimally weighted grayscale image. Finally, a deep embedding model with cross-domain compatibility, generates discriminative and robust representations in a unified embedding space compatible with both contactless and contact-based fingerprints for verification. F2P enhances ridge clarity and achieves superior recognition performance (AUC=0.999, EER=1.12%) over single-capture baselines (Verifinger, DeepPrint).", "date": "2026-01-05", "categories": []}
{"id": "2604.04183", "title": "Scale-Aware Vision-Language Adaptation for Extreme Far-Distance Video Person Re-identification", "abstract": "Extreme far-distance video person re-identification (ReID) is particularly challenging due to scale compression, resolution degradation, motion blur, and aerial-ground viewpoint mismatch. As camera altitude and subject distance increase, models trained on close-range imagery degrade significantly. In this work, we investigate how large-scale vision-language models can be adapted to operate reliably under these conditions. Starting from a CLIP-based baseline, we upgrade the visual backbone from ViT-B/16 to ViT-L/14 and introduce backbone-aware selective fine-tuning to stabilize adaptation of the larger transformer. To address noisy and low-resolution tracklets, we incorporate a lightweight temporal attention pooling mechanism that suppresses degraded frames and emphasizes informative observations. We retain adapter-based and prompt-conditioned cross-view learning to mitigate aerial-ground domain shifts, and further refine retrieval using improved optimization and k-reciprocal re-ranking. Experiments on the DetReIDX stress-test benchmark show that our approach achieves mAP scores of 46.69 (A2G), 41.23 (G2A), and 22.98 (A2A), corresponding to an overall mAP of 35.73. These results show that large-scale vision-language backbones, when combined with stability-focused adaptation, significantly enhance robustness in extreme far-distance video person ReID.", "date": "2026-04-05", "categories": []}
{"id": "2604.04485", "title": "ECG Biometrics with ArcFace-Inception: External Validation on MIMIC and HEEDB", "abstract": "ECG biometrics has been studied mainly on small cohorts and short inter-session intervals, leaving open how identification behaves under large galleries, external domain shift, and multi-year temporal gaps. We evaluated a 1D\n  Inception-v1 model trained with ArcFace on an internal clinical corpus of 164,440 12-lead ECGs from 53,079 patients and tested it on larger cohorts derived from MIMIC-IV-ECG and HEEDB. The study used a unified closed-set leave-one-out\n  protocol with Rank@K and TAR@FAR metrics, together with scale, temporal-stress, reranking, and confidence analyses. Under general comparability, the system achieved Rank@1 of 0.9506 on ASUGI-DB, 0.8291 on MIMIC-GC, and 0.6884 on\n  HEEDB-GC. In the temporal stress test at constant gallery size, Rank@1 declined from 0.7853 to 0.6433 on MIMIC and from 0.6864 to 0.5560 on HEEDB from 1 to 5 years. Scale analysis on HEEDB showed monotonic degradation as gallery size\n  increased and recovery as more examinations per patient became available. On HEEDB-RR, post-hoc reranking further improved retrieval, with AS-norm reaching Rank@1 = 0.8005 from a 0.7765 baseline. ECG identity information therefore\n  remains measurable under externally validated large-scale closed-set conditions, but its operational quality is strongly affected by domain heterogeneity, longitudinal drift, gallery size, and second-stage score processing.", "date": "2026-04-06", "categories": []}
{"id": "2604.02808", "title": "CMCC-ReID: Cross-Modality Clothing-Change Person Re-Identification", "abstract": "Person Re-Identification (ReID) faces severe challenges from modality discrepancy and clothing variation in long-term surveillance scenario. While existing studies have made significant progress in either Visible-Infrared ReID (VI-ReID) or Clothing-Change ReID (CC-ReID), real-world surveillance system often face both challenges simultaneously. To address this overlooked yet realistic problem, we define a new task, termed Cross-Modality Clothing-Change Re-Identification (CMCC-ReID), which targets pedestrian matching across variations in both modality and clothing. To advance research in this direction, we construct a new benchmark SYSU-CMCC, where each identity is captured in both visible and infrared domains with distinct outfits, reflecting the dual heterogeneity of long-term surveillance. To tackle CMCC-ReID, we propose a Progressive Identity Alignment Network (PIA) that progressively mitigates the issues of clothing variation and modality discrepancy. Specifically, a Dual-Branch Disentangling Learning (DBDL) module separates identity-related cues from clothing-related factors to achieve clothing-agnostic representation, and a Bi-Directional Prototype Learning (BPL) module performs intra-modality and inter-modality contrast in the embedding space to bridge the modality gap while further suppressing clothing interference. Extensive experiments on the SYSU-CMCC dataset demonstrate that PIA establishes a strong baseline for this new task and significantly outperforms existing methods.", "date": "2026-04-03", "categories": []}
{"id": "2509.10969", "title": "Gaze Authentication: Factors Influencing Authentication Performance", "abstract": "This paper examines the key factors that influence the performance of state-of-the-art gaze-based authentication. Experiments were conducted on a large-scale, in-house dataset comprising 8,849 subjects collected with Meta Quest Pro equivalent hardware running a video oculography-driven gaze estimation pipeline at 72~Hz. State of the neural network architecture was employed to study the influence of the following factors on authentication performance: eye tracking signal quality, various aspects of eye tracking calibration, and simple filtering on estimated raw gaze. This report provides performance results and their analysis.", "date": "2025-09-13", "categories": []}
{"id": "2603.21482", "title": "ALADIN:Attribute-Language Distillation Network for Person Re-Identification", "abstract": "Recent vision-language models such as CLIP provide strong cross-modal alignment, but current CLIP-guided ReID pipelines rely on global features and fixed prompts. This limits their ability to capture fine-grained attribute cues and adapt to diverse appearances. We propose ALADIN, an attribute-language distillation network that distills knowledge from a frozen CLIP teacher to a lightweight ReID student. ALADIN introduces fine-grained attribute-local alignment to establish adaptive text-visual correspondence and robust representation learning. A Scene-Aware Prompt Generator produces image-specific soft prompts to facilitate adaptive alignment. Attribute-local distillation enforces consistency between textual attributes and local visual features, significantly enhancing robustness under occlusions. Furthermore, we employ cross-modal contrastive and relation distillation to preserve the inherent structural relationships among attributes. To provide precise supervision, we leverage Multimodal LLMs to generate structured attribute descriptions, which are then converted into localized attention maps via CLIP. At inference, only the student is used. Experiments on Market-1501, DukeMTMC-reID, and MSMT17 show improvements over CNN-, Transformer-, and CLIP-based methods, with better generalization and interpretability.", "date": "2026-03-23", "categories": []}
{"id": "2603.29422", "title": "Multimodal Models Meet Presentation Attack Detection on ID Documents", "abstract": "The integration of multimodal models into Presentation Attack Detection (PAD) for ID Documents represents a significant advancement in biometric security. Traditional PAD systems rely solely on visual features, which often fail to detect sophisticated spoofing attacks. This study explores the combination of visual and textual modalities by utilizing pre-trained multimodal models, such as Paligemma, Llava, and Qwen, to enhance the detection of presentation attacks on ID Documents. This approach merges deep visual embeddings with contextual metadata (e.g., document type, issuer, and date). However, experimental results indicate that these models struggle to accurately detect PAD on ID Documents.", "date": "2026-03-31", "categories": []}
{"id": "2506.23555", "title": "LH2Face: Loss function for Hard High-quality Face", "abstract": "In current practical face authentication systems, most face recognition (FR) algorithms are based on cosine similarity with softmax classification. Despite its reliable classification performance, this method struggles with hard samples. A popular strategy to improve FR performance is incorporating angular or cosine margins. However, it does not take face quality or recognition hardness into account, simply increasing the margin value and thus causing an overly uniform training strategy. To address this problem, a novel loss function is proposed, named Loss function for Hard High-quality Face (LH2Face). Firstly, a similarity measure based on the von Mises-Fisher (vMF) distribution is stated, specifically focusing on the logarithm of the Probability Density Function (PDF), which represents the distance between a probability distribution and a vector. Then, an adaptive margin-based multi-classification method using softmax, called the Uncertainty-Aware Margin Function, is implemented in the article. Furthermore, proxy-based loss functions are used to apply extra constraints between the proxy and sample to optimize their representation space distribution. Finally, a renderer is constructed that optimizes FR through face reconstruction and vice versa. Our LH2Face is superior to similiar schemes on hard high-quality face datasets, achieving 49.39% accuracy on the IJB-B dataset, which surpasses the second-place method by 2.37%.", "date": "2025-06-30", "categories": []}
{"id": "2602.02914", "title": "FaceLinkGen: Rethinking Identity Leakage in Privacy-Preserving Face Recognition with Identity Extraction", "abstract": "Transformation-based privacy-preserving face recognition (PPFR) aims to verify identities while hiding facial data from attackers and malicious service providers. Existing evaluations mostly treat privacy as resistance to pixel-level reconstruction, measured by PSNR and SSIM. We show that this reconstruction-centric view fails. We present FaceLinkGen, an identity extraction attack that performs linkage/matching and face regeneration directly from protected templates without recovering original pixels. On three recent PPFR systems, FaceLinkGen reaches over 98.5\\% matching accuracy and above 96\\% regeneration success, and still exceeds 92\\% matching and 94\\% regeneration in a near zero knowledge setting. These results expose a structural gap between pixel distortion metrics, which are widely used in PPFR evaluation, and real privacy. We show that visual obfuscation leaves identity information broadly exposed to both external intruders and untrusted service providers.", "date": "2026-02-02", "categories": []}
{"id": "2603.26890", "title": "Privacy-Preserving Iris Recognition: Performance Challenges and Outlook", "abstract": "Iris-based biometric identification is increasingly recognized for its significant accuracy and long-term stability compared to other biometric modalities such as fingerprints or facial features. However, all biometric modalities are highly sensitive data that raise serious privacy and security concerns, particularly in decentralized and untrusted environments. While Fully Homomorphic Encryption (FHE) has emerged as a promising solution for protecting sensitive data during computation, existing privacy-preserving iris recognition systems face significant performance limitations that hinder their practical deployment. This paper investigates the performance challenges of the current landscape of privacy-preserving iris recognition systems using FHE. Based on these insights, we outline a scalable privacy-preserving framework that aligns with all the requirements specified in the ISO/IEC 24745 standard. Leveraging the Open Iris library, our approach starts with robust iris segmentation, followed by normalization and feature extraction using Gabor filters to generate iris codes. We then apply binary masking to filter out unreliable regions and perform matching using Hamming distance on encrypted iris codes. The accuracy and performance of our proposed privacy-preserving framework is evaluated on the CASIA-Iris-Thousand dataset. Results show that our privacy-preserving framework yields very similar accuracy to the cleartext equivalent, but a much higher computational overhead with respect to pairwise iris template comparisons, of $\\sim 120\\,000 \\times$. This points towards the need for the deployment of two-level schemes in the context of scalable $1-N$ template comparisons.", "date": "2026-03-27", "categories": []}
{"id": "2603.26908", "title": "FusionAgent: A Multimodal Agent with Dynamic Model Selection for Human Recognition", "abstract": "Model fusion is a key strategy for robust recognition in unconstrained scenarios, as different models provide complementary strengths. This is especially important for whole-body human recognition, where biometric cues such as face, gait, and body shape vary across samples and are typically integrated via score-fusion. However, existing score-fusion strategies are usually static, invoking all models for every test sample regardless of sample quality or modality reliability. To overcome these limitations, we propose \\textbf{FusionAgent}, a novel agentic framework that leverages a Multimodal Large Language Model (MLLM) to perform dynamic, sample-specific model selection. Each expert model is treated as a tool, and through Reinforcement Fine-Tuning (RFT) with a metric-based reward, the agent learns to adaptively determine the optimal model combination for each test input. To address the model score misalignment and embedding heterogeneity, we introduce Anchor-based Confidence Top-k (ACT) score-fusion, which anchors on the most confident model and integrates complementary predictions in a confidence-aware manner. Extensive experiments on multiple whole-body biometric benchmarks demonstrate that FusionAgent significantly outperforms SoTA methods while achieving higher efficiency through fewer model invocations, underscoring the critical role of dynamic, explainable, and robust model fusion in real-world recognition systems. Project page: \\href{https://fusionagent.github.io/}{FusionAgent}.", "date": "2026-03-27", "categories": []}
{"id": "2603.26934", "title": "Leveraging Avatar Fingerprinting: A Multi-Generator Photorealistic Talking-Head Public Database and Benchmark", "abstract": "Recent advances in photorealistic avatar generation have enabled highly realistic talking-head avatars, raising security concerns regarding identity impersonation in AI-mediated communication. To advance in this challenging problem, the task of avatar fingerprinting aims to determine whether two avatar videos are driven by the same human operator or not. However, current public databases in the literature are scarce and based solely on old-fashioned talking-head avatar generators, not representing realistic scenarios for the current task of avatar fingerprinting. To overcome this situation, the present article introduces AVAPrintDB, a new publicly available multi-generator talking-head avatar database for avatar fingerprinting. AVAPrintDB is constructed from two audiovisual corpora and three state-of-the-art avatar generators (GAGAvatar, LivePortrait, HunyuanPortrait), representing different synthesis paradigms, and includes both self- and cross-reenactments to simulate legitimate usage and impersonation scenarios.\n  Building on this database, we also define a standardized and reproducible benchmark for avatar fingerprinting, considering public state-of-the-art avatar fingerprinting systems and exploring novel methods based on Foundation Models (DINOv2 and CLIP). Also, we conduct a comprehensive analysis under generator and dataset shift. Our results show that, while identity-related motion cues persist across synthetic avatars, current avatar fingerprinting systems remain highly sensitive to changes in the synthesis pipeline and source domain. The AVAPrintDB, benchmark protocols, and avatar fingerprinting systems are publicly available to facilitate reproducible research.", "date": "2026-03-27", "categories": []}
{"id": "2603.26976", "title": "Beyond Mortality: Advancements in Post-Mortem Iris Recognition through Data Collection and Computer-Aided Forensic Examination", "abstract": "Post-mortem iris recognition brings both hope to the forensic community (a short-term but accurate and fast means of verifying identity) as well as concerns to society (its potential illicit use in post-mortem impersonation). These hopes and concerns have grown along with the volume of research in post-mortem iris recognition. Barriers to further progress in post-mortem iris recognition include the difficult nature of data collection, and the resulting small number of approaches designed specifically for comparing iris images of deceased subjects. This paper makes several unique contributions to mitigate these barriers. First, we have collected and we offer a new dataset of NIR (compliant with ISO/IEC 19794-6 where possible) and visible-light iris images collected after demise from 259 subjects, with the largest PMI (post-mortem interval) being 1,674 hours. For one subject, the data has been collected before and after death, the first such case ever published. Second, the collected dataset was combined with publicly-available post-mortem samples to assess the current state of the art in automatic forensic iris recognition with five iris recognition methods and data originating from 338 deceased subjects. These experiments include analyses of how selected demographic factors influence recognition performance. Thirdly, this study implements a model for detecting post-mortem iris images, which can be considered as presentation attacks. Finally, we offer an open-source forensic tool integrating three post-mortem iris recognition methods with explainability elements added to make the comparison process more human-interpretable.", "date": "2026-03-27", "categories": []}
{"id": "2603.28322", "title": "SFDemorpher: Generalizable Face Demorphing for Operational Morphing Attack Detection", "abstract": "Face morphing attacks compromise biometric security by creating document images that verify against multiple identities, posing significant risks from document issuance to border control. Differential Morphing Attack Detection (D-MAD) offers an effective countermeasure, particularly when employing face demorphing to disentangle identities blended in the morph. However, existing methods lack operational generalizability due to limited training data and the assumption that all document inputs are morphs. This paper presents SFDemorpher, a framework designed for the operational deployment of face demorphing for D-MAD that performs identity disentanglement within joint StyleGAN latent and high-dimensional feature spaces. We introduce a dual-pass training strategy handling both morphed and bona fide documents, leveraging a hybrid corpus with predominantly synthetic identities to enhance robustness against unseen distributions. Extensive evaluation confirms state-of-the-art generalizability across unseen identities, diverse capture conditions, and 13 morphing techniques, spanning both border verification and the challenging document enrollment stage. Our framework achieves superior D-MAD performance by widening the margin between the score distributions of bona fide and morphed samples while providing high-fidelity visual reconstructions facilitating explainability.", "date": "2026-03-30", "categories": []}
{"id": "2603.16629", "title": "MLLM-based Textual Explanations for Face Comparison", "abstract": "Multimodal Large Language Models (MLLMs) have recently been proposed as a means to generate natural-language explanations for face recognition decisions. While such explanations facilitate human interpretability, their reliability on unconstrained face images remains underexplored. In this work, we systematically analyze MLLM-generated explanations for the unconstrained face verification task on the challenging IJB-S dataset, with a particular focus on extreme pose variation and surveillance imagery. Our results show that even when MLLMs produce correct verification decisions, the accompanying explanations frequently rely on non-verifiable or hallucinated facial attributes that are not supported by visual evidence. We further study the effect of incorporating information from traditional face recognition systems, viz., scores and decisions, alongside the input images. Although such information improves categorical verification performance, it does not consistently lead to faithful explanations. To evaluate the explanations beyond decision accuracy, we introduce a likelihood-ratio-based framework that measures the evidential strength of textual explanations. Our findings highlight fundamental limitations of current MLLMs for explainable face recognition and underscore the need for a principled evaluation of reliable and trustworthy explanations in biometric applications. Code is available at https://github.com/redwankarimsony/LR-MLLMFR-Explainability.", "date": "2026-03-17", "categories": []}
{"id": "2603.23003", "title": "On the use of Aggregation Operators to improve Human Identification using Dental Records", "abstract": "The comparison of dental records is a standardized technique in forensic dentistry used to speed up the identification of individuals in multiple-comparison scenarios. Specifically, the odontogram comparison is a procedure to compute criteria that will be used to perform a ranking. State-of-the-art automatic methods either make use of simple techniques, without utilizing the full potential of the information obtained from a comparison, or their internal behavior is not known due to the lack of peer-reviewed publications. This work aims to design aggregation mechanisms to automatically compare pairs of dental records that can be understood and validated by experts, improving the current methods. To do so, we introduce different aggregation approaches using the state-of-the-art codification, based on seven different criteria. In particular, we study the performance of i) data-driven lexicographical order-based aggregations, ii) well-known fuzzy logic aggregation methods and iii) machine learning techniques as aggregation mechanisms. To validate our proposals, 215 forensic cases from two different populations have been used. The results obtained show how the use of white-box machine learning techniques as aggregation models (average ranking from 2.02 to 2.21) are able to improve the state-of-the-art (average ranking of 3.91) without compromising the explainability and interpretability of the method.", "date": "2026-03-24", "categories": []}
{"id": "2603.21867", "title": "Adversarial Camouflage", "abstract": "While the rapid development of facial recognition algorithms has enabled numerous beneficial applications, their widespread deployment has raised significant concerns about the risks of mass surveillance and threats to individual privacy. In this paper, we introduce \\textit{Adversarial Camouflage} as a novel solution for protecting users' privacy. This approach is designed to be efficient and simple to reproduce for users in the physical world. The algorithm starts by defining a low-dimensional pattern space parameterized by color, shape, and angle. Optimized patterns, once found, are projected onto semantically valid facial regions for evaluation. Our method maximizes recognition error across multiple architectures, ensuring high cross-model transferability even against black-box systems. It significantly degrades the performance of all tested state-of-the-art face recognition models during simulations and demonstrates promising results in real-world human experiments, while revealing differences in model robustness and evidence of attack transferability across architectures.", "date": "2026-03-23", "categories": []}
{"id": "2603.19678", "title": "Vision-Language Attribute Disentanglement and Reinforcement for Lifelong Person Re-Identification", "abstract": "Lifelong person re-identification (LReID) aims to learn from varying domains to obtain a unified person retrieval model. Existing LReID approaches typically focus on learning from scratch or a visual classification-pretrained model, while the Vision-Language Model (VLM) has shown generalizable knowledge in a variety of tasks. Although existing methods can be directly adapted to the VLM, since they only consider global-aware learning, the fine-grained attribute knowledge is underleveraged, leading to limited acquisition and anti-forgetting capacity. To address this problem, we introduce a novel VLM-driven LReID approach named Vision-Language Attribute Disentanglement and Reinforcement (VLADR). Our key idea is to explicitly model the universally shared human attributes to improve inter-domain knowledge transfer, thereby effectively utilizing historical knowledge to reinforce new knowledge learning and alleviate forgetting. Specifically, VLADR includes a Multi-grain Text Attribute Disentanglement mechanism that mines the global and diverse local text attributes of an image. Then, an Inter-domain Cross-modal Attribute Reinforcement scheme is developed, which introduces cross-modal attribute alignment to guide visual attribute extraction and adopts inter-domain attribute alignment to achieve fine-grained knowledge transfer. Experimental results demonstrate that our VLADR outperforms the state-of-the-art methods by 1.9\\%-2.2\\% and 2.1\\%-2.5\\% on anti-forgetting and generalization capacity. Our source code is available at https://github.com/zhoujiahuan1991/CVPR2026-VLADR", "date": "2026-03-20", "categories": []}
{"id": "2509.04050", "title": "A Re-ranking Method using K-nearest Weighted Fusion for Person Re-identification", "abstract": "In person re-identification, re-ranking is a crucial step to enhance the overall accuracy by refining the initial ranking of retrieved results. Previous studies have mainly focused on features from single-view images, which can cause view bias and issues like pose variation, viewpoint changes, and occlusions. Using multi-view features to present a person can help reduce view bias. In this work, we present an efficient re-ranking method that generates multi-view features by aggregating neighbors' features using K-nearest Weighted Fusion (KWF) method. Specifically, we hypothesize that features extracted from re-identification models are highly similar when representing the same identity. Thus, we select K neighboring features in an unsupervised manner to generate multi-view features. Additionally, this study explores the weight selection strategies during feature aggregation, allowing us to identify an effective strategy. Our re-ranking approach does not require model fine-tuning or extra annotations, making it applicable to large-scale datasets. We evaluate our method on the person re-identification datasets Market1501, MSMT17, and Occluded-DukeMTMC. The results show that our method significantly improves Rank@1 and mAP when re-ranking the top M candidates from the initial ranking results. Specifically, compared to the initial results, our re-ranking method achieves improvements of 9.8%/22.0% in Rank@1 on the challenging datasets: MSMT17 and Occluded-DukeMTMC, respectively. Furthermore, our approach demonstrates substantial enhancements in computational efficiency compared to other re-ranking methods. Code is available at https://github.com/chequanghuy/Enhancing-Person-Re-Identification-via-UFFM-and-AMC.", "date": "2025-09-04", "categories": []}
{"id": "2603.19004", "title": "Unleashing the Power of Simplicity: A Minimalist Strategy for State-of-the-Art Fingerprint Enhancement", "abstract": "Fingerprint recognition systems, which rely on the unique characteristics of human fingerprints, are essential in modern security and verification applications. Accurate minutiae extraction, a critical step in these systems, depends on the quality of fingerprint images. Despite recent improvements in fingerprint enhancement techniques, state-of-the-art methods often struggle with low-quality fingerprints and can be computationally demanding. This paper presents a minimalist approach to fingerprint enhancement, prioritizing simplicity and effectiveness. Two novel methods are introduced: a contextual filtering method and a learning-based method. These techniques consistently outperform complex state-of-the-art methods, producing clearer, more accurate, and less noisy images. The effectiveness of these methods is validated using a challenging latent fingerprint database. The open-source implementation of these techniques not only fosters reproducibility but also encourages further advancements in the field. The findings underscore the importance of simplicity in achieving high-quality fingerprint enhancement and suggest that future research should balance complexity and practical benefits.", "date": "2026-03-19", "categories": []}
{"id": "2603.17173", "title": "Generalist Multimodal LLMs Gain Biometric Expertise via Human Salience", "abstract": "Iris presentation attack detection (PAD) is critical for secure biometric deployments, yet developing specialized models faces significant practical barriers: collecting data representing future unknown attacks is impossible, and collecting diverse-enough data, yet still limited in terms of its predictive power, is expensive. Additionally, sharing biometric data raises privacy concerns. Due to rapid emergence of new attack vectors demanding adaptable solutions, we thus investigate in this paper whether general-purpose multimodal large language models (MLLMs) can perform iris PAD when augmented with human expert knowledge, operating under strict privacy constraints that prohibit sending biometric data to public cloud MLLM services. Through analysis of vision encoder embeddings applied to our dataset, we demonstrate that pre-trained vision transformers in MLLMs inherently cluster many iris attack types despite never being explicitly trained for this task. However, where clustering shows overlap between attack classes, we find that structured prompts incorporating human salience (verbal descriptions from subjects identifying attack indicators) enable these models to resolve ambiguities. Testing on an IRB-restricted dataset of 224 iris images spanning seven attack types, using only university-approved services (Gemini 2.5 Pro) or locally-hosted models (e.g., Llama 3.2-Vision), we show that Gemini with expert-informed prompts outperforms both a specialized convolutional neural networks (CNN)-based baseline and human examiners, while the locally-deployable Llama achieves near-human performance. Our results establish that MLLMs deployable within institutional privacy constraints offer a viable path for iris PAD.", "date": "2026-03-17", "categories": []}
{"id": "2603.16165", "title": "Homogeneous and Heterogeneous Consistency progressive Re-ranking for Visible-Infrared Person Re-identification", "abstract": "Visible-infrared person re-identification faces greater challenges than traditional person re-identification due to the significant differences between modalities. In particular, the differences between these modalities make effective matching even more challenging, mainly because existing re-ranking algorithms cannot simultaneously address the intra-modal variations and inter-modal discrepancy in cross-modal person re-identification. To address this problem, we propose a novel Progressive Modal Relationship Re-ranking method consisting of two modules, called heterogeneous and homogeneous consistency re-ranking(HHCR). The first module, heterogeneous consistency re-ranking, explores the relationship between the query and the gallery modalities in the test set. The second module, homogeneous consistency reranking, investigates the intrinsic relationship within each modality between the query and the gallery in the test set. Based on this, we propose a baseline for cross-modal person re-identification, called a consistency re-ranking inference network (CRI). We conducted comprehensive experiments demonstrating that our proposed re-ranking method is generalized, and both the re-ranking and the baseline achieve state-of-the-art performance.", "date": "2026-03-17", "categories": []}
{"id": "2505.17343", "title": "Ocular Authentication: Fusion of Gaze and Periocular Modalities", "abstract": "This paper investigates the feasibility of fusing two eye-centric authentication modalities-eye movements and periocular images-within a calibration-free authentication system. While each modality has independently shown promise for user authentication, their combination within a unified gaze-estimation pipeline has not been thoroughly explored at scale. In this report, we propose a multimodal authentication system and evaluate it using a large-scale in-house dataset comprising 9202 subjects with an eye tracking (ET) signal quality equivalent to a consumer-facing virtual reality (VR) device. Our results show that the multimodal approach consistently outperforms both unimodal systems across all scenarios, surpassing the FIDO benchmark. The integration of a state-of-the-art machine learning architecture contributed significantly to the overall authentication performance at scale, driven by the model's ability to capture authentication representations and the complementary discriminative characteristics of the fused modalities.", "date": "2025-05-22", "categories": []}
{"id": "2512.15774", "title": "Two-Step Data Augmentation for Masked Face Detection and Recognition: Turning Fake Masks to Real", "abstract": "The absence of large-scale masked face datasets challenges masked face detection and recognition. We propose a two-step generative data augmentation framework combining rule-based mask warping with unpaired image-to-image translation via GANs, producing masked face samples that go beyond rule-based overlays. Trained on about 19,100 images in the target domain (3.8% of IAMGAN's scale), or, including out-of-domain transfer pretraining, 59,600 and 11.8%, the proposed approach yields consistent improvements over rule-based warping alone and achieves results complementary to IAMGAN's, showing that both steps contribute. Evaluation is conducted directly on the generated samples and is qualitative; quantitative metrics like FID and KID were not applied as any real reference distribution would unfairly favor the model with closer training data. We introduce a non-mask preservation loss to reduce non-mask distortions and stabilize training, and stochastic noise injection to enhance sample diversity.\n  Note: The paper originated as a coursework submission completed under resource constraints. Following an inexplicable scholarship termination, the author took on part-time employment to maintain research continuity, which led to a mid-semester domain pivot from medical imaging to masked face tasks due to company data restrictions. The work was completed alongside concurrent coursework with delayed compute access and without any AI assistance. It was submitted to a small venue at the semester end under an obligatory publication requirement and accepted without revision requests. Subsequent invitations to submit to first-tier venues were not pursued due to continued funding absence. Downstream evaluation on recognition or detection performance was not completed by the submission deadline. The note is added in response to subsequent comparisons and criticisms that did not account for these conditions.", "date": "2025-12-13", "categories": []}
{"id": "2603.13877", "title": "Scribe Verification in Chinese manuscripts using Siamese, Triplet, and Vision Transformer Neural Networks", "abstract": "The paper examines deep learning models for scribe verification in Chinese manuscripts. That is, to automatically determine whether two manuscript fragments were written by the same scribe using deep metric learning methods. Two datasets were used: the Tsinghua Bamboo Slips Dataset and a selected subset of the Multi-Attribute Chinese Calligraphy Dataset, focusing on the calligraphers with a large number of samples. Siamese and Triplet neural network architectures are implemented, including convolutional and Transformer-based models. The experimental results show that the MobileNetV3+ Custom Siamese model trained with contrastive loss achieves either the best or the second-best overall accuracy and area under the Receiver Operating Characteristic Curve on both datasets.", "date": "2026-03-14", "categories": []}
{"id": "2603.14012", "title": "Multi-Grained Vision-Language Alignment for Domain Generalized Person Re-Identification", "abstract": "Domain Generalized person Re-identification (DG Re-ID) is a challenging task, where models are trained on source domains but tested on unseen target domains. Although previous pure vision-based models have achieved significant progress, the performance remains further improved. Recently, Vision-Language Models (VLMs) present outstanding generalization capabilities in various visual applications. However, directly adapting a VLM to Re-ID shows limited generalization improvement. This is because the VLM only produces with global features that are insensitive to ID nuances. To tacle this problem, we propose a CLIP-based multi-grained vision-language alignment framework in this work. Specifically, several multi-grained prompts are introduced in language modality to describe different body parts and align with their counterparts in vision modality. To obtain fine-grained visual information, an adaptively masked multi-head self-attention module is employed to precisely extract specific part features. To train the proposed module, an MLLM-based visual grounding expert is employed to automatically generate pseudo labels of body parts for supervision. Extensive experiments conducted on both single- and multi-source generalization protocols demonstrate the superior performance of our approach. The implementation code will be released at https://github.com/RikoLi/MUVA.", "date": "2026-03-14", "categories": []}
{"id": "2603.14189", "title": "Walking Further: Semantic-aware Multimodal Gait Recognition Under Long-Range Conditions", "abstract": "Gait recognition is an emerging biometric technology that enables non-intrusive and hard-to-spoof human identification. However, most existing methods are confined to short-range, unimodal settings and fail to generalize to long-range and cross-distance scenarios under real-world conditions. To address this gap, we present \\textbf{LRGait}, the first LiDAR-Camera multimodal benchmark designed for robust long-range gait recognition across diverse outdoor distances and environments. We further propose \\textbf{EMGaitNet}, an end-to-end framework tailored for long-range multimodal gait recognition. To bridge the modality gap between RGB images and point clouds, we introduce a semantic-guided fusion pipeline. A CLIP-based Semantic Mining (SeMi) module first extracts human body-part-aware semantic cues, which are then employed to align 2D and 3D features via a Semantic-Guided Alignment (SGA) module within a unified embedding space. A Symmetric Cross-Attention Fusion (SCAF) module hierarchically integrates visual contours and 3D geometric features, and a Spatio-Temporal (ST) module captures global gait dynamics. Extensive experiments on various gait datasets validate the effectiveness of our method.", "date": "2026-03-15", "categories": []}
{"id": "2603.15062", "title": "The Good, the Better, and the Best: Improving the Discriminability of Face Embeddings through Attribute-aware Learning", "abstract": "Despite recent advances in face recognition, robust performance remains challenging under large variations in age, pose, and occlusion. A common strategy to address these issues is to guide representation learning with auxiliary supervision from facial attributes, encouraging the visual encoder to focus on identity-relevant regions. However, existing approaches typically rely on heterogeneous and fixed sets of attributes, implicitly assuming equal relevance across attributes. This assumption is suboptimal, as different attributes exhibit varying discriminative power for identity recognition, and some may even introduce harmful biases. In this paper, we propose an attribute-aware face recognition architecture that supervises the learning of facial embeddings using identity class labels, identity-relevant facial attributes, and non-identity-related attributes. Facial attributes are organized into interpretable groups, making it possible to decompose and analyze their individual contributions in a human-understandable manner. Experiments on standard face verification benchmarks demonstrate that joint learning of identity and facial attributes improves the discriminability of face embeddings with two major conclusions: (i) using identity-relevant subsets of facial attributes consistently outperforms supervision with a broader attribute set, and (ii) explicitly forcing embeddings to unlearn non-identity-related attributes yields further performance gains compared to leaving such attributes unsupervised. Additionally, our method serves as a diagnostic tool for assessing the trustworthiness of face recognition encoders by allowing for the measurement of accuracy gains with suppression of non-identity-relevant attributes, with such gains suggesting shortcut learning from redundant attributes associated with each identity.", "date": "2026-03-16", "categories": []}
{"id": "2508.01587", "title": "Distilling the Past: Information-Dense and Style-Aware Replay for Lifelong Person Re-Identification", "abstract": "Lifelong person re-identification (LReID) aims to continuously adapt to new domains while mitigating catastrophic forgetting. While replay-based methods effectively alleviate forgetting, they are constrained by strict memory budgets, leading to limited sample diversity. Conversely, exemplar-free approaches bypass memory constraints entirely but struggle to preserve the fine-grained identity semantics crucial for Re-ID tasks. To resolve this fundamental dilemma, we propose an Information-Dense and Style-Aware Replay framework. Instead of storing a sparse set of raw historical images, we fuse the knowledge of sequential data into the pixel space of a compact replay buffer via multi-stage gradient matching and identity supervision. This condensation process not only maximizes the semantic representativeness of limited memory but also naturally conceals original visual details, inherently preserving data privacy. Furthermore, to combat forgetting induced by cross-domain shifts, we introduce a dual-alignment style replay strategy that adapts both current and fused replay samples, harmonizing feature representations across disparate domains. Extensive experiments on multiple LReID benchmarks demonstrate that our method significantly outperforms existing approaches, achieving improvements of +5.0% and +6.0% in Seen-Avg mAP over current state-of-the-art and traditional replay-based methods, respectively, thereby establishing an efficient and robust new baseline for lifelong learning.", "date": "2025-08-03", "categories": []}
{"id": "2603.12912", "title": "FedBPrompt: Federated Domain Generalization Person Re-Identification via Body Distribution Aware Visual Prompts", "abstract": "Federated Domain Generalization for Person Re-Identification (FedDG-ReID) learns domain-invariant representations from decentralized data. While Vision Transformer (ViT) is widely adopted, its global attention often fails to distinguish pedestrians from high similarity backgrounds or diverse viewpoints -- a challenge amplified by cross-client distribution shifts in FedDG-ReID. To address this, we propose Federated Body Distribution Aware Visual Prompt (FedBPrompt), introducing learnable visual prompts to guide Transformer attention toward pedestrian-centric regions. FedBPrompt employs a Body Distribution Aware Visual Prompts Mechanism (BAPM) comprising: Holistic Full Body Prompts to suppress cross-client background noise, and Body Part Alignment Prompts to capture fine-grained details robust to pose and viewpoint variations. To mitigate high communication costs, we design a Prompt-based Fine-Tuning Strategy (PFTS) that freezes the ViT backbone and updates only lightweight prompts, significantly reducing communication overhead while maintaining adaptability. Extensive experiments demonstrate that BAPM effectively enhances feature discrimination and cross-domain generalization, while PFTS achieves notable performance gains within only a few aggregation rounds. Moreover, both BAPM and PFTS can be easily integrated into existing ViT-based FedDG-ReID frameworks, making FedBPrompt a flexible and effective solution for federated person re-identification. The code is available at https://github.com/leavlong/FedBPrompt.", "date": "2026-03-13", "categories": []}
{"id": "2602.18990", "title": "IDSelect: A RL-Based Cost-Aware Selection Agent for Video-based Multi-Modal Person Recognition", "abstract": "Video-based person recognition achieves robust identification by integrating face, body, and gait. However, current systems waste computational resources by processing all modalities with fixed heavyweight ensembles regardless of input complexity. To address these limitations, we propose IDSelect, a reinforcement learning-based cost-aware selector that chooses one pre-trained model per modality per-sequence to optimize the accuracy-efficiency trade-off. Our key insight is that an input-conditioned selector can discover complementary model choices that surpass fixed ensembles while using substantially fewer resources. IDSelect trains a lightweight agent end-to-end using actor-critic reinforcement learning with budget-aware optimization. The reward balances recognition accuracy with computational cost, while entropy regularization prevents premature convergence. At inference, the policy selects the most probable model per modality and fuses modality-specific similarities for the final score. Extensive experiments on challenging video-based datasets demonstrate IDSelect's superior efficiency: on CCVID, it achieves 95.9% Rank-1 accuracy with 92.4% less computation than strong baselines while improving accuracy by 1.8%; on MEVID, it reduces computation by 41.3% while maintaining competitive performance.", "date": "2026-02-22", "categories": []}
{"id": "2311.10605", "title": "CA-Jaccard: Camera-aware Jaccard Distance for Person Re-identification", "abstract": "Person re-identification (re-ID) is a challenging task that aims to learn discriminative features for person retrieval. In person re-ID, Jaccard distance is a widely used distance metric, especially in re-ranking and clustering scenarios. However, we discover that camera variation has a significant negative impact on the reliability of Jaccard distance. In particular, Jaccard distance calculates the distance based on the overlap of relevant neighbors. Due to camera variation, intra-camera samples dominate the relevant neighbors, which reduces the reliability of the neighbors by introducing intra-camera negative samples and excluding inter-camera positive samples. To overcome this problem, we propose a novel camera-aware Jaccard (CA-Jaccard) distance that leverages camera information to enhance the reliability of Jaccard distance. Specifically, we design camera-aware k-reciprocal nearest neighbors (CKRNNs) to find k-reciprocal nearest neighbors on the intra-camera and inter-camera ranking lists, which improves the reliability of relevant neighbors and guarantees the contribution of inter-camera samples in the overlap. Moreover, we propose a camera-aware local query expansion (CLQE) to mine reliable samples in relevant neighbors by exploiting camera variation as a strong constraint and assign these samples higher weights in overlap, further improving the reliability. Our CA-Jaccard distance is simple yet effective and can serve as a general distance metric for person re-ID methods with high reliability and low computational cost. Extensive experiments demonstrate the effectiveness of our method.", "date": "2023-11-17", "categories": []}
{"id": "2405.15965", "title": "Goldilocks Test Sets for Face Verification", "abstract": "Reported face verification accuracy has reached a plateau on current well-known test sets. As a result, some difficult test sets have been assembled by reducing the image quality or adding artifacts to the image. However, we argue that test sets can be challenging without artificially reducing the image quality because the face recognition (FR) models suffer from correctly recognizing 1) the pairs from the same identity (i.e., genuine pairs) with a large face attribute difference, 2) the pairs from different identities (i.e., impostor pairs) with a small face attribute difference, and 3) the pairs of similar-looking identities (e.g., twins and relatives). We propose three challenging test sets to reveal important but ignored weaknesses of the existing FR algorithms. To challenge models on variation of facial attributes, we propose Hadrian and Eclipse to address facial hair differences and face exposure differences. The images in both test sets are high-quality and collected in a controlled environment. To challenge FR models on similar-looking persons, we propose ND-Twins, which contains images from a dedicated twins dataset. The LFW test protocol is used to structure the proposed test sets. Moreover, we introduce additional rules to assemble ``Goldilocks\\footnote{https://en.wikipedia.org/wiki/Goldilocks_and_the_Three_Bears}\" level test sets, including 1) restricted number of occurrence of hard samples, 2) equal chance evaluation across demographic groups, and 3) constrained identity overlap across validation folds. Quantitatively, without further processing the images, the proposed test sets have on-par or higher difficulties than the existing test sets that add artifacts to the images. The datasets are available at: https://github.com/HaiyuWu/SOTA-Face-Recognition-Train-and-Test.", "date": "2024-05-24", "categories": []}
{"id": "2603.01640", "title": "MSP-ReID: Hairstyle-Robust Cloth-Changing Person Re-Identification", "abstract": "Cloth-Changing Person Re-Identification (CC-ReID) aims to match the same individual across cameras under varying clothing conditions. Existing approaches often remove apparel and focus on the head region to reduce clothing bias. However, treating the head holistically without distinguishing between face and hair leads to over-reliance on volatile hairstyle cues, causing performance degradation under hairstyle changes. To address this issue, we propose the Mitigating Hairstyle Distraction and Structural Preservation (MSP) framework. Specifically, MSP introduces Hairstyle-Oriented Augmentation (HSOA), which generates intra-identity hairstyle diversity to reduce hairstyle dependence and enhance attention to stable facial and body cues. To prevent the loss of structural information, we design Cloth-Preserved Random Erasing (CPRE), which performs ratio-controlled erasing within clothing regions to suppress texture bias while retaining body shape and context. Furthermore, we employ Region-based Parsing Attention (RPA) to incorporate parsing-guided priors that highlight face and limb regions while suppressing hair features. Extensive experiments on multiple CC-ReID benchmarks demonstrate that MSP achieves state-of-the-art performance, providing a robust and practical solution for long-term person re-identification.", "date": "2026-03-02", "categories": []}
{"id": "2603.06122", "title": "FedARKS: Federated Aggregation via Robust and Discriminative Knowledge Selection and Integration for Person Re-identification", "abstract": "The application of federated domain generalization in person re-identification (FedDG-ReID) aims to enhance the model's generalization ability in unseen domains while protecting client data privacy. However, existing mainstream methods typically rely on global feature representations and simple averaging operations for model aggregation, leading to two limitations in domain generalization: (1) Using only global features makes it difficult to capture subtle, domain-invariant local details (such as accessories or textures); (2) Uniform parameter averaging treats all clients as equivalent, ignoring their differences in robust feature extraction capabilities, thereby diluting the contributions of high quality clients. To address these issues, we propose a novel federated learning framework, Federated Aggregation via Robust and Discriminative Knowledge Selection and Integration (FedARKS), comprising two mechanisms: RK (Robust Knowledge) and KS (Knowledge Selection).", "date": "2026-03-06", "categories": []}
{"id": "2502.17244", "title": "A dataset of high-resolution plantar pressures for gait analysis across varying footwear and walking speeds", "abstract": "Gait refers to the patterns of limb movement generated during walking, which are unique to each individual due to both physical and behavioral traits. Walking patterns have been widely studied in biometrics, biomechanics, sports, and rehabilitation. While traditional methods rely on video and motion capture, advances in plantar pressure sensing technology now offer deeper insights into gait. However, underfoot pressures during walking remain underexplored due to the lack of large, publicly accessible datasets. To address this, we introduce the UNB StepUP-P150 dataset: a footStep database for gait analysis and recognition using Underfoot Pressure, including data from 150 individuals. This dataset comprises high-resolution plantar pressure data (4 sensors per cm-squared) collected using a 1.2m by 3.6m pressure-sensing walkway. It contains over 200,000 footsteps from participants walking with various speeds (preferred, slow-to-stop, fast, and slow) and footwear conditions (barefoot, standard shoes, and two personal shoes), supporting advancements in biometric gait recognition and presenting new research opportunities in biomechanics and deep learning. UNB StepUP-P150 establishes a new benchmark for plantar pressure-based gait analysis and recognition.", "date": "2025-02-24", "categories": []}
{"id": "2602.11086", "title": "First International StepUP Competition for Biometric Footstep Recognition: Methods, Results and Remaining Challenges", "abstract": "Biometric footstep recognition, based on a person's unique pressure patterns under their feet during walking, is an emerging field with growing applications in security and safety. However, progress in this area has been limited by the lack of large, diverse datasets necessary to address critical challenges such as generalization to new users and robustness to shifts in factors like footwear or walking speed. The recent release of the UNB StepUP-P150 dataset, the largest and most comprehensive collection of high-resolution footstep pressure recordings to date, opens new opportunities for addressing these challenges through deep learning. To mark this milestone, the First International StepUP Competition for Biometric Footstep Recognition was launched. Competitors were tasked with developing robust recognition models using the StepUP-P150 dataset that were then evaluated on a separate, dedicated test set designed to assess verification performance under challenging variations, given limited and relatively homogeneous reference data. The competition attracted global participation, with 23 registered teams from academia and industry. The top-performing team, Saeid_UCC, achieved the best equal error rate (EER) of 10.77% using a generative reward machine (GRM) optimization strategy. Overall, the competition showcased strong solutions, but persistent challenges in generalizing to unfamiliar footwear highlight a critical area for future work.", "date": "2026-02-11", "categories": []}
{"id": "2509.16557", "title": "Person Identification from Egocentric Human-Object Interactions using 3D Hand Pose", "abstract": "Human-Object Interaction Recognition (HOIR) and user identification play a crucial role in advancing augmented reality (AR)-based personalized assistive technologies. These systems are increasingly being deployed in high-stakes, human-centric environments such as aircraft cockpits, aerospace maintenance, and surgical procedures. This research introduces I2S (Interact2Sign), a multi stage framework designed for unobtrusive user identification through human object interaction recognition, leveraging 3D hand pose analysis in egocentric videos. I2S utilizes handcrafted features extracted from 3D hand poses and per forms sequential feature augmentation: first identifying the object class, followed by HOI recognition, and ultimately, user identification. A comprehensive feature extraction and description process was carried out for 3D hand poses, organizing the extracted features into semantically meaningful categories: Spatial, Frequency, Kinematic, Orientation, and a novel descriptor introduced in this work, the Inter-Hand Spatial Envelope (IHSE). Extensive ablation studies were conducted to determine the most effective combination of features. The optimal configuration achieved an impressive average F1-score of 97.52% for user identification, evaluated on a bimanual object manipulation dataset derived from the ARCTIC and H2O datasets. I2S demonstrates state-of-the-art performance while maintaining a lightweight model size of under 4 MB and a fast inference time of 0.1 seconds. These characteristics make the proposed framework highly suitable for real-time, on-device authentication in security-critical, AR-based systems.", "date": "2025-09-20", "categories": []}
{"id": "2602.23133", "title": "From Calibration to Refinement: Seeking Certainty via Probabilistic Evidence Propagation for Noisy-Label Person Re-Identification", "abstract": "With the increasing demand for robust person Re-ID in unconstrained environments, learning from datasets with noisy labels and sparse per-identity samples remains a critical challenge. Existing noise-robust person Re-ID methods primarily rely on loss-correction or sample-selection strategies using softmax outputs. However, these methods suffer from two key limitations: 1) Softmax exhibits translation invariance, leading to over-confident and unreliable predictions on corrupted labels. 2) Conventional sample selection based on small-loss criteria often discards valuable hard positives that are crucial for learning discriminative features. To overcome these issues, we propose the CAlibration-to-REfinement (CARE) method, a two-stage framework that seeks certainty through probabilistic evidence propagation from calibration to refinement. In the calibration stage, we propose the probabilistic evidence calibration (PEC) that dismantles softmax translation invariance by injecting adaptive learnable parameters into the similarity function, and employs an evidential calibration loss to mitigate overconfidence on mislabeled samples. In the refinement stage, we design the evidence propagation refinement (EPR) that can more accurately distinguish between clean and noisy samples. Specifically, the EPR contains two steps: Firstly, the composite angular margin (CAM) metric is proposed to precisely distinguish clean but hard-to-learn positive samples from mislabeled ones in a hyperspherical space; Secondly, the certainty-oriented sphere weighting (COSW) is developed to dynamically allocate the importance of samples according to CAM, ensuring clean instances drive model updates. Extensive experimental results on Market1501, DukeMTMC-ReID, and CUHK03 datasets under both random and patterned noises show that CARE achieves competitive performance.", "date": "2026-02-26", "categories": []}
{"id": "2511.13065", "title": "RobustGait: Robustness Analysis for Appearance Based Gait Recognition", "abstract": "Appearance-based gait recognition have achieved strong performance on controlled datasets, yet systematic evaluation of its robustness to real-world corruptions and silhouette variability remains lacking. We present RobustGait, a framework for fine-grained robustness evaluation of appearance-based gait recognition systems. RobustGait evaluation spans four dimensions: the type of perturbation (digital, environmental, temporal, occlusion), the silhouette extraction method (segmentation and parsing networks), the architectural capacities of gait recognition models, and various deployment scenarios. The benchmark introduces 15 corruption types at 5 severity levels across CASIA-B, CCPG, and SUSTech1K, with in-the-wild validation on MEVID, and evaluates six state-of-the-art gait systems. We came across several exciting insights. First, applying noise at the RGB level better reflects real-world degradation, and reveal how distortions propagate through silhouette extraction to the downstream gait recognition systems. Second, gait accuracy is highly sensitive to silhouette extractor biases, revealing an overlooked source of benchmark bias. Third, robustness is dependent on both the type of perturbation and the architectural design. Finally, we explore robustness-enhancing strategies, showing that noise-aware training and knowledge distillation improve performance and move toward deployment-ready systems. Code is available at https://reeshoon.github.io/robustgaitbenchmark", "date": "2025-11-17", "categories": []}
{"id": "2601.21405", "title": "Rectifying Geometry-Induced Similarity Distortions for Real-World Aerial-Ground Person Re-Identification", "abstract": "Aerial-ground person re-identification (AG-ReID) is fundamentally challenged by extreme viewpoint and distance discrepancies between aerial and ground cameras, which induce severe geometric distortions and invalidate the assumption of a shared similarity space across views. Existing methods primarily rely on geometry-aware feature learning or appearance-conditioned prompting, while implicitly assuming that the geometry-invariant dot-product similarity used in attention mechanisms remains reliable under large viewpoint and scale variations. We argue that this assumption does not hold. Extreme camera geometry systematically distorts the query-key similarity space and degrades attention-based matching, even when feature representations are partially aligned. To address this issue, we introduce Geometry-Induced Query-Key Transformation (GIQT), a lightweight low-rank module that explicitly rectifies the similarity space by conditioning query-key interactions on camera geometry. Rather than modifying feature representations or the attention formulation itself, GIQT adapts the similarity computation to compensate for dominant geometry-induced anisotropic distortions. Building on this local similarity rectification, we further incorporate a geometry-conditioned prompt generation mechanism that provides global, view-adaptive representation priors derived directly from camera geometry.Experiments on four aerial-ground person re-identification benchmarks demonstrate that the proposed framework consistently improves robustness under extreme and previously unseen geometric conditions, while introducing minimal computational overhead compared to state-of-the-art methods.", "date": "2026-01-29", "categories": []}
{"id": "2602.21503", "title": "AHAN: Asymmetric Hierarchical Attention Network for Identical Twin Face Verification", "abstract": "Identical twin face verification represents an extreme fine-grained recognition challenge where even state-of-the-art systems fail due to overwhelming genetic similarity. Current face recognition methods achieve over 99.8% accuracy on standard benchmarks but drop dramatically to 88.9% when distinguishing identical twins, exposing critical vulnerabilities in biometric security systems. The difficulty lies in learning features that capture subtle, non-genetic variations that uniquely identify individuals. We propose the Asymmetric Hierarchical Attention Network (AHAN), a novel architecture specifically designed for this challenge through multi-granularity facial analysis. AHAN introduces a Hierarchical Cross-Attention (HCA) module that performs multi-scale analysis on semantic facial regions, enabling specialized processing at optimal resolutions. We further propose a Facial Asymmetry Attention Module (FAAM) that learns unique biometric signatures by computing cross-attention between left and right facial halves, capturing subtle asymmetric patterns that differ even between twins. To ensure the network learns truly individuating features, we introduce Twin-Aware Pair-Wise Cross-Attention (TA-PWCA), a training-only regularization strategy that uses each subject's own twin as the hardest possible distractor. Extensive experiments on the ND_TWIN dataset demonstrate that AHAN achieves 92.3% twin verification accuracy, representing a 3.4% improvement over state-of-the-art methods.", "date": "2026-02-25", "categories": []}
{"id": "2210.11974", "title": "Face Pyramid Vision Transformer", "abstract": "A novel Face Pyramid Vision Transformer (FPVT) is proposed to learn a discriminative multi-scale facial representations for face recognition and verification. In FPVT, Face Spatial Reduction Attention (FSRA) and Dimensionality Reduction (FDR) layers are employed to make the feature maps compact, thus reducing the computations. An Improved Patch Embedding (IPE) algorithm is proposed to exploit the benefits of CNNs in ViTs (e.g., shared weights, local context, and receptive fields) to model lower-level edges to higher-level semantic primitives. Within FPVT framework, a Convolutional Feed-Forward Network (CFFN) is proposed that extracts locality information to learn low level facial information. The proposed FPVT is evaluated on seven benchmark datasets and compared with ten existing state-of-the-art methods, including CNNs, pure ViTs, and Convolutional ViTs. Despite fewer parameters, FPVT has demonstrated excellent performance over the compared methods. Project page is available at https://khawar-islam.github.io/fpvt/", "date": "2022-10-21", "categories": []}
{"id": "2503.12047", "title": "PSGait: Gait Recognition using Parsing Skeleton", "abstract": "Gait recognition has emerged as a robust biometric modality due to its non-intrusive nature. Conventional gait recognition methods mainly rely on silhouettes or skeletons. While effective in controlled laboratory settings, their limited information entropy restricts generalization to real-world scenarios. To overcome this, we propose a novel representation called \\textbf{Parsing Skeleton}, which uses a skeleton-guided human parsing method to capture fine-grained body dynamics with much higher information entropy. To effectively explore the capability of the Parsing Skeleton, we also introduce \\textbf{PSGait}, a framework that fuses Parsing Skeleton with silhouettes to enhance individual differentiation. Comprehensive benchmarks demonstrate that PSGait outperforms state-of-the-art multimodal methods while significantly reducing computational resources. As a plug-and-play method, it achieves an improvement of up to 15.7\\% in the accuracy of Rank-1 in various models. These results validate the Parsing Skeleton as a \\textbf{lightweight}, \\textbf{effective}, and highly \\textbf{generalizable} representation for gait recognition in the wild. Code is available at https://github.com/realHarryX/PSGait.", "date": "2025-03-15", "categories": []}
{"id": "2602.18965", "title": "Face Presentation Attack Detection via Content-Adaptive Spatial Operators", "abstract": "Face presentation attack detection (FacePAD) is critical for securing facial authentication against print, replay, and mask-based spoofing. This paper proposes CASO-PAD, an RGB-only, single-frame model that enhances MobileNetV3 with content-adaptive spatial operators (involution) to better capture localized spoof cues. Unlike spatially shared convolution kernels, the proposed operator generates location-specific, channel-shared kernels conditioned on the input, improving spatial selectivity with minimal overhead. CASO-PAD remains lightweight (3.6M parameters; 0.64 GFLOPs at $256\\times256$) and is trained end-to-end using a standard binary cross-entropy objective. Extensive experiments on Replay-Attack, Replay-Mobile, ROSE-Youtu, and OULU-NPU demonstrate strong performance, achieving 100/100/98.9/99.7\\% test accuracy, AUC of 1.00/1.00/0.9995/0.9999, and HTER of 0.00/0.00/0.82/0.44\\%, respectively. On the large-scale SiW-Mv2 Protocol-1 benchmark, CASO-PAD further attains 95.45\\% accuracy with 3.11\\% HTER and 3.13\\% EER, indicating improved robustness under diverse real-world attacks. Ablation studies show that placing the adaptive operator near the network head and using moderate group sharing yields the best accuracy--efficiency balance. Overall, CASO-PAD provides a practical pathway for robust, on-device FacePAD with mobile-class compute and without auxiliary sensors or temporal stacks.", "date": "2026-02-21", "categories": []}
{"id": "2602.15493", "title": "LEADER: Lightweight End-to-End Attention-Gated Dual Autoencoder for Robust Minutiae Extraction", "abstract": "Minutiae extraction, a fundamental stage in fingerprint recognition, is increasingly shifting toward deep learning. However, truly end-to-end methods that eliminate separate preprocessing and postprocessing steps remain scarce. This paper introduces LEADER (Lightweight End-to-end Attention-gated Dual autoencodER), a neural network that maps raw fingerprint images to minutiae descriptors, including location, direction, and type. The proposed architecture integrates non-maximum suppression and angular decoding to enable complete end-to-end inference using only 0.9M parameters. It employs a novel \"Castle-Moat-Rampart\" ground-truth encoding and a dual-autoencoder structure, interconnected through an attention-gating mechanism. Experimental evaluations demonstrate state-of-the-art accuracy on plain fingerprints and robust cross-domain generalization to latent impressions. Specifically, LEADER attains a 34% higher F1-score on the NIST SD27 dataset compared to specialized latent minutiae extractors. Sample-level analysis on this challenging benchmark reveals an average rank of 2.07 among all compared methods, with LEADER securing the first-place position in 47% of the samples-more than doubling the frequency of the second-best extractor. The internal representations learned by the model align with established fingerprint domain features, such as segmentation masks, orientation fields, frequency maps, and skeletons. Inference requires 15ms on GPU and 322ms on CPU, outperforming leading commercial software in computational efficiency. The source code and pre-trained weights are publicly released to facilitate reproducibility.", "date": "2026-02-17", "categories": []}
{"id": "2602.15535", "title": "Advanced Acceptance Score: A Holistic Measure for Biometric Quantification", "abstract": "Quantifying biometric characteristics within hand gestures involve derivation of fitness scores from a gesture and identity aware feature space. However, evaluating the quality of these scores remains an open question. Existing biometric capacity estimation literature relies upon error rates. But these rates do not indicate goodness of scores. Thus, in this manuscript we present an exhaustive set of evaluation measures. We firstly identify ranking order and relevance of output scores as the primary basis for evaluation. In particular, we consider both rank deviation as well as rewards for: (i) higher scores of high ranked gestures and (ii) lower scores of low ranked gestures. We also compensate for correspondence between trends of output and ground truth scores. Finally, we account for disentanglement between identity features of gestures as a discounting factor. Integrating these elements with adequate weighting, we formulate advanced acceptance score as a holistic evaluation measure. To assess effectivity of the proposed we perform in-depth experimentation over three datasets with five state-of-the-art (SOTA) models. Results show that the optimal score selected with our measure is more appropriate than existing other measures. Also, our proposed measure depicts correlation with existing measures. This further validates its reliability. We have made our \\href{https://github.com/AmanVerma2307/MeasureSuite}{code} public.", "date": "2026-02-17", "categories": []}
{"id": "2506.07304", "title": "FANVID: A Benchmark for Face and License Plate Recognition in Low-Resolution Videos", "abstract": "Real-world surveillance often renders faces and license plates unrecognizable in individual low-resolution (LR) frames, hindering reliable identification. To advance temporal recognition models, we present FANVID, a novel video-based benchmark comprising nearly 1,463 LR clips (180 x 320, 20--60 FPS) featuring 63 identities and 49 license plates from three English-speaking countries. Each video includes distractor faces and plates, increasing task difficulty and realism. The dataset contains 31,096 manually verified bounding boxes and labels.\n  FANVID defines two tasks: (1) face matching -- detecting LR faces and matching them to high-resolution mugshots, and (2) license plate recognition -- extracting text from LR plates without a predefined database. Videos are downsampled from high-resolution sources to ensure that faces and text are indecipherable in single frames, requiring models to exploit temporal information. We introduce evaluation metrics adapted from mean Average Precision at IoU > 0.5, prioritizing identity correctness for faces and character-level accuracy for text.\n  A baseline method with pre-trained video super-resolution, detection, and recognition achieved performance scores of 0.58 (face matching) and 0.42 (plate recognition), highlighting both the feasibility and challenge of the tasks. FANVID's selection of faces and plates balances diversity with recognition challenge. We release the software for data access, evaluation, baseline, and annotation to support reproducibility and extension. FANVID aims to catalyze innovation in temporal modeling for LR recognition, with applications in surveillance, forensics, and autonomous vehicles.", "date": "2025-06-08", "categories": []}
{"id": "2510.10753", "title": "Restricted Receptive Fields for Face Verification", "abstract": "Understanding how deep neural networks make decisions is crucial for analyzing their behavior and diagnosing failure cases. In computer vision, a common approach to improve interpretability is to assign importance to individual pixels using post-hoc methods. Although they are widely used to explain black-box models, their fidelity to the model's actual reasoning is uncertain due to the lack of reliable evaluation metrics. This limitation motivates an alternative approach, which is to design models whose decision processes are inherently interpretable. To this end, we propose a face similarity metric that breaks down global similarity into contributions from restricted receptive fields. Our method defines the similarity between two face images as the sum of patch-level similarity scores, providing a locally additive explanation without relying on post-hoc analysis. We show that the proposed approach achieves competitive verification performance even with patches as small as 28x28 within 112x112 face images, and surpasses state-of-the-art methods when using 56x56 patches.", "date": "2025-10-12", "categories": []}
{"id": "2602.07565", "title": "Human Identification at a Distance: Challenges, Methods and Results on the Competition HID 2025", "abstract": "Human identification at a distance (HID) is challenging because traditional biometric modalities such as face and fingerprints are often difficult to acquire in real-world scenarios. Gait recognition provides a practical alternative, as it can be captured reliably at a distance. To promote progress in gait recognition and provide a fair evaluation platform, the International Competition on Human Identification at a Distance (HID) has been organized annually since 2020. Since 2023, the competition has adopted the challenging SUSTech-Competition dataset, which features substantial variations in clothing, carried objects, and view angles. No dedicated training data are provided, requiring participants to train their models using external datasets. Each year, the competition applies a different random seed to generate distinct evaluation splits, which reduces the risk of overfitting and supports a fair assessment of cross-domain generalization. While HID 2023 and HID 2024 already used this dataset, HID 2025 explicitly examined whether algorithmic advances could surpass the accuracy limits observed previously. Despite the heightened difficulty, participants achieved further improvements, and the best-performing method reached 94.2% accuracy, setting a new benchmark on this dataset. We also analyze key technical trends and outline potential directions for future research in gait recognition.", "date": "2026-02-07", "categories": []}
{"id": "2602.07566", "title": "Cross-Camera Cow Identification via Disentangled Representation Learning", "abstract": "Precise identification of individual cows is a fundamental prerequisite for comprehensive digital management in smart livestock farming. While existing animal identification methods excel in controlled, single-camera settings, they face severe challenges regarding cross-camera generalization. When models trained on source cameras are deployed to new monitoring nodes characterized by divergent illumination, backgrounds, viewpoints, and heterogeneous imaging properties, recognition performance often degrades dramatically. This limits the large-scale application of non-contact technologies in dynamic, real-world farming environments. To address this challenge, this study proposes a cross-camera cow identification framework based on disentangled representation learning. This framework leverages the Subspace Identifiability Guarantee (SIG) theory in the context of bovine visual recognition. By modeling the underlying physical data generation process, we designed a principle-driven feature disentanglement module that decomposes observed images into multiple orthogonal latent subspaces. This mechanism effectively isolates stable, identity-related biometric features that remain invariant across cameras, thereby substantially improving generalization to unseen cameras. We constructed a high-quality dataset spanning five distinct camera nodes, covering heterogeneous acquisition devices and complex variations in lighting and angles. Extensive experiments across seven cross-camera tasks demonstrate that the proposed method achieves an average accuracy of 86.0%, significantly outperforming the Source-only Baseline (51.9%) and the strongest cross-camera baseline method (79.8%). This work establishes a subspace-theoretic feature disentanglement framework for collaborative cross-camera cow identification, offering a new paradigm for precise animal monitoring in uncontrolled smart farming environments.", "date": "2026-02-07", "categories": []}
{"id": "2602.08809", "title": "Efficient Deep Learning for Biometrics: Overview, Challenges and Trends in Ear of Frugal AI", "abstract": "Recent advances in deep learning, whether on discriminative or generative tasks have been beneficial for various applications, among which security and defense. However, their increasing computational demands during training and deployment translates directly into high energy consumption. As a consequence, this induces a heavy carbon footprint which hinders their widespread use and scalability, but also a limitation when deployed on resource-constrained edge devices for real-time use. In this paper, we briefly survey efficient deep learning methods for biometric applications. Specifically, we tackle the challenges one might incur when training and deploying deep learning approaches, and provide a taxonomy of the various efficient deep learning families. Additionally, we discuss complementary metrics for evaluating the efficiency of these models such as memory, computation, latency, throughput, and advocate for universal and reproducible metrics for better comparison. Last, we give future research directions to consider.", "date": "2026-02-09", "categories": []}
{"id": "2602.05785", "title": "ReText: Text Boosts Generalization in Image-Based Person Re-identification", "abstract": "Generalizable image-based person re-identification (Re-ID) aims to recognize individuals across cameras in unseen domains without retraining. While multiple existing approaches address the domain gap through complex architectures, recent findings indicate that better generalization can be achieved by stylistically diverse single-camera data. Although this data is easy to collect, it lacks complexity due to minimal cross-view variation. We propose ReText, a novel method trained on a mixture of multi-camera Re-ID data and single-camera data, where the latter is complemented by textual descriptions to enrich semantic cues. During training, ReText jointly optimizes three tasks: (1) Re-ID on multi-camera data, (2) image-text matching, and (3) image reconstruction guided by text on single-camera data. Experiments demonstrate that ReText achieves strong generalization and significantly outperforms state-of-the-art methods on cross-domain Re-ID benchmarks. To the best of our knowledge, this is the first work to explore multimodal joint learning on a mixture of multi-camera and single-camera data in image-based person Re-ID.", "date": "2026-02-05", "categories": []}
{"id": "2602.02776", "title": "Verification and Identification in ECG biometric on large-scale", "abstract": "This work studies electrocardiogram (ECG) biometrics at large scale, directly addressing a critical gap in the literature: the scarcity of large-scale evaluations with operational metrics and protocols that enable meaningful standardization and comparison across studies. We show that identity information is already present in tabular representations (fiducial features): even a simple MLP-based embedding network yields non-trivial performance, establishing a strong baseline before waveform modeling. We then adopt embedding-based deep learning models (ArcFace), first on features and then on ECG waveforms, showing a clear performance jump when moving from tabular inputs to waveforms, and a further gain with larger training sets and consistent normalization across train/val/test. On a large-scale test set, verification achieves high TAR at strict FAR thresholds (TAR=0.908 @ FAR=1e-3; TAR=0.820 @ FAR=1e-4) with EER=2.53\\% (all-vs-all); closed-set identification yields Rank@1=0.812 and Rank@10=0.910. In open-set, a two-stage pipeline (top-$K$ shortlist on embeddings + re-ranking) reaches DIR@FAR up to 0.976 at FAR=1e-3 and 1e-4. Overall, the results show that ECG carries a measurable individual signature and that large-scale testing is essential to obtain realistic, comparable metrics. The study provides an operationally grounded benchmark that helps standardize evaluation across protocols.", "date": "2026-02-02", "categories": []}
{"id": "2508.03516", "title": "DSKC: Domain Style Modeling with Adaptive Knowledge Consolidation for Exemplar-free Lifelong Person Re-Identification", "abstract": "Lifelong Person Re-identification (LReID) aims to continuously match individuals across camera views from sequential data streams. Existing LReID methods often ignore domain-specific style awareness and unified knowledge consolidation, which are crucial for mitigating forgetting when adapting to new information. We propose DSKC, a novel rehearsal-free and distillation-free framework for LReID. DSKC designs a domain-style encoder (DSE) to dynamically model domain-specific styles, and a unified knowledge consolidation (UKC) mechanism to adaptively integrate instance-level representations with domain-specific style into a cross-domain unified representation. By leveraging unified representation as a bridge, DSKC explicitly models inter-domain associations at both instance and domain levels to enhance anti-forgetting and generalization. Experimental results demonstrate that our DSKC outperforms state-of-the-art methods in two training orders and enhances the model's strong performance. Our code is available at https://github.com/LiuShiBen/DKUA.", "date": "2025-08-05", "categories": []}
{"id": "2602.03035", "title": "Generalizable and Interpretable RF Fingerprinting with Shapelet-Enhanced Large Language Models", "abstract": "Deep neural networks (DNNs) have achieved remarkable success in radio frequency (RF) fingerprinting for wireless device authentication. However, their practical deployment faces two major limitations: domain shift, where models trained in one environment struggle to generalize to others, and the black-box nature of DNNs, which limits interpretability. To address these issues, we propose a novel framework that integrates a group of variable-length two-dimensional (2D) shapelets with a pre-trained large language model (LLM) to achieve efficient, interpretable, and generalizable RF fingerprinting. The 2D shapelets explicitly capture diverse local temporal patterns across the in-phase and quadrature (I/Q) components, providing compact and interpretable representations. Complementarily, the pre-trained LLM captures more long-range dependencies and global contextual information, enabling strong generalization with minimal training overhead. Moreover, our framework also supports prototype generation for few-shot inference, enhancing cross-domain performance without additional retraining. To evaluate the effectiveness of our proposed method, we conduct extensive experiments on six datasets across various protocols and domains. The results show that our method achieves superior standard and few-shot performance across both source and unseen domains.", "date": "2026-02-03", "categories": []}
{"id": "2505.11884", "title": "Facial Recognition Leveraging Generative Adversarial Networks", "abstract": "Face recognition performance based on deep learning heavily relies on large-scale training data, which is often difficult to acquire in practical applications. To address this challenge, this paper proposes a GAN-based data augmentation method with three key contributions: (1) a residual-embedded generator to alleviate gradient vanishing/exploding problems, (2) an Inception ResNet-V1 based FaceNet discriminator for improved adversarial training, and (3) an end-to-end framework that jointly optimizes data generation and recognition performance. Experimental results demonstrate that our approach achieves stable training dynamics and significantly improves face recognition accuracy by 12.7% on the LFW benchmark compared to baseline methods, while maintaining good generalization capability with limited training samples.", "date": "2025-05-17", "categories": []}
{"id": "2602.00109", "title": "Robustness of Presentation Attack Detection in Remote Identity Validation Scenarios", "abstract": "Presentation attack detection (PAD) subsystems are an important part of effective and user-friendly remote identity validation (RIV) systems. However, ensuring robust performance across diverse environmental and procedural conditions remains a critical challenge. This paper investigates the impact of low-light conditions and automated image acquisition on the robustness of commercial PAD systems using a scenario test of RIV. Our results show that PAD systems experience a significant decline in performance when utilized in low-light or auto-capture scenarios, with a model-predicted increase in error rates by a factor of about four under low-light conditions and a doubling of those odds under auto-capture workflows. Specifically, only one of the tested systems was robust to these perturbations, maintaining a maximum bona fide presentation classification error rate below 3% across all scenarios. Our findings emphasize the importance of testing across diverse environments to ensure robust and reliable PAD performance in real-world applications.", "date": "2026-01-26", "categories": []}
{"id": "2602.01012", "title": "LocalScore: Local Density-Aware Similarity Scoring for Biometrics", "abstract": "Open-set biometrics faces challenges with probe subjects who may not be enrolled in the gallery, as traditional biometric systems struggle to detect these non-mated probes. Despite the growing prevalence of multi-sample galleries in real-world deployments, most existing methods collapse intra-subject variability into a single global representation, leading to suboptimal decision boundaries and poor open-set robustness. To address this issue, we propose LocalScore, a simple yet effective scoring algorithm that explicitly incorporates the local density of the gallery feature distribution using the k-th nearest neighbors. LocalScore is architecture-agnostic, loss-independent, and incurs negligible computational overhead, making it a plug-and-play solution for existing biometric systems. Extensive experiments across multiple modalities demonstrate that LocalScore consistently achieves substantial gains in open-set retrieval (FNIR@FPIR reduced from 53% to 40%) and verification (TAR@FAR improved from 51% to 74%). We further provide theoretical analysis and empirical validation explaining when and why the method achieves the most significant gains based on dataset characteristics.", "date": "2026-02-01", "categories": []}
{"id": "2602.01059", "title": "DRFormer: A Dual-Regularized Bidirectional Transformer for Person Re-identification", "abstract": "Both fine-grained discriminative details and global semantic features can contribute to solving person re-identification challenges, such as occlusion and pose variations. Vision foundation models (\\textit{e.g.}, DINO) excel at mining local textures, and vision-language models (\\textit{e.g.}, CLIP) capture strong global semantic difference. Existing methods predominantly rely on a single paradigm, neglecting the potential benefits of their integration. In this paper, we analyze the complementary roles of these two architectures and propose a framework to synergize their strengths by a \\textbf{D}ual-\\textbf{R}egularized Bidirectional \\textbf{Transformer} (\\textbf{DRFormer}). The dual-regularization mechanism ensures diverse feature extraction and achieves a better balance in the contributions of the two models. Extensive experiments on five benchmarks show that our method effectively harmonizes local and global representations, achieving competitive performance against state-of-the-art methods.", "date": "2026-02-01", "categories": []}
{"id": "2501.09700", "title": "Cueless EEG imagined speech for subject identification: dataset and benchmarks", "abstract": "Electroencephalogram (EEG) signals have emerged as a promising modality for biometric identification. While previous studies have explored the use of imagined speech with semantically meaningful words for subject identification, most have relied on additional visual or auditory cues. In this study, we introduce a cueless EEG-based imagined speech paradigm, where subjects imagine the pronunciation of semantically meaningful words without any external cues. This innovative approach addresses the limitations of prior methods by requiring subjects to select and imagine words from a predefined list naturally. The dataset comprises over 4,350 trials from 11 subjects across five sessions. We assess a variety of classification methods, including traditional machine learning techniques such as Support Vector Machines (SVM) and XGBoost, as well as time-series foundation models and deep learning architectures specifically designed for EEG classification, such as EEG Conformer and Shallow ConvNet. A session-based hold-out validation strategy was employed to ensure reliable evaluation and prevent data leakage. Our results demonstrate outstanding classification accuracy, reaching 97.93%. These findings highlight the potential of cueless EEG paradigms for secure and reliable subject identification in real-world applications, such as brain-computer interfaces (BCIs).", "date": "2025-01-16", "categories": []}
{"id": "2601.20351", "title": "PalmBridge: A Plug-and-Play Feature Alignment Framework for Open-Set Palmprint Verification", "abstract": "Palmprint recognition is widely used in biometric systems, yet real-world performance often degrades due to feature distribution shifts caused by heterogeneous deployment conditions. Most deep palmprint models assume a closed and stationary distribution, leading to overfitting to dataset-specific textures rather than learning domain-invariant representations. Although data augmentation is commonly used to mitigate this issue, it assumes augmented samples can approximate the target deployment distribution, an assumption that often fails under significant domain mismatch. To address this limitation, we propose PalmBridge, a plug-and-play feature-space alignment framework for open-set palmprint verification based on vector quantization. Rather than relying solely on data-level augmentation, PalmBridge learns a compact set of representative vectors directly from training features. During enrollment and verification, each feature vector is mapped to its nearest representative vector under a minimum-distance criterion, and the mapped vector is then blended with the original vector. This design suppresses nuisance variation induced by domain shifts while retaining discriminative identity cues. The representative vectors are jointly optimized with the backbone network using task supervision, a feature-consistency objective, and an orthogonality regularization term to form a stable and well-structured shared embedding space. Furthermore, we analyze feature-to-representative mappings via assignment consistency and collision rate to assess model's sensitivity to blending weights. Experiments on multiple palmprint datasets and backbone architectures show that PalmBridge consistently reduces EER in intra-dataset open-set evaluation and improves cross-dataset generalization with negligible to modest runtime overhead.", "date": "2026-01-28", "categories": []}
{"id": "2601.20598", "title": "Person Re-ID in 2025: Supervised, Self-Supervised, and Language-Aligned. What Works?", "abstract": "Person Re-Identification (ReID) remains a challenging problem in computer vision. This work reviews various training paradigm and evaluates the robustness of state-of-the-art ReID models in cross-domain applications and examines the role of foundation models in improving generalization through richer, more transferable visual representations. We compare three training paradigms, supervised, self-supervised, and language-aligned models. Through the study the aim is to answer the following questions: Can supervised models generalize in cross-domain scenarios? How does foundation models like SigLIP2 perform for the ReID tasks? What are the weaknesses of current supervised and foundational models for ReID? We have conducted the analysis across 11 models and 9 datasets. Our results show a clear split: supervised models dominate their training domain but crumble on cross-domain data. Language-aligned models, however, show surprising robustness cross-domain for ReID tasks, even though they are not explicitly trained to do so. Code and data available at: https://github.com/moiiai-tech/object-reid-benchmark.", "date": "2026-01-28", "categories": []}
{"id": "2601.20656", "title": "FD-MAD: Frequency-Domain Residual Analysis for Face Morphing Attack Detection", "abstract": "Face morphing attacks present a significant threat to face recognition systems used in electronic identity enrolment and border control, particularly in single-image morphing attack detection (S-MAD) scenarios where no trusted reference is available. In spite of the vast amount of research on this problem, morph detection systems struggle in cross-dataset scenarios. To address this problem, we introduce a region-aware frequency-based morph detection strategy that drastically improves over strong baseline methods in challenging cross-dataset and cross-morph settings using a lightweight approach. Having observed the separability of bona fide and morph samples in the frequency domain of different facial parts, our approach 1) introduces the concept of residual frequency domain, where the frequency of the signal is decoupled from the natural spectral decay to easily discriminate between morph and bona fide data; 2) additionally, we reason in a global and local manner by combining the evidence from different facial regions in a Markov Random Field, which infers a globally consistent decision. The proposed method, trained exclusively on the synthetic morphing attack detection development dataset (SMDD), is evaluated in challenging cross-dataset and cross-morph settings on FRLL-Morph and MAD22 sets. Our approach achieves an average equal error rate (EER) of 1.85\\% on FRLL-Morph and ranks second on MAD22 with an average EER of 6.12\\%, while also obtaining a good bona fide presentation classification error rate (BPCER) at a low attack presentation classification error rate (APCER) using only spectral features. These findings indicate that Fourier-domain residual modeling with structured regional fusion offers a competitive alternative to deep S-MAD architectures.", "date": "2026-01-28", "categories": []}
{"id": "2512.14961", "title": "Adaptive Multimodal Person Recognition: A Robust Framework for Handling Missing Modalities", "abstract": "Person identification systems often rely on audio, visual, or behavioral cues, but real-world conditions frequently present with missing or degraded modalities. To address this challenge, we propose a multimodal person identification framework incorporating upper-body motion, face, and voice. Experimental results demonstrate that body motion outperforms traditional modalities such as face and voice in within-session evaluations, while serving as a complementary cue that enhances performance in multi-session scenarios. Our model employs a unified hybrid fusion strategy, fusing both feature-level and score-level information to maximize representational richness and decision accuracy. Specifically, it leverages multi-task learning to process modalities independently, followed by cross-attention and gated fusion mechanisms to exploit both unimodal information and cross-modal interactions. Finally, a confidence-weighted strategy and mistake-correction mechanism dynamically adapt to missing data, ensuring that our single classification head achieves optimal performance even in unimodal and bimodal scenarios. We evaluate our method on CANDOR, a newly introduced interview-based multimodal dataset, which we benchmark in this work for the first time. Our results demonstrate that the proposed trimodal system achieves 99.51% Top-1 accuracy on person identification tasks. In addition, we evaluate our model on the VoxCeleb1 dataset as a widely used evaluation protocol and reach 99.92% accuracy in bimodal mode, outperforming conventional approaches. Moreover, we show that our system maintains high accuracy even when one or two modalities are unavailable, making it a robust solution for real-world person recognition applications. The code and data for this work are publicly available.", "date": "2025-12-16", "categories": []}
{"id": "2601.19133", "title": "QA-ReID: Quality-Aware Query-Adaptive Convolution Leveraging Fused Global and Structural Cues for Clothes-Changing ReID", "abstract": "Unlike conventional person re-identification (ReID), clothes-changing ReID (CC-ReID) presents severe challenges due to substantial appearance variations introduced by clothing changes. In this work, we propose the Quality-Aware Dual-Branch Matching (QA-ReID), which jointly leverages RGB-based features and parsing-based representations to model both global appearance and clothing-invariant structural cues. These heterogeneous features are adaptively fused through a multi-modal attention module. At the matching stage, we further design the Quality-Aware Query Adaptive Convolution (QAConv-QA), which incorporates pixel-level importance weighting and bidirectional consistency constraints to enhance robustness against clothing variations. Extensive experiments demonstrate that QA-ReID achieves state-of-the-art performance on multiple benchmarks, including PRCC, LTCC, and VC-Clothes, and significantly outperforms existing approaches under cross-clothing scenarios.", "date": "2026-01-27", "categories": []}
{"id": "2601.19771", "title": "PaW-ViT: A Patch-based Warping Vision Transformer for Robust Ear Verification", "abstract": "The rectangular tokens common to vision transformer methods for visual recognition can strongly affect performance of these methods due to incorporation of information outside the objects to be recognized. This paper introduces PaW-ViT, Patch-based Warping Vision Transformer, a preprocessing approach rooted in anatomical knowledge that normalizes ear images to enhance the efficacy of ViT. By accurately aligning token boundaries to detected ear feature boundaries, PaW-ViT obtains greater robustness to shape, size, and pose variation. By aligning feature boundaries to natural ear curvature, it produces more consistent token representations for various morphologies. Experiments confirm the effectiveness of PaW-ViT on various ViT models (ViT-T, ViT-S, ViT-B, ViT-L) and yield reasonable alignment robustness to variation in shape, size, and pose. Our work aims to solve the disconnect between ear biometric morphological variation and transformer architecture positional sensitivity, presenting a possible avenue for authentication schemes.", "date": "2026-01-27", "categories": []}
{"id": "2601.19795", "title": "Diffusion for De-Occlusion: Accessory-Aware Diffusion Inpainting for Robust Ear Biometric Recognition", "abstract": "Ear occlusions (arising from the presence of ear accessories such as earrings and earphones) can negatively impact performance in ear-based biometric recognition systems, especially in unconstrained imaging circumstances. In this study, we assess the effectiveness of a diffusion-based ear inpainting technique as a pre-processing aid to mitigate the issues of ear accessory occlusions in transformer-based ear recognition systems. Given an input ear image and an automatically derived accessory mask, the inpainting model reconstructs clean and anatomically plausible ear regions by synthesizing missing pixels while preserving local geometric coherence along key ear structures, including the helix, antihelix, concha, and lobule. We evaluate the effectiveness of this pre-processing aid in transformer-based recognition systems for several vision transformer models and different patch sizes for a range of benchmark datasets. Experiments show that diffusion-based inpainting can be a useful pre-processing aid to alleviate ear accessory occlusions to improve overall recognition performance.", "date": "2026-01-27", "categories": []}
{"id": "2601.16440", "title": "Masked Face Recognition under Different Backbones", "abstract": "Erratum to the paper (Zhang et al., 2025): corrections to Table IV and the data in Page 3, Section A. In the post-pandemic era, a high proportion of civil aviation passengers wear masks during security checks, posing significant challenges to traditional face recognition models. The backbone network serves as the core component of face recognition models. In standard tests, r100 series models excelled (98%+ accuracy at 0.01% FAR in face comparison, high top1/top5 in search). r50 ranked second, r34_mask_v1 lagged. In masked tests, r100_mask_v2 led (90.07% accuracy), r50_mask_v3 performed best among r50 but trailed r100. Vit-Small/Tiny showed strong masked performance with gains in effectiveness. Through extensive comparative experiments, this paper conducts a comprehensive evaluation of several core backbone networks, aiming to reveal the impacts of different models on face recognition with and without masks, and provide specific deployment recommendations.", "date": "2026-01-23", "categories": []}
{"id": "2601.15406", "title": "Evaluating Multimodal Large Language Models for Heterogeneous Face Recognition", "abstract": "Multimodal Large Language Models (MLLMs) have recently demonstrated strong performance on a wide range of vision-language tasks, raising interest in their potential use for biometric applications. In this paper, we conduct a systematic evaluation of state-of-the-art MLLMs for heterogeneous face recognition (HFR), where enrollment and probe images are from different sensing modalities, including visual (VIS), near infrared (NIR), short-wave infrared (SWIR), and thermal camera. We benchmark multiple open-source MLLMs across several cross-modality scenarios, including VIS-NIR, VIS-SWIR, and VIS-THERMAL face recognition. The recognition performance of MLLMs is evaluated using biometric protocols and based on different metrics, including Acquire Rate, Equal Error Rate (EER), and True Accept Rate (TAR). Our results reveal substantial performance gaps between MLLMs and classical face recognition systems, particularly under challenging cross-spectral conditions, in spite of recent advances in MLLMs. Our findings highlight the limitations of current MLLMs for HFR and also the importance of rigorous biometric evaluation when considering their deployment in face recognition systems.", "date": "2026-01-21", "categories": []}
{"id": "2406.14261", "title": "Unleashing the Potential of Tracklets for Unsupervised Video Person Re-Identification", "abstract": "With rich temporal-spatial information, video-based person re-identification methods have shown broad prospects. Although tracklets can be easily obtained with ready-made tracking models, annotating identities is still expensive and impractical. Therefore, some video-based methods propose using only a few identity annotations or camera labels to facilitate feature learning. They also simply average the frame features of each tracklet, overlooking unexpected variations and inherent identity consistency within tracklets. In this paper, we propose the Self-Supervised Refined Clustering (SSR-C) framework without relying on any annotation or auxiliary information to promote unsupervised video person re-identification. Specifically, we first propose the Noise-Filtered Tracklet Partition (NFTP) module to reduce the feature bias of tracklets caused by noisy tracking results, and sequentially partition the noise-filtered tracklets into \"sub-tracklets\". Then, we cluster and further merge sub-tracklets using the self-supervised signal from the tracklet partition, which is enhanced through a progressive strategy to generate reliable pseudo labels, facilitating intra-class cross-tracklet aggregation. Moreover, we propose the Class Smoothing Classification (CSC) loss to efficiently promote model learning. Extensive experiments on the MARS and DukeMTMC-VideoReID datasets demonstrate that our proposed SSR-C for unsupervised video person re-identification achieves state-of-the-art results and is comparable to advanced supervised methods. The code is available at https://github.com/Darylmeng/SSRC-Reid.", "date": "2024-06-20", "categories": []}
{"id": "2601.11700", "title": "Telling Human and Machine Handwriting Apart", "abstract": "Handwriting movements can be leveraged as a unique form of behavioral biometrics, to verify whether a real user is operating a device or application. This task can be framed as a reverse Turing test in which a computer has to detect if an input instance has been generated by a human or artificially. To tackle this task, we study ten public datasets of handwritten symbols (isolated characters, digits, gestures, pointing traces, and signatures) that are artificially reproduced using seven different synthesizers, including, among others, the Kinematic Theory (Sigma h model), generative adversarial networks, Transformers, and Diffusion models. We train a shallow recurrent neural network that achieves excellent performance (98.3 percent Area Under the ROC Curve (AUC) score and 1.4 percent equal error rate on average across all synthesizers and datasets) using nonfeaturized trajectory data as input. In few-shot settings, we show that our classifier achieves such an excellent performance when trained on just 10 percent of the data, as evaluated on the remaining 90% of the data as a test set. We further challenge our classifier in out-of-domain settings, and observe very competitive results as well. Our work has implications for computerized systems that need to verify human presence, and adds an additional layer of security to keep attackers at bay.", "date": "2026-01-16", "categories": []}
{"id": "2601.12062", "title": "Learning Language-Driven Sequence-Level Modal-Invariant Representations for Video-Based Visible-Infrared Person Re-Identification", "abstract": "The core of video-based visible-infrared person re-identification (VVI-ReID) lies in learning sequence-level modal-invariant representations across different modalities. Recent research tends to use modality-shared language prompts generated by CLIP to guide the learning of modal-invariant representations. Despite achieving optimal performance, such methods still face limitations in efficient spatial-temporal modeling, sufficient cross-modal interaction, and explicit modality-level loss guidance. To address these issues, we propose the language-driven sequence-level modal-invariant representation learning (LSMRL) method, which includes spatial-temporal feature learning (STFL) module, semantic diffusion (SD) module and cross-modal interaction (CMI) module. To enable parameter- and computation-efficient spatial-temporal modeling, the STFL module is built upon CLIP with minimal modifications. To achieve sufficient cross-modal interaction and enhance the learning of modal-invariant features, the SD module is proposed to diffuse modality-shared language prompts into visible and infrared features to establish preliminary modal consistency. The CMI module is further developed to leverage bidirectional cross-modal self-attention to eliminate residual modality gaps and refine modal-invariant representations. To explicitly enhance the learning of modal-invariant representations, two modality-level losses are introduced to improve the features' discriminative ability and their generalization to unseen categories. Extensive experiments on large-scale VVI-ReID datasets demonstrate the superiority of LSMRL over AOTA methods.", "date": "2026-01-17", "categories": []}
{"id": "2601.11243", "title": "Image-Text Knowledge Modeling for Unsupervised Multi-Scenario Person Re-Identification", "abstract": "We propose unsupervised multi-scenario (UMS) person re-identification (ReID) as a new task that expands ReID across diverse scenarios (cross-resolution, clothing change, etc.) within a single coherent framework. To tackle UMS-ReID, we introduce image-text knowledge modeling (ITKM) -- a three-stage framework that effectively exploits the representational power of vision-language models. We start with a pre-trained CLIP model with an image encoder and a text encoder. In Stage I, we introduce a scenario embedding in the image encoder and fine-tune the encoder to adaptively leverage knowledge from multiple scenarios. In Stage II, we optimize a set of learned text embeddings to associate with pseudo-labels from Stage I and introduce a multi-scenario separation loss to increase the divergence between inter-scenario text representations. In Stage III, we first introduce cluster-level and instance-level heterogeneous matching modules to obtain reliable heterogeneous positive pairs (e.g., a visible image and an infrared image of the same person) within each scenario. Next, we propose a dynamic text representation update strategy to maintain consistency between text and image supervision signals. Experimental results across multiple scenarios demonstrate the superiority and generalizability of ITKM; it not only outperforms existing scenario-specific methods but also enhances overall performance by integrating knowledge from multiple scenarios.", "date": "2026-01-16", "categories": []}
{"id": "2510.17035", "title": "Conditional Synthetic Live and Spoof Fingerprint Generation", "abstract": "Large fingerprint datasets, while important for training and evaluation, are time-consuming and expensive to collect and require strict privacy measures. Researchers are exploring the use of synthetic fingerprint data to address these issues. This paper presents a novel approach for generating synthetic fingerprint images (both spoof and live), addressing concerns related to privacy, cost, and accessibility in biometric data collection. Our approach utilizes conditional StyleGAN2-ADA and StyleGAN3 architectures to produce high-resolution synthetic live fingerprints, conditioned on specific finger identities (thumb through little finger). Additionally, we employ CycleGANs to translate these into realistic spoof fingerprints, simulating a variety of presentation attack materials (e.g., EcoFlex, Play-Doh). These synthetic spoof fingerprints are crucial for developing robust spoof detection systems. Through these generative models, we created two synthetic datasets (DB2 and DB3), each containing 1,500 fingerprint images of all ten fingers with multiple impressions per finger, and including corresponding spoofs in eight material types. The results indicate robust performance: our StyleGAN3 model achieves a Fr\\'echet Inception Distance (FID) as low as 5, and the generated fingerprints achieve a True Accept Rate of 99.47% at a 0.01% False Accept Rate. The StyleGAN2-ADA model achieved a TAR of 98.67% at the same 0.01% FAR. We assess fingerprint quality using standard metrics (NFIQ2, MINDTCT), and notably, matching experiments confirm strong privacy preservation, with no significant evidence of identity leakage, confirming the strong privacy-preserving properties of our synthetic datasets.", "date": "2025-10-19", "categories": []}
{"id": "2601.09229", "title": "SPOT-Face: Forensic Face Identification using Attention Guided Optimal Transport", "abstract": "Person identification in forensic investigations becomes very challenging when common identification means for DNA (i.e., hair strands, soft tissue) are not available. Current methods utilize deep learning methods for face recognition. However, these methods lack effective mechanisms to model cross-domain structural correspondence between two different forensic modalities. In this paper, we introduce a SPOT-Face, a superpixel graph-based framework designed for cross-domain forensic face identification of victims using their skeleton and sketch images. Our unified framework involves constructing a superpixel-based graph from an image and then using different graph neural networks(GNNs) backbones to extract the embeddings of these graphs, while cross-domain correspondence is established through attention-guided optimal transport mechanism. We have evaluated our proposed framework on two publicly available dataset: IIT\\_Mandi\\_S2F (S2F) and CUFS. Extensive experiments were conducted to evaluate our proposed framework. The experimental results show significant improvement in identification metrics ( i.e., Recall, mAP) over existing graph-based baselines. Furthermore, our framework demonstrates to be highly effective for matching skulls and sketches to faces in forensic investigations.", "date": "2026-01-14", "categories": []}
{"id": "2601.08807", "title": "S3-CLIP: Video Super Resolution for Person-ReID", "abstract": "Tracklet quality is often treated as an afterthought in most person re-identification (ReID) methods, with the majority of research presenting architectural modifications to foundational models. Such approaches neglect an important limitation, posing challenges when deploying ReID systems in real-world, difficult scenarios. In this paper, we introduce S3-CLIP, a video super-resolution-based CLIP-ReID framework developed for the VReID-XFD challenge at WACV 2026. The proposed method integrates recent advances in super-resolution networks with task-driven super-resolution pipelines, adapting them to the video-based person re-identification setting. To the best of our knowledge, this work represents the first systematic investigation of video super-resolution as a means of enhancing tracklet quality for person ReID, particularly under challenging cross-view conditions. Experimental results demonstrate performance competitive with the baseline, achieving 37.52% mAP in aerial-to-ground and 29.16% mAP in ground-to-aerial scenarios. In the ground-to-aerial setting, S3-CLIP achieves substantial gains in ranking accuracy, improving Rank-1, Rank-5, and Rank-10 performance by 11.24%, 13.48%, and 17.98%, respectively.", "date": "2026-01-13", "categories": []}
{"id": "2601.06218", "title": "Two-step Authentication: Multi-biometric System Using Voice and Facial Recognition", "abstract": "We present a cost-effective two-step authentication system that integrates face identification and speaker verification using only a camera and microphone available on common devices. The pipeline first performs face recognition to identify a candidate user from a small enrolled group, then performs voice recognition only against the matched identity to reduce computation and improve robustness. For face recognition, a pruned VGG-16 based classifier is trained on an augmented dataset of 924 images from five subjects, with faces localized by MTCNN; it achieves 95.1% accuracy. For voice recognition, a CNN speaker-verification model trained on LibriSpeech (train-other-360) attains 98.9% accuracy and 3.456% EER on test-clean. Source code and trained models are available at https://github.com/NCUE-EE-AIAL/Two-step-Authentication-Multi-biometric-System.", "date": "2026-01-09", "categories": []}
{"id": "2601.06239", "title": "A survey of facial recognition techniques", "abstract": "As multimedia content is quickly growing, the field of facial recognition has become one of the major research fields, particularly in the recent years. The most problematic area to researchers in image processing and computer vision is the human face which is a complex object with myriads of distinctive features that can be used to identify the face. The survey of this survey is particularly focused on most challenging facial characteristics, including differences in the light, ageing, variation in poses, partial occlusion, and facial expression and presents methodological solutions. The factors, therefore, are inevitable in the creation of effective facial recognition mechanisms used on facial images. This paper reviews the most sophisticated methods of facial detection which are Hidden Markov Models, Principal Component Analysis (PCA), Elastic Cluster Plot Matching, Support Vector Machine (SVM), Gabor Waves, Artificial Neural Networks (ANN), Eigenfaces, Independent Component Analysis (ICA), and 3D Morphable Model. Alongside the works mentioned above, we have also analyzed the images of a number of facial databases, namely JAFEE, FEI, Yale, LFW, AT&T (then called ORL), and AR (created by Martinez and Benavente), to analyze the results. However, this survey is aimed at giving a thorough literature review of face recognition, and its applications, and some experimental results are provided at the end after a detailed discussion.", "date": "2026-01-09", "categories": []}
{"id": "2601.06725", "title": "When Humans Judge Irises: Pupil Size Normalization as an Aid and Synthetic Irises as a Challenge", "abstract": "Iris recognition is a mature biometric technology offering remarkable precision and speed, and allowing for large-scale deployments to populations exceeding a billion enrolled users (e.g., AADHAAR in India). However, in forensic applications, a human expert may be needed to review and confirm a positive identification before an iris matching result can be presented as evidence in court, especially in cases where processed samples are degraded (e.g., in post-mortem cases) or where there is a need to judge whether the sample is authentic, rather than a result of a presentation attack.\n  This paper presents a study that examines human performance in iris verification in two controlled scenarios: (a) under varying pupil sizes, with and without a linear/nonlinear alignment of the pupil size between compared images, and (b) when both genuine and impostor iris image pairs are synthetically generated. The results demonstrate that pupil size normalization carried out by a modern autoencoder-based identity-preserving image-to-image translation model significantly improves verification accuracy. Participants were also able to determine whether iris pairs corresponded to the same or different eyes when both images were either authentic or synthetic. However, accuracy declined when subjects were comparing authentic irises against high-quality, same-eye synthetic counterparts. These findings (a) demonstrate the importance of pupil-size alignment for iris matching tasks in which humans are involved, and (b) indicate that despite the high fidelity of modern generative models, same-eye synthetic iris images are more often judged by humans as different-eye images, compared to same-eye authentic image pairs.\n  We offer data and human judgments along with this paper to allow full replicability of this study and future works.", "date": "2026-01-11", "categories": []}
{"id": "2601.05535", "title": "SAS-VPReID: A Scale-Adaptive Framework with Shape Priors for Video-based Person Re-Identification at Extreme Far Distances", "abstract": "Video-based Person Re-IDentification (VPReID) aims to retrieve the same person from videos captured by non-overlapping cameras. At extreme far distances, VPReID is highly challenging due to severe resolution degradation, drastic viewpoint variation and inevitable appearance noise. To address these issues, we propose a Scale-Adaptive framework with Shape Priors for VPReID, named SAS-VPReID. The framework is built upon three complementary modules. First, we deploy a Memory-Enhanced Visual Backbone (MEVB) to extract discriminative feature representations, which leverages the CLIP vision encoder and multi-proxy memory. Second, we propose a Multi-Granularity Temporal Modeling (MGTM) to construct sequences at multiple temporal granularities and adaptively emphasize motion cues across scales. Third, we incorporate Prior-Regularized Shape Dynamics (PRSD) to capture body structure dynamics. With these modules, our framework can obtain more discriminative feature representations. Experiments on the VReID-XFD benchmark demonstrate the effectiveness of each module and our final framework ranks the first on the VReID-XFD challenge leaderboard. The source code is available at https://github.com/YangQiWei3/SAS-VPReID.", "date": "2026-01-09", "categories": []}
{"id": "2601.01312", "title": "VReID-XFD: Video-based Person Re-identification at Extreme Far Distance Challenge Results", "abstract": "Person re-identification (ReID) across aerial and ground views at extreme far distances introduces a distinct operating regime where severe resolution degradation, extreme viewpoint changes, unstable motion cues, and clothing variation jointly undermine the appearance-based assumptions of existing ReID systems. To study this regime, we introduce VReID-XFD, a video-based benchmark and community challenge for extreme far-distance (XFD) aerial-to-ground person re-identification. VReID-XFD is derived from the DetReIDX dataset and comprises 371 identities, 11,288 tracklets, and 11.75 million frames, captured across altitudes from 5.8 m to 120 m, viewing angles from oblique (30 degrees) to nadir (90 degrees), and horizontal distances up to 120 m. The benchmark supports aerial-to-aerial, aerial-to-ground, and ground-to-aerial evaluation under strict identity-disjoint splits, with rich physical metadata. The VReID-XFD-25 Challenge attracted 10 teams with hundreds of submissions. Systematic analysis reveals monotonic performance degradation with altitude and distance, a universal disadvantage of nadir views, and a trade-off between peak performance and robustness. Even the best-performing SAS-PReID method achieves only 43.93 percent mAP in the aerial-to-ground setting. The dataset, annotations, and official evaluation protocols are publicly available at https://www.it.ubi.pt/DetReIDX/ .", "date": "2026-01-04", "categories": []}
{"id": "2601.01356", "title": "Advanced Machine Learning Approaches for Enhancing Person Re-Identification Performance", "abstract": "Person re-identification (ReID) plays a critical role in intelligent surveillance systems by linking identities across multiple cameras in complex environments. However, ReID faces significant challenges such as appearance variations, domain shifts, and limited labeled data. This dissertation proposes three advanced approaches to enhance ReID performance under supervised, unsupervised domain adaptation (UDA), and fully unsupervised settings. First, SCM-ReID integrates supervised contrastive learning with hybrid loss optimization (classification, center, triplet, and centroid-triplet losses), improving discriminative feature representation and achieving state-of-the-art accuracy on Market-1501 and CUHK03 datasets. Second, for UDA, IQAGA and DAPRH combine GAN-based image augmentation, domain-invariant mapping, and pseudo-label refinement to mitigate domain discrepancies and enhance cross-domain generalization. Experiments demonstrate substantial gains over baseline methods, with mAP and Rank-1 improvements up to 12% in challenging transfer scenarios. Finally, ViTC-UReID leverages Vision Transformer-based feature encoding and camera-aware proxy learning to boost unsupervised ReID. By integrating global and local attention with camera identity constraints, this method significantly outperforms existing unsupervised approaches on large-scale benchmarks. Comprehensive evaluations across CUHK03, Market-1501, DukeMTMC-reID, and MSMT17 confirm the effectiveness of the proposed methods. The contributions advance ReID research by addressing key limitations in feature learning, domain adaptation, and label noise handling, paving the way for robust deployment in real-world surveillance systems.", "date": "2026-01-04", "categories": []}
{"id": "2601.01680", "title": "Evaluating Deep Learning-Based Face Recognition for Infants and Toddlers: Impact of Age Across Developmental Stages", "abstract": "Face recognition for infants and toddlers presents unique challenges due to rapid facial morphology changes, high inter-class similarity, and limited dataset availability. This study evaluates the performance of four deep learning-based face recognition models FaceNet, ArcFace, MagFace, and CosFace on a newly developed longitudinal dataset collected over a 24 month period in seven sessions involving children aged 0 to 3 years. Our analysis examines recognition accuracy across developmental stages, showing that the True Accept Rate (TAR) is only 30.7% at 0.1% False Accept Rate (FAR) for infants aged 0 to 6 months, due to unstable facial features. Performance improves significantly in older children, reaching 64.7% TAR at 0.1% FAR in the 2.5 to 3 year age group. We also evaluate verification performance over different time intervals, revealing that shorter time gaps result in higher accuracy due to reduced embedding drift. To mitigate this drift, we apply a Domain Adversarial Neural Network (DANN) approach that improves TAR by over 12%, yielding features that are more temporally stable and generalizable. These findings are critical for building biometric systems that function reliably over time in smart city applications such as public healthcare, child safety, and digital identity services. The challenges observed in early age groups highlight the importance of future research on privacy preserving biometric authentication systems that can address temporal variability, particularly in secure and regulated urban environments where child verification is essential.", "date": "2026-01-04", "categories": []}
{"id": "2601.01689", "title": "Mitigating Longitudinal Performance Degradation in Child Face Recognition Using Synthetic Data", "abstract": "Longitudinal face recognition in children remains challenging due to rapid and nonlinear facial growth, which causes template drift and increasing verification errors over time. This work investigates whether synthetic face data can act as a longitudinal stabilizer by improving temporal robustness of child face recognition models. Using an identity disjoint protocol on the Young Face Aging (YFA) dataset, we evaluate three settings: (i) pretrained MagFace embeddings without dataset specific fine-tuning, (ii) MagFace fine-tuned using authentic training faces only, and (iii) MagFace fine-tuned using a combination of authentic and synthetically generated training faces. Synthetic data is generated using StyleGAN2 ADA and incorporated exclusively within the training identities; a post generation filtering step is applied to mitigate identity leakage and remove artifact affected samples. Experimental results across enrollment verification gaps from 6 to 36 months show that synthetic-augmented fine tuning substantially reduces error rates relative to both the pretrained baseline and real only fine tuning. These findings provide a risk aware assessment of synthetic augmentation for improving identity persistence in pediatric face recognition.", "date": "2026-01-04", "categories": []}
{"id": "2601.00170", "title": "Hear the Heartbeat in Phases: Physiologically Grounded Phase-Aware ECG Biometrics", "abstract": "Electrocardiography (ECG) is adopted for identity authentication in wearable devices due to its individual-specific characteristics and inherent liveness. However, existing methods often treat heartbeats as homogeneous signals, overlooking the phase-specific characteristics within the cardiac cycle. To address this, we propose a Hierarchical Phase-Aware Fusion~(HPAF) framework that explicitly avoids cross-feature entanglement through a three-stage design. In the first stage, Intra-Phase Representation (IPR) independently extracts representations for each cardiac phase, ensuring that phase-specific morphological and variation cues are preserved without interference from other phases. In the second stage, Phase-Grouped Hierarchical Fusion (PGHF) aggregates physiologically related phases in a structured manner, enabling reliable integration of complementary phase information. In the final stage, Global Representation Fusion (GRF) further combines the grouped representations and adaptively balances their contributions to produce a unified and discriminative identity representation. Moreover, considering ECG signals are continuously acquired, multiple heartbeats can be collected for each individual. We propose a Heartbeat-Aware Multi-prototype (HAM) enrollment strategy, which constructs a multi-prototype gallery template set to reduce the impact of heartbeat-specific noise and variability. Extensive experiments on three public datasets demonstrate that HPAF achieves state-of-the-art results in the comparison with other methods under both closed and open-set settings.", "date": "2026-01-01", "categories": []}
{"id": "2601.00307", "title": "VisNet: Efficient Person Re-Identification via Alpha-Divergence Loss, Feature Fusion and Dynamic Multi-Task Learning", "abstract": "Person re-identification (ReID) is an extremely important area in both surveillance and mobile applications, requiring strong accuracy with minimal computational cost. State-of-the-art methods give good accuracy but with high computational budgets. To remedy this, this paper proposes VisNet, a computationally efficient and effective re-identification model suitable for real-world scenarios. It is the culmination of conceptual contributions, including feature fusion at multiple scales with automatic attention on each, semantic clustering with anatomical body partitioning, a dynamic weight averaging technique to balance classification semantic regularization, and the use of loss function FIDI for improved metric learning tasks. The multiple scales fuse ResNet50's stages 1 through 4 without the use of parallel paths, with semantic clustering introducing spatial constraints through the use of rule-based pseudo-labeling. VisNet achieves 87.05% Rank-1 and 77.65% mAP on the Market-1501 dataset, having 32.41M parameters and 4.601 GFLOPs, hence, proposing a practical approach for real-time deployment in surveillance and mobile applications where computational resources are limited.", "date": "2026-01-01", "categories": []}
{"id": "2206.13356", "title": "Effective Online Exam Proctoring by Combining Lightweight Face Detection and Deep Recognition", "abstract": "Online exams conducted via video conferencing platforms such as Zoom have become widespread, yet ensuring exam integrity remains challenging due to the difficulty of monitoring multiple video feeds in real time. We present iExam, an online exam proctoring and analysis system that combines lightweight real-time face detection with deep face recognition for postexam analysis. iExam assists invigilators by monitoring student presence during exams and identifies abnormal behaviors, such as face disappearance, face rotation, and identity substitution, from recorded videos. The system addresses three key challenges: (i)efficient real-time video capture and analysis, (ii) automated student identity labeling using enhanced OCR on dynamic Zoom name tags, and (iii) resource-efficient training and inference on standard teacher devices. Extensive experiments show that iExam achieves 90.4% accuracy in real-time face detection and 98.4% accuracy in post-exam recognition with low overhead, demonstrating its practicality and effectiveness for online exam proctoring.", "date": "2022-06-27", "categories": []}
{"id": "2512.25000", "title": "Bi-C2R: Bidirectional Continual Compatible Representation for Re-indexing Free Lifelong Person Re-identification", "abstract": "Lifelong person Re-IDentification (L-ReID) exploits sequentially collected data to continuously train and update a ReID model, focusing on the overall performance of all data. Its main challenge is to avoid the catastrophic forgetting problem of old knowledge while training on new data. Existing L-ReID methods typically re-extract new features for all historical gallery images for inference after each update, known as \"re-indexing\". However, historical gallery data typically suffers from direct saving due to the data privacy issue and the high re-indexing costs for large-scale gallery images. As a result, it inevitably leads to incompatible retrieval between query features extracted by the updated model and gallery features extracted by those before the update, greatly impairing the re-identification performance. To tackle the above issue, this paper focuses on a new task called Re-index Free Lifelong person Re-IDentification (RFL-ReID), which requires performing lifelong person re-identification without re-indexing historical gallery images. Therefore, RFL-ReID is more challenging than L-ReID, requiring continuous learning and balancing new and old knowledge in diverse streaming data, and making the features output by the new and old models compatible with each other. To this end, we propose a Bidirectional Continuous Compatible Representation (Bi-C2R) framework to continuously update the gallery features extracted by the old model to perform efficient L-ReID in a compatible manner. We verify our proposed Bi-C2R method through theoretical analysis and extensive experiments on multiple benchmarks, which demonstrate that the proposed method can achieve leading performance on both the introduced RFL-ReID task and the traditional L-ReID task.", "date": "2025-12-31", "categories": []}
{"id": "2412.02198", "title": "Age-Defying Face Recognition with Transformer-Enhanced Loss", "abstract": "Aging presents a significant challenge in face recognition, as changes in skin texture and tone can alter facial features over time, making it particularly difficult to compare images of the same individual taken years apart, such as in long-term identification scenarios. Transformer networks have the strength to preserve sequential spatial relationships caused by aging effect. This paper presents a technique for loss evaluation that uses a transformer network as an additive loss in the face recognition domain. The standard metric loss function typically takes the final embedding of the main CNN backbone as its input. Here, we employ a transformer-metric loss, a combined approach that integrates both transformer-loss and metric-loss. This research intends to analyze the transformer behavior on the convolution output when the CNN outcome is arranged in a sequential vector. These sequential vectors have the potential to overcome the texture or regional structure referred to as wrinkles or sagging skin affected by aging. The transformer encoder takes input from the contextual vectors obtained from the final convolution layer of the network. The learned features can be more age-invariant, complementing the discriminative power of the standard metric loss embedding. With this technique, we use transformer loss with various base metric-loss functions to evaluate the effect of the combined loss functions. We observe that such a configuration allows the network to achieve SoTA results in LFW and age-variant datasets (CA-LFW and AgeDB). This research expands the role of transformers in the machine vision domain and opens new possibilities for exploring transformers as a loss function.", "date": "2024-12-03", "categories": []}
{"id": "2511.19067", "title": "DynaMix: Generalizable Person Re-identification via Dynamic Relabeling and Mixed Data Sampling", "abstract": "Generalizable person re-identification (Re-ID) aims to recognize individuals across unseen cameras and environments. While existing methods rely heavily on limited labeled multi-camera data, we propose DynaMix, a novel method that effectively combines manually labeled multi-camera and large-scale pseudo-labeled single-camera data. Unlike prior works, DynaMix dynamically adapts to the structure and noise of the training data through three core components: (1) a Relabeling Module that refines pseudo-labels of single-camera identities on-the-fly; (2) an Efficient Centroids Module that maintains robust identity representations under a large identity space; and (3) a Data Sampling Module that carefully composes mixed data mini-batches to balance learning complexity and intra-batch diversity. All components are specifically designed to operate efficiently at scale, enabling effective training on millions of images and hundreds of thousands of identities. Extensive experiments demonstrate that DynaMix consistently outperforms state-of-the-art methods in generalizable person Re-ID.", "date": "2025-11-24", "categories": []}
{"id": "2505.08961", "title": "Learning Informative Attention Weights for Person Re-Identification", "abstract": "Attention mechanisms have been widely used in deep learning, and recent efforts have been devoted to incorporating attention modules into deep neural networks (DNNs) for person Re-Identification (Re-ID) to enhance their discriminative feature learning capabilities. Existing attention modules, including self-attention and channel attention, learn attention weights that quantify the importance of feature tokens or feature channels. However, existing attention methods do not explicitly ensure that the attention weights are informative for predicting the identity of the person in the input image, and may consequently introduce noisy information from the input image. To address this issue, we propose a novel method termed Reduction of Information Bottleneck loss (RIB), motivated by the principle of the Information Bottleneck (IB). A novel distribution-free and efficient variational upper bound for the IB loss (IBB), which can be optimized by standard SGD, is derived and incorporated into the training loss of the RIB models. RIB is applied to DNNs with self-attention modules through a novel Differentiable Channel Selection Attention module, or DCS-Attention, that selects the most informative channels for computing attention weights, leading to competitive models termed RIB-DCS. RIB is also incorporated into DNNs with existing channel attention modules to promote the learning of informative channel attention weights, leading to models termed RIB-CA. Both RIB-DCS and RIB-CA are applied to fixed neural network backbones and learnable backbones with Differentiable Neural Architecture Search (DNAS). Extensive experiments on multiple person Re-ID benchmarks show that RIB significantly enhances the prediction accuracy of DNNs for person Re-ID, even for the occluded person Re-ID.", "date": "2025-05-13", "categories": []}
{"id": "2508.06831", "title": "Low-Rank Expert Merging for Multi-Source Domain Adaptation in Person Re-Identification", "abstract": "Adapting person re-identification (reID) models to new target environments remains a challenging problem that is typically addressed using unsupervised domain adaptation (UDA) methods. Recent works show that when labeled data originates from several distinct sources (e.g., datasets and cameras), considering each source separately and applying multi-source domain adaptation (MSDA) typically yields higher accuracy and robustness compared to blending the sources and performing conventional UDA. However, state-of-the-art MSDA methods learn domain-specific backbone models or require access to source domain data during adaptation, resulting in significant growth in training parameters and computational cost. In this paper, a Source-free Adaptive Gated Experts (SAGE-reID) method is introduced for person reID. Our SAGE-reID is a cost-effective, source-free MSDA method that first trains individual source-specific low-rank adapters (LoRA) through source-free UDA. Next, a lightweight gating network is introduced and trained to dynamically assign optimal merging weights for fusion of LoRA experts, enabling effective cross-domain knowledge transfer. While the number of backbone parameters remains constant across source domains, LoRA experts scale linearly but remain negligible in size (<= 2% of the backbone), reducing both the memory consumption and risk of overfitting. Extensive experiments conducted on three challenging benchmarks: Market-1501, DukeMTMC-reID, and MSMT17 indicate that SAGE-reID outperforms state-of-the-art methods while being computationally efficient.", "date": "2025-08-09", "categories": []}
{"id": "2512.18057", "title": "FOODER: Real-time Facial Authentication and Expression Recognition", "abstract": "Out-of-distribution (OOD) detection is essential for the safe deployment of neural networks, as it enables the identification of samples outside the training domain. We present FOODER, a real-time, privacy-preserving radar-based framework that integrates OOD-based facial authentication with facial expression recognition. FOODER operates using low-cost frequency-modulated continuous-wave (FMCW) radar and exploits both range-Doppler and micro range-Doppler representations. The authentication module employs a multi-encoder multi-decoder architecture with Body Part (BP) and Intermediate Linear Encoder-Decoder (ILED) components to classify a single enrolled individual as in-distribution while detecting all other faces as OOD. Upon successful authentication, an expression recognition module is activated. Concatenated radar representations are processed by a ResNet block to distinguish between dynamic and static facial expressions. Based on this categorization, two specialized MobileViT networks are used to classify dynamic expressions (smile, shock) and static expressions (neutral, anger). This hierarchical design enables robust facial authentication and fine-grained expression recognition while preserving user privacy by relying exclusively on radar data. Experiments conducted on a dataset collected with a 60 GHz short-range FMCW radar demonstrate that FOODER achieves an AUROC of 94.13% and an FPR95 of 18.12% for authentication, along with an average expression recognition accuracy of 94.70%. FOODER outperforms state-of-the-art OOD detection methods and several transformer-based architectures while operating efficiently in real time.", "date": "2025-12-19", "categories": []}
{"id": "2512.19275", "title": "Is Visual Realism Enough? Evaluating Gait Biometric Fidelity in Generative AI Human Animation", "abstract": "Generative AI (GenAI) models have revolutionized animation, enabling the synthesis of humans and motion patterns with remarkable visual fidelity. However, generating truly realistic human animation remains a formidable challenge, where even minor inconsistencies can make a subject appear unnatural. This limitation is particularly critical when AI-generated videos are evaluated for behavioral biometrics, where subtle motion cues that define identity are easily lost or distorted. The present study investigates whether state-of-the-art GenAI human animation models can preserve the subtle spatio-temporal details needed for person identification through gait biometrics. Specifically, we evaluate four different GenAI models across two primary evaluation tasks to assess their ability to i) restore gait patterns from reference videos under varying conditions of complexity, and ii) transfer these gait patterns to different visual identities. Our results show that while visual quality is mostly high, biometric fidelity remains low in tasks focusing on identification, suggesting that current GenAI models struggle to disentangle identity from motion. Furthermore, through an identity transfer task, we expose a fundamental flaw in appearance-based gait recognition: when texture is disentangled from motion, identification collapses, proving current GenAI models rely on visual attributes rather than temporal dynamics.", "date": "2025-12-22", "categories": []}
{"id": "2512.16648", "title": "Exploiting Radio Frequency Fingerprints for Device Identification: Tackling Cross-receiver Challenges in the Source-data-free Scenario", "abstract": "With the rapid proliferation of edge computing, Radio Frequency Fingerprint Identification (RFFI) has become increasingly important for secure device authentication. However, practical deployment of deep learning-based RFFI models is hindered by a critical challenge: their performance often degrades significantly when applied across receivers with different hardware characteristics due to distribution shifts introduced by receiver variation. To address this, we investigate the source-data-free cross-receiver RFFI (SCRFFI) problem, where a model pretrained on labeled signals from a source receiver must adapt to unlabeled signals from a target receiver, without access to any source-domain data during adaptation. We first formulate a novel constrained pseudo-labeling-based SCRFFI adaptation framework, and provide a theoretical analysis of its generalization performance. Our analysis highlights a key insight: the target-domain performance is highly sensitive to the quality of the pseudo-labels generated during adaptation. Motivated by this, we propose Momentum Soft pseudo-label Source Hypothesis Transfer (MS-SHOT), a new method for SCRFFI that incorporates momentum-center-guided soft pseudo-labeling and enforces global structural constraints to encourage confident and diverse predictions. Notably, MS-SHOT effectively addresses scenarios involving label shift or unknown, non-uniform class distributions in the target domain -- a significant limitation of prior methods. Extensive experiments on real-world datasets demonstrate that MS-SHOT consistently outperforms existing approaches in both accuracy and robustness, offering a practical and scalable solution for source-data-free cross-receiver adaptation in RFFI.", "date": "2025-12-18", "categories": []}
{"id": "2512.14878", "title": "Visual-textual Dermatoglyphic Animal Biometrics: A First Case Study on Panthera tigris", "abstract": "Biologists have long combined visuals with textual field notes to re-identify (Re-ID) animals. Contemporary AI tools automate this for species with distinctive morphological features but remain largely image-based. Here, we extend Re-ID methodologies by incorporating precise dermatoglyphic textual descriptors-an approach used in forensics but new to ecology. We demonstrate that these specialist semantics abstract and encode animal coat topology using human-interpretable language tags. Drawing on 84,264 manually labelled minutiae across 3,355 images of 185 tigers (Panthera tigris), we evaluate this visual-textual methodology, revealing novel capabilities for cross-modal identity retrieval. To optimise performance, we developed a text-image co-synthesis pipeline to generate 'virtual individuals', each comprising dozens of life-like visuals paired with dermatoglyphic text. Benchmarking against real-world scenarios shows this augmentation significantly boosts AI accuracy in cross-modal retrieval while alleviating data scarcity. We conclude that dermatoglyphic language-guided biometrics can overcome vision-only limitations, enabling textual-to-visual identity recovery underpinned by human-verifiable matchings. This represents a significant advance towards explainability in Re-ID and a language-driven unification of descriptive modalities in ecological monitoring.", "date": "2025-12-16", "categories": []}
{"id": "2512.13317", "title": "Face Identity Unlearning for Retrieval via Embedding Dispersion", "abstract": "Face recognition systems rely on learning highly discriminative and compact identity clusters to enable accurate retrieval. However, as with other surveillance-oriented technologies, such systems raise serious privacy concerns due to their potential for unauthorized identity tracking. While several works have explored machine unlearning as a means of privacy protection, their applicability to face retrieval - especially for modern embedding-based recognition models - remains largely unexplored. In this work, we study the problem of face identity unlearning for retrieval systems and present its inherent challenges. The goal is to make selected identities unretrievable by dispersing their embeddings on the hypersphere and preventing the formation of compact identity clusters that enable re-identification in the gallery. The primary challenge is to achieve this forgetting effect while preserving the discriminative structure of the embedding space and the retrieval performance of the model for the remaining identities. To address this, we evaluate several existing approximate class unlearning methods (e.g., Random Labeling, Gradient Ascent, Boundary Unlearning, and other recent approaches) in the context of face retrieval and propose a simple yet effective dispersion-based unlearning approach. Extensive experiments on standard benchmarks (VGGFace2, CelebA) demonstrate that our method achieves superior forgetting behavior while preserving retrieval utility.", "date": "2025-12-15", "categories": []}
{"id": "2512.13361", "title": "Automated User Identification from Facial Thermograms with Siamese Networks", "abstract": "The article analyzes the use of thermal imaging technologies for biometric identification based on facial thermograms. It presents a comparative analysis of infrared spectral ranges (NIR, SWIR, MWIR, and LWIR). The paper also defines key requirements for thermal cameras used in biometric systems, including sensor resolution, thermal sensitivity, and a frame rate of at least 30 Hz. Siamese neural networks are proposed as an effective approach for automating the identification process. In experiments conducted on a proprietary dataset, the proposed method achieved an accuracy of approximately 80%. The study also examines the potential of hybrid systems that combine visible and infrared spectra to overcome the limitations of individual modalities. The results indicate that thermal imaging is a promising technology for developing reliable security systems.", "date": "2025-12-15", "categories": []}
{"id": "2512.10653", "title": "Virtual camera detection: Catching video injection attacks in remote biometric systems", "abstract": "Face anti-spoofing (FAS) is a vital component of remote biometric authentication systems based on facial recognition, increasingly used across web-based applications. Among emerging threats, video injection attacks -- facilitated by technologies such as deepfakes and virtual camera software -- pose significant challenges to system integrity. While virtual camera detection (VCD) has shown potential as a countermeasure, existing literature offers limited insight into its practical implementation and evaluation. This study introduces a machine learning-based approach to VCD, with a focus on its design and validation. The model is trained on metadata collected during sessions with authentic users. Empirical results demonstrate its effectiveness in identifying video injection attempts and reducing the risk of malicious users bypassing FAS systems.", "date": "2025-12-11", "categories": []}
{"id": "2404.10378", "title": "Second Edition FRCSyn Challenge at CVPR 2024: Face Recognition Challenge in the Era of Synthetic Data", "abstract": "Synthetic data is gaining increasing relevance for training machine learning models. This is mainly motivated due to several factors such as the lack of real data and intra-class variability, time and errors produced in manual labeling, and in some cases privacy concerns, among others. This paper presents an overview of the 2nd edition of the Face Recognition Challenge in the Era of Synthetic Data (FRCSyn) organized at CVPR 2024. FRCSyn aims to investigate the use of synthetic data in face recognition to address current technological limitations, including data privacy concerns, demographic biases, generalization to novel scenarios, and performance constraints in challenging situations such as aging, pose variations, and occlusions. Unlike the 1st edition, in which synthetic data from DCFace and GANDiffFace methods was only allowed to train face recognition systems, in this 2nd edition we propose new sub-tasks that allow participants to explore novel face generative methods. The outcomes of the 2nd FRCSyn Challenge, along with the proposed experimental protocol and benchmarking contribute significantly to the application of synthetic data to face recognition.", "date": "2024-04-16", "categories": []}
{"id": "2412.01383", "title": "Second FRCSyn-onGoing: Winning Solutions and Post-Challenge Analysis to Improve Face Recognition with Synthetic Data", "abstract": "Synthetic data is gaining increasing popularity for face recognition technologies, mainly due to the privacy concerns and challenges associated with obtaining real data, including diverse scenarios, quality, and demographic groups, among others. It also offers some advantages over real data, such as the large amount of data that can be generated or the ability to customize it to adapt to specific problem-solving needs. To effectively use such data, face recognition models should also be specifically designed to exploit synthetic data to its fullest potential. In order to promote the proposal of novel Generative AI methods and synthetic data, and investigate the application of synthetic data to better train face recognition systems, we introduce the 2nd FRCSyn-onGoing challenge, based on the 2nd Face Recognition Challenge in the Era of Synthetic Data (FRCSyn), originally launched at CVPR 2024. This is an ongoing challenge that provides researchers with an accessible platform to benchmark i) the proposal of novel Generative AI methods and synthetic data, and ii) novel face recognition systems that are specifically proposed to take advantage of synthetic data. We focus on exploring the use of synthetic data both individually and in combination with real data to solve current challenges in face recognition such as demographic bias, domain adaptation, and performance constraints in demanding situations, such as age disparities between training and testing, changes in the pose, or occlusions. Very interesting findings are obtained in this second edition, including a direct comparison with the first one, in which synthetic databases were restricted to DCFace and GANDiffFace.", "date": "2024-12-02", "categories": []}
{"id": "2512.08981", "title": "Mitigating Bias with Words: Inducing Demographic Ambiguity in Face Recognition Templates by Text Encoding", "abstract": "Face recognition (FR) systems are often prone to demographic biases, partially due to the entanglement of demographic-specific information with identity-relevant features in facial embeddings. This bias is extremely critical in large multicultural cities, especially where biometrics play a major role in smart city infrastructure. The entanglement can cause demographic attributes to overshadow identity cues in the embedding space, resulting in disparities in verification performance across different demographic groups. To address this issue, we propose a novel strategy, Unified Text-Image Embedding (UTIE), which aims to induce demographic ambiguity in face embeddings by enriching them with information related to other demographic groups. This encourages face embeddings to emphasize identity-relevant features and thus promotes fairer verification performance across groups. UTIE leverages the zero-shot capabilities and cross-modal semantic alignment of Vision-Language Models (VLMs). Given that VLMs are naturally trained to align visual and textual representations, we enrich the facial embeddings of each demographic group with text-derived demographic features extracted from other demographic groups. This encourages a more neutral representation in terms of demographic attributes. We evaluate UTIE using three VLMs, CLIP, OpenCLIP, and SigLIP, on two widely used benchmarks, RFW and BFW, designed to assess bias in FR. Experimental results show that UTIE consistently reduces bias metrics while maintaining, or even improving in several cases, the face verification accuracy.", "date": "2025-12-05", "categories": []}
{"id": "2505.24247", "title": "50 Years of Automated Face Recognition", "abstract": "Over the past five decades, automated face recognition (FR) has progressed from handcrafted geometric and statistical approaches to advanced deep learning architectures that now approach, and in many cases exceed, human performance. This paper traces the historical and technological evolution of FR, encompassing early algorithmic paradigms through to contemporary neural systems trained on extensive real and synthetically generated datasets. We examine pivotal innovations that have driven this progression, including advances in dataset construction, loss function formulation, network architecture design, and feature fusion strategies. Furthermore, we analyze the relationship between data scale, diversity, and model generalization, highlighting how dataset expansion correlates with benchmark performance gains. Recent systems have achieved near-perfect large-scale identification accuracy, with the leading algorithm in the latest NIST FRTE 1:N benchmark reporting a FNIR of 0.15 percent at FPIR of 0.001 on a gallery of over 10 million identities. We delineate key open problems and emerging directions, including scalable training, multi-modal fusion, synthetic data, and interpretable recognition frameworks.", "date": "2025-05-30", "categories": []}
{"id": "2512.08697", "title": "What really matters for person re-identification? A Mixture-of-Experts Framework for Semantic Attribute Importance", "abstract": "State-of-the-art person re-identification methods achieve impressive accuracy but remain largely opaque, leaving open the question: which high-level semantic attributes do these models actually rely on? We propose MoSAIC-ReID, a Mixture-of-Experts framework that systematically quantifies the importance of pedestrian attributes for re-identification. Our approach uses LoRA-based experts, each linked to a single attribute, and an oracle router that enables controlled attribution analysis. While MoSAIC-ReID achieves competitive performance on Market-1501 and DukeMTMC under the assumption that attribute annotations are available at test time, its primary value lies in providing a large-scale, quantitative study of attribute importance across intrinsic and extrinsic cues. Using generalized linear models, statistical tests, and feature-importance analyses, we reveal which attributes, such as clothing colors and intrinsic characteristics, contribute most strongly, while infrequent cues (e.g. accessories) have limited effect. This work offers a principled framework for interpretable ReID and highlights the requirements for integrating explicit semantic knowledge in practice. Code is available at https://github.com/psaltaath/MoSAIC-ReID", "date": "2025-12-09", "categories": []}
{"id": "2512.06103", "title": "SpectraIrisPAD: Leveraging Vision Foundation Models for Spectrally Conditioned Multispectral Iris Presentation Attack Detection", "abstract": "Iris recognition is widely recognized as one of the most accurate biometric modalities. However, its growing deployment in real-world applications raises significant concerns regarding its vulnerability to Presentation Attacks (PAs). Effective Presentation Attack Detection (PAD) is therefore critical to ensure the integrity and security of iris-based biometric systems. While conventional iris recognition systems predominantly operate in the near-infrared (NIR) spectrum, multispectral imaging across multiple NIR bands provides complementary reflectance information that can enhance the generalizability of PAD methods. In this work, we propose \\textbf{SpectraIrisPAD}, a novel deep learning-based framework for robust multispectral iris PAD. The SpectraIrisPAD leverages a DINOv2 Vision Transformer (ViT) backbone equipped with learnable spectral positional encoding, token fusion, and contrastive learning to extract discriminative, band-specific features that effectively distinguish bona fide samples from various spoofing artifacts. Furthermore, we introduce a new comprehensive dataset Multispectral Iris PAD (\\textbf{MSIrPAD}) with diverse PAIs, captured using a custom-designed multispectral iris sensor operating at five distinct NIR wavelengths (800\\,nm, 830\\,nm, 850\\,nm, 870\\,nm, and 980\\,nm). The dataset includes 18,848 iris images encompassing eight diverse PAI categories, including five textured contact lenses, print attacks, and display-based attacks. We conduct comprehensive experiments under unseen attack evaluation protocols to assess the generalization capability of the proposed method. SpectraIrisPAD consistently outperforms several state-of-the-art baselines across all performance metrics, demonstrating superior robustness and generalizability in detecting a wide range of presentation attacks.", "date": "2025-12-05", "categories": []}
{"id": "2506.10361", "title": "FaceLiVT: Face Recognition using Linear Vision Transformer with Structural Reparameterization For Mobile Device", "abstract": "This paper introduces FaceLiVT, a lightweight yet powerful face recognition model that integrates a hybrid Convolution Neural Network (CNN)-Transformer architecture with an innovative and lightweight Multi-Head Linear Attention (MHLA) mechanism. By combining MHLA alongside a reparameterized token mixer, FaceLiVT effectively reduces computational complexity while preserving competitive accuracy. Extensive evaluations on challenging benchmarks; including LFW, CFP-FP, AgeDB-30, IJB-B, and IJB-C; highlight its superior performance compared to state-of-the-art lightweight models. MHLA notably improves inference speed, allowing FaceLiVT to deliver high accuracy with lower latency on mobile devices. Specifically, FaceLiVT is 8.6 faster than EdgeFace, a recent hybrid CNN-Transformer model optimized for edge devices, and 21.2 faster than a pure ViT-Based model. With its balanced design, FaceLiVT offers an efficient and practical solution for real-time face recognition on resource-constrained platforms.", "date": "2025-06-12", "categories": []}
{"id": "2512.05928", "title": "A Comparative Study on Synthetic Facial Data Generation Techniques for Face Recognition", "abstract": "Facial recognition has become a widely used method for authentication and identification, with applications for secure access and locating missing persons. Its success is largely attributed to deep learning, which leverages large datasets and effective loss functions to learn discriminative features. Despite these advances, facial recognition still faces challenges in explainability, demographic bias, privacy, and robustness to aging, pose variations, lighting changes, occlusions, and facial expressions. Privacy regulations have also led to the degradation of several datasets, raising legal, ethical, and privacy concerns. Synthetic facial data generation has been proposed as a promising solution. It mitigates privacy issues, enables experimentation with controlled facial attributes, alleviates demographic bias, and provides supplementary data to improve models trained on real data. This study compares the effectiveness of synthetic facial datasets generated using different techniques in facial recognition tasks. We evaluate accuracy, rank-1, rank-5, and the true positive rate at a false positive rate of 0.01% on eight leading datasets, offering a comparative analysis not extensively explored in the literature. Results demonstrate the ability of synthetic data to capture realistic variations while emphasizing the need for further research to close the performance gap with real data. Techniques such as diffusion models, GANs, and 3D models show substantial progress; however, challenges remain.", "date": "2025-12-05", "categories": []}
{"id": "2512.04522", "title": "Identity Clue Refinement and Enhancement for Visible-Infrared Person Re-Identification", "abstract": "Visible-Infrared Person Re-Identification (VI-ReID) is a challenging cross-modal matching task due to significant modality discrepancies. While current methods mainly focus on learning modality-invariant features through unified embedding spaces, they often focus solely on the common discriminative semantics across modalities while disregarding the critical role of modality-specific identity-aware knowledge in discriminative feature learning. To bridge this gap, we propose a novel Identity Clue Refinement and Enhancement (ICRE) network to mine and utilize the implicit discriminative knowledge inherent in modality-specific attributes. Initially, we design a Multi-Perception Feature Refinement (MPFR) module that aggregates shallow features from shared branches, aiming to capture modality-specific attributes that are easily overlooked. Then, we propose a Semantic Distillation Cascade Enhancement (SDCE) module, which distills identity-aware knowledge from the aggregated shallow features and guide the learning of modality-invariant features. Finally, an Identity Clues Guided (ICG) Loss is proposed to alleviate the modality discrepancies within the enhanced features and promote the learning of a diverse representation space. Extensive experiments across multiple public datasets clearly show that our proposed ICRE outperforms existing SOTA methods.", "date": "2025-12-04", "categories": []}
{"id": "2403.14114", "title": "Test-time Similarity Modification for Person Re-identification toward Temporal Distribution Shift", "abstract": "Person re-identification (re-id), which aims to retrieve images of the same person in a given image from a database, is one of the most practical image recognition applications. In the real world, however, the environments that the images are taken from change over time. This causes a distribution shift between training and testing and degrades the performance of re-id. To maintain re-id performance, models should continue adapting to the test environment's temporal changes. Test-time adaptation (TTA), which aims to adapt models to the test environment with only unlabeled test data, is a promising way to handle this problem because TTA can adapt models instantly in the test environment. However, the previous TTA methods are designed for classification and cannot be directly applied to re-id. This is because the set of people's identities in the dataset differs between training and testing in re-id, whereas the set of classes is fixed in the current TTA methods designed for classification. To improve re-id performance in changing test environments, we propose TEst-time similarity Modification for Person re-identification (TEMP), a novel TTA method for re-id. TEMP is the first fully TTA method for re-id, which does not require any modification to pre-training. Inspired by TTA methods that refine the prediction uncertainty in classification, we aim to refine the uncertainty in re-id. However, the uncertainty cannot be computed in the same way as classification in re-id since it is an open-set task, which does not share person labels between training and testing. Hence, we propose re-id entropy, an alternative uncertainty measure for re-id computed based on the similarity between the feature vectors. Experiments show that the re-id entropy can measure the uncertainty on re-id and TEMP improves the performance of re-id in online settings where the distribution changes over time.", "date": "2024-03-21", "categories": []}
{"id": "2512.02554", "title": "OmniPerson: Unified Identity-Preserving Pedestrian Generation", "abstract": "Person re-identification (ReID) suffers from a lack of large-scale high-quality training data due to challenges in data privacy and annotation costs. While previous approaches have explored pedestrian generation for data augmentation, they often fail to ensure identity consistency and suffer from insufficient controllability, thereby limiting their effectiveness in dataset augmentation. To address this, We introduce OmniPerson, the first unified identity-preserving pedestrian generation pipeline for visible/infrared image/video ReID tasks. Our contributions are threefold: 1) We proposed OmniPerson, a unified generation model, offering holistic and fine-grained control over all key pedestrian attributes. Supporting RGB/IR modality image/video generation with any number of reference images, two kinds of person poses, and text. Also including RGB-to-IR transfer and image super-resolution abilities.2) We designed Multi-Refer Fuser for robust identity preservation with any number of reference images as input, making OmniPerson could distill a unified identity from a set of multi-view reference images, ensuring our generated pedestrians achieve high-fidelity pedestrian generation.3) We introduce PersonSyn, the first large-scale dataset for multi-reference, controllable pedestrian generation, and present its automated curation pipeline which transforms public, ID-only ReID benchmarks into a richly annotated resource with the dense, multi-modal supervision required for this task. Experimental results demonstrate that OmniPerson achieves SoTA in pedestrian generation, excelling in both visual fidelity and identity consistency. Furthermore, augmenting existing datasets with our generated data consistently improves the performance of ReID models. We will open-source the full codebase, pretrained model, and the PersonSyn dataset.", "date": "2025-12-02", "categories": []}
{"id": "2512.00691", "title": "Silhouette-based Gait Foundation Model", "abstract": "Gait patterns play a critical role in human identification and healthcare analytics, yet current progress remains constrained by small, narrowly designed models that fail to scale or generalize. Building a unified gait foundation model requires addressing two longstanding barriers: (a) Scalability. Why have gait models historically failed to follow scaling laws? (b) Generalization. Can one model serve the diverse gait tasks that have traditionally been studied in isolation? We introduce FoundationGait, the first scalable, self-supervised pretraining framework for gait understanding. Its largest version has nearly 0.13 billion parameters and is pretrained on 12 public gait datasets comprising over 2 million walking sequences. Extensive experiments demonstrate that FoundationGait, with or without fine-tuning, performs robustly across a wide spectrum of gait datasets, conditions, tasks (e.g., human identification, scoliosis screening, depression prediction, and attribute estimation), and even input modality. Notably, it achieves 48.0% zero-shot rank-1 accuracy on the challenging in-the-wild Gait3D dataset (1,000 test subjects) and 64.5% on the largest in-the-lab OU-MVLP dataset (5,000+ test subjects), setting a new milestone in robust gait recognition. Coming code and model: https://github.com/ShiqiYu/OpenGait.", "date": "2025-11-30", "categories": []}
{"id": "2511.22994", "title": "Optimizer Sensitivity In Vision Transformerbased Iris Recognition: Adamw Vs Sgd Vs Rmsprop", "abstract": "The security of biometric authentication is increasingly critical as digital identity systems expand. Iris recognition offers high reliability due to its distinctive and stable texture patterns. Recent progress in deep learning, especially Vision Transformers ViT, has improved visual recognition performance. Yet, the effect of optimizer choice on ViT-based biometric systems remains understudied. This work evaluates how different optimizers influence the accuracy and stability of ViT for iris recognition, providing insights to enhance the robustness of biometric identification models.", "date": "2025-11-28", "categories": []}
{"id": "2508.02143", "title": "TrackletGait: A Robust Framework for Gait Recognition in the Wild", "abstract": "Gait recognition aims to identify individuals based on their body shape and walking patterns. Though much progress has been achieved driven by deep learning, gait recognition in real-world surveillance scenarios remains quite challenging to current methods. Conventional approaches, which rely on periodic gait cycles and controlled environments, struggle with the non-periodic and occluded silhouette sequences encountered in the wild. In this paper, we propose a novel framework, TrackletGait, designed to address these challenges in the wild. We propose Random Tracklet Sampling, a generalization of existing sampling methods, which strikes a balance between robustness and representation in capturing diverse walking patterns. Next, we introduce Haar Wavelet-based Downsampling to preserve information during spatial downsampling. Finally, we present a Hardness Exclusion Triplet Loss, designed to exclude low-quality silhouettes by discarding hard triplet samples. TrackletGait achieves state-of-the-art results, with 77.8 and 80.4 rank-1 accuracy on the Gait3D and GREW datasets, respectively, while using only 10.3M backbone parameters. Extensive experiments are also conducted to further investigate the factors affecting gait recognition in the wild.", "date": "2025-08-04", "categories": []}
{"id": "2511.19958", "title": "GFT-GCN: Privacy-Preserving 3D Face Mesh Recognition with Spectral Diffusion", "abstract": "3D face recognition offers a robust biometric solution by capturing facial geometry, providing resilience to variations in illumination, pose changes, and presentation attacks. Its strong spoof resistance makes it suitable for high-security applications, but protecting stored biometric templates remains critical. We present GFT-GCN, a privacy-preserving 3D face recognition framework that combines spectral graph learning with diffusion-based template protection. Our approach integrates the Graph Fourier Transform (GFT) and Graph Convolutional Networks (GCN) to extract compact, discriminative spectral features from 3D face meshes. To secure these features, we introduce a spectral diffusion mechanism that produces irreversible, renewable, and unlinkable templates. A lightweight client-server architecture ensures that raw biometric data never leaves the client device. Experiments on the BU-3DFE and FaceScape datasets demonstrate high recognition accuracy and strong resistance to reconstruction attacks. Results show that GFT-GCN effectively balances privacy and performance, offering a practical solution for secure 3D face authentication.", "date": "2025-11-25", "categories": []}
{"id": "2511.21097", "title": "CLRecogEye : Curriculum Learning towards exploiting convolution features for Dynamic Iris Recognition", "abstract": "Iris authentication algorithms have achieved impressive recognition performance, making them highly promising for real-world applications such as border control, citizen identification, and both criminal investigations and commercial systems. However, their robustness is still challenged by variations in rotation, scale, specular reflections, and defocus blur. In addition, most existing approaches rely on straightforward point-to-point comparisons, typically using cosine or L2 distance, without effectively leveraging the spatio-spatial-temporal structure of iris patterns. To address these limitations, we propose a novel and generalized matching pipeline that learns rich spatio-spatial-temporal representations of iris features. Our approach first splits each iris image along one dimension, generating a sequence of sub-images that serve as input to a 3D-CNN, enabling the network to capture both spatial and spatio-spatial-temporal cues. To further enhance the modeling of spatio-spatial-temporal feature dynamics, we train the model in curriculum manner. This design allows the network to embed temporal dependencies directly into the feature space, improving discriminability in the deep metric domain. The framework is trained end-to-end with triplet and ArcFace loss in a curriculum manner, enforcing highly discriminative embeddings despite challenges like rotation, scale, reflections, and blur. This design yields a robust and generalizable solution for iris authentication.Github code: https://github.com/GeetanjaliGTZ/CLRecogEye", "date": "2025-11-26", "categories": []}
{"id": "2510.17201", "title": "Optimizing DINOv2 with Registers for Face Anti-Spoofing", "abstract": "Face recognition systems are designed to be robust against variations in head pose, illumination, and image blur during capture. However, malicious actors can exploit these systems by presenting a face photo of a registered user, potentially bypassing the authentication process. Such spoofing attacks must be detected prior to face recognition. In this paper, we propose a DINOv2-based spoofing attack detection method to discern minute differences between live and spoofed face images. Specifically, we employ DINOv2 with registers to extract generalizable features and to suppress perturbations in the attention mechanism, which enables focused attention on essential and minute features. We demonstrate the effectiveness of the proposed method through experiments conducted on the dataset provided by ``The 6th Face Anti-Spoofing Workshop: Unified Physical-Digital Attacks Detection@ICCV2025'' and SiW dataset. The project page is available at: https://gsisaoki.github.io/FAS-DINOv2-ICCVW/ .", "date": "2025-10-20", "categories": []}
{"id": "2511.17964", "title": "X-ReID: Multi-granularity Information Interaction for Video-Based Visible-Infrared Person Re-Identification", "abstract": "Large-scale vision-language models (e.g., CLIP) have recently achieved remarkable performance in retrieval tasks, yet their potential for Video-based Visible-Infrared Person Re-Identification (VVI-ReID) remains largely unexplored. The primary challenges are narrowing the modality gap and leveraging spatiotemporal information in video sequences. To address the above issues, in this paper, we propose a novel cross-modality feature learning framework named X-ReID for VVI-ReID. Specifically, we first propose a Cross-modality Prototype Collaboration (CPC) to align and integrate features from different modalities, guiding the network to reduce the modality discrepancy. Then, a Multi-granularity Information Interaction (MII) is designed, incorporating short-term interactions from adjacent frames, long-term cross-frame information fusion, and cross-modality feature alignment to enhance temporal modeling and further reduce modality gaps. Finally, by integrating multi-granularity information, a robust sequence-level representation is achieved. Extensive experiments on two large-scale VVI-ReID benchmarks (i.e., HITSZ-VCM and BUPTCampus) demonstrate the superiority of our method over state-of-the-art methods. The source code is released at https://github.com/AsuradaYuci/X-ReID.", "date": "2025-11-22", "categories": []}
{"id": "2511.19846", "title": "Face, Whole-Person, and Object Classification in a Unified Space Via The Interleaved Multi-Domain Identity Curriculum", "abstract": "Vision foundation models can perform generalized object classification in zero-shot mode, and face/person recognition when they are fine-tuned. However, fine-tuned models suffer from catastrophic forgetting. We create models that perform four tasks (object recognition, face recognition from high- and low-quality images, and person recognition from whole-body images) in a single embedding space -- without incurring substantial catastrophic forgetting. To accomplish this, we introduce two variants of the Interleaved Multi-Domain Identity Curriculum (IMIC): a gradient-coupled, interleaving training schedule that fine-tunes a foundation backbone simultaneously on all four tasks. The IMIC method proved effective with three foundation model bases: DINOv3, CLIP, and EVA-02. Two of these (EVA-02 and CLIP) performed comparably with domain experts on all four tasks concurrently and were more accurate than humans at multi-tasking across face, body, and object datasets. Further, we demonstrate that our approach does not substantially harm out-of-distribution generalization, thus maintaining a key property of foundation models. Analysis of the most accurate model variants (EVA-02 + IMIC A and B) showed linearly separable representations of the four tasks in the unified embedding space, but with substantial sharing of features across tasks. Fewer than 100 PCs calculated from any one task could perform all other tasks with nearly zero performance degradation.", "date": "2025-11-25", "categories": []}
{"id": "2508.06908", "title": "Find Them All: Unveiling MLLMs for Versatile Person Re-identification", "abstract": "Person re-identification (ReID) aims to retrieve images of a target person from the gallery set, with wide applications in medical rehabilitation and public security. However, traditional person ReID models are typically uni-modal, resulting in limited generalizability across heterogeneous data modalities. Recently, the emergence of multi-modal large language models (MLLMs) has shown a promising avenue for addressing this issue. Despite this potential, existing methods merely regard MLLMs as feature extractors or caption generators, leaving their capabilities in person ReID tasks largely unexplored. To bridge this gap, we introduce a novel benchmark for \\underline{\\textbf{V}}ersatile \\underline{\\textbf{P}}erson \\underline{\\textbf{Re}}-\\underline{\\textbf{ID}}entification, termed VP-ReID. The benchmark includes 257,310 multi-modal queries and gallery images, covering ten diverse person ReID tasks. In addition, we propose two task-oriented evaluation schemes for MLLM-based person ReID. Extensive experiments demonstrate the impressive versatility, effectiveness, and interpretability of MLLMs in various person ReID tasks. Nevertheless, they also have limitations in handling a few modalities, particularly thermal and infrared data. We hope that VP-ReID can facilitate the community in developing more robust and generalizable cross-modal foundation models for person ReID.", "date": "2025-08-09", "categories": []}
{"id": "2511.13621", "title": "Alpha Divergence Losses for Biometric Verification", "abstract": "Performance in face and speaker verification is largely driven by margin-based softmax losses such as CosFace and ArcFace. Recently introduced $\\alpha$-divergence loss functions offer a compelling alternative, particularly due to their ability to induce sparse solutions (when $\\alpha>1$). However, integrating an angular margin-crucial for verification tasks-is not straightforward. We find that this integration can be achieved in at least two distinct ways: via the reference measure (prior probabilities) or via the logits (unnormalized log-likelihoods). In this paper, we explore both pathways, deriving two novel margin-based $\\alpha$-divergence losses: Q-Margin (margin in the reference measure) and A3M (margin in the logits). We identify and address a training instability in A3M-caused by sparsity-with a simple yet effective prototype re-initialization strategy. Our methods achieve significant performance gains on the challenging IJB-B and IJB-C face verification benchmarks. We demonstrate similarly strong performance in speaker verification on VoxCeleb. Crucially, our models significantly outperform strong baselines at low false acceptance rates (FAR). This capability is critical for practical high-security applications, such as banking authentication, when minimizing false authentications is paramount. Finally, the sparsity of $\\alpha$-divergence-based posteriors enables memory-efficient training, which is crucial for datasets with millions of identities.", "date": "2025-11-17", "categories": []}
{"id": "2511.18677", "title": "A Theory-Inspired Framework for Few-Shot Cross-Modal Sketch Person Re-Identification", "abstract": "Sketch based person re-identification aims to match hand-drawn sketches with RGB surveillance images, but remains challenging due to significant modality gaps and limited annotated data. To address this, we introduce KTCAA, a theoretically grounded framework for few-shot cross-modal generalization. Motivated by generalization theory, we identify two key factors influencing target domain risk: (1) domain discrepancy, which quantifies the alignment difficulty between source and target distributions; and (2) perturbation invariance, which evaluates the model's robustness to modality shifts. Based on these insights, we propose two components: (1) Alignment Augmentation (AA), which applies localized sketch-style transformations to simulate target distributions and facilitate progressive alignment; and (2) Knowledge Transfer Catalyst (KTC), which enhances invariance by introducing worst-case perturbations and enforcing consistency. These modules are jointly optimized under a meta-learning paradigm that transfers alignment knowledge from data-rich RGB domains to sketch-based scenarios. Experiments on multiple benchmarks demonstrate that KTCAA achieves state-of-the-art performance, particularly in data-scarce conditions.", "date": "2025-11-24", "categories": []}
{"id": "2511.11381", "title": "SoK: Security Evaluation of Wi-Fi CSI Biometrics: Attacks, Metrics, and Open Challenges", "abstract": "Wi-Fi Channel State Information (CSI) has been repeatedly proposed as a biometric modality, often with reports of high accuracy and operational feasibility. However, the field lacks a consolidated understanding of its security properties, adversarial resilience, and methodological consistency. This Systematization of Knowledge (SoK) examines CSI-based biometric authentication through a security lens, analyzing how existing works diverge in sensing infrastructure, signal representations, feature pipelines, learning models, and evaluation methodologies. Our synthesis reveals systemic inconsistencies: reliance on aggregate accuracy metrics, limited reporting of FAR/FRR/EER, absence of per-user risk analysis, and scarce consideration of threat models or adversarial feasibility.\n  To this end, we construct a unified evaluation framework to expose these issues empirically and demonstrate how security-relevant metrics such as per-class EER, Frequency Count of Scores (FCS), and the Gini Coefficient uncover risk concentration that remains hidden under traditional reporting practices. The resulting analysis highlights concrete attack surfaces--including replay, geometric mimicry, and environmental perturbation--and shows how methodological choices materially influence vulnerability profiles. Based on these findings, we articulate the security boundaries of current CSI biometrics and provide guidelines for rigorous evaluation, reproducible experimentation, and future research directions. This SoK offers the security community a structured, evidence-driven reassessment of Wi-Fi CSI biometrics and their suitability as an authentication primitive.", "date": "2025-11-14", "categories": []}
{"id": "2511.17393", "title": "Designing and Generating Diverse, Equitable Face Image Datasets for Face Verification Tasks", "abstract": "Face verification is a significant component of identity authentication in various applications including online banking and secure access to personal devices. The majority of the existing face image datasets often suffer from notable biases related to race, gender, and other demographic characteristics, limiting the effectiveness and fairness of face verification systems. In response to these challenges, we propose a comprehensive methodology that integrates advanced generative models to create varied and diverse high-quality synthetic face images. This methodology emphasizes the representation of a diverse range of facial traits, ensuring adherence to characteristics permissible in identity card photographs. Furthermore, we introduce the Diverse and Inclusive Faces for Verification (DIF-V) dataset, comprising 27,780 images of 926 unique identities, designed as a benchmark for future research in face verification. Our analysis reveals that existing verification models exhibit biases toward certain genders and races, and notably, applying identity style modifications negatively impacts model performance. By tackling the inherent inequities in existing datasets, this work not only enriches the discussion on diversity and ethics in artificial intelligence but also lays the foundation for developing more inclusive and reliable face verification technologies", "date": "2025-11-21", "categories": []}
{"id": "2511.15016", "title": "CKDA: Cross-modality Knowledge Disentanglement and Alignment for Visible-Infrared Lifelong Person Re-identification", "abstract": "Lifelong person Re-IDentification (LReID) aims to match the same person employing continuously collected individual data from different scenarios. To achieve continuous all-day person matching across day and night, Visible-Infrared Lifelong person Re-IDentification (VI-LReID) focuses on sequential training on data from visible and infrared modalities and pursues average performance over all data. To this end, existing methods typically exploit cross-modal knowledge distillation to alleviate the catastrophic forgetting of old knowledge. However, these methods ignore the mutual interference of modality-specific knowledge acquisition and modality-common knowledge anti-forgetting, where conflicting knowledge leads to collaborative forgetting. To address the above problems, this paper proposes a Cross-modality Knowledge Disentanglement and Alignment method, called CKDA, which explicitly separates and preserves modality-specific knowledge and modality-common knowledge in a balanced way. Specifically, a Modality-Common Prompting (MCP) module and a Modality-Specific Prompting (MSP) module are proposed to explicitly disentangle and purify discriminative information that coexists and is specific to different modalities, avoiding the mutual interference between both knowledge. In addition, a Cross-modal Knowledge Alignment (CKA) module is designed to further align the disentangled new knowledge with the old one in two mutually independent inter- and intra-modality feature spaces based on dual-modality prototypes in a balanced manner. Extensive experiments on four benchmark datasets verify the effectiveness and superiority of our CKDA against state-of-the-art methods. The source code of this paper is available at https://github.com/PKU-ICST-MIPL/CKDA-AAAI2026.", "date": "2025-11-19", "categories": []}
{"id": "2511.15322", "title": "Adaptive thresholding pattern for fingerprint forgery detection", "abstract": "Fingerprint liveness detection systems have been affected by spoofing, which is a severe threat for fingerprint-based biometric systems. Therefore, it is crucial to develop some techniques to distinguish the fake fingerprints from the real ones. The software based techniques can detect the fingerprint forgery automatically. Also, the scheme shall be resistant against various distortions such as noise contamination, pixel missing and block missing, so that the forgers cannot deceive the detector by adding some distortions to the faked fingerprint. In this paper, we propose a fingerprint forgery detection algorithm based on a suggested adaptive thresholding pattern. The anisotropic diffusion of the input image is passed through three levels of the wavelet transform. The coefficients of different layers are adaptively thresholded and concatenated to produce the feature vector which is classified using the SVM classifier. Another contribution of the paper is to investigate the effect of various distortions such as pixel missing, block missing, and noise contamination. Our suggested approach includes a novel method that exhibits improved resistance against a range of distortions caused by environmental phenomena or manipulations by malicious users. In quantitative comparisons, our proposed method outperforms its counterparts by approximately 8% and 5% in accuracy for missing pixel scenarios of 90% and block missing scenarios of size 70x70 , respectively. This highlights the novelty approach in addressing such challenges.", "date": "2025-11-19", "categories": []}
{"id": "2511.14376", "title": "A Quantitative Method for Shoulder Presentation Evaluation in Biometric Identity Documents", "abstract": "International standards for biometric identity documents mandate strict compliance with pose requirements, including the square presentation of a subject's shoulders. However, the literature on automated quality assessment offers few quantitative methods for evaluating this specific attribute. This paper proposes a Shoulder Presentation Evaluation (SPE) algorithm to address this gap. The method quantifies shoulder yaw and roll using only the 3D coordinates of two shoulder landmarks provided by common pose estimation frameworks. The algorithm was evaluated on a dataset of 121 portrait images. The resulting SPE scores demonstrated a strong Pearson correlation (r approx. 0.80) with human-assigned labels. An analysis of the metric's filtering performance, using an adapted Error-versus-Discard methodology, confirmed its utility in identifying non-compliant samples. The proposed algorithm is a viable lightweight tool for automated compliance checking in enrolment systems.", "date": "2025-11-18", "categories": []}
{"id": "2511.14411", "title": "Cranio-ID: Graph-Based Craniofacial Identification via Automatic Landmark Annotation in 2D Multi-View X-rays", "abstract": "In forensic craniofacial identification and in many biomedical applications, craniometric landmarks are important. Traditional methods for locating landmarks are time-consuming and require specialized knowledge and expertise. Current methods utilize superimposition and deep learning-based methods that employ automatic annotation of landmarks. However, these methods are not reliable due to insufficient large-scale validation studies. In this paper, we proposed a novel framework Cranio-ID: First, an automatic annotation of landmarks on 2D skulls (which are X-ray scans of faces) with their respective optical images using our trained YOLO-pose models. Second, cross-modal matching by formulating these landmarks into graph representations and then finding semantic correspondence between graphs of these two modalities using cross-attention and optimal transport framework. Our proposed framework is validated on the S2F and CUHK datasets (CUHK dataset resembles with S2F dataset). Extensive experiments have been conducted to evaluate the performance of our proposed framework, which demonstrates significant improvements in both reliability and accuracy, as well as its effectiveness in cross-domain skull-to-face and sketch-to-face matching in forensic science.", "date": "2025-11-18", "categories": []}
{"id": "2508.11334", "title": "GANDiff FR: Hybrid GAN Diffusion Synthesis for Causal Bias Attribution in Face Recognition", "abstract": "We introduce GANDiff FR, the first synthetic framework that precisely controls demographic and environmental factors to measure, explain, and reduce bias with reproducible rigor. GANDiff FR unifies StyleGAN3-based identity-preserving generation with diffusion-based attribute control, enabling fine-grained manipulation of pose around 30 degrees, illumination (four directions), and expression (five levels) under ceteris paribus conditions. We synthesize 10,000 demographically balanced faces across five cohorts validated for realism via automated detection (98.2%) and human review (89%) to isolate and quantify bias drivers. Benchmarking ArcFace, CosFace, and AdaFace under matched operating points shows AdaFace reduces inter-group TPR disparity by 60% (2.5% vs. 6.3%), with illumination accounting for 42% of residual bias. Cross-dataset evaluation on RFW, BUPT, and CASIA WebFace confirms strong synthetic-to-real transfer (r 0.85). Despite around 20% computational overhead relative to pure GANs, GANDiff FR yields three times more attribute-conditioned variants, establishing a reproducible, regulation-aligned (EU AI Act) standard for fairness auditing. Code and data are released to support transparent, scalable bias evaluation.", "date": "2025-08-15", "categories": []}
{"id": "2511.05893", "title": "Hybrid second-order gradient histogram based global low-rank sparse regression for robust face recognition", "abstract": "Low-rank sparse regression models have been widely adopted in face recognition due to their robustness against occlusion and illumination variations. However, existing methods often suffer from insufficient feature representation and limited modeling of structured corruption across samples. To address these issues, this paper proposes a Hybrid second-order gradient Histogram based Global Low-Rank Sparse Regression (H2H-GLRSR) model. First, we propose the Histogram of Oriented Hessian (HOH) to capture second-order geometric characteristics such as curvature and ridge patterns. By fusing HOH and first-order gradient histograms, we construct a unified local descriptor, termed the Hybrid second-order gradient Histogram (H2H), which enhances structural discriminability under challenging conditions. Subsequently, the H2H features are incorporated into an extended version of the Sparse Regularized Nuclear Norm based Matrix Regression (SR\\_NMR) model, where a global low-rank constraint is imposed on the residual matrix to exploit cross-sample correlations in structured noise. The resulting H2H-GLRSR model achieves superior discrimination and robustness. Experimental results on benchmark datasets demonstrate that the proposed method significantly outperforms state-of-the-art regression-based classifiers in both recognition accuracy and computational efficiency.", "date": "2025-11-08", "categories": []}
{"id": "2511.11754", "title": "Batch Transformer Architecture: Case of Synthetic Image Generation for Emotion Expression Facial Recognition", "abstract": "A novel Transformer variation architecture is proposed in the implicit sparse style. Unlike \"traditional\" Transformers, instead of attention to sequential or batch entities in their entirety of whole dimensionality, in the proposed Batch Transformers, attention to the \"important\" dimensions (primary components) is implemented. In such a way, the \"important\" dimensions or feature selection allows for a significant reduction of the bottleneck size in the encoder-decoder ANN architectures. The proposed architecture is tested on the synthetic image generation for the face recognition task in the case of the makeup and occlusion data set, allowing for increased variability of the limited original data set.", "date": "2025-11-13", "categories": []}
{"id": "2511.13575", "title": "Hierarchical Prompt Learning for Image- and Text-Based Person Re-Identification", "abstract": "Person re-identification (ReID) aims to retrieve target pedestrian images given either visual queries (image-to-image, I2I) or textual descriptions (text-to-image, T2I). Although both tasks share a common retrieval objective, they pose distinct challenges: I2I emphasizes discriminative identity learning, while T2I requires accurate cross-modal semantic alignment. Existing methods often treat these tasks separately, which may lead to representation entanglement and suboptimal performance. To address this, we propose a unified framework named Hierarchical Prompt Learning (HPL), which leverages task-aware prompt modeling to jointly optimize both tasks. Specifically, we first introduce a Task-Routed Transformer, which incorporates dual classification tokens into a shared visual encoder to route features for I2I and T2I branches respectively. On top of this, we develop a hierarchical prompt generation scheme that integrates identity-level learnable tokens with instance-level pseudo-text tokens. These pseudo-tokens are derived from image or text features via modality-specific inversion networks, injecting fine-grained, instance-specific semantics into the prompts. Furthermore, we propose a Cross-Modal Prompt Regularization strategy to enforce semantic alignment in the prompt token space, ensuring that pseudo-prompts preserve source-modality characteristics while enhancing cross-modal transferability. Extensive experiments on multiple ReID benchmarks validate the effectiveness of our method, achieving state-of-the-art performance on both I2I and T2I tasks.", "date": "2025-11-17", "categories": []}
{"id": "2511.11235", "title": "Neural Network-Powered Finger-Drawn Biometric Authentication", "abstract": "This paper investigates neural network-based biometric authentication using finger-drawn digits on touchscreen devices. We evaluated CNN and autoencoder architectures for user authentication through simple digit patterns (0-9) traced with finger input. Twenty participants contributed 2,000 finger-drawn digits each on personal touchscreen devices. We compared two CNN architectures: a modified Inception-V1 network and a lightweight shallow CNN for mobile environments. Additionally, we examined Convolutional and Fully Connected autoencoders for anomaly detection. Both CNN architectures achieved ~89% authentication accuracy, with the shallow CNN requiring fewer parameters. Autoencoder approaches achieved ~75% accuracy. The results demonstrate that finger-drawn symbol authentication provides a viable, secure, and user-friendly biometric solution for touchscreen devices. This approach can be integrated with existing pattern-based authentication methods to create multi-layered security systems for mobile applications.", "date": "2025-11-14", "categories": []}
{"id": "2511.10309", "title": "CLIP4VI-ReID: Learning Modality-shared Representations via CLIP Semantic Bridge for Visible-Infrared Person Re-identification", "abstract": "This paper proposes a novel CLIP-driven modality-shared representation learning network named CLIP4VI-ReID for VI-ReID task, which consists of Text Semantic Generation (TSG), Infrared Feature Embedding (IFE), and High-level Semantic Alignment (HSA). Specifically, considering the huge gap in the physical characteristics between natural images and infrared images, the TSG is designed to generate text semantics only for visible images, thereby enabling preliminary visible-text modality alignment. Then, the IFE is proposed to rectify the feature embeddings of infrared images using the generated text semantics. This process injects id-related semantics into the shared image encoder, enhancing its adaptability to the infrared modality. Besides, with text serving as a bridge, it enables indirect visible-infrared modality alignment. Finally, the HSA is established to refine the high-level semantic alignment. This process ensures that the fine-tuned text semantics only contain id-related information, thereby achieving more accurate cross-modal alignment and enhancing the discriminability of the learned modal-shared representations. Extensive experimental results demonstrate that the proposed CLIP4VI-ReID achieves superior performance than other state-of-the-art methods on some widely used VI-ReID datasets.", "date": "2025-11-13", "categories": []}
{"id": "2510.12988", "title": "Behavioral Biometrics for Automatic Detection of User Familiarity in VR", "abstract": "As virtual reality (VR) devices become increasingly integrated into everyday settings, a growing number of users without prior experience will engage with VR systems. Automatically detecting a user's familiarity with VR as an interaction medium enables real-time, adaptive training and interface adjustments, minimizing user frustration and improving task performance. In this study, we explore the automatic detection of VR familiarity by analyzing hand movement patterns during a passcode-based door-opening task, which is a well-known interaction in collaborative virtual environments such as meeting rooms, offices, and healthcare spaces. While novice users may lack prior VR experience, they are likely to be familiar with analogous real-world tasks involving keypad entry. We conducted a pilot study with 26 participants, evenly split between experienced and inexperienced VR users, who performed tasks using both controller-based and hand-tracking interactions. Our approach uses state-of-the-art deep classifiers for automatic VR familiarity detection, achieving the highest accuracies of 92.05% and 83.42% for hand-tracking and controller-based interactions, respectively. In the cross-device evaluation, where classifiers trained on controller data were tested using hand-tracking data, the model achieved an accuracy of 78.89%. The integration of both modalities in the mixed-device evaluation obtained an accuracy of 94.19%. Our results underline the promise of using hand movement biometrics for the real-time detection of user familiarity in critical VR applications, paving the way for personalized and adaptive VR experiences.", "date": "2025-10-14", "categories": []}
{"id": "2511.07948", "title": "ReIDMamba: Learning Discriminative Features with Visual State Space Model for Person Re-Identification", "abstract": "Extracting robust discriminative features is a critical challenge in person re-identification (ReID). While Transformer-based methods have successfully addressed some limitations of convolutional neural networks (CNNs), such as their local processing nature and information loss resulting from convolution and downsampling operations, they still face the scalability issue due to the quadratic increase in memory and computational requirements with the length of the input sequence. To overcome this, we propose a pure Mamba-based person ReID framework named ReIDMamba. Specifically, we have designed a Mamba-based strong baseline that effectively leverages fine-grained, discriminative global features by introducing multiple class tokens. To further enhance robust features learning within Mamba, we have carefully designed two novel techniques. First, the multi-granularity feature extractor (MGFE) module, designed with a multi-branch architecture and class token fusion, effectively forms multi-granularity features, enhancing both discrimination ability and fine-grained coverage. Second, the ranking-aware triplet regularization (RATR) is introduced to reduce redundancy in features from multiple branches, enhancing the diversity of multi-granularity features by incorporating both intra-class and inter-class diversity constraints, thus ensuring the robustness of person features. To our knowledge, this is the pioneering work that integrates a purely Mamba-driven approach into ReID research. Our proposed ReIDMamba model boasts only one-third the parameters of TransReID, along with lower GPU memory usage and faster inference throughput. Experimental results demonstrate ReIDMamba's superior and promising performance, achieving state-of-the-art performance on five person ReID benchmarks. Code is available at https://github.com/GuHY777/ReIDMamba.", "date": "2025-11-11", "categories": []}
{"id": "2511.08090", "title": "StableMorph: High-Quality Face Morph Generation with Stable Diffusion", "abstract": "Face morphing attacks threaten the integrity of biometric identity systems by enabling multiple individuals to share a single identity. To develop and evaluate effective morphing attack detection (MAD) systems, we need access to high-quality, realistic morphed images that reflect the challenges posed in real-world scenarios. However, existing morph generation methods often produce images that are blurry, riddled with artifacts, or poorly constructed making them easy to detect and not representative of the most dangerous attacks. In this work, we introduce StableMorph, a novel approach that generates highly realistic, artifact-free morphed face images using modern diffusion-based image synthesis. Unlike prior methods, StableMorph produces full-head images with sharp details, avoids common visual flaws, and offers unmatched control over visual attributes. Through extensive evaluation, we show that StableMorph images not only rival or exceed the quality of genuine face images but also maintain a strong ability to fool face recognition systems posing a greater challenge to existing MAD solutions and setting a new standard for morph quality in research and operational testing. StableMorph improves the evaluation of biometric security by creating more realistic and effective attacks and supports the development of more robust detection systems.", "date": "2025-11-11", "categories": []}
{"id": "2511.08114", "title": "Introducing Nylon Face Mask Attacks: A Dataset for Evaluating Generalised Face Presentation Attack Detection", "abstract": "Face recognition systems are increasingly deployed across a wide range of applications, including smartphone authentication, access control, and border security. However, these systems remain vulnerable to presentation attacks (PAs), which can significantly compromise their reliability. In this work, we introduce a new dataset focused on a novel and realistic presentation attack instrument called Nylon Face Masks (NFMs), designed to simulate advanced 3D spoofing scenarios. NFMs are particularly concerning due to their elastic structure and photorealistic appearance, which enable them to closely mimic the victim's facial geometry when worn by an attacker. To reflect real-world smartphone-based usage conditions, we collected the dataset using an iPhone 11 Pro, capturing 3,760 bona fide samples from 100 subjects and 51,281 NFM attack samples across four distinct presentation scenarios involving both humans and mannequins. We benchmark the dataset using five state-of-the-art PAD methods to evaluate their robustness under unseen attack conditions. The results demonstrate significant performance variability across methods, highlighting the challenges posed by NFMs and underscoring the importance of developing PAD techniques that generalise effectively to emerging spoofing threats.", "date": "2025-11-11", "categories": []}
{"id": "2511.08119", "title": "LatentPrintFormer: A Hybrid CNN-Transformer with Spatial Attention for Latent Fingerprint identification", "abstract": "Latent fingerprint identification remains a challenging task due to low image quality, background noise, and partial impressions. In this work, we propose a novel identification approach called LatentPrintFormer. The proposed model integrates a CNN backbone (EfficientNet-B0) and a Transformer backbone (Swin Tiny) to extract both local and global features from latent fingerprints. A spatial attention module is employed to emphasize high-quality ridge regions while suppressing background noise. The extracted features are fused and projected into a unified 512-dimensional embedding, and matching is performed using cosine similarity in a closed-set identification setting. Extensive experiments on two publicly available datasets demonstrate that LatentPrintFormer consistently outperforms three state-of-the-art latent fingerprint recognition techniques, achieving higher identification rates across Rank-10.", "date": "2025-11-11", "categories": []}
{"id": "2409.08782", "title": "Improving Contactless Fingerprint Recognition with Robust 3D Feature Extraction and Graph Embedding", "abstract": "Contactless fingerprint has gained lots of attention in recent fingerprint studies. However, most existing contactless fingerprint algorithms treat contactless fingerprints as 2D plain fingerprints, and still utilize traditional contact-based 2D fingerprints recognition methods. This recognition approach lacks consideration of the modality difference between contactless and contact fingerprints, especially the intrinsic 3D features in contactless fingerprints. This paper proposes a novel contactless fingerprint recognition algorithm that captures the revealed 3D feature of contactless fingerprints rather than the plain 2D feature. The proposed method first recovers 3D features from the input contactless fingerprint, including the 3D shape model and 3D fingerprint feature (minutiae, orientation, etc.). Then, a novel 3D graph matching method is proposed according to the extracted 3D feature. Additionally, the proposed method is able to perform robust 3D feature extractions on various contactless fingerprints across multiple finger poses. The results of the experiments on contactless fingerprint databases show that the proposed method successfully improves the matching accuracy of contactless fingerprints. Exceptionally, our method performs stably across multiple poses of contactless fingerprints due to 3D embeddings, which is a great advantage compared to 2D-based previous contactless fingerprint recognition algorithms.", "date": "2024-09-13", "categories": []}
{"id": "2507.13659", "title": "When Person Re-Identification Meets Event Camera: A Benchmark Dataset and An Attribute-guided Re-Identification Framework", "abstract": "Recent researchers have proposed using event cameras for person re-identification (ReID) due to their promising performance and better balance in terms of privacy protection, event camera-based person ReID has attracted significant attention. Currently, mainstream event-based person ReID algorithms primarily focus on fusing visible light and event stream, as well as preserving privacy. Although significant progress has been made, these methods are typically trained and evaluated on small-scale or simulated event camera datasets, making it difficult to assess their real identification performance and generalization ability. To address the issue of data scarcity, this paper introduces a large-scale RGB-event based person ReID dataset, called EvReID. The dataset contains 118,988 image pairs and covers 1200 pedestrian identities, with data collected across multiple seasons, scenes, and lighting conditions. We also evaluate 15 state-of-the-art person ReID algorithms, laying a solid foundation for future research in terms of both data and benchmarking. Based on our newly constructed dataset, this paper further proposes a pedestrian attribute-guided contrastive learning framework to enhance feature learning for person re-identification, termed TriPro-ReID. This framework not only effectively explores the visual features from both RGB frames and event streams, but also fully utilizes pedestrian attributes as mid-level semantic features. Extensive experiments on the EvReID dataset and MARS datasets fully validated the effectiveness of our proposed RGB-Event person ReID framework. The benchmark dataset and source code will be released on https://github.com/Event-AHU/Neuromorphic_ReID", "date": "2025-07-18", "categories": []}
{"id": "2511.05574", "title": "Elements of Active Continuous Learning and Uncertainty Self-Awareness: a Narrow Implementation for Face and Facial Expression Recognition", "abstract": "Reflection on one's thought process and making corrections to it if there exists dissatisfaction in its performance is, perhaps, one of the essential traits of intelligence. However, such high-level abstract concepts mandatory for Artificial General Intelligence can be modelled even at the low level of narrow Machine Learning algorithms. Here, we present the self-awareness mechanism emulation in the form of a supervising artificial neural network (ANN) observing patterns in activations of another underlying ANN in a search for indications of the high uncertainty of the underlying ANN and, therefore, the trustworthiness of its predictions. The underlying ANN is a convolutional neural network (CNN) ensemble employed for face recognition and facial expression tasks. The self-awareness ANN has a memory region where its past performance information is stored, and its learnable parameters are adjusted during the training to optimize the performance. The trustworthiness verdict triggers the active learning mode, giving elements of agency to the machine learning algorithm that asks for human help in high uncertainty and confusion conditions.", "date": "2025-11-04", "categories": []}
{"id": "2501.01032", "title": "DynamicLip: Shape-Independent Continuous Authentication via Lip Articulator Dynamics", "abstract": "Biometrics authentication has become increasingly popular due to its security and convenience; however, traditional biometrics are becoming less desirable in scenarios such as new mobile devices, Virtual Reality, and Smart Vehicles. For example, while face authentication is widely used, it suffers from significant privacy concerns. The collection of complete facial data makes it less desirable for privacy-sensitive applications. Lip authentication, on the other hand, has emerged as a promising biometrics method. However, existing lip-based authentication methods heavily depend on static lip shape when the mouth is closed, which can be less robust due to lip shape dynamic motion and can barely work when the user is speaking. In this paper, we revisit the nature of lip biometrics and extract shape-independent features from the lips. We study the dynamic characteristics of lip biometrics based on articulator motion. Building on the knowledge, we propose a system for shape-independent continuous authentication via lip articulator dynamics. This system enables robust, shape-independent and continuous authentication, making it particularly suitable for scenarios with high security and privacy requirements. We conducted comprehensive experiments in different environments and attack scenarios and collected a dataset of 50 subjects. The results indicate that our system achieves an overall accuracy of 99.06% and demonstrates robustness under advanced mimic attacks and AI deepfake attacks, making it a viable solution for continuous biometric authentication in various applications.", "date": "2025-01-02", "categories": []}
{"id": "2511.05092", "title": "A Dual-stage Prompt-driven Privacy-preserving Paradigm for Person Re-Identification", "abstract": "With growing concerns over data privacy, researchers have started using virtual data as an alternative to sensitive real-world images for training person re-identification (Re-ID) models. However, existing virtual datasets produced by game engines still face challenges such as complex construction and poor domain generalization, making them difficult to apply in real scenarios. To address these challenges, we propose a Dual-stage Prompt-driven Privacy-preserving Paradigm (DPPP). In the first stage, we generate rich prompts incorporating multi-dimensional attributes such as pedestrian appearance, illumination, and viewpoint that drive the diffusion model to synthesize diverse data end-to-end, building a large-scale virtual dataset named GenePerson with 130,519 images of 6,641 identities. In the second stage, we propose a Prompt-driven Disentanglement Mechanism (PDM) to learn domain-invariant generalization features. With the aid of contrastive learning, we employ two textual inversion networks to map images into pseudo-words representing style and content, respectively, thereby constructing style-disentangled content prompts to guide the model in learning domain-invariant content features at the image level. Experiments demonstrate that models trained on GenePerson with PDM achieve state-of-the-art generalization performance, surpassing those on popular real and virtual Re-ID datasets.", "date": "2025-11-07", "categories": []}
{"id": "2503.18469", "title": "CFReID: Continual Few-shot Person Re-Identification", "abstract": "Real-world surveillance systems are dynamically evolving, requiring a person Re-identification model to continuously handle newly incoming data from various domains. To cope with these dynamics, Lifelong ReID (LReID) has been proposed to learn and accumulate knowledge across multiple domains incrementally. However, LReID models need to be trained on large-scale labeled data for each unseen domain, which are typically inaccessible due to privacy and cost concerns. In this paper, we propose a new paradigm called Continual Few-shot ReID (CFReID), which requires models to be incrementally trained using few-shot data and tested on all seen domains. Under few-shot conditions, CFREID faces two core challenges: 1) learning knowledge from few-shot data of unseen domain, and 2) avoiding catastrophic forgetting of seen domains. To tackle these two challenges, we propose a Stable Distribution Alignment (SDA) framework from feature distribution perspective. Specifically, our SDA is composed of two modules, i.e., Meta Distribution Alignment (MDA) and Prototype-based Few-shot Adaptation (PFA). To support the study of CFReID, we establish an evaluation benchmark for CFReID on five publicly available ReID datasets. Extensive experiments demonstrate that our SDA can enhance the few-shot learning and anti-forgetting capabilities under few-shot conditions. Notably, our approach, using only 5\\% of the data, i.e., 32 IDs, significantly outperforms LReID's state-of-the-art performance, which requires 700 to 1,000 IDs.", "date": "2025-03-24", "categories": []}
{"id": "2511.04037", "title": "A Hybrid Deep Learning Model for Robust Biometric Authentication from Low-Frame-Rate PPG Signals", "abstract": "Photoplethysmography (PPG) signals, which measure changes in blood volume in the skin using light, have recently gained attention in biometric authentication because of their non-invasive acquisition, inherent liveness detection, and suitability for low-cost wearable devices. However, PPG signal quality is challenged by motion artifacts, illumination changes, and inter-subject physiological variability, making robust feature extraction and classification crucial. This study proposes a lightweight and cost-effective biometric authentication framework based on PPG signals extracted from low-frame-rate fingertip videos. The CFIHSR dataset, comprising PPG recordings from 46 subjects at a sampling rate of 14 Hz, is employed for evaluation. The raw PPG signals undergo a standard preprocessing pipeline involving baseline drift removal, motion artifact suppression using Principal Component Analysis (PCA), bandpass filtering, Fourier-based resampling, and amplitude normalization. To generate robust representations, each one-dimensional PPG segment is converted into a two-dimensional time-frequency scalogram via the Continuous Wavelet Transform (CWT), effectively capturing transient cardiovascular dynamics. We developed a hybrid deep learning model, termed CVT-ConvMixer-LSTM, by combining spatial features from the Convolutional Vision Transformer (CVT) and ConvMixer branches with temporal features from a Long Short-Term Memory network (LSTM). The experimental results on 46 subjects demonstrate an authentication accuracy of 98%, validating the robustness of the model to noise and variability between subjects. Due to its efficiency, scalability, and inherent liveness detection capability, the proposed system is well-suited for real-world mobile and embedded biometric security applications.", "date": "2025-11-06", "categories": []}
{"id": "2506.10564", "title": "Balancing Tails when Comparing Distributions: Comprehensive Equity Index (CEI) with Application to Bias Evaluation in Operational Face Biometrics", "abstract": "Demographic bias in high-performance face recognition (FR) systems often eludes detection by existing metrics, especially with respect to subtle disparities in the tails of the score distribution. We introduce the Comprehensive Equity Index (CEI), a novel metric designed to address this limitation. CEI uniquely analyzes genuine and impostor score distributions separately, enabling a configurable focus on tail probabilities while also considering overall distribution shapes. Our extensive experiments (evaluating state-of-the-art FR systems, intentionally biased models, and diverse datasets) confirm CEI's superior ability to detect nuanced biases where previous methods fall short. Furthermore, we present CEI^A, an automated version of the metric that enhances objectivity and simplifies practical application. CEI provides a robust and sensitive tool for operational FR fairness assessment. The proposed methods have been developed particularly for bias evaluation in face biometrics but, in general, they are applicable for comparing statistical distributions in any problem where one is interested in analyzing the distribution tails.", "date": "2025-06-12", "categories": []}
{"id": "2511.02564", "title": "Seeing Across Time and Views: Multi-Temporal Cross-View Learning for Robust Video Person Re-Identification", "abstract": "Video-based person re-identification (ReID) in cross-view domains (for example, aerial-ground surveillance) remains an open problem because of extreme viewpoint shifts, scale disparities, and temporal inconsistencies. To address these challenges, we propose MTF-CVReID, a parameter-efficient framework that introduces seven complementary modules over a ViT-B/16 backbone. Specifically, we include: (1) Cross-Stream Feature Normalization (CSFN) to correct camera and view biases; (2) Multi-Resolution Feature Harmonization (MRFH) for scale stabilization across altitudes; (3) Identity-Aware Memory Module (IAMM) to reinforce persistent identity traits; (4) Temporal Dynamics Modeling (TDM) for motion-aware short-term temporal encoding; (5) Inter-View Feature Alignment (IVFA) for perspective-invariant representation alignment; (6) Hierarchical Temporal Pattern Learning (HTPL) to capture multi-scale temporal regularities; and (7) Multi-View Identity Consistency Learning (MVICL) that enforces cross-view identity coherence using a contrastive learning paradigm. Despite adding only about 2 million parameters and 0.7 GFLOPs over the baseline, MTF-CVReID maintains real-time efficiency (189 FPS) and achieves state-of-the-art performance on the AG-VPReID benchmark across all altitude levels, with strong cross-dataset generalization to G2A-VReID and MARS datasets. These results show that carefully designed adapter-based modules can substantially enhance cross-view robustness and temporal consistency without compromising computational efficiency. The source code is available at https://github.com/MdRashidunnabi/MTF-CVReID", "date": "2025-11-04", "categories": []}
{"id": "2511.02685", "title": "Modality-Transition Representation Learning for Visible-Infrared Person Re-Identification", "abstract": "Visible-infrared person re-identification (VI-ReID) technique could associate the pedestrian images across visible and infrared modalities in the practical scenarios of background illumination changes. However, a substantial gap inherently exists between these two modalities. Besides, existing methods primarily rely on intermediate representations to align cross-modal features of the same person. The intermediate feature representations are usually create by generating intermediate images (kind of data enhancement), or fusing intermediate features (more parameters, lack of interpretability), and they do not make good use of the intermediate features. Thus, we propose a novel VI-ReID framework via Modality-Transition Representation Learning (MTRL) with a middle generated image as a transmitter from visible to infrared modals, which are fully aligned with the original visible images and similar to the infrared modality. After that, using a modality-transition contrastive loss and a modality-query regularization loss for training, which could align the cross-modal features more effectively. Notably, our proposed framework does not need any additional parameters, which achieves the same inference speed to the backbone while improving its performance on VI-ReID task. Extensive experimental results illustrate that our model significantly and consistently outperforms existing SOTAs on three typical VI-ReID datasets.", "date": "2025-11-04", "categories": []}
{"id": "2505.12580", "title": "Coarse Attribute Prediction with Task Agnostic Distillation for Real World Clothes Changing ReID", "abstract": "This work focuses on Clothes Changing Re-IDentification (CC-ReID) for the real world. Existing works perform well with high-quality (HQ) images, but struggle with low-quality (LQ) where we can have artifacts like pixelation, out-of-focus blur, and motion blur. These artifacts introduce noise to not only external biometric attributes (e.g. pose, body shape, etc.) but also corrupt the model's internal feature representation. Models usually cluster LQ image features together, making it difficult to distinguish between them, leading to incorrect matches. We propose a novel framework Robustness against Low-Quality (RLQ) to improve CC-ReID model on real-world data. RLQ relies on Coarse Attributes Prediction (CAP) and Task Agnostic Distillation (TAD) operating in alternate steps in a novel training mechanism. CAP enriches the model with external fine-grained attributes via coarse predictions, thereby reducing the effect of noisy inputs. On the other hand, TAD enhances the model's internal feature representation by bridging the gap between HQ and LQ features, via an external dataset through task-agnostic self-supervision and distillation. RLQ outperforms the existing approaches by 1.6%-2.9% Top-1 on real-world datasets like LaST, and DeepChange, while showing consistent improvement of 5.3%-6% Top-1 on PRCC with competitive performance on LTCC. *The code will be made public soon.*", "date": "2025-05-19", "categories": []}
{"id": "2511.01498", "title": "EPAN: Robust Pedestrian Re-Identification via Enhanced Alignment Network for IoT Surveillance", "abstract": "Person re-identification (ReID) plays a pivotal role in computer vision, particularly in surveillance and security applications within IoT-enabled smart environments. This study introduces the Enhanced Pedestrian Alignment Network (EPAN), tailored for robust ReID across diverse IoT surveillance conditions. EPAN employs a dual-branch architecture to mitigate the impact of perspective and environmental changes, extracting alignment information under varying scales and viewpoints. Here, we demonstrate EPAN's strong feature extraction capabilities, achieving outstanding performance on the Inspection-Personnel dataset with a Rank-1 accuracy of 90.09% and a mean Average Precision (mAP) of 78.82%. This highlights EPAN's potential for real-world IoT applications, enabling effective and reliable person ReID across diverse cameras in surveillance and security systems. The code and data are available at: https://github.com/ggboy2580/EPAN", "date": "2025-11-03", "categories": []}
{"id": "2511.01546", "title": "PCD-ReID: Occluded Person Re-Identification for Base Station Inspection", "abstract": "Occluded pedestrian re-identification (ReID) in base station environments is a critical task in computer vision, particularly for surveillance and security applications. This task faces numerous challenges, as occlusions often obscure key body features, increasing the complexity of identification. Traditional ResNet-based ReID algorithms often fail to address occlusions effectively, necessitating new ReID methods. We propose the PCD-ReID (Pedestrian Component Discrepancy) algorithm to address these issues. The contributions of this work are as follows: To tackle the occlusion problem, we design a Transformer-based PCD network capable of extracting shared component features, such as helmets and uniforms. To mitigate overfitting on public datasets, we collected new real-world patrol surveillance images for model training, covering six months, 10,000 individuals, and over 50,000 images. Comparative experiments with existing ReID algorithms demonstrate that our model achieves a mean Average Precision (mAP) of 79.0% and a Rank-1 accuracy of 82.7%, marking a 15.9% Rank-1 improvement over ResNet50-based methods. Experimental evaluations indicate that PCD-ReID effectively achieves occlusion-aware ReID performance for personnel in tower inspection scenarios, highlighting its potential for practical deployment in surveillance and security applications.", "date": "2025-11-03", "categories": []}
{"id": "2510.27677", "title": "Vision Transformer for Robust Occluded Person Reidentification in Complex Surveillance Scenes", "abstract": "Person re-identification (ReID) in surveillance is challenged by occlusion, viewpoint distortion, and poor image quality. Most existing methods rely on complex modules or perform well only on clear frontal images. We propose Sh-ViT (Shuffling Vision Transformer), a lightweight and robust model for occluded person ReID. Built on ViT-Base, Sh-ViT introduces three components: First, a Shuffle module in the final Transformer layer to break spatial correlations and enhance robustness to occlusion and blur; Second, scenario-adapted augmentation (geometric transforms, erasing, blur, and color adjustment) to simulate surveillance conditions; Third, DeiT-based knowledge distillation to improve learning with limited labels.To support real-world evaluation, we construct the MyTT dataset, containing over 10,000 pedestrians and 30,000+ images from base station inspections, with frequent equipment occlusion and camera variations. Experiments show that Sh-ViT achieves 83.2% Rank-1 and 80.1% mAP on MyTT, outperforming CNN and ViT baselines, and 94.6% Rank-1 and 87.5% mAP on Market1501, surpassing state-of-the-art methods.In summary, Sh-ViT improves robustness to occlusion and blur without external modules, offering a practical solution for surveillance-based personnel monitoring.", "date": "2025-10-31", "categories": []}
{"id": "2311.07734", "title": "Quality-Aware Prototype Memory for Face Representation Learning", "abstract": "Prototype Memory is a powerful model for face representation learning. It enables training face recognition models on datasets of any size by generating prototypes (classifier weights) on the fly and efficiently utilizing them. Prototype Memory demonstrated strong results in many face recognition benchmarks. However, the algorithm of prototype generation, used in it, is prone to the problems of imperfectly calculated prototypes in case of low-quality or poorly recognizable faces in the images, selected for the prototype creation. All images of the same person presented in the mini-batch are used with equal weights, and the resulting averaged prototype can be contaminated by imperfect embeddings of low-quality face images. This may lead to misleading training signals and degrade the performance of the trained models. In this paper, we propose a simple and effective way to improve Prototype Memory with quality-aware prototype generation. Quality-Aware Prototype Memory uses different weights for images of different quality in the process of prototype generation. With this improvement, prototypes receive more informative signals from high-quality images and are less affected by low-quality ones. We propose and compare several methods of quality estimation and usage, perform extensive experiments on the different face recognition benchmarks and demonstrate the advantages of the proposed model compared to the basic version of Prototype Memory.", "date": "2023-11-13", "categories": []}
{"id": "2510.26282", "title": "Exploring Complementarity and Explainability in CNNs for Periocular Verification Across Acquisition Distances", "abstract": "We study the complementarity of different CNNs for periocular verification at different distances on the UBIPr database. We train three architectures of increasing complexity (SqueezeNet, MobileNetv2, and ResNet50) on a large set of eye crops from VGGFace2. We analyse performance with cosine and chi2 metrics, compare different network initialisations, and apply score-level fusion via logistic regression. In addition, we use LIME heatmaps and Jensen-Shannon divergence to compare attention patterns of the CNNs. While ResNet50 consistently performs best individually, the fusion provides substantial gains, especially when combining all three networks. Heatmaps show that networks usually focus on distinct regions of a given image, which explains their complementarity. Our method significantly outperforms previous works on UBIPr, achieving a new state-of-the-art.", "date": "2025-10-30", "categories": []}
{"id": "2510.26294", "title": "Leveraging Large-Scale Face Datasets for Deep Periocular Recognition via Ocular Cropping", "abstract": "We focus on ocular biometrics, specifically the periocular region (the area around the eye), which offers high discrimination and minimal acquisition constraints. We evaluate three Convolutional Neural Network architectures of varying depth and complexity to assess their effectiveness for periocular recognition. The networks are trained on 1,907,572 ocular crops extracted from the large-scale VGGFace2 database. This significantly contrasts with existing works, which typically rely on small-scale periocular datasets for training having only a few thousand images. Experiments are conducted with ocular images from VGGFace2-Pose, a subset of VGGFace2 containing in-the-wild face images, and the UFPR-Periocular database, which consists of selfies captured via mobile devices with user guidance on the screen. Due to the uncontrolled conditions of VGGFace2, the Equal Error Rates (EERs) obtained with ocular crops range from 9-15%, noticeably higher than the 3-6% EERs achieved using full-face images. In contrast, UFPR-Periocular yields significantly better performance (EERs of 1-2%), thanks to higher image quality and more consistent acquisition protocols. To the best of our knowledge, these are the lowest reported EERs on the UFPR dataset to date.", "date": "2025-10-30", "categories": []}
{"id": "2510.25184", "title": "Mask-Robust Face Verification for Online Learning via YOLOv5 and Residual Networks", "abstract": "In the contemporary landscape, the fusion of information technology and the rapid advancement of artificial intelligence have ushered school education into a transformative phase characterized by digitization and heightened intelligence. Concurrently, the global paradigm shift caused by the Covid-19 pandemic has catalyzed the evolution of e-learning, accentuating its significance. Amidst these developments, one pivotal facet of the online education paradigm that warrants attention is the authentication of identities within the digital learning sphere. Within this context, our study delves into a solution for online learning authentication, utilizing an enhanced convolutional neural network architecture, specifically the residual network model. By harnessing the power of deep learning, this technological approach aims to galvanize the ongoing progress of online education, while concurrently bolstering its security and stability. Such fortification is imperative in enabling online education to seamlessly align with the swift evolution of the educational landscape. This paper's focal proposition involves the deployment of the YOLOv5 network, meticulously trained on our proprietary dataset. This network is tasked with identifying individuals' faces culled from images captured by students' open online cameras. The resultant facial information is then channeled into the residual network to extract intricate features at a deeper level. Subsequently, a comparative analysis of Euclidean distances against students' face databases is performed, effectively ascertaining the identity of each student.", "date": "2025-10-29", "categories": []}
{"id": "2504.06131", "title": "FaceCloak: Learning to Protect Face Templates", "abstract": "Generative models can reconstruct face images from encoded representations (templates) bearing remarkable likeness to the original face, raising security and privacy concerns. We present \\textsc{FaceCloak}, a neural network framework that protects face templates by generating smart, renewable binary cloaks. Our method proactively thwarts inversion attacks by cloaking face templates with unique disruptors synthesized from a single face template on the fly while provably retaining biometric utility and unlinkability. Our cloaked templates can suppress sensitive attributes while generalizing to novel feature extraction schemes and outperform leading baselines in terms of biometric matching and resiliency to reconstruction attacks. \\textsc{FaceCloak}-based matching is extremely fast (inference time =0.28 ms) and light (0.57 MB). We have released our \\href{https://github.com/sudban3089/FaceCloak.git}{code} for reproducible research.", "date": "2025-04-08", "categories": []}
{"id": "2505.08423", "title": "DArFace: Deformation Aware Robustness for Low Quality Face Recognition", "abstract": "Facial recognition systems have achieved remarkable success by leveraging deep neural networks, advanced loss functions, and large-scale datasets. However, their performance often deteriorates in real-world scenarios involving low-quality facial images. Such degradations, common in surveillance footage or standoff imaging include low resolution, motion blur, and various distortions, resulting in a substantial domain gap from the high-quality data typically used during training. While existing approaches attempt to address robustness by modifying network architectures or modeling global spatial transformations, they frequently overlook local, non-rigid deformations that are inherently present in real-world settings. In this work, we introduce \\textbf{DArFace}, a \\textbf{D}eformation-\\textbf{A}ware \\textbf{r}obust \\textbf{Face} recognition framework that enhances robustness to such degradations without requiring paired high- and low-quality training samples. Our method adversarially integrates both global transformations (e.g., rotation, translation) and local elastic deformations during training to simulate realistic low-quality conditions. Moreover, we introduce a contrastive objective to enforce identity consistency across different deformed views. Extensive evaluations on low-quality benchmarks including TinyFace, IJB-B, and IJB-C demonstrate that DArFace surpasses state-of-the-art methods, with significant gains attributed to the inclusion of local deformation modeling.", "date": "2025-05-13", "categories": []}
{"id": "2302.10883", "title": "Blockchain and Biometrics: Survey, GDPR Analysis, and Future Directions", "abstract": "Biometric recognition as an efficient and hard-to-forge way of identification and verification has become an indispensable part of the current digital world. The fast evolution of this technology has been a strong incentive for integration into many applications. Meanwhile, blockchain, the decentralized ledger technology, has been widely received by both research and industry in the past few years, and it is being increasingly deployed today in many different applications, such as money transfer, IoT, healthcare, or logistics. Recently, researchers have started to speculate on the pros and cons and what the best applications would be when these two technologies cross paths. This paper provides a survey of the research literature on the combination of blockchain and biometrics and includes a first legal analysis of this integration based on GDPR to shed light on challenges and potentials. Although the integration of blockchain technology into the biometric sector is still in its infancy, with a growing body of literature discussing specific applications and advanced technological setups, this paper aims to provide a holistic understanding of blockchain applicability in biometrics. Based on published studies, this article discusses, among others, practical examples combining blockchain and biometrics for novel applications in PKI systems, distributed trusted services, and identity management. Challenges and limitations when combining blockchain and biometrics that motivate future work will also be discussed; e.g., blockchain networks at their current stage may not be efficient or economical for some real-time biometric applications. Finally, we also discuss key legal aspects of the EU General Data Protection Regulation (GDPR) related to this combination of technologies (blockchain and biometrics); for example, accountability, immutability, anonymity, and data protection elements.", "date": "2023-02-21", "categories": []}
{"id": "2308.10133", "title": "TransFace++: Rethinking the Face Recognition Paradigm with a Focus on Accuracy, Efficiency, and Security", "abstract": "Face Recognition (FR) technology has made significant strides with the emergence of deep learning. Typically, most existing FR models are built upon Convolutional Neural Networks (CNN) and take RGB face images as the model's input. In this work, we take a closer look at existing FR paradigms from high-efficiency, security, and precision perspectives, and identify the following three problems: (i) CNN frameworks are vulnerable in capturing global facial features and modeling the correlations between local facial features. (ii) Selecting RGB face images as the model's input greatly degrades the model's inference efficiency, increasing the extra computation costs. (iii) In the real-world FR system that operates on RGB face images, the integrity of user privacy may be compromised if hackers successfully penetrate and gain access to the input of this model. To solve these three issues, we propose two novel FR frameworks, i.e., TransFace and TransFace++, which successfully explore the feasibility of applying ViTs and image bytes to FR tasks, respectively. Experiments on popular face benchmarks demonstrate the superiority of our TransFace and TransFace++. Code is available at https://github.com/DanJun6737/TransFace_pp.", "date": "2023-08-20", "categories": []}
{"id": "2510.21778", "title": "Ageing Drift in Binary Face Templates: A Bits-per-Decade Analysis", "abstract": "We study the longitudinal stability of compact binary face templates and quantify ageing drift directly in bits per decade. Float embeddings from a modern face CNN are compressed with PCA-ITQ into 64- and 128-bit codes. For each identity in AgeDB with at least three distinct ages, we form all genuine pairs and fit a per-identity linear model of Hamming distance versus absolute age gap. Across 566 identities, the median slope is 1.357 bits per decade for 64-bit templates and 2.571 bits per decade for 128-bit templates, with tight non-parametric 95 percent bootstrap confidence intervals. The distributions are predominantly positive, indicating a small but systematic increase in intra-class distance over time. Because drift scales with code length, shorter codes are inherently more age-stable at a fixed decision threshold. We connect these slopes to operating characteristics by reporting EER and TPR at FAR = 1 percent in three age bins. We discuss implications for smart-card and match-on-card deployments, including simple mitigations such as periodic re-enrolment and targeted parity on empirically unstable bit positions. Code and CSV artifacts are provided to support reproducibility.", "date": "2025-10-17", "categories": []}
{"id": "2510.22133", "title": "HandPass: A Wi-Fi CSI Palm Authentication Approach for Access Control", "abstract": "Wi-Fi Channel State Information (CSI) has been extensively studied for sensing activities. However, its practical application in user authentication still needs to be explored. This study presents a novel approach to biometric authentication using Wi-Fi Channel State Information (CSI) data for palm recognition. The research delves into utilizing a Raspberry Pi encased in a custom-built box with antenna power reduced to 1dBm, which was used to capture CSI data from the right hands of 20 participants (10 men and 10 women). The dataset was normalized using MinMax scaling to ensure uniformity and accuracy. By focusing on biophysical aspects such as hand size, shape, angular spread between fingers, and finger phalanx lengths, among other characteristics, the study explores how these features affect electromagnetic signals, which are then reflected in Wi-Fi CSI, allowing for precise user identification. Five classification algorithms were evaluated, with the Random Forest classifier achieving an average F1-Score of 99.82\\% using 10-fold cross-validation. Amplitude and Phase data were used, with each capture session recording approximately 1000 packets per second in five 5-second intervals for each User. This high accuracy highlights the potential of Wi-Fi CSI in developing robust and reliable user authentication systems based on palm biometric data.", "date": "2025-10-25", "categories": []}
{"id": "2510.22937", "title": "Bi-Encoder Contrastive Learning for Fingerprint and Iris Biometrics", "abstract": "There has been a historic assumption that the biometrics of an individual are statistically uncorrelated. We test this assumption by training Bi-Encoder networks on three verification tasks, including fingerprint-to-fingerprint matching, iris-to-iris matching, and cross-modal fingerprint-to-iris matching using 274 subjects with $\\sim$100k fingerprints and 7k iris images. We trained ResNet-50 and Vision Transformer backbones in Bi-Encoder architectures such that the contrastive loss between images sampled from the same individual is minimized. The iris ResNet architecture reaches 91 ROC AUC score for iris-to-iris matching, providing clear evidence that the left and right irises of an individual are correlated. Fingerprint models reproduce the positive intra-subject suggested by prior work in this space. This is the first work attempting to use Vision Transformers for this matching. Cross-modal matching rises only slightly above chance, which suggests that more data and a more sophisticated pipeline is needed to obtain compelling results. These findings continue challenge independence assumptions of biometrics and we plan to extend this work to other biometrics in the future. Code available: https://github.com/MatthewSo/bio_fingerprints_iris.", "date": "2025-10-27", "categories": []}
{"id": "2410.10587", "title": "TopoFR: A Closer Look at Topology Alignment on Face Recognition", "abstract": "The field of face recognition (FR) has undergone significant advancements with the rise of deep learning. Recently, the success of unsupervised learning and graph neural networks has demonstrated the effectiveness of data structure information. Considering that the FR task can leverage large-scale training data, which intrinsically contains significant structure information, we aim to investigate how to encode such critical structure information into the latent space. As revealed from our observations, directly aligning the structure information between the input and latent spaces inevitably suffers from an overfitting problem, leading to a structure collapse phenomenon in the latent space. To address this problem, we propose TopoFR, a novel FR model that leverages a topological structure alignment strategy called PTSA and a hard sample mining strategy named SDE. Concretely, PTSA uses persistent homology to align the topological structures of the input and latent spaces, effectively preserving the structure information and improving the generalization performance of FR model. To mitigate the impact of hard samples on the latent space structure, SDE accurately identifies hard samples by automatically computing structure damage score (SDS) for each sample, and directs the model to prioritize optimizing these samples. Experimental results on popular face benchmarks demonstrate the superiority of our TopoFR over the state-of-the-art methods. Code and models are available at: https://github.com/modelscope/facechain/tree/main/face_module/TopoFR.", "date": "2024-10-14", "categories": []}
{"id": "2503.11544", "title": "AugGen: Synthetic Augmentation using Diffusion Models Can Improve Recognition", "abstract": "The increasing reliance on large-scale datasets in machine learning poses significant privacy and ethical challenges, particularly in sensitive domains such as face recognition. Synthetic data generation offers a promising alternative; however, most existing methods depend heavily on external datasets or pre-trained models, increasing complexity and resource demands. In this paper, we introduce AugGen, a self-contained synthetic augmentation technique. AugGen strategically samples from a class-conditional generative model trained exclusively on the target FR dataset, eliminating the need for external resources. Evaluated across 8 FR benchmarks, including IJB-C and IJB-B, our method achieves 1-12% performance improvements, outperforming models trained solely on real data and surpassing state-of-the-art synthetic data generation approaches, while using less real data. Notably, these gains often exceed those from architectural enhancements, underscoring the value of synthetic augmentation in data-limited scenarios. Our findings demonstrate that carefully integrated synthetic data can both mitigate privacy constraints and substantially enhance recognition performance. Paper website: https://parsa-ra.github.io/auggen/.", "date": "2025-03-14", "categories": []}
{"id": "2510.19321", "title": "Online Handwritten Signature Verification Based on Temporal-Spatial Graph Attention Transformer", "abstract": "Handwritten signature verification is a crucial aspect of identity authentication, with applications in various domains such as finance and e-commerce. However, achieving high accuracy in signature verification remains challenging due to intra-user variability and the risk of forgery. This paper introduces a novel approach for dynamic signature verification: the Temporal-Spatial Graph Attention Transformer (TS-GATR). TS-GATR combines the Graph Attention Network (GAT) and the Gated Recurrent Unit (GRU) to model both spatial and temporal dependencies in signature data. TS-GATR enhances verification performance by representing signatures as graphs, where each node captures dynamic features (e.g. position, velocity, pressure), and by using attention mechanisms to model their complex relationships. The proposed method further employs a Dual-Graph Attention Transformer (DGATR) module, which utilizes k-step and k-nearest neighbor adjacency graphs to model local and global spatial features, respectively. To capture long-term temporal dependencies, the model integrates GRU, thereby enhancing its ability to learn dynamic features during signature verification. Comprehensive experiments conducted on benchmark datasets such as MSDS and DeepSignDB show that TS-GATR surpasses current state-of-the-art approaches, consistently achieving lower Equal Error Rates (EER) across various scenarios.", "date": "2025-10-22", "categories": []}
{"id": "2501.01166", "title": "Deep Learning in Palmprint Recognition-A Comprehensive Survey", "abstract": "Palmprint recognition has emerged as a prominent biometric technology, widely applied in diverse scenarios. Traditional handcrafted methods for palmprint recognition often fall short in representation capability, as they heavily depend on researchers' prior knowledge. Deep learning (DL) has been introduced to address this limitation, leveraging its remarkable successes across various domains. While existing surveys focus narrowly on specific tasks within palmprint recognition-often grounded in traditional methodologies-there remains a significant gap in comprehensive research exploring DL-based approaches across all facets of palmprint recognition. This paper bridges that gap by thoroughly reviewing recent advancements in DL-powered palmprint recognition. The paper systematically examines progress across key tasks, including region-of-interest segmentation, feature extraction, and security/privacy-oriented challenges. Beyond highlighting these advancements, the paper identifies current challenges and uncovers promising opportunities for future research. By consolidating state-of-the-art progress, this review serves as a valuable resource for researchers, enabling them to stay abreast of cutting-edge technologies and drive innovation in palmprint recognition.", "date": "2025-01-02", "categories": []}
{"id": "2506.09385", "title": "ReID5o: Achieving Omni Multi-modal Person Re-identification in a Single Model", "abstract": "In real-word scenarios, person re-identification (ReID) expects to identify a person-of-interest via the descriptive query, regardless of whether the query is a single modality or a combination of multiple modalities. However, existing methods and datasets remain constrained to limited modalities, failing to meet this requirement. Therefore, we investigate a new challenging problem called Omni Multi-modal Person Re-identification (OM-ReID), which aims to achieve effective retrieval with varying multi-modal queries. To address dataset scarcity, we construct ORBench, the first high-quality multi-modal dataset comprising 1,000 unique identities across five modalities: RGB, infrared, color pencil, sketch, and textual description. This dataset also has significant superiority in terms of diversity, such as the painting perspectives and textual information. It could serve as an ideal platform for follow-up investigations in OM-ReID. Moreover, we propose ReID5o, a novel multi-modal learning framework for person ReID. It enables synergistic fusion and cross-modal alignment of arbitrary modality combinations in a single model, with a unified encoding and multi-expert routing mechanism proposed. Extensive experiments verify the advancement and practicality of our ORBench. A wide range of possible models have been evaluated and compared on it, and our proposed ReID5o model gives the best performance. The dataset and code will be made publicly available at https://github.com/Zplusdragon/ReID5o_ORBench.", "date": "2025-06-11", "categories": []}
{"id": "2505.18200", "title": "CrossRF: A Domain-Invariant Deep Learning Approach for RF Fingerprinting", "abstract": "Radio Frequency (RF) fingerprinting offers a promising approach for drone identification and security, although it suffers from significant performance degradation when operating on different transmission channels. This paper presents CrossRF, a domain-invariant deep learning approach that addresses the problem of cross-channel RF fingerprinting for Unmanned Aerial Vehicle (UAV) identification. Our approach aims to minimize the domain gap between different RF channels by using adversarial learning to train a more robust model that maintains consistent identification performance despite channel variations. We validate our approach using the UAVSig dataset, comprising real-world over-the-air RF signals from identical drone models operating across several frequency channels, ensuring that the findings correspond to real-world scenarios. The experimental results show CrossRF's efficiency, achieving up to 99.03% accuracy when adapting from Channel 3 to Channel 4, compared to only 26.39% using conventional methods. The model maintains robust performance in more difficult multi-channel scenarios (87.57% accuracy adapting from Channels 1,3 to 2,4) and achieves 89.45% accuracy with 0.9 precision for controller classification. These results confirm CrossRF's ability to significantly reduce performance degradation due to cross-channel variations while maintaining high identification accuracy with minimal training data requirements, making it particularly suitable for practical drone security applications.", "date": "2025-05-21", "categories": []}
{"id": "2510.17169", "title": "Investigating Adversarial Robustness against Preprocessing used in Blackbox Face Recognition", "abstract": "Face Recognition (FR) models have been shown to be vulnerable to adversarial examples that subtly alter benign facial images, exposing blind spots in these systems, as well as protecting user privacy. End-to-end FR systems first obtain preprocessed faces from diverse facial imagery prior to computing the similarity of the deep feature embeddings. Whilst face preprocessing is a critical component of FR systems, and hence adversarial attacks against them, we observe that this preprocessing is often overlooked in blackbox settings. Our study seeks to investigate the transferability of several out-of-the-box state-of-the-art adversarial attacks against FR when applied against different preprocessing techniques used in a blackbox setting. We observe that the choice of face detection model can degrade the attack success rate by up to 78%, whereas choice of interpolation method during downsampling has relatively minimal impacts. Furthermore, we find that the requirement for facial preprocessing even degrades attack strength in a whitebox setting, due to the unintended interaction of produced noise vectors against face detection models. Based on these findings, we propose a preprocessing-invariant method using input transformations that improves the transferability of the studied attacks by up to 27%. Our findings highlight the importance of preprocessing in FR systems, and the need for its consideration towards improving the adversarial generalisation of facial adversarial examples.", "date": "2025-10-20", "categories": []}
{"id": "2510.17250", "title": "A Prototypical Network with an Attention-based Encoder for Drivers Identification Application", "abstract": "Driver identification has become an area of increasing interest in recent years, especially for data- driven applications, because biometric-based technologies may incur privacy issues. This study proposes a deep learning neural network architecture, an attention-based encoder (AttEnc), which uses an attention mechanism for driver identification and uses fewer model parameters than current methods. Most studies do not address the issue of data shortages for driver identification, and most of them are inflexible when encountering unknown drivers. In this study, an architecture that combines a prototypical network and an attention-based encoder (P-AttEnc) is proposed. It applies few-shot learning to overcome the data shortage issues and to enhance model generalizations. The experiments showed that the attention-based encoder can identify drivers with accuracies of 99.3%, 99.0% and 99.9% in three different datasets and has a prediction time that is 44% to 79% faster because it significantly reduces, on average, 87.6% of the model parameters. P-AttEnc identifies drivers based on few shot data, extracts driver fingerprints to address the issue of data shortages, and is able to classify unknown drivers. The first experiment showed that P-AttEnc can identify drivers with an accuracy of 69.8% in the one-shot scenario. The second experiment showed that P-AttEnc, in the 1-shot scenario, can classify unknown drivers with an average accuracy of 65.7%.", "date": "2025-10-20", "categories": []}
{"id": "2510.17372", "title": "Beyond Real Faces: Synthetic Datasets Can Achieve Reliable Recognition Performance without Privacy Compromise", "abstract": "The deployment of facial recognition systems has created an ethical dilemma: achieving high accuracy requires massive datasets of real faces collected without consent, leading to dataset retractions and potential legal liabilities under regulations like GDPR. While synthetic facial data presents a promising privacy-preserving alternative, the field lacks comprehensive empirical evidence of its viability. This study addresses this critical gap through extensive evaluation of synthetic facial recognition datasets. We present a systematic literature review identifying 25 synthetic facial recognition datasets (2018-2025), combined with rigorous experimental validation. Our methodology examines seven key requirements for privacy-preserving synthetic data: identity leakage prevention, intra-class variability, identity separability, dataset scale, ethical data sourcing, bias mitigation, and benchmark reliability. Through experiments involving over 10 million synthetic samples, extended by a comparison of results reported on five standard benchmarks, we provide the first comprehensive empirical assessment of synthetic data's capability to replace real datasets. Best-performing synthetic datasets (VariFace, VIGFace) achieve recognition accuracies of 95.67% and 94.91% respectively, surpassing established real datasets including CASIA-WebFace (94.70%). While those images remain private, publicly available alternatives Vec2Face (93.52%) and CemiFace (93.22%) come close behind. Our findings reveal that they ensure proper intra-class variability while maintaining identity separability. Demographic bias analysis shows that, even though synthetic data inherits limited biases, it offers unprecedented control for bias mitigation through generation parameters. These results establish synthetic facial data as a scientifically viable and ethically imperative alternative for facial recognition research.", "date": "2025-10-20", "categories": []}
{"id": "2510.17467", "title": "CrossStateECG: Multi-Scale Deep Convolutional Network with Attention for Rest-Exercise ECG Biometrics", "abstract": "Current research in Electrocardiogram (ECG) biometrics mainly emphasizes resting-state conditions, leaving the performance decline in rest-exercise scenarios largely unresolved. This paper introduces CrossStateECG, a robust ECG-based authentication model explicitly tailored for cross-state (rest-exercise) conditions. The proposed model creatively combines multi-scale deep convolutional feature extraction with attention mechanisms to ensure strong identification across different physiological states. Experimental results on the exercise-ECGID dataset validate the effectiveness of CrossStateECG, achieving an identification accuracy of 92.50% in the Rest-to-Exercise scenario (training on resting ECG and testing on post-exercise ECG) and 94.72% in the Exercise-to-Rest scenario (training on post-exercise ECG and testing on resting ECG). Furthermore, CrossStateECG demonstrates exceptional performance across both state combinations, reaching an accuracy of 99.94% in Rest-to-Rest scenarios and 97.85% in Mixed-to-Mixed scenarios. Additional validations on the ECG-ID and MIT-BIH datasets further confirmed the generalization abilities of CrossStateECG, underscoring its potential as a practical solution for post-exercise ECG-based authentication in dynamic real-world settings.", "date": "2025-10-20", "categories": []}
{"id": "2502.06619", "title": "Unleashing the Potential of Pre-Trained Diffusion Models for Generalizable Person Re-Identification", "abstract": "Domain-generalizable re-identification (DG Re-ID) aims to train a model on one or more source domains and evaluate its performance on unseen target domains, a task that has attracted growing attention due to its practical relevance. While numerous methods have been proposed, most rely on discriminative or contrastive learning frameworks to learn generalizable feature representations. However, these approaches often fail to mitigate shortcut learning, leading to suboptimal performance. In this work, we propose a novel method called diffusion model-assisted representation learning with a correlation-aware conditioning scheme (DCAC) to enhance DG Re-ID. Our method integrates a discriminative and contrastive Re-ID model with a pre-trained diffusion model through a correlation-aware conditioning scheme. By incorporating ID classification probabilities generated from the Re-ID model with a set of learnable ID-wise prompts, the conditioning scheme injects dark knowledge that captures ID correlations to guide the diffusion process. Simultaneously, feedback from the diffusion model is back-propagated through the conditioning scheme to the Re-ID model, effectively improving the generalization capability of Re-ID features. Extensive experiments on both single-source and multi-source DG Re-ID tasks demonstrate that our method achieves state-of-the-art performance. Comprehensive ablation studies further validate the effectiveness of the proposed approach, providing insights into its robustness. Codes will be available at https://github.com/RikoLi/DCAC.", "date": "2025-02-10", "categories": []}
{"id": "2510.15289", "title": "QCFace: Image Quality Control for boosting Face Representation & Recognition", "abstract": "Recognizability, a key perceptual factor in human face processing, strongly affects the performance of face recognition (FR) systems in both verification and identification tasks. Effectively using recognizability to enhance feature representation remains challenging. In deep FR, the loss function plays a crucial role in shaping how features are embedded. However, current methods have two main drawbacks: (i) recognizability is only partially captured through soft margin constraints, resulting in weaker quality representation and lower discrimination, especially for low-quality or ambiguous faces; (ii) mutual overlapping gradients between feature direction and magnitude introduce undesirable interactions during optimization, causing instability and confusion in hypersphere planning, which may result in poor generalization, and entangled representations where recognizability and identity are not cleanly separated. To address these issues, we introduce a hard margin strategy - Quality Control Face (QCFace), which overcomes the mutual overlapping gradient problem and enables the clear decoupling of recognizability from identity representation. Based on this strategy, a novel hard-margin-based loss function employs a guidance factor for hypersphere planning, simultaneously optimizing for recognition ability and explicit recognizability representation. Extensive experiments confirm that QCFace not only provides robust and quantifiable recognizability encoding but also achieves state-of-the-art performance in both verification and identification benchmarks compared to existing recognizability-based losses.", "date": "2025-10-17", "categories": []}
{"id": "2510.15595", "title": "FlexiReID: Adaptive Mixture of Expert for Multi-Modal Person Re-Identification", "abstract": "Multimodal person re-identification (Re-ID) aims to match pedestrian images across different modalities. However, most existing methods focus on limited cross-modal settings and fail to support arbitrary query-retrieval combinations, hindering practical deployment. We propose FlexiReID, a flexible framework that supports seven retrieval modes across four modalities: rgb, infrared, sketches, and text. FlexiReID introduces an adaptive mixture-of-experts (MoE) mechanism to dynamically integrate diverse modality features and a cross-modal query fusion module to enhance multimodal feature extraction. To facilitate comprehensive evaluation, we construct CIRS-PEDES, a unified dataset extending four popular Re-ID datasets to include all four modalities. Extensive experiments demonstrate that FlexiReID achieves state-of-the-art performance and offers strong generalization in complex scenarios.", "date": "2025-10-17", "categories": []}
{"id": "2508.01427", "title": "Capturing More: Learning Multi-Domain Representations for Robust Online Handwriting Verification", "abstract": "In this paper, we propose SPECTRUM, a temporal-frequency synergistic model that unlocks the untapped potential of multi-domain representation learning for online handwriting verification (OHV). SPECTRUM comprises three core components: (1) a multi-scale interactor that finely combines temporal and frequency features through dual-modal sequence interaction and multi-scale aggregation, (2) a self-gated fusion module that dynamically integrates global temporal and frequency features via self-driven balancing. These two components work synergistically to achieve micro-to-macro spectral-temporal integration. (3) A multi-domain distance-based verifier then utilizes both temporal and frequency representations to improve discrimination between genuine and forged handwriting, surpassing conventional temporal-only approaches. Extensive experiments demonstrate SPECTRUM's superior performance over existing OHV methods, underscoring the effectiveness of temporal-frequency multi-domain learning. Furthermore, we reveal that incorporating multiple handwritten biometrics fundamentally enhances the discriminative power of handwriting representations and facilitates verification. These findings not only validate the efficacy of multi-domain learning in OHV but also pave the way for future research in multi-domain approaches across both feature and biometric domains. Code is publicly available at https://github.com/NiceRingNode/SPECTRUM.", "date": "2025-08-02", "categories": []}
{"id": "2510.09645", "title": "AdaptAuth: Multi-Layered Behavioral and Credential Analysis for a Secure and Adaptive Authentication Framework for Password Security", "abstract": "Password security has been compelled to evolve in response to the growing computational capabilities of modern systems. However, this evolution has often resulted in increasingly complex security practices that alienate users, leading to poor compliance and heightened vulnerability. Consequently, individuals remain exposed to attackers through weak or improperly managed passwords, underscoring the urgent need for a comprehensive defense mechanism that effectively addresses password-related risks and threats. In this paper, we propose a multifaceted solution designed to revolutionize password security by integrating diverse attributes such as the Password Dissection Mechanism, Dynamic Password Policy Mechanism, human behavioral patterns, device characteristics, network parameters, geographical context, and other relevant factors. By leveraging learning-based models, our framework constructs detailed user profiles capable of recognizing individuals and preventing nearly all forms of unauthorized access or device possession. The proposed framework enhances the usability-security paradigm by offering stronger protection than existing standards while simultaneously engaging users in the policy-setting process through a novel, adaptive approach.", "date": "2025-10-04", "categories": []}
{"id": "2510.09663", "title": "Adversarial-Resilient RF Fingerprinting: A CNN-GAN Framework for Rogue Transmitter Detection", "abstract": "Radio Frequency Fingerprinting (RFF) has evolved as an effective solution for authenticating devices by leveraging the unique imperfections in hardware components involved in the signal generation process. In this work, we propose a Convolutional Neural Network (CNN) based framework for detecting rogue devices and identifying genuine ones using softmax probability thresholding. We emulate an attack scenario in which adversaries attempt to mimic the RF characteristics of genuine devices by training a Generative Adversarial Network (GAN) using In-phase and Quadrature (IQ) samples from genuine devices. The proposed approach is verified using IQ samples collected from ten different ADALM-PLUTO Software Defined Radios (SDRs), with seven devices considered genuine, two as rogue, and one used for validation to determine the threshold.", "date": "2025-10-07", "categories": []}
{"id": "2510.10663", "title": "Scalable Face Security Vision Foundation Model for Deepfake, Diffusion, and Spoofing Detection", "abstract": "With abundant, unlabeled real faces, how can we learn robust and transferable facial representations to boost generalization across various face security tasks? We make the first attempt and propose FS-VFM, a scalable self-supervised pre-training framework, to learn fundamental representations of real face images. We introduce three learning objectives, namely 3C, that synergize masked image modeling (MIM) and instance discrimination (ID), empowering FS-VFM to encode both local patterns and global semantics of real faces. Specifically, we formulate various facial masking strategies for MIM and devise a simple yet effective CRFR-P masking, which explicitly prompts the model to pursue meaningful intra-region Consistency and challenging inter-region Coherency. We present a reliable self-distillation mechanism that seamlessly couples MIM with ID to establish underlying local-to-global Correspondence. After pre-training, vanilla vision transformers (ViTs) serve as universal Vision Foundation Models for downstream Face Security tasks: cross-dataset deepfake detection, cross-domain face anti-spoofing, and unseen diffusion facial forensics. To efficiently transfer the pre-trained FS-VFM, we further propose FS-Adapter, a lightweight plug-and-play bottleneck atop the frozen backbone with a novel real-anchor contrastive objective. Extensive experiments on 11 public benchmarks demonstrate that our FS-VFM consistently generalizes better than diverse VFMs, spanning natural and facial domains, fully, weakly, and self-supervised paradigms, small, base, and large ViT scales, and even outperforms SOTA task-specific methods, while FS-Adapter offers an excellent efficiency-performance trade-off. The code and models are available on https://fsfm-3c.github.io/fsvfm.html.", "date": "2025-10-12", "categories": []}
{"id": "2510.11223", "title": "Investigating Identity Signals in Conversational Facial Dynamics via Disentangled Expression Features", "abstract": "This work investigates whether individuals can be identified solely through the pure dynamical components of their facial expressions, independent of static facial appearance. We leverage the FLAME 3D morphable model to achieve explicit disentanglement between facial shape and expression dynamics, extracting frame-by-frame parameters from conversational videos while retaining only expression and jaw coefficients. On the CANDOR dataset of 1,429 speakers in naturalistic conversations, our Conformer model with supervised contrastive learning achieves 61.14\\%accuracy on 1,429-way classification -- 458 times above chance -- demonstrating that facial dynamics carry strong identity signatures. We introduce a drift-to-noise ratio (DNR) that quantifies the reliability of shape expression separation by measuring across-session shape changes relative to within-session variability. DNR strongly negatively correlates with recognition performance, confirming that unstable shape estimation compromises dynamic identification. Our findings reveal person-specific signatures in conversational facial dynamics, with implications for social perception and clinical assessment.", "date": "2025-10-13", "categories": []}
{"id": "2510.09509", "title": "Diagonal Artifacts in Samsung Images: PRNU Challenges and Solutions", "abstract": "We investigate diagonal artifacts present in images captured by several Samsung smartphones and their impact on PRNU-based camera source verification. We first show that certain Galaxy S series models share a common pattern causing fingerprint collisions, with a similar issue also found in some Galaxy A models. Next, we demonstrate that reliable PRNU verification remains feasible for devices supporting PRO mode with raw capture, since raw images bypass the processing pipeline that introduces artifacts. This option, however, is not available for the mid-range A series models or in forensic cases without access to raw images. Finally, we outline potential forensic applications of the diagonal artifacts, such as reducing misdetections in HDR images and localizing regions affected by synthetic bokeh in portrait-mode images.", "date": "2025-10-10", "categories": []}
{"id": "2412.12032", "title": "FSFM: A Generalizable Face Security Foundation Model via Self-Supervised Facial Representation Learning", "abstract": "This work asks: with abundant, unlabeled real faces, how to learn a robust and transferable facial representation that boosts various face security tasks with respect to generalization performance? We make the first attempt and propose a self-supervised pretraining framework to learn fundamental representations of real face images, FSFM, that leverages the synergy between masked image modeling (MIM) and instance discrimination (ID). We explore various facial masking strategies for MIM and present a simple yet powerful CRFR-P masking, which explicitly forces the model to capture meaningful intra-region consistency and challenging inter-region coherency. Furthermore, we devise the ID network that naturally couples with MIM to establish underlying local-to-global correspondence via tailored self-distillation. These three learning objectives, namely 3C, empower encoding both local features and global semantics of real faces. After pretraining, a vanilla ViT serves as a universal vision foundation model for downstream face security tasks: cross-dataset deepfake detection, cross-domain face anti-spoofing, and unseen diffusion facial forgery detection. Extensive experiments on 10 public datasets demonstrate that our model transfers better than supervised pretraining, visual and facial self-supervised learning arts, and even outperforms task-specialized SOTA methods.", "date": "2024-12-16", "categories": []}
{"id": "2510.08160", "title": "Beyond Sub-6 GHz: Leveraging mmWave Wi-Fi for Gait-Based Person Identification", "abstract": "Person identification plays a vital role in enabling intelligent, personalized, and secure human-computer interaction. Recent research has demonstrated the feasibility of leveraging Wi-Fi signals for passive person identification using a person's unique gait pattern. Although most existing work focuses on sub-6 GHz frequencies, the emergence of mmWave offers new opportunities through its finer spatial resolution, though its comparative advantages for person identification remain unexplored. This work presents the first comparative study between sub-6 GHz and mmWave Wi-Fi signals for person identification with commercial off-the-shelf (COTS) Wi-Fi, using a novel dataset of synchronized measurements from the two frequency bands in an indoor environment. To ensure a fair comparison, we apply identical training pipelines and model configurations across both frequency bands. Leveraging end-to-end deep learning, we show that even at low sampling rates (10 Hz), mmWave Wi-Fi signals can achieve high identification accuracy (91.2% on 20 individuals) when combined with effective background subtraction.", "date": "2025-10-09", "categories": []}
{"id": "2510.05163", "title": "Deep Learning-Based Multi-Factor Authentication: A Survey of Biometric and Smart Card Integration Approaches", "abstract": "In the era of pervasive cyber threats and exponential growth in digital services, the inadequacy of single-factor authentication has become increasingly evident. Multi-Factor Authentication (MFA), which combines knowledge-based factors (passwords, PINs), possession-based factors (smart cards, tokens), and inherence-based factors (biometric traits), has emerged as a robust defense mechanism. Recent breakthroughs in deep learning have transformed the capabilities of biometric systems, enabling higher accuracy, resilience to spoofing, and seamless integration with hardware-based solutions. At the same time, smart card technologies have evolved to include on-chip biometric verification, cryptographic processing, and secure storage, thereby enabling compact and secure multi-factor devices. This survey presents a comprehensive synthesis of recent work (2019-2025) at the intersection of deep learning, biometrics, and smart card technologies for MFA. We analyze biometric modalities (face, fingerprint, iris, voice), review hardware-based approaches (smart cards, NFC, TPMs, secure enclaves), and highlight integration strategies for real-world applications such as digital banking, healthcare IoT, and critical infrastructure. Furthermore, we discuss the major challenges that remain open, including usability-security tradeoffs, adversarial attacks on deep learning models, privacy concerns surrounding biometric data, and the need for standardization in MFA deployment. By consolidating current advancements, limitations, and research opportunities, this survey provides a roadmap for designing secure, scalable, and user-friendly authentication frameworks.", "date": "2025-10-04", "categories": []}
{"id": "2312.17494", "title": "QGFace: Quality-Guided Joint Training For Mixed-Quality Face Recognition", "abstract": "The quality of a face crop in an image is decided by many factors such as camera resolution, distance, and illumination condition. This makes the discrimination of face images with different qualities a challenging problem in realistic applications. However, most existing approaches are designed specifically for high-quality (HQ) or low-quality (LQ) images, and the performances would degrade for the mixed-quality images. Besides, many methods ask for pre-trained feature extractors or other auxiliary structures to support the training and the evaluation. In this paper, we point out that the key to better understand both the HQ and the LQ images simultaneously is to apply different learning methods according to their qualities. We propose a novel quality-guided joint training approach for mixed-quality face recognition, which could simultaneously learn the images of different qualities with a single encoder. Based on quality partition, classification-based method is employed for HQ data learning. Meanwhile, for the LQ images which lack identity information, we learn them with self-supervised image-image contrastive learning. To effectively catch up the model update and improve the discriminability of contrastive learning in our joint training scenario, we further propose a proxy-updated real-time queue to compose the contrastive pairs with features from the genuine encoder. Experiments on the low-quality datasets SCface and Tinyface, the mixed-quality dataset IJB-B, and five high-quality datasets demonstrate the effectiveness of our proposed approach in recognizing face images of different qualities.", "date": "2023-12-29", "categories": []}
{"id": "2510.04753", "title": "Beyond Appearance: Transformer-based Person Identification from Conversational Dynamics", "abstract": "This paper investigates the performance of transformer-based architectures for person identification in natural, face-to-face conversation scenario. We implement and evaluate a two-stream framework that separately models spatial configurations and temporal motion patterns of 133 COCO WholeBody keypoints, extracted from a subset of the CANDOR conversational corpus. Our experiments compare pre-trained and from-scratch training, investigate the use of velocity features, and introduce a multi-scale temporal transformer for hierarchical motion modeling. Results demonstrate that domain-specific training significantly outperforms transfer learning, and that spatial configurations carry more discriminative information than temporal dynamics. The spatial transformer achieves 95.74% accuracy, while the multi-scale temporal transformer achieves 93.90%. Feature-level fusion pushes performance to 98.03%, confirming that postural and dynamic information are complementary. These findings highlight the potential of transformer architectures for person identification in natural interactions and provide insights for future multimodal and cross-cultural studies.", "date": "2025-10-06", "categories": []}
{"id": "2510.02570", "title": "Unlocking the power of partnership: How humans and machines can work together to improve face recognition", "abstract": "Human review of consequential decisions by face recognition algorithms creates a \"collaborative\" human-machine system. Individual differences between people and machines, however, affect whether collaboration improves or degrades accuracy in any given case. We establish the circumstances under which combining human and machine face identification decisions improves accuracy. Using data from expert and non-expert face identifiers, we examined the benefits of human-human and human-machine collaborations. The benefits of collaboration increased as the difference in baseline accuracy between collaborators decreased-following the Proximal Accuracy Rule (PAR). This rule predicted collaborative (fusion) benefit across a wide range of baseline abilities, from people with no training to those with extensive training. Using the PAR, we established a critical fusion zone, where humans are less accurate than the machine, but fusing the two improves system accuracy. This zone was surprisingly large. We implemented \"intelligent human-machine fusion\" by selecting people with the potential to increase the accuracy of a high-performing machine. Intelligent fusion was more accurate than the machine operating alone and more accurate than combining all human and machine judgments. The highest system-wide accuracy achievable with human-only partnerships was found by graph theory. This fully human system approximated the average performance achieved by intelligent human-machine collaboration. However, intelligent human-machine collaboration more effectively minimized the impact of low-performing humans on system-wide accuracy. The results demonstrate a meaningful role for both humans and machines in assuring accurate face identification. This study offers an evidence-based road map for the intelligent use of AI in face identification.", "date": "2025-10-02", "categories": []}
{"id": "2510.02197", "title": "Cross-Breed Pig Identification Using Auricular Vein Pattern Recognition: A Machine Learning Approach for Small-Scale Farming Applications", "abstract": "Accurate livestock identification is a cornerstone of modern farming: it supports health monitoring, breeding programs, and productivity tracking. However, common pig identification methods, such as ear tags and microchips, are often unreliable, costly, target pure breeds, and thus impractical for small-scale farmers. To address this gap, we propose a noninvasive biometric identification approach that leverages uniqueness of the auricular vein patterns. To this end, we have collected 800 ear images from 20 mixed-breed pigs (Landrace cross Pietrain and Duroc cross Pietrain), captured using a standard smartphone and simple back lighting. A multistage computer vision pipeline was developed to enhance vein visibility, extract structural and spatial features, and generate biometric signatures. These features were then classified using machine learning models. Support Vector Machines (SVM) achieved the highest accuracy: correctly identifying pigs with 98.12% precision across mixed-breed populations. The entire process from image processing to classification was completed in an average of 8.3 seconds, demonstrating feasibility for real-time farm deployment. We believe that by replacing fragile physical identifiers with permanent biological markers, this system provides farmers with a cost-effective and stress-free method of animal identification. More broadly, the findings confirm the practicality of auricular vein biometrics for digitizing livestock management, reinforcing its potential to extend the benefits of precision farming to resource-constrained agricultural communities.", "date": "2025-10-02", "categories": []}
{"id": "2509.20537", "title": "Innovative Deep Learning Architecture for Enhanced Altered Fingerprint Recognition", "abstract": "Altered fingerprint recognition (AFR) is challenging for biometric verification in applications such as border control, forensics, and fiscal admission. Adversaries can deliberately modify ridge patterns to evade detection, so robust recognition of altered prints is essential. We present DeepAFRNet, a deep learning recognition model that matches and recognizes distorted fingerprint samples. The approach uses a VGG16 backbone to extract high-dimensional features and cosine similarity to compare embeddings. We evaluate on the SOCOFing Real-Altered subset with three difficulty levels (Easy, Medium, Hard). With strict thresholds, DeepAFRNet achieves accuracies of 96.7 percent, 98.76 percent, and 99.54 percent for the three levels. A threshold-sensitivity study shows that relaxing the threshold from 0.92 to 0.72 sharply degrades accuracy to 7.86 percent, 27.05 percent, and 29.51 percent, underscoring the importance of threshold selection in biometric systems. By using real altered samples and reporting per-level metrics, DeepAFRNet addresses limitations of prior work based on synthetic alterations or limited verification protocols, and indicates readiness for real-world deployments where both security and recognition resilience are critical.", "date": "2025-09-24", "categories": []}
{"id": "2509.20382", "title": "Lightweight MobileNetV1+GRU for ECG Biometric Authentication: Federated and Adversarial Evaluation", "abstract": "ECG biometrics offer a unique, secure authentication method, yet their deployment on wearable devices faces real-time processing, privacy, and spoofing vulnerability challenges. This paper proposes a lightweight deep learning model (MobileNetV1+GRU) for ECG-based authentication, injection of 20dB Gaussian noise & custom preprocessing. We simulate wearable conditions and edge deployment using the ECGID, MIT-BIH, CYBHi, and PTB datasets, achieving accuracies of 99.34%, 99.31%, 91.74%, and 98.49%, F1-scores of 0.9869, 0.9923, 0.9125, and 0.9771, Precision of 0.9866, 0.9924, 0.9180 and 0.9845, Recall of 0.9878, 0.9923, 0.9129, and 0.9756, equal error rates (EER) of 0.0009, 0.00013, 0.0091, and 0.0009, and ROC-AUC values of 0.9999, 0.9999, 0.9985, and 0.9998, while under FGSM adversarial attacks, accuracy drops from 96.82% to as low as 0.80%. This paper highlights federated learning, adversarial testing, and the need for diverse wearable physiological datasets to ensure secure and scalable biometrics.", "date": "2025-09-21", "categories": []}
{"id": "2509.19753", "title": "ExpFace: Exponential Angular Margin Loss for Deep Face Recognition", "abstract": "Face recognition is an open-set problem requiring high discriminative power to ensure that intra-class distances remain smaller than inter-class distances. Margin-based softmax losses, such as SphereFace, CosFace, and ArcFace, have been widely adopted to enhance intra-class compactness and inter-class separability, yet they overlook the impact of noisy samples. By examining the distribution of samples in the angular space, we observe that clean samples predominantly cluster in the center region, whereas noisy samples tend to shift toward the peripheral region. Motivated by this observation, we propose the Exponential Angular Margin Loss (ExpFace), which introduces an angular exponential term as the margin. This design applies a larger penalty in the center region and a smaller penalty in the peripheral region within the angular space, thereby emphasizing clean samples while suppressing noisy samples. We present a unified analysis of ExpFace and classical margin-based softmax losses in terms of margin embedding forms, similarity curves, and gradient curves, showing that ExpFace not only avoids the training instability of SphereFace and the non-monotonicity of ArcFace, but also exhibits a similarity curve that applies penalties in the same manner as the decision boundary in the angular space. Extensive experiments demonstrate that ExpFace achieves state-of-the-art performance. To facilitate future research, we have released the source code at: https://github.com/dfr-code/ExpFace.", "date": "2025-09-24", "categories": []}
{"id": "2509.20024", "title": "Generative Adversarial Networks Applied for Privacy Preservation in Biometric-Based Authentication and Identification", "abstract": "Biometric-based authentication systems are getting broadly adopted in many areas. However, these systems do not allow participating users to influence the way their data is used. Furthermore, the data may leak and can be misused without the users' knowledge. In this paper, we propose a new authentication method that preserves the privacy of individuals and is based on a generative adversarial network (GAN). Concretely, we suggest using the GAN for translating images of faces to a visually private domain (e.g., flowers or shoes). Classifiers, which are used for authentication purposes, are then trained on the images from the visually private domain. Based on our experiments, the method is robust against attacks and still provides meaningful utility.", "date": "2025-09-24", "categories": []}
{"id": "2509.17457", "title": "Explainable AI for Analyzing Person-Specific Patterns in Facial Recognition Tasks", "abstract": "The proliferation of facial recognition systems presents major privacy risks, driving the need for effective countermeasures. Current adversarial techniques apply generalized methods rather than adapting to individual facial characteristics, limiting their effectiveness and inconspicuousness. In this work, we introduce Layer Embedding Activation Mapping (LEAM), a novel technique that identifies which facial areas contribute most to recognition at an individual level. Unlike adversarial attack methods that aim to fool recognition systems, LEAM is an explainability technique designed to understand how these systems work, providing insights that could inform future privacy protection research. We integrate LEAM with a face parser to analyze data from 1000 individuals across 9 pre-trained facial recognition models.\n  Our analysis reveals that while different layers within facial recognition models vary significantly in their focus areas, these models generally prioritize similar facial regions across architectures when considering their overall activation patterns, which show significantly higher similarity between images of the same individual (Bhattacharyya Coefficient: 0.32-0.57) vs. different individuals (0.04-0.13), validating the existence of person-specific recognition patterns. Our results show that facial recognition models prioritize the central region of face images (with nose areas accounting for 18.9-29.7% of critical recognition regions), while still distributing attention across multiple facial fragments. Proper selection of relevant facial areas was confirmed using validation occlusions, based on just 1% of the most relevant, LEAM-identified, image pixels, which proved to be transferable across different models. Our findings establish the foundation for future individually tailored privacy protection systems centered around LEAM's choice of areas to be perturbed.", "date": "2025-09-22", "categories": []}
{"id": "2509.15648", "title": "FingerSplat: Contactless Fingerprint 3D Reconstruction and Generation based on 3D Gaussian Splatting", "abstract": "Researchers have conducted many pioneer researches on contactless fingerprints, yet the performance of contactless fingerprint recognition still lags behind contact-based methods primary due to the insufficient contactless fingerprint data with pose variations and lack of the usage of implicit 3D fingerprint representations. In this paper, we introduce a novel contactless fingerprint 3D registration, reconstruction and generation framework by integrating 3D Gaussian Splatting, with the goal of offering a new paradigm for contactless fingerprint recognition that integrates 3D fingerprint reconstruction and generation. To our knowledge, this is the first work to apply 3D Gaussian Splatting to the field of fingerprint recognition, and the first to achieve effective 3D registration and complete reconstruction of contactless fingerprints with sparse input images and without requiring camera parameters information. Experiments on 3D fingerprint registration, reconstruction, and generation prove that our method can accurately align and reconstruct 3D fingerprints from 2D images, and sequentially generates high-quality contactless fingerprints from 3D model, thus increasing the performances for contactless fingerprint recognition.", "date": "2025-09-19", "categories": []}
{"id": "2405.16600", "title": "Image-Text-Image Knowledge Transfer for Lifelong Person Re-Identification with Hybrid Clothing States", "abstract": "With the continuous expansion of intelligent surveillance networks, lifelong person re-identification (LReID) has received widespread attention, pursuing the need of self-evolution across different domains. However, existing LReID studies accumulate knowledge with the assumption that people would not change their clothes. In this paper, we propose a more practical task, namely lifelong person re-identification with hybrid clothing states (LReID-Hybrid), which takes a series of cloth-changing and same-cloth domains into account during lifelong learning. To tackle the challenges of knowledge granularity mismatch and knowledge presentation mismatch in LReID-Hybrid, we take advantage of the consistency and generalization capabilities of the text space, and propose a novel framework, dubbed $Teata$, to effectively align, transfer, and accumulate knowledge in an \"image-text-image\" closed loop. Concretely, to achieve effective knowledge transfer, we design a Structured Semantic Prompt (SSP) learning to decompose the text prompt into several structured pairs to distill knowledge from the image space with a unified granularity of text description. Then, we introduce a Knowledge Adaptation and Projection (KAP) strategy, which tunes text knowledge via a slow-paced learner to adapt to different tasks without catastrophic forgetting. Extensive experiments demonstrate the superiority of our proposed $Teata$ for LReID-Hybrid as well as on conventional LReID benchmarks over advanced methods.", "date": "2024-05-26", "categories": []}
{"id": "2509.15177", "title": "A Race Bias Free Face Aging Model for Reliable Kinship Verification", "abstract": "The age gap in kinship verification addresses the time difference between the photos of the parent and the child. Moreover, their same-age photos are often unavailable, and face aging models are racially biased, which impacts the likeness of photos. Therefore, we propose a face aging GAN model, RA-GAN, consisting of two new modules, RACEpSp and a feature mixer, to produce racially unbiased images. The unbiased synthesized photos are used in kinship verification to investigate the results of verifying same-age parent-child images. The experiments demonstrate that our RA-GAN outperforms SAM-GAN on an average of 13.14\\% across all age groups, and CUSP-GAN in the 60+ age group by 9.1\\% in terms of racial accuracy. Moreover, RA-GAN can preserve subjects' identities better than SAM-GAN and CUSP-GAN across all age groups. Additionally, we demonstrate that transforming parent and child images from the KinFaceW-I and KinFaceW-II datasets to the same age can enhance the verification accuracy across all age groups. The accuracy increases with our RA-GAN for the kinship relationships of father-son and father-daughter, mother-son, and mother-daughter, which are 5.22, 5.12, 1.63, and 0.41, respectively, on KinFaceW-I. Additionally, the accuracy for the relationships of father-daughter, father-son, and mother-son is 2.9, 0.39, and 1.6 on KinFaceW-II, respectively. The code is available at~\\href{https://github.com/bardiya2254kariminia/An-Age-Transformation-whitout-racial-bias-for-Kinship-verification}{Github}", "date": "2025-09-18", "categories": []}
{"id": "2504.08272", "title": "Palmprint De-Identification Using Diffusion Model for High-Quality and Diverse Synthesis", "abstract": "Palmprint recognition techniques have advanced significantly in recent years, enabling reliable recognition even when palmprints are captured in uncontrolled or challenging environments. However, this strength also introduces new risks, as publicly available palmprint images can be misused by adversaries for malicious activities. Despite this growing concern, research on methods to obscure or anonymize palmprints remains largely unexplored. Thus, it is essential to develop a palmprint de-identification technique capable of removing identity-revealing features while retaining the image's utility and preserving non-sensitive information. In this paper, we propose a training-free framework that utilizes pre-trained diffusion models to generate diverse, high-quality palmprint images that conceal identity features for de-identification purposes. To ensure greater stability and controllability in the synthesis process, we incorporate a semantic-guided embedding fusion alongside a prior interpolation mechanism. We further propose the de-identification ratio, a novel metric for intuitive de-identification assessment. Extensive experiments across multiple palmprint datasets and recognition methods demonstrate that our method effectively conceals identity-related traits with significant diversity across de-identified samples. The de-identified samples preserve high visual fidelity and maintain excellent usability, achieving a balance between de-identification and retaining non-identity information.", "date": "2025-04-11", "categories": []}
{"id": "2307.12255", "title": "ResWCAE: Biometric Pattern Image Denoising Using Residual Wavelet-Conditioned Autoencoder", "abstract": "The utilization of biometric authentication with pattern images is increasingly popular in compact Internet of Things (IoT) devices. However, the reliability of such systems can be compromised by image quality issues, particularly in the presence of high levels of noise. While state-of-the-art deep learning algorithms designed for generic image denoising have shown promise, their large number of parameters and lack of optimization for unique biometric pattern retrieval make them unsuitable for these devices and scenarios. In response to these challenges, this paper proposes a lightweight and robust deep learning architecture, the Residual Wavelet-Conditioned Convolutional Autoencoder (Res-WCAE) with a Kullback-Leibler divergence (KLD) regularization, designed specifically for fingerprint image denoising. Res-WCAE comprises two encoders - an image encoder and a wavelet encoder - and one decoder. Residual connections between the image encoder and decoder are leveraged to preserve fine-grained spatial features, where the bottleneck layer conditioned on the compressed representation of features obtained from the wavelet encoder using approximation and detail subimages in the wavelet-transform domain. The effectiveness of Res-WCAE is evaluated against several state-of-the-art denoising methods, and the experimental results demonstrate that Res-WCAE outperforms these methods, particularly for heavily degraded fingerprint images in the presence of high levels of noise. Overall, Res-WCAE shows promise as a solution to the challenges faced by biometric authentication systems in compact IoT devices.", "date": "2023-07-23", "categories": []}
{"id": "2509.11247", "title": "Contextualized Multimodal Lifelong Person Re-Identification in Hybrid Clothing States", "abstract": "Person Re-Identification (ReID) has several challenges in real-world surveillance systems due to clothing changes (CCReID) and the need for maintaining continual learning (LReID). Previous existing methods either develop models specifically for one application, which is mostly a same-cloth (SC) setting or treat CCReID as its own separate sub-problem. In this work, we will introduce the LReID-Hybrid task with the goal of developing a model to achieve both SC and CC while learning in a continual setting. Mismatched representations and forgetting from one task to the next are significant issues, we address this with CMLReID, a CLIP-based framework composed of two novel tasks: (1) Context-Aware Semantic Prompt (CASP) that generates adaptive prompts, and also incorporates context to align richly multi-grained visual cues with semantic text space; and (2) Adaptive Knowledge Fusion and Projection (AKFP) which produces robust SC/CC prototypes through the use of a dual-path learner that aligns features with our Clothing-State-Aware Projection Loss. Experiments performed on a wide range of datasets and illustrate that CMLReID outperforms all state-of-the-art methods with strong robustness and generalization despite clothing variations and a sophisticated process of sequential learning.", "date": "2025-09-14", "categories": []}
{"id": "2509.11587", "title": "Hierarchical Identity Learning for Unsupervised Visible-Infrared Person Re-Identification", "abstract": "Unsupervised visible-infrared person re-identification (USVI-ReID) aims to learn modality-invariant image features from unlabeled cross-modal person datasets by reducing the modality gap while minimizing reliance on costly manual annotations. Existing methods typically address USVI-ReID using cluster-based contrastive learning, which represents a person by a single cluster center. However, they primarily focus on the commonality of images within each cluster while neglecting the finer-grained differences among them. To address the limitation, we propose a Hierarchical Identity Learning (HIL) framework. Since each cluster may contain several smaller sub-clusters that reflect fine-grained variations among images, we generate multiple memories for each existing coarse-grained cluster via a secondary clustering. Additionally, we propose Multi-Center Contrastive Learning (MCCL) to refine representations for enhancing intra-modal clustering and minimizing cross-modal discrepancies. To further improve cross-modal matching quality, we design a Bidirectional Reverse Selection Transmission (BRST) mechanism, which establishes reliable cross-modal correspondences by performing bidirectional matching of pseudo-labels. Extensive experiments conducted on the SYSU-MM01 and RegDB datasets demonstrate that the proposed method outperforms existing approaches. The source code is available at: https://github.com/haonanshi0125/HIL.", "date": "2025-09-15", "categories": []}
{"id": "2412.17541", "title": "Spoof Trace Discovery for Deep Learning Based Explainable Face Anti-Spoofing", "abstract": "With the rapid growth usage of face recognition in people's daily life, face anti-spoofing becomes increasingly important to avoid malicious attacks. Recent face anti-spoofing models can reach a high classification accuracy on multiple datasets but these models can only tell people \"this face is fake\" while lacking the explanation to answer \"why it is fake\". Such a system undermines trustworthiness and causes user confusion, as it denies their requests without providing any explanations. In this paper, we incorporate XAI into face anti-spoofing and propose a new problem termed X-FAS (eXplainable Face Anti-Spoofing) empowering face anti-spoofing models to provide an explanation. We propose SPTD (SPoof Trace Discovery), an X-FAS method which can discover spoof concepts and provide reliable explanations on the basis of discovered concepts. To evaluate the quality of X-FAS methods, we propose an X-FAS benchmark with annotated spoof traces by experts. We analyze SPTD explanations on face anti-spoofing dataset and compare SPTD quantitatively and qualitatively with previous XAI methods on proposed X-FAS benchmark. Experimental results demonstrate SPTD's ability to generate reliable explanations.", "date": "2024-12-23", "categories": []}
{"id": "2403.08557", "title": "OC4-ReID: Occluded Cloth-Changing Person Re-Identification", "abstract": "The study of Cloth-Changing Person Re-identification (CC-ReID) focuses on retrieving specific pedestrians when their clothing has changed, typically under the assumption that the entire pedestrian images are visible. Pedestrian images in real-world scenarios, however, are often partially obscured by obstacles, presenting a significant challenge to existing CC-ReID systems. In this paper, we introduce a more challenging task termed Occluded Cloth-Changing Person Re-Identification (OC4-ReID), which simultaneously addresses two challenges of clothing changes and occlusion. Concretely, we construct two new datasets, Occ-LTCC and Occ-PRCC, based on original CC-ReID datasets to include random occlusions of key pedestrians components (e.g., head, torso). Moreover, a novel benchmark is proposed for OC4-ReID incorporating a Train-Test Micro Granularity Screening (T2MGS) module to mitigate the influence of occlusion and proposing a Part-Robust Triplet (PRT) loss for partial features learning. Comprehensive experiments on the proposed datasets, as well as on two CC-ReID benchmark datasets demonstrate the superior performance of proposed method against other state-of-the-art methods. The codes and datasets are available at: https://github.com/1024AILab/OC4-ReID.", "date": "2024-03-13", "categories": []}
{"id": "2506.01806", "title": "Ridgeformer: Mutli-Stage Contrastive Training For Fine-grained Cross-Domain Fingerprint Recognition", "abstract": "The increasing demand for hygienic and portable biometric systems has underscored the critical need for advancements in contactless fingerprint recognition. Despite its potential, this technology faces notable challenges, including out-of-focus image acquisition, reduced contrast between fingerprint ridges and valleys, variations in finger positioning, and perspective distortion. These factors significantly hinder the accuracy and reliability of contactless fingerprint matching. To address these issues, we propose a novel multi-stage transformer-based contactless fingerprint matching approach that first captures global spatial features and subsequently refines localized feature alignment across fingerprint samples. By employing a hierarchical feature extraction and matching pipeline, our method ensures fine-grained, cross-sample alignment while maintaining the robustness of global feature representation. We perform extensive evaluations on publicly available datasets such as HKPolyU and RidgeBase under different evaluation protocols, such as contactless-to-contact matching and contactless-to-contactless matching and demonstrate that our proposed approach outperforms existing methods, including COTS solutions.", "date": "2025-06-02", "categories": []}
{"id": "2409.19954", "title": "Domain Consistency Representation Learning for Lifelong Person Re-Identification", "abstract": "Lifelong person re-identification (LReID) exhibits a contradictory relationship between intra-domain discrimination and inter-domain gaps when learning from continuous data. Intra-domain discrimination focuses on individual nuances (i.e., clothing type, accessories, etc.), while inter-domain gaps emphasize domain consistency. Achieving a trade-off between maximizing intra-domain discrimination and minimizing inter-domain gaps is a crucial challenge for improving LReID performance. Most existing methods strive to reduce inter-domain gaps through knowledge distillation to maintain domain consistency. However, they often ignore intra-domain discrimination. To address this challenge, we propose a novel domain consistency representation learning (DCR) model that explores global and attribute-wise representations as a bridge to balance intra-domain discrimination and inter-domain gaps. At the intra-domain level, we explore the complementary relationship between global and attribute-wise representations to improve discrimination among similar identities. Excessive learning intra-domain discrimination can lead to catastrophic forgetting. We further develop an attribute-oriented anti-forgetting (AF) strategy that explores attribute-wise representations to enhance inter-domain consistency, and propose a knowledge consolidation (KC) strategy to facilitate knowledge transfer. Extensive experiments show that our DCR achieves superior performance compared to state-of-the-art LReID methods. Our code is available at https://github.com/LiuShiBen/DCR.", "date": "2024-09-30", "categories": []}
{"id": "2509.03032", "title": "Background Matters Too: A Language-Enhanced Adversarial Framework for Person Re-Identification", "abstract": "Person re-identification faces two core challenges: precisely locating the foreground target while suppressing background noise and extracting fine-grained features from the target region. Numerous visual-only approaches address these issues by partitioning an image and applying attention modules, yet they rely on costly manual annotations and struggle with complex occlusions. Recent multimodal methods, motivated by CLIP, introduce semantic cues to guide visual understanding. However, they focus solely on foreground information, but overlook the potential value of background cues. Inspired by human perception, we argue that background semantics are as important as the foreground semantics in ReID, as humans tend to eliminate background distractions while focusing on target appearance. Therefore, this paper proposes an end-to-end framework that jointly models foreground and background information within a dual-branch cross-modal feature extraction pipeline. To help the network distinguish between the two domains, we propose an intra-semantic alignment and inter-semantic adversarial learning strategy. Specifically, we align visual and textual features that share the same semantics across domains, while simultaneously penalizing similarity between foreground and background features to enhance the network's discriminative power. This strategy drives the model to actively suppress noisy background regions and enhance attention toward identity-relevant foreground cues. Comprehensive experiments on two holistic and two occluded ReID benchmarks demonstrate the effectiveness and generality of the proposed method, with results that match or surpass those of current state-of-the-art approaches.", "date": "2025-09-03", "categories": []}
{"id": "2308.00428", "title": "Multiscale Feature Learning Using Co-Tuplet Loss for Offline Handwritten Signature Verification", "abstract": "Handwritten signature verification, crucial for legal and financial institutions, faces challenges including inter-writer similarity, intra-writer variations, and limited signature samples. To address these, we introduce the MultiScale Signature feature learning Network (MS-SigNet) with the co-tuplet loss, a novel metric learning loss designed for offline handwritten signature verification. MS-SigNet learns both global and regional signature features from multiple spatial scales, enhancing feature discrimination. This approach effectively distinguishes genuine signatures from skilled forgeries by capturing overall strokes and detailed local differences. The co-tuplet loss, focusing on multiple positive and negative examples, overcomes the limitations of typical metric learning losses by addressing inter-writer similarity and intra-writer variations and emphasizing informative examples. The code is available at https://github.com/ashleyfhh/MS-SigNet. We also present HanSig, a large-scale Chinese signature dataset to support robust system development for this language. The dataset is accessible at https://github.com/hsinmin/HanSig. Experimental results on four benchmark datasets in different languages demonstrate the promising performance of our method in comparison to state-of-the-art approaches.", "date": "2023-08-01", "categories": []}
{"id": "2407.06096", "title": "Muzzle-Based Cattle Identification System Using Artificial Intelligence (AI)", "abstract": "Absence of tamper-proof cattle identification technology was a significant problem preventing insurance companies from providing livestock insurance. This lack of technology had devastating financial consequences for marginal farmers as they did not have the opportunity to claim compensation for any unexpected events such as the accidental death of cattle in Bangladesh. Using machine learning and deep learning algorithms, we have solved the bottleneck of cattle identification by developing and introducing a muzzle-based cattle identification system. The uniqueness of cattle muzzles has been scientifically established, which resembles human fingerprints. This is the fundamental premise that prompted us to develop a cattle identification system that extracts the uniqueness of cattle muzzles. For this purpose, we collected 32,374 images from 826 cattle. Contrast-limited adaptive histogram equalization (CLAHE) with sharpening filters was applied in the preprocessing steps to remove noise from images. We used the YOLO algorithm for cattle muzzle detection in the image and the FaceNet architecture to learn unified embeddings from muzzle images using squared $L_2$ distances. Our system performs with an accuracy of $96.489\\%$, $F_1$ score of $97.334\\%$, and a true positive rate (tpr) of $87.993\\%$ at a remarkably low false positive rate (fpr) of $0.098\\%$. This reliable and efficient system for identifying cattle can significantly advance livestock insurance and precision farming.", "date": "2024-07-08", "categories": []}
{"id": "2503.21595", "title": "PS-ReID: Advancing Person Re-Identification and Precise Segmentation with Multimodal Retrieval", "abstract": "Person re-identification (ReID) plays a critical role in applications such as security surveillance and criminal investigations. Most traditional image-based ReID methods face challenges including occlusions and lighting changes, while text provides complementary information to mitigate these issues. However, the integration of both image and text modalities remains underexplored. To address this gap, we propose {\\bf PS-ReID}, a multimodal model that combines image and text inputs to enhance ReID performance. In contrast to existing ReID methods limited by cropped pedestrian images, our PS-ReID focuses on full-scene settings and introduces a multimodal ReID task that incorporates segmentation, enabling precise feature extraction of the queried individual, even under challenging conditions such as occlusion. To this end, our model adopts a dual-path asymmetric encoding scheme that explicitly separates query and target roles: the query branch captures identity-discriminative cues, while the target branch performs holistic scene reasoning. Additionally, a token-level ReID loss supervises identity-aware tokens, coupling retrieval and segmentation to yield masks that are both spatially precise and identity-consistent. To facilitate systematic evaluation, we construct M2ReID, currently the largest full-scene multimodal ReID dataset, with over 200K images and 4,894 identities, featuring multimodal queries and high-quality segmentation masks. Experimental results demonstrate that PS-ReID significantly outperforms unimodal query-based models in both ReID and segmentation tasks. The model excels in challenging real-world scenarios such as occlusion, low lighting, and background clutter, offering a robust and flexible solution for person retrieval and segmentation. All code, models, and datasets will be publicly available.", "date": "2025-03-27", "categories": []}
{"id": "2509.00332", "title": "CryptoFace: End-to-End Encrypted Face Recognition", "abstract": "Face recognition is central to many authentication, security, and personalized applications. Yet, it suffers from significant privacy risks, particularly arising from unauthorized access to sensitive biometric data. This paper introduces CryptoFace, the first end-to-end encrypted face recognition system with fully homomorphic encryption (FHE). It enables secure processing of facial data across all stages of a face-recognition process--feature extraction, storage, and matching--without exposing raw images or features. We introduce a mixture of shallow patch convolutional networks to support higher-dimensional tensors via patch-based processing while reducing the multiplicative depth and, thus, inference latency. Parallel FHE evaluation of these networks ensures near-resolution-independent latency. On standard face recognition benchmarks, CryptoFace significantly accelerates inference and increases verification accuracy compared to the state-of-the-art FHE neural networks adapted for face recognition. CryptoFace will facilitate secure face recognition systems requiring robust and provable security. The code is available at https://github.com/human-analysis/CryptoFace.", "date": "2025-08-30", "categories": []}
{"id": "2509.00781", "title": "Secure and Scalable Face Retrieval via Cancelable Product Quantization", "abstract": "Despite the ubiquity of modern face retrieval systems, their retrieval stage is often outsourced to third-party entities, posing significant risks to user portrait privacy. Although homomorphic encryption (HE) offers strong security guarantees by enabling arithmetic computations in the cipher space, its high computational inefficiency makes it unsuitable for real-time, real-world applications. To address this issue, we propose Cancelable Product Quantization, a highly efficient framework for secure face representation retrieval. Our hierarchical two-stage framework comprises: (i) a high-throughput cancelable PQ indexing module for fast candidate filtering, and (ii) a fine-grained cipher-space retrieval module for final precise face ranking. A tailored protection mechanism is designed to secure the indexing module for cancelable biometric authentication while ensuring efficiency. Experiments on benchmark datasets demonstrate that our method achieves an decent balance between effectiveness, efficiency and security.", "date": "2025-08-31", "categories": []}
{"id": "2508.19312", "title": "Sistema de Reconocimiento Facial Federado en Conjuntos Abiertos basado en OpenMax", "abstract": "Facial recognition powered by Artificial Intelligence has achieved high accuracy in specific scenarios and applications. Nevertheless, it faces significant challenges regarding privacy and identity management, particularly when unknown individuals appear in the operational context. This paper presents the design, implementation, and evaluation of a facial recognition system within a federated learning framework tailored to open-set scenarios. The proposed approach integrates the OpenMax algorithm into federated learning, leveraging the exchange of mean activation vectors and local distance measures to reliably distinguish between known and unknown subjects. Experimental results validate the effectiveness of the proposed solution, demonstrating its potential for enhancing privacy-aware and robust facial recognition in distributed environments.\n  --\n  El reconocimiento facial impulsado por Inteligencia Artificial ha demostrado una alta precisi\\'on en algunos escenarios y aplicaciones. Sin embargo, presenta desaf\\'ios relacionados con la privacidad y la identificaci\\'on de personas, especialmente considerando que pueden aparecer sujetos desconocidos para el sistema que lo implementa. En este trabajo, se propone el dise\\~no, implementaci\\'on y evaluaci\\'on de un sistema de reconocimiento facial en un escenario de aprendizaje federado, orientado a conjuntos abiertos. Concretamente, se dise\\~na una soluci\\'on basada en el algoritmo OpenMax para escenarios de aprendizaje federado. La propuesta emplea el intercambio de los vectores de activaci\\'on promedio y distancias locales para identificar de manera eficaz tanto personas conocidas como desconocidas. Los experimentos realizados demuestran la implementaci\\'on efectiva de la soluci\\'on propuesta.", "date": "2025-08-26", "categories": []}
{"id": "2508.21371", "title": "Print2Volume: Generating Synthetic OCT-based 3D Fingerprint Volume from 2D Fingerprint Image", "abstract": "Optical Coherence Tomography (OCT) enables the acquisition of high-resolution, three-dimensional fingerprint data, capturing rich subsurface structures for robust biometric recognition. However, the high cost and time-consuming nature of OCT data acquisition have led to a scarcity of large-scale public datasets, significantly hindering the development of advanced algorithms, particularly data-hungry deep learning models. To address this critical bottleneck, this paper introduces Print2Volume, a novel framework for generating realistic, synthetic OCT-based 3D fingerprints from 2D fingerprint image. Our framework operates in three sequential stages: (1) a 2D style transfer module that converts a binary fingerprint into a grayscale images mimicking the style of a Z-direction mean-projected OCT scan; (2) a 3D Structure Expansion Network that extrapolates the 2D im-age into a plausible 3D anatomical volume; and (3) an OCT Realism Refiner, based on a 3D GAN, that renders the structural volume with authentic textures, speckle noise, and other imaging characteristics. Using Print2Volume, we generated a large-scale synthetic dataset of 420,000 samples. Quantitative experiments demonstrate the high quality of our synthetic data and its significant impact on recognition performance. By pre-training a recognition model on our synthetic data and fine-tuning it on a small real-world dataset, we achieved a remarkable reduction in the Equal Error Rate (EER) from 15.62% to 2.50% on the ZJUT-EIFD benchmark, proving the effectiveness of our approach in overcoming data scarcity.", "date": "2025-08-29", "categories": []}
{"id": "2503.12232", "title": "L2RW+: A Comprehensive Benchmark Towards Privacy-Preserved Visible-Infrared Person Re-Identification", "abstract": "Visible-infrared person re-identification (VI-ReID) is a challenging task that aims to match pedestrian images captured under varying lighting conditions, which has drawn intensive research attention and achieved promising results. However, existing methods adopt the centralized training, ignoring the potential privacy concerns as the data is distributed across multiple devices or entities in reality. In this paper, we propose L2RW+, a benchmark that brings VI-ReID closer to real-world applications. The core rationale behind L2RW+ is that incorporating decentralized training into VI-ReID can address privacy concerns in scenarios with limited data-sharing constrains. Specifically, we design protocols and corresponding algorithms for different privacy sensitivity levels. In our new benchmark, we simulate the training under real-world data conditions that: 1) data from each camera is completely isolated, or 2) different data entities (e.g., data controllers of a certain region) can selectively share the data. In this way, we simulate scenarios with strict privacy restrictions, which is closer to real-world conditions. Comprehensive experiments show the feasibility and potential of decentralized VI-ReID training at both image and video levels. In particular, with increasing data scales, the performance gap between decentralized and centralized training decreases, especially in video-level VI-ReID. In unseen domains, decentralized training even achieves performance comparable to SOTA centralized methods. This work offers a novel research entry for deploying VI-ReID into real-world scenarios and can benefit the community. Code is available at: https://github.com/Joey623/L2RW.", "date": "2025-03-15", "categories": []}
{"id": "2508.20626", "title": "ArtFace: Towards Historical Portrait Face Identification via Model Adaptation", "abstract": "Identifying sitters in historical paintings is a key task for art historians, offering insight into their lives and how they chose to be seen. However, the process is often subjective and limited by the lack of data and stylistic variations. Automated facial recognition is capable of handling challenging conditions and can assist, but while traditional facial recognition models perform well on photographs, they struggle with paintings due to domain shift and high intra-class variation. Artistic factors such as style, skill, intent, and influence from other works further complicate recognition. In this work, we investigate the potential of foundation models to improve facial recognition in artworks. By fine-tuning foundation models and integrating their embeddings with those from conventional facial recognition networks, we demonstrate notable improvements over current state-of-the-art methods. Our results show that foundation models can bridge the gap where traditional methods are ineffective. Paper page at https://www.idiap.ch/paper/artface/", "date": "2025-08-28", "categories": []}
{"id": "2508.05038", "title": "HAMoBE: Hierarchical and Adaptive Mixture of Biometric Experts for Video-based Person ReID", "abstract": "Recently, research interest in person re-identification (ReID) has increasingly focused on video-based scenarios, which are essential for robust surveillance and security in varied and dynamic environments. However, existing video-based ReID methods often overlook the necessity of identifying and selecting the most discriminative features from both videos in a query-gallery pair for effective matching. To address this issue, we propose a novel Hierarchical and Adaptive Mixture of Biometric Experts (HAMoBE) framework, which leverages multi-layer features from a pre-trained large model (e.g., CLIP) and is designed to mimic human perceptual mechanisms by independently modeling key biometric features--appearance, static body shape, and dynamic gait--and adaptively integrating them. Specifically, HAMoBE includes two levels: the first level extracts low-level features from multi-layer representations provided by the frozen large model, while the second level consists of specialized experts focusing on long-term, short-term, and temporal features. To ensure robust matching, we introduce a new dual-input decision gating network that dynamically adjusts the contributions of each expert based on their relevance to the input scenarios. Extensive evaluations on benchmarks like MEVID demonstrate that our approach yields significant performance improvements (e.g., +13.0% Rank-1 accuracy).", "date": "2025-08-07", "categories": []}
{"id": "2508.18415", "title": "Securing Face and Fingerprint Templates in Humanitarian Biometric Systems", "abstract": "In humanitarian and emergency scenarios, the use of biometrics can dramatically improve the efficiency of operations, but it poses risks for the data subjects, which are exacerbated in contexts of vulnerability. To address this, we present a mobile biometric system implementing a biometric template protection (BTP) scheme suitable for these scenarios. After rigorously formulating the functional, operational, and security and privacy requirements of these contexts, we perform a broad comparative analysis of the BTP landscape. PolyProtect, a method designed to operate on neural network face embeddings, is identified as the most suitable method due to its effectiveness, modularity, and lightweight computational burden. We evaluate PolyProtect in terms of verification and identification accuracy, irreversibility, and unlinkability, when this BTP method is applied to face embeddings extracted using EdgeFace, a novel state-of-the-art efficient feature extractor, on a real-world face dataset from a humanitarian field project in Ethiopia. Moreover, as PolyProtect promises to be modality-independent, we extend its evaluation to fingerprints. To the best of our knowledge, this is the first time that PolyProtect has been evaluated for the identification scenario and for fingerprint biometrics. Our experimental results are promising, and we plan to release our code", "date": "2025-08-25", "categories": []}
{"id": "2404.09376", "title": "$\\textit{sweet}$- An Open Source Modular Platform for Contactless Hand Vascular Biometric Experiments", "abstract": "Current finger-vein or palm-vein recognition systems usually require direct contact of the subject with the apparatus. This can be problematic in environments where hygiene is of primary importance. In this work we present a contactless vascular biometrics sensor platform named \\sweet which can be used for hand vascular biometrics studies (wrist, palm, and finger-vein) and surface features such as palmprint. It supports several acquisition modalities such as multi-spectral Near-Infrared (NIR), RGB-color, Stereo Vision (SV) and Photometric Stereo (PS). Using this platform we collect a dataset consisting of the fingers, palm and wrist vascular data of 120 subjects and develop a powerful 3D pipeline for the pre-processing of this data. We then present biometric experimental results, focusing on Finger-Vein Recognition (FVR). Finally, we discuss fusion of multiple modalities, such palm-vein combined with palm-print biometrics. The acquisition software, parts of the hardware design, the new FV dataset, as well as source-code for our experiments are publicly available for research purposes.", "date": "2024-04-14", "categories": []}
{"id": "2508.15850", "title": "Linkage Attacks Expose Identity Risks in Public ECG Data Sharing", "abstract": "The increasing availability of publicly shared electrocardiogram (ECG) data raises critical privacy concerns, as its biometric properties make individuals vulnerable to linkage attacks. Unlike prior studies that assume idealized adversarial capabilities, we evaluate ECG privacy risks under realistic conditions where attackers operate with partial knowledge. Using data from 109 participants across diverse real-world datasets, our approach achieves 85% accuracy in re-identifying individuals in public datasets while maintaining a 14.2% overall misclassification rate at an optimal confidence threshold, with 15.6% of unknown individuals misclassified as known and 12.8% of known individuals misclassified as unknown. These results highlight the inadequacy of simple anonymization techniques in preventing re-identification, demonstrating that even limited adversarial knowledge enables effective identity linkage. Our findings underscore the urgent need for privacy-preserving strategies, such as differential privacy, access control, and encrypted computation, to mitigate re-identification risks while ensuring the utility of shared biosignal data in healthcare applications.", "date": "2025-08-20", "categories": []}
{"id": "2508.13617", "title": "Two-Factor Authentication Smart Entryway Using Modified LBPH Algorithm", "abstract": "Face mask detection has become increasingly important recently, particularly during the COVID-19 pandemic. Many face detection models have been developed in smart entryways using IoT. However, there is a lack of IoT development on face mask detection. This paper proposes a two-factor authentication system for smart entryway access control using facial recognition and passcode verification and an automation process to alert the owner and activate the surveillance system when a stranger is detected and controls the system remotely via Telegram on a Raspberry Pi platform. The system employs the Local Binary Patterns Histograms for the full face recognition algorithm and modified LBPH algorithm for occluded face detection. On average, the system achieved an Accuracy of approximately 70%, a Precision of approximately 80%, and a Recall of approximately 83.26% across all tested users. The results indicate that the system is capable of conducting face recognition and mask detection, automating the operation of the remote control to register users, locking or unlocking the door, and notifying the owner. The sample participants highly accept it for future use in the user acceptance test.", "date": "2025-08-19", "categories": []}
{"id": "2508.13690", "title": "Know Me by My Pulse: Toward Practical Continuous Authentication on Wearable Devices via Wrist-Worn PPG", "abstract": "Biometric authentication using physiological signals offers a promising path toward secure and user-friendly access control in wearable devices. While electrocardiogram (ECG) signals have shown high discriminability, their intrusive sensing requirements and discontinuous acquisition limit practicality. Photoplethysmography (PPG), on the other hand, enables continuous, non-intrusive authentication with seamless integration into wrist-worn wearable devices. However, most prior work relies on high-frequency PPG (e.g., 75 - 500 Hz) and complex deep models, which incur significant energy and computational overhead, impeding deployment in power-constrained real-world systems. In this paper, we present the first real-world implementation and evaluation of a continuous authentication system on a smartwatch, We-Be Band, using low-frequency (25 Hz) multi-channel PPG signals. Our method employs a Bi-LSTM with attention mechanism to extract identity-specific features from short (4 s) windows of 4-channel PPG. Through extensive evaluations on both public datasets (PTTPPG) and our We-Be Dataset (26 subjects), we demonstrate strong classification performance with an average test accuracy of 88.11%, macro F1-score of 0.88, False Acceptance Rate (FAR) of 0.48%, False Rejection Rate (FRR) of 11.77%, and Equal Error Rate (EER) of 2.76%. Our 25 Hz system reduces sensor power consumption by 53% compared to 512 Hz and 19% compared to 128 Hz setups without compromising performance. We find that sampling at 25 Hz preserves authentication accuracy, whereas performance drops sharply at 20 Hz while offering only trivial additional power savings, underscoring 25 Hz as the practical lower bound. Additionally, we find that models trained exclusively on resting data fail under motion, while activity-diverse training improves robustness across physiological states.", "date": "2025-08-19", "categories": []}
{"id": "2508.13874", "title": "A Comprehensive Re-Evaluation of Biometric Modality Properties in the Modern Era", "abstract": "The rapid advancement of authentication systems and their increasing reliance on biometrics for faster and more accurate user verification experience, highlight the critical need for a reliable framework to evaluate the suitability of biometric modalities for specific applications. Currently, the most widely known evaluation framework is a comparative table from 1998, which no longer adequately captures recent technological developments or emerging vulnerabilities in biometric systems. To address these challenges, this work revisits the evaluation of biometric modalities through an expert survey involving 24 biometric specialists. The findings indicate substantial shifts in property ratings across modalities. For example, face recognition, shows improved ratings due to technological progress, while fingerprint, shows decreased reliability because of emerging vulnerabilities and attacks. Further analysis of expert agreement levels across rated properties highlighted the consistency of the provided evaluations and ensured the reliability of the ratings. Finally, expert assessments are compared with dataset-level uncertainty across 55 biometric datasets, revealing strong alignment in most modalities and underscoring the importance of integrating empirical evidence with expert insight. Moreover, the identified expert disagreements reveal key open challenges and help guide future research toward resolving them.", "date": "2025-08-19", "categories": []}
{"id": "2508.14012", "title": "Evaluating Identity Leakage in Speaker De-Identification Systems", "abstract": "Speaker de-identification aims to conceal a speaker's identity while preserving intelligibility of the underlying speech. We introduce a benchmark that quantifies residual identity leakage with three complementary error rates: equal error rate, cumulative match characteristic hit rate, and embedding-space similarity measured via canonical correlation analysis and Procrustes analysis. Evaluation results reveal that all state-of-the-art speaker de-identification systems leak identity information. The highest performing system in our evaluation performs only slightly better than random guessing, while the lowest performing system achieves a 45% hit rate within the top 50 candidates based on CMC. These findings highlight persistent privacy risks in current speaker de-identification technologies.", "date": "2025-08-19", "categories": []}
{"id": "2503.18830", "title": "DAGait: Generalized Skeleton-Guided Data Alignment for Gait Recognition", "abstract": "Gait recognition is emerging as a promising and innovative area within the field of computer vision, widely applied to remote person identification. Although existing gait recognition methods have achieved substantial success in controlled laboratory datasets, their performance often declines significantly when transitioning to wild datasets.We argue that the performance gap can be primarily attributed to the spatio-temporal distribution inconsistencies present in wild datasets, where subjects appear at varying angles, positions, and distances across the frames. To achieve accurate gait recognition in the wild, we propose a skeleton-guided silhouette alignment strategy, which uses prior knowledge of the skeletons to perform affine transformations on the corresponding silhouettes.To the best of our knowledge, this is the first study to explore the impact of data alignment on gait recognition. We conducted extensive experiments across multiple datasets and network architectures, and the results demonstrate the significant advantages of our proposed alignment strategy.Specifically, on the challenging Gait3D dataset, our method achieved an average performance improvement of 7.9% across all evaluated networks. Furthermore, our method achieves substantial improvements on cross-domain datasets, with accuracy improvements of up to 24.0%.", "date": "2025-03-24", "categories": []}
{"id": "2508.10672", "title": "Hybrid Generative Fusion for Efficient and Privacy-Preserving Face Recognition Dataset Generation", "abstract": "In this paper, we present our approach to the DataCV ICCV Challenge, which centers on building a high-quality face dataset to train a face recognition model. The constructed dataset must not contain identities overlapping with any existing public face datasets. To handle this challenge, we begin with a thorough cleaning of the baseline HSFace dataset, identifying and removing mislabeled or inconsistent identities through a Mixture-of-Experts (MoE) strategy combining face embedding clustering and GPT-4o-assisted verification. We retain the largest consistent identity cluster and apply data augmentation up to a fixed number of images per identity. To further diversify the dataset, we generate synthetic identities using Stable Diffusion with prompt engineering. As diffusion models are computationally intensive, we generate only one reference image per identity and efficiently expand it using Vec2Face, which rapidly produces 49 identity-consistent variants. This hybrid approach fuses GAN-based and diffusion-based samples, enabling efficient construction of a diverse and high-quality dataset. To address the high visual similarity among synthetic identities, we adopt a curriculum learning strategy by placing them early in the training schedule, allowing the model to progress from easier to harder samples. Our final dataset contains 50 images per identity, and all newly generated identities are checked with mainstream face datasets to ensure no identity leakage. Our method achieves \\textbf{1st place} in the competition, and experimental results show that our dataset improves model performance across 10K, 20K, and 100K identity scales. Code is available at https://github.com/Ferry-Li/datacv_fr.", "date": "2025-08-14", "categories": []}
{"id": "2501.13420", "title": "LVFace: Progressive Cluster Optimization for Large Vision Models in Face Recognition", "abstract": "Vision Transformers (ViTs) have revolutionized large-scale visual modeling, yet remain underexplored in face recognition (FR) where CNNs still dominate. We identify a critical bottleneck: CNN-inspired training paradigms fail to unlock ViT's potential, leading to suboptimal performance and convergence instability.To address this challenge, we propose LVFace, a ViT-based FR model that integrates Progressive Cluster Optimization (PCO) to achieve superior results. Specifically, PCO sequentially applies negative class sub-sampling (NCS) for robust and fast feature alignment from random initialization, feature expectation penalties for centroid stabilization, performing cluster boundary refinement through full-batch training without NCS constraints. LVFace establishes a new state-of-the-art face recognition baseline, surpassing leading approaches such as UniFace and TopoFR across multiple benchmarks. Extensive experiments demonstrate that LVFace delivers consistent performance gains, while exhibiting scalability to large-scale datasets and compatibility with mainstream VLMs and LLMs. Notably, LVFace secured 1st place in the ICCV 2021 Masked Face Recognition (MFR)-Ongoing Challenge (March 2025), proving its efficacy in real-world scenarios. Project is available at https://github.com/bytedance/LVFace.", "date": "2025-01-23", "categories": []}
{"id": "2508.11218", "title": "A CLIP-based Uncertainty Modal Modeling (UMM) Framework for Pedestrian Re-Identification in Autonomous Driving", "abstract": "Re-Identification (ReID) is a critical technology in intelligent perception systems, especially within autonomous driving, where onboard cameras must identify pedestrians across views and time in real-time to support safe navigation and trajectory prediction. However, the presence of uncertain or missing input modalities--such as RGB, infrared, sketches, or textual descriptions--poses significant challenges to conventional ReID approaches. While large-scale pre-trained models offer strong multimodal semantic modeling capabilities, their computational overhead limits practical deployment in resource-constrained environments. To address these challenges, we propose a lightweight Uncertainty Modal Modeling (UMM) framework, which integrates a multimodal token mapper, synthetic modality augmentation strategy, and cross-modal cue interactive learner. Together, these components enable unified feature representation, mitigate the impact of missing modalities, and extract complementary information across different data types. Additionally, UMM leverages CLIP's vision-language alignment ability to fuse multimodal inputs efficiently without extensive finetuning. Experimental results demonstrate that UMM achieves strong robustness, generalization, and computational efficiency under uncertain modality conditions, offering a scalable and practical solution for pedestrian re-identification in autonomous driving scenarios.", "date": "2025-08-15", "categories": []}
{"id": "2508.11376", "title": "Unified Knowledge Distillation Framework: Fine-Grained Alignment and Geometric Relationship Preservation for Deep Face Recognition", "abstract": "Knowledge Distillation is crucial for optimizing face recognition models for deployment in computationally limited settings, such as edge devices. Traditional KD methods, such as Raw L2 Feature Distillation or Feature Consistency loss, often fail to capture both fine-grained instance-level details and complex relational structures, leading to suboptimal performance. We propose a unified approach that integrates two novel loss functions, Instance-Level Embedding Distillation and Relation-Based Pairwise Similarity Distillation. Instance-Level Embedding Distillation focuses on aligning individual feature embeddings by leveraging a dynamic hard mining strategy, thereby enhancing learning from challenging examples. Relation-Based Pairwise Similarity Distillation captures relational information through pairwise similarity relationships, employing a memory bank mechanism and a sample mining strategy. This unified framework ensures both effective instance-level alignment and preservation of geometric relationships between samples, leading to a more comprehensive distillation process. Our unified framework outperforms state-of-the-art distillation methods across multiple benchmark face recognition datasets, as demonstrated by extensive experimental evaluations. Interestingly, when using strong teacher networks compared to the student, our unified KD enables the student to even surpass the teacher's accuracy.", "date": "2025-08-15", "categories": []}
{"id": "2508.11419", "title": "Training-free Dimensionality Reduction via Feature Truncation: Enhancing Efficiency in Privacy-preserving Multi-Biometric Systems", "abstract": "Biometric recognition is widely used, making the privacy and security of extracted templates a critical concern. Biometric Template Protection schemes, especially those utilizing Homomorphic Encryption, introduce significant computational challenges due to increased workload. Recent advances in deep neural networks have enabled state-of-the-art feature extraction for face, fingerprint, and iris modalities. The ubiquity and affordability of biometric sensors further facilitate multi-modal fusion, which can enhance security by combining features from different modalities. This work investigates the biometric performance of reduced multi-biometric template sizes. Experiments are conducted on an in-house virtual multi-biometric database, derived from DNN-extracted features for face, fingerprint, and iris, using the FRGC, MCYT, and CASIA databases. The evaluated approaches are (i) explainable and straightforward to implement under encryption, (ii) training-free, and (iii) capable of generalization. Dimensionality reduction of feature vectors leads to fewer operations in the Homomorphic Encryption (HE) domain, enabling more efficient encrypted processing while maintaining biometric accuracy and security at a level equivalent to or exceeding single-biometric recognition. Our results demonstrate that, by fusing feature vectors from multiple modalities, template size can be reduced by 67 % with no loss in Equal Error Rate (EER) compared to the best-performing single modality.", "date": "2025-08-15", "categories": []}
{"id": "2412.19160", "title": "A Lightweight Transformer with Phase-Only Cross-Attention for Illumination-Invariant Biometric Authentication", "abstract": "Traditional biometric systems have encountered significant setbacks due to various unavoidable factors, for example, wearing of face masks in face recognition-based biometrics and hygiene concerns in fingerprint-based biometrics. This paper proposes a novel lightweight vision transformer with phase-only cross-attention (POC-ViT) using dual biometric traits of forehead and periocular portions of the face, capable of performing well even with face masks and without any physical touch, offering a promising alternative to traditional methods. The POC-ViT framework is designed to handle two biometric traits and to capture inter-dependencies in terms of relative structural patterns. Each channel consists of a Cross-Attention using phase-only correlation (POC) that captures both their individual and correlated structural patterns. The computation of cross-attention using POC extracts the phase correlation in the spatial features. Therefore, it is robust against variations in resolution and intensity, as well as illumination changes in the input images. The lightweight model is suitable for edge device deployment. The performance of the proposed framework was successfully demonstrated using the Forehead Subcutaneous Vein Pattern and Periocular Biometric Pattern (FSVP-PBP) database, having 350 subjects. The POC-ViT framework outperformed state-of-the-art methods with an outstanding classification accuracy of $98.8\\%$ with the dual biometric traits.", "date": "2024-12-26", "categories": []}
{"id": "2508.10786", "title": "Cooperative Face Liveness Detection from Optical Flow", "abstract": "In this work, we proposed a novel cooperative video-based face liveness detection method based on a new user interaction scenario where participants are instructed to slowly move their frontal-oriented face closer to the camera. This controlled approaching face protocol, combined with optical flow analysis, represents the core innovation of our approach. By designing a system where users follow this specific movement pattern, we enable robust extraction of facial volume information through neural optical flow estimation, significantly improving discrimination between genuine faces and various presentation attacks (including printed photos, screen displays, masks, and video replays). Our method processes both the predicted optical flows and RGB frames through a neural classifier, effectively leveraging spatial-temporal features for more reliable liveness detection compared to passive methods.", "date": "2025-08-14", "categories": []}
{"id": "2508.09555", "title": "Topological Invariant-Based Iris Identification via Digital Homology and Machine Learning", "abstract": "Objective - This study presents a biometric identification method based on topological invariants from 2D iris images, representing iris texture via formally defined digital homology and evaluating classification performance.\n  Methods - Each normalized iris image (48x482 pixels) is divided into grids (e.g., 6x54 or 3x27). For each subregion, we compute Betti0, Betti1, and their ratio using a recent algorithm for homology groups in 2D digital images. The resulting invariants form a feature matrix used with logistic regression, KNN, and SVM (with PCA and 100 randomized repetitions). A convolutional neural network (CNN) is trained on raw images for comparison.\n  Results - Logistic regression achieved 97.78 +/- 0.82% accuracy, outperforming CNN (96.44 +/- 1.32%) and other feature-based models. The topological features showed high accuracy with low variance.\n  Conclusion - This is the first use of topological invariants from formal digital homology for iris recognition. The method offers a compact, interpretable, and accurate alternative to deep learning, useful when explainability or limited data is important. Beyond iris recognition, it can apply to other biometrics, medical imaging, materials science, remote sensing, and interpretable AI. It runs efficiently on CPU-only systems and produces robust, explainable features valuable for security-critical domains.", "date": "2025-08-13", "categories": []}
{"id": "2508.04614", "title": "How Does Bilateral Ear Symmetry Affect Deep Ear Features?", "abstract": "Ear recognition has gained attention as a reliable biometric technique due to the distinctive characteristics of human ears. With the increasing availability of large-scale datasets, convolutional neural networks (CNNs) have been widely adopted to learn features directly from raw ear images, outperforming traditional hand-crafted methods. However, the effect of bilateral ear symmetry on the features learned by CNNs has received little attention in recent studies. In this paper, we investigate how bilateral ear symmetry influences the effectiveness of CNN-based ear recognition. To this end, we first develop an ear side classifier to automatically categorize ear images as either left or right. We then explore the impact of incorporating this side information during both training and test. Cross-dataset evaluations are conducted on five datasets. Our results suggest that treating left and right ears separately during training and testing can lead to notable performance improvements. Furthermore, our ablation studies on alignment strategies, input sizes, and various hyperparameter settings provide practical insights into training CNN-based ear recognition systems on large-scale datasets to achieve higher verification rates.", "date": "2025-08-06", "categories": []}
{"id": "2507.03541", "title": "Foundation versus Domain-specific Models: Performance Comparison, Fusion, and Explainability in Face Recognition", "abstract": "In this paper, we address the following question: How do generic foundation models (e.g., CLIP, BLIP, GPT-4o, Grok-4) compare against a domain-specific face recognition model (viz., AdaFace or ArcFace) on the face recognition task? Through a series of experiments involving several foundation models and benchmark datasets, we report the following findings: (a) In all face benchmark datasets considered, domain-specific models outperformed zero-shot foundation models. (b) The performance of zero-shot generic foundation models improved on over-segmented face images compared to tightly cropped faces, thereby suggesting the importance of contextual clues. (c) A simple score-level fusion of a foundation model with a domain-specific face recognition model improved the accuracy at low false match rates. (d) Foundation models, such as GPT-4o and Grok-4, are able to provide explainability to the face recognition pipeline. In some instances, foundation models are even able to resolve low-confidence decisions made by AdaFace, thereby reiterating the importance of combining domain-specific face recognition models with generic foundation models in a judicious manner.", "date": "2025-07-04", "categories": []}
{"id": "2508.07960", "title": "VOIDFace: A Privacy-Preserving Multi-Network Face Recognition With Enhanced Security", "abstract": "Advancement of machine learning techniques, combined with the availability of large-scale datasets, has significantly improved the accuracy and efficiency of facial recognition. Modern facial recognition systems are trained using large face datasets collected from diverse individuals or public repositories. However, for training, these datasets are often replicated and stored in multiple workstations, resulting in data replication, which complicates database management and oversight. Currently, once a user submits their face for dataset preparation, they lose control over how their data is used, raising significant privacy and ethical concerns. This paper introduces VOIDFace, a novel framework for facial recognition systems that addresses two major issues. First, it eliminates the need of data replication and improves data control to securely store training face data by using visual secret sharing. Second, it proposes a patch-based multi-training network that uses this novel training data storage mechanism to develop a robust, privacy-preserving facial recognition system. By integrating these advancements, VOIDFace aims to improve the privacy, security, and efficiency of facial recognition training, while ensuring greater control over sensitive personal face data. VOIDFace also enables users to exercise their Right-To-Be-Forgotten property to control their personal data. Experimental evaluations on the VGGFace2 dataset show that VOIDFace provides Right-To-Be-Forgotten, improved data control, security, and privacy while maintaining competitive facial recognition performance. Code is available at: https://github.com/ajnasmuhammed89/VOIDFace", "date": "2025-08-11", "categories": []}
{"id": "2508.05246", "title": "A Study of Gender Classification Techniques Based on Iris Images: A Deep Survey and Analysis", "abstract": "Gender classification is attractive in a range of applications, including surveillance and monitoring, corporate profiling, and human-computer interaction. Individuals' identities may be gleaned from information about their gender, which is a kind of soft biometric. Over the years, several methods for determining a person's gender have been devised. Some of the most well-known ones are based on physical characteristics like face, fingerprint, palmprint, DNA, ears, gait, and iris. On the other hand, facial features account for the vast majority of gender classification methods. Also, the iris is a significant biometric trait because the iris, according to research, remains basically constant during an individual's life. Besides that, the iris is externally visible and is non-invasive to the user, which is important for practical applications. Furthermore, there are already high-quality methods for segmenting and encoding iris images, and the current methods facilitate selecting and extracting attribute vectors from iris textures. This study discusses several approaches to determining gender. The previous works of literature are briefly reviewed. Additionally, there are a variety of methodologies for different steps of gender classification. This study provides researchers with knowledge and analysis of the existing gender classification approaches. Also, it will assist researchers who are interested in this specific area, as well as highlight the gaps and challenges in the field, and finally provide suggestions and future paths for improvement.", "date": "2025-08-07", "categories": []}
{"id": "2508.06357", "title": "Are you In or Out (of gallery)? Wisdom from the Same-Identity Crowd", "abstract": "A central problem in one-to-many facial identification is that the person in the probe image may or may not have enrolled image(s) in the gallery; that is, may be In-gallery or Out-of-gallery. Past approaches to detect when a rank-one result is Out-of-gallery have mostly focused on finding a suitable threshold on the similarity score. We take a new approach, using the additional enrolled images of the identity with the rank-one result to predict if the rank-one result is In-gallery / Out-of-gallery. Given a gallery of identities and images, we generate In-gallery and Out-of-gallery training data by extracting the ranks of additional enrolled images corresponding to the rank-one identity. We then train a classifier to utilize this feature vector to predict whether a rank-one result is In-gallery or Out-of-gallery. Using two different datasets and four different matchers, we present experimental results showing that our approach is viable for mugshot quality probe images, and also, importantly, for probes degraded by blur, reduced resolution, atmospheric turbulence and sunglasses. We also analyze results across demographic groups, and show that In-gallery / Out-of-gallery classification accuracy is similar across demographics. Our approach has the potential to provide an objective estimate of whether a one-to-many facial identification is Out-of-gallery, and thereby to reduce false positive identifications, wrongful arrests, and wasted investigative time. Interestingly, comparing the results of older deep CNN-based face matchers with newer ones suggests that the effectiveness of our Out-of-gallery detection approach emerges only with matchers trained using advanced margin-based loss functions.", "date": "2025-08-08", "categories": []}
{"id": "2407.16133", "title": "Open-Set Biometrics: Beyond Good Closed-Set Models", "abstract": "Biometric recognition has primarily addressed closed-set identification, assuming all probe subjects are in the gallery. However, most practical applications involve open-set biometrics, where probe subjects may or may not be present in the gallery. This poses distinct challenges in effectively distinguishing individuals in the gallery while minimizing false detections. While it is commonly believed that powerful biometric models can excel in both closed- and open-set scenarios, existing loss functions are inconsistent with open-set evaluation. They treat genuine (mated) and imposter (non-mated) similarity scores symmetrically and neglect the relative magnitudes of imposter scores. To address these issues, we simulate open-set evaluation using minibatches during training and introduce novel loss functions: (1) the identification-detection loss optimized for open-set performance under selective thresholds and (2) relative threshold minimization to reduce the maximum negative score for each probe. Across diverse biometric tasks, including face recognition, gait recognition, and person re-identification, our experiments demonstrate the effectiveness of the proposed loss functions, significantly enhancing open-set performance while positively impacting closed-set performance. Our code and models are available at https://github.com/prevso1088/open-set-biometrics.", "date": "2024-07-23", "categories": []}
{"id": "2508.00135", "title": "Exploring the Feasibility of Deep Learning Techniques for Accurate Gender Classification from Eye Images", "abstract": "Gender classification has emerged as a crucial aspect in various fields, including security, human-machine interaction, surveillance, and advertising. Nonetheless, the accuracy of this classification can be influenced by factors such as cosmetics and disguise. Consequently, our study is dedicated to addressing this concern by concentrating on gender classification using color images of the periocular region. The periocular region refers to the area surrounding the eye, including the eyelids, eyebrows, and the region between them. It contains valuable visual cues that can be used to extract key features for gender classification. This paper introduces a sophisticated Convolutional Neural Network (CNN) model that utilizes color image databases to evaluate the effectiveness of the periocular region for gender classification. To validate the model's performance, we conducted tests on two eye datasets, namely CVBL and (Female and Male). The recommended architecture achieved an outstanding accuracy of 99% on the previously unused CVBL dataset while attaining a commendable accuracy of 96% with a small number of learnable parameters (7,235,089) on the (Female and Male) dataset. To ascertain the effectiveness of our proposed model for gender classification using the periocular region, we evaluated its performance through an extensive range of metrics and compared it with other state-of-the-art approaches. The results unequivocally demonstrate the efficacy of our model, thereby suggesting its potential for practical application in domains such as security and surveillance.", "date": "2025-07-31", "categories": []}
{"id": "2508.04998", "title": "Attribute Guidance With Inherent Pseudo-label For Occluded Person Re-identification", "abstract": "Person re-identification (Re-ID) aims to match person images across different camera views, with occluded Re-ID addressing scenarios where pedestrians are partially visible. While pre-trained vision-language models have shown effectiveness in Re-ID tasks, they face significant challenges in occluded scenarios by focusing on holistic image semantics while neglecting fine-grained attribute information. This limitation becomes particularly evident when dealing with partially occluded pedestrians or when distinguishing between individuals with subtle appearance differences. To address this limitation, we propose Attribute-Guide ReID (AG-ReID), a novel framework that leverages pre-trained models' inherent capabilities to extract fine-grained semantic attributes without additional data or annotations. Our framework operates through a two-stage process: first generating attribute pseudo-labels that capture subtle visual characteristics, then introducing a dual-guidance mechanism that combines holistic and fine-grained attribute information to enhance image feature extraction. Extensive experiments demonstrate that AG-ReID achieves state-of-the-art results on multiple widely-used Re-ID datasets, showing significant improvements in handling occlusions and subtle attribute differences while maintaining competitive performance on standard Re-ID scenarios.", "date": "2025-08-07", "categories": []}
{"id": "2508.05409", "title": "From Detection to Correction: Backdoor-Resilient Face Recognition via Vision-Language Trigger Detection and Noise-Based Neutralization", "abstract": "Biometric systems, such as face recognition systems powered by deep neural networks (DNNs), rely on large and highly sensitive datasets. Backdoor attacks can subvert these systems by manipulating the training process. By inserting a small trigger, such as a sticker, make-up, or patterned mask, into a few training images, an adversary can later present the same trigger during authentication to be falsely recognized as another individual, thereby gaining unauthorized access. Existing defense mechanisms against backdoor attacks still face challenges in precisely identifying and mitigating poisoned images without compromising data utility, which undermines the overall reliability of the system. We propose a novel and generalizable approach, TrueBiometric: Trustworthy Biometrics, which accurately detects poisoned images using a majority voting mechanism leveraging multiple state-of-the-art large vision language models. Once identified, poisoned samples are corrected using targeted and calibrated corrective noise. Our extensive empirical results demonstrate that TrueBiometric detects and corrects poisoned images with 100\\% accuracy without compromising accuracy on clean images. Compared to existing state-of-the-art approaches, TrueBiometric offers a more practical, accurate, and effective solution for mitigating backdoor attacks in face recognition systems.", "date": "2025-08-07", "categories": []}
{"id": "2508.05636", "title": "FaceAnonyMixer: Cancelable Faces via Identity Consistent Latent Space Mixing", "abstract": "Advancements in face recognition (FR) technologies have amplified privacy concerns, necessitating methods that protect identity while maintaining recognition utility. Existing face anonymization methods typically focus on obscuring identity but fail to meet the requirements of biometric template protection, including revocability, unlinkability, and irreversibility. We propose FaceAnonyMixer, a cancelable face generation framework that leverages the latent space of a pre-trained generative model to synthesize privacy-preserving face images. The core idea of FaceAnonyMixer is to irreversibly mix the latent code of a real face image with a synthetic code derived from a revocable key. The mixed latent code is further refined through a carefully designed multi-objective loss to satisfy all cancelable biometric requirements. FaceAnonyMixer is capable of generating high-quality cancelable faces that can be directly matched using existing FR systems without requiring any modifications. Extensive experiments on benchmark datasets demonstrate that FaceAnonyMixer delivers superior recognition accuracy while providing significantly stronger privacy protection, achieving over an 11% gain on commercial API compared to recent cancelable biometric methods. Code is available at: https://github.com/talha-alam/faceanonymixer.", "date": "2025-08-07", "categories": []}
{"id": "2508.04381", "title": "ProtoN: Prototype Node Graph Neural Network for Unconstrained Multi-Impression Ear Recognition", "abstract": "Ear biometrics offer a stable and contactless modality for identity recognition, yet their effectiveness remains limited by the scarcity of annotated data and significant intra-class variability. Existing methods typically extract identity features from individual impressions in isolation, restricting their ability to capture consistent and discriminative representations. To overcome these limitations, a few-shot learning framework, ProtoN, is proposed to jointly process multiple impressions of an identity using a graph-based approach. Each impression is represented as a node in a class-specific graph, alongside a learnable prototype node that encodes identity-level information. This graph is processed by a Prototype Graph Neural Network (PGNN) layer, specifically designed to refine both impression and prototype representations through a dual-path message-passing mechanism. To further enhance discriminative power, the PGNN incorporates a cross-graph prototype alignment strategy that improves class separability by enforcing intra-class compactness while maintaining inter-class distinction. Additionally, a hybrid loss function is employed to balance episodic and global classification objectives, thereby improving the overall structure of the embedding space. Extensive experiments on five benchmark ear datasets demonstrate that ProtoN achieves state-of-the-art performance, with Rank-1 identification accuracy of up to 99.60% and an Equal Error Rate (EER) as low as 0.025, showing the effectiveness for few-shot ear recognition under limited data conditions.", "date": "2025-08-06", "categories": []}
{"id": "2508.03064", "title": "CORE-ReID: Comprehensive Optimization and Refinement through Ensemble fusion in Domain Adaptation for person re-identification", "abstract": "This study introduces a novel framework, \"Comprehensive Optimization and Refinement through Ensemble Fusion in Domain Adaptation for Person Re-identification (CORE-ReID)\", to address an Unsupervised Domain Adaptation (UDA) for Person Re-identification (ReID). The framework utilizes CycleGAN to generate diverse data that harmonizes differences in image characteristics from different camera sources in the pre-training stage. In the fine-tuning stage, based on a pair of teacher-student networks, the framework integrates multi-view features for multi-level clustering to derive diverse pseudo labels. A learnable Ensemble Fusion component that focuses on fine-grained local information within global features is introduced to enhance learning comprehensiveness and avoid ambiguity associated with multiple pseudo-labels. Experimental results on three common UDAs in Person ReID demonstrate significant performance gains over state-of-the-art approaches. Additional enhancements, such as Efficient Channel Attention Block and Bidirectional Mean Feature Normalization mitigate deviation effects and adaptive fusion of global and local features using the ResNet-based model, further strengthening the framework. The proposed framework ensures clarity in fusion features, avoids ambiguity, and achieves high ac-curacy in terms of Mean Average Precision, Top-1, Top-5, and Top-10, positioning it as an advanced and effective solution for the UDA in Person ReID. Our codes and models are available at https://github.com/TrinhQuocNguyen/CORE-ReID.", "date": "2025-08-05", "categories": []}
{"id": "2501.01760", "title": "From Age Estimation to Age-Invariant Face Recognition: Generalized Age Feature Extraction Using Order-Enhanced Contrastive Learning", "abstract": "Generalized age feature extraction is crucial for age-related facial analysis tasks, such as age estimation and age-invariant face recognition (AIFR). Despite the recent successes of models in homogeneous-dataset experiments, their performance drops significantly in cross-dataset evaluations. Most of these models fail to extract generalized age features as they only attempt to map extracted features with training age labels directly without explicitly modeling the natural ordinal progression of aging. In this paper, we propose Order-Enhanced Contrastive Learning (OrdCon), a novel contrastive learning framework designed explicitly for ordinal attributes like age. Specifically, to extract generalized features, OrdCon aligns the direction vector of two features with either the natural aging direction or its reverse to model the ordinal process of aging. To further enhance generalizability, OrdCon leverages a novel soft proxy matching loss as a second contrastive objective, ensuring that features are positioned around the center of each age cluster with minimal intra-class variance and proportionally away from other clusters. By modeling the ageing process, the framework can enhance generalizability by improving the alignment of samples from the same class and reducing the divergence of direction vectors. We demonstrate that our proposed method achieves comparable results to state-of-the-art methods on various benchmark datasets in homogeneous-dataset evaluations for both age estimation and AIFR. In cross-dataset experiments, OrdCon outperforms other methods by reducing the mean absolute error by approximately 1.38 on average for the age estimation task and boosts the average accuracy for AIFR by 1.87%.", "date": "2025-01-03", "categories": []}
{"id": "2507.12869", "title": "WhoFi: Deep Person Re-Identification via Wi-Fi Channel Signal Encoding", "abstract": "Person Re-Identification is a key and challenging task in video surveillance. While traditional methods rely on visual data, issues like poor lighting, occlusion, and suboptimal angles often hinder performance. To address these challenges, we introduce WhoFi, a novel pipeline that utilizes Wi-Fi signals for person re-identification. Biometric features are extracted from Channel State Information (CSI) and processed through a modular Deep Neural Network (DNN) featuring a Transformer-based encoder. The network is trained using an in-batch negative loss function to learn robust and generalizable biometric signatures. Experiments on the NTU-Fi dataset show that our approach achieves competitive results compared to state-of-the-art methods, confirming its effectiveness in identifying individuals via Wi-Fi signals.", "date": "2025-07-17", "categories": []}
{"id": "2508.00841", "title": "Inclusive Review on Advances in Masked Human Face Recognition Technologies", "abstract": "Masked Face Recognition (MFR) is an increasingly important area in biometric recognition technologies, especially with the widespread use of masks as a result of the COVID-19 pandemic. This development has created new challenges for facial recognition systems due to the partial concealment of basic facial features. This paper aims to provide a comprehensive review of the latest developments in the field, with a focus on deep learning techniques, especially convolutional neural networks (CNNs) and twin networks (Siamese networks), which have played a pivotal role in improving the accuracy of covering face recognition. The paper discusses the most prominent challenges, which include changes in lighting, different facial positions, partial concealment, and the impact of mask types on the performance of systems. It also reviews advanced technologies developed to overcome these challenges, including data enhancement using artificial databases and multimedia methods to improve the ability of systems to generalize. In addition, the paper highlights advance in deep network design, feature extraction techniques, evaluation criteria, and data sets used in this area. Moreover, it reviews the various applications of masked face recognition in the fields of security and medicine, highlighting the growing importance of these systems in light of recurrent health crises and increasing security threats. Finally, the paper focuses on future research trends such as developing more efficient algorithms and integrating multimedia technologies to improve the performance of recognition systems in real-world environments and expand their applications.", "date": "2025-07-04", "categories": []}
{"id": "2508.00941", "title": "Latent Diffusion Based Face Enhancement under Degraded Conditions for Forensic Face Recognition", "abstract": "Face recognition systems experience severe performance degradation when processing low-quality forensic evidence imagery. This paper presents an evaluation of latent diffusion-based enhancement for improving face recognition under forensically relevant degradations. Using a dataset of 3,000 individuals from LFW with 24,000 recognition attempts, we implement the Flux.1 Kontext Dev pipeline with Facezoom LoRA adaptation to test against seven degradation categories, including compression artefacts, blur effects, and noise contamination. Our approach demonstrates substantial improvements, increasing overall recognition accuracy from 29.1% to 84.5% (55.4 percentage point improvement, 95% CI: [54.1, 56.7]). Statistical analysis reveals significant performance gains across all degradation types, with effect sizes exceeding conventional thresholds for practical significance. These findings establish the potential of sophisticated diffusion based enhancement in forensic face recognition applications.", "date": "2025-07-31", "categories": []}
{"id": "2508.00053", "title": "A Quality-Guided Mixture of Score-Fusion Experts Framework for Human Recognition", "abstract": "Whole-body biometric recognition is a challenging multimodal task that integrates various biometric modalities, including face, gait, and body. This integration is essential for overcoming the limitations of unimodal systems. Traditionally, whole-body recognition involves deploying different models to process multiple modalities, achieving the final outcome by score-fusion (e.g., weighted averaging of similarity matrices from each model). However, these conventional methods may overlook the variations in score distributions of individual modalities, making it challenging to improve final performance. In this work, we present \\textbf{Q}uality-guided \\textbf{M}ixture of score-fusion \\textbf{E}xperts (QME), a novel framework designed for improving whole-body biometric recognition performance through a learnable score-fusion strategy using a Mixture of Experts (MoE). We introduce a novel pseudo-quality loss for quality estimation with a modality-specific Quality Estimator (QE), and a score triplet loss to improve the metric performance. Extensive experiments on multiple whole-body biometric datasets demonstrate the effectiveness of our proposed approach, achieving state-of-the-art results across various metrics compared to baseline methods. Our method is effective for multimodal and multi-model, addressing key challenges such as model misalignment in the similarity score domain and variability in data quality.", "date": "2025-07-31", "categories": []}
{"id": "2506.12447", "title": "CLIP-HandID: Vision-Language Model for Hand-Based Person Identification", "abstract": "This paper introduces a novel approach to person identification using hand images, designed specifically for criminal investigations. The method is particularly valuable in serious crimes such as sexual abuse, where hand images are often the only identifiable evidence available. Our proposed method, CLIP-HandID, leverages a pre-trained foundational vision-language model - CLIP - to efficiently learn discriminative deep feature representations from hand images (input to CLIP's image encoder) using textual prompts as semantic guidance. Since hand images are labeled with indexes rather than text descriptions, we employ a textual inversion network to learn pseudo-tokens that encode specific visual contexts or appearance attributes. These learned pseudo-tokens are then incorporated into textual prompts, which are fed into CLIP's text encoder to leverage its multi-modal reasoning and enhance generalization for identification. Through extensive evaluations on two large, publicly available hand datasets with multi-ethnic representation, we demonstrate that our method significantly outperforms existing approaches.", "date": "2025-06-14", "categories": []}
{"id": "2504.15041", "title": "Distribution-aware Forgetting Compensation for Exemplar-Free Lifelong Person Re-identification", "abstract": "Lifelong Person Re-identification (LReID) suffers from a key challenge in preserving old knowledge while adapting to new information. The existing solutions include rehearsal-based and rehearsal-free methods to address this challenge. Rehearsal-based approaches rely on knowledge distillation, continuously accumulating forgetting during the distillation process. Rehearsal-free methods insufficiently learn the distribution of each domain, leading to forgetfulness over time. To solve these issues, we propose a novel Distribution-aware Forgetting Compensation (DAFC) model that explores cross-domain shared representation learning and domain-specific distribution integration without using old exemplars or knowledge distillation. We propose a Text-driven Prompt Aggregation (TPA) that utilizes text features to enrich prompt elements and guide the prompt model to learn fine-grained representations for each instance. This can enhance the differentiation of identity information and establish the foundation for domain distribution awareness. Then, Distribution-based Awareness and Integration (DAI) is designed to capture each domain-specific distribution by a dedicated expert network and adaptively consolidate them into a shared region in high-dimensional space. In this manner, DAI can consolidate and enhance cross-domain shared representation learning while alleviating catastrophic forgetting. Furthermore, we develop a Knowledge Consolidation Mechanism (KCM) that comprises instance-level discrimination and cross-domain consistency alignment strategies to facilitate model adaptive learning of new knowledge from the current domain and promote knowledge consolidation learning between acquired domain-specific distributions, respectively. Experimental results show that our DAFC outperforms state-of-the-art methods. Our code is available at https://github.com/LiuShiBen/DAFC.", "date": "2025-04-21", "categories": []}
{"id": "2507.17192", "title": "Vec2Face+ for Face Dataset Generation", "abstract": "When synthesizing identities as face recognition training data, it is generally believed that large inter-class separability and intra-class attribute variation are essential for synthesizing a quality dataset. % This belief is generally correct, and this is what we aim for. However, when increasing intra-class variation, existing methods overlook the necessity of maintaining intra-class identity consistency. % To address this and generate high-quality face training data, we propose Vec2Face+, a generative model that creates images directly from image features and allows for continuous and easy control of face identities and attributes. Using Vec2Face+, we obtain datasets with proper inter-class separability and intra-class variation and identity consistency using three strategies: 1) we sample vectors sufficiently different from others to generate well-separated identities; 2) we propose an AttrOP algorithm for increasing general attribute variations; 3) we propose LoRA-based pose control for generating images with profile head poses, which is more efficient and identity-preserving than AttrOP. % Our system generates VFace10K, a synthetic face dataset with 10K identities, which allows an FR model to achieve state-of-the-art accuracy on seven real-world test sets. Scaling the size to 4M and 12M images, the corresponding VFace100K and VFace300K datasets yield higher accuracy than the real-world training dataset, CASIA-WebFace, on five real-world test sets. This is the first time a synthetic dataset beats the CASIA-WebFace in average accuracy. In addition, we find that only 1 out of 11 synthetic datasets outperforms random guessing (\\emph{i.e., 50\\%}) in twin verification and that models trained with synthetic identities are more biased than those trained with real identities. Both are important aspects for future investigation. Code is available at https://github.com/HaiyuWu/Vec2Face_plus", "date": "2025-07-23", "categories": []}
{"id": "2507.20404", "title": "Second Competition on Presentation Attack Detection on ID Card", "abstract": "This work summarises and reports the results of the second Presentation Attack Detection competition on ID cards. This new version includes new elements compared to the previous one. (1) An automatic evaluation platform was enabled for automatic benchmarking; (2) Two tracks were proposed in order to evaluate algorithms and datasets, respectively; and (3) A new ID card dataset was shared with Track 1 teams to serve as the baseline dataset for the training and optimisation. The Hochschule Darmstadt, Fraunhofer-IGD, and Facephi company jointly organised this challenge. 20 teams were registered, and 74 submitted models were evaluated. For Track 1, the \"Dragons\" team reached first place with an Average Ranking and Equal Error rate (EER) of AV-Rank of 40.48% and 11.44% EER, respectively. For the more challenging approach in Track 2, the \"Incode\" team reached the best results with an AV-Rank of 14.76% and 6.36% EER, improving on the results of the first edition of 74.30% and 21.87% EER, respectively. These results suggest that PAD on ID cards is improving, but it is still a challenging problem related to the number of images, especially of bona fide images.", "date": "2025-07-27", "categories": []}
{"id": "2507.20782", "title": "Investigation of Accuracy and Bias in Face Recognition Trained with Synthetic Data", "abstract": "Synthetic data has emerged as a promising alternative for training face recognition (FR) models, offering advantages in scalability, privacy compliance, and potential for bias mitigation. However, critical questions remain on whether both high accuracy and fairness can be achieved with synthetic data. In this work, we evaluate the impact of synthetic data on bias and performance of FR systems. We generate balanced face dataset, FairFaceGen, using two state of the art text-to-image generators, Flux.1-dev and Stable Diffusion v3.5 (SD35), and combine them with several identity augmentation methods, including Arc2Face and four IP-Adapters. By maintaining equal identity count across synthetic and real datasets, we ensure fair comparisons when evaluating FR performance on standard (LFW, AgeDB-30, etc.) and challenging IJB-B/C benchmarks and FR bias on Racial Faces in-the-Wild (RFW) dataset. Our results demonstrate that although synthetic data still lags behind the real datasets in the generalization on IJB-B/C, demographically balanced synthetic datasets, especially those generated with SD35, show potential for bias mitigation. We also observe that the number and quality of intra-class augmentations significantly affect FR accuracy and fairness. These findings provide practical guidelines for constructing fairer FR systems using synthetic data.", "date": "2025-07-28", "categories": []}
{"id": "2507.01884", "title": "Self-Reinforcing Prototype Evolution with Dual-Knowledge Cooperation for Semi-Supervised Lifelong Person Re-Identification", "abstract": "Current lifelong person re-identification (LReID) methods predominantly rely on fully labeled data streams. However, in real-world scenarios where annotation resources are limited, a vast amount of unlabeled data coexists with scarce labeled samples, leading to the Semi-Supervised LReID (Semi-LReID) problem where LReID methods suffer severe performance degradation. Existing LReID methods, even when combined with semi-supervised strategies, suffer from limited long-term adaptation performance due to struggling with the noisy knowledge occurring during unlabeled data utilization. In this paper, we pioneer the investigation of Semi-LReID, introducing a novel Self-Reinforcing Prototype Evolution with Dual-Knowledge Cooperation framework (SPRED). Our key innovation lies in establishing a self-reinforcing cycle between dynamic prototype-guided pseudo-label generation and new-old knowledge collaborative purification to enhance the utilization of unlabeled data. Specifically, learnable identity prototypes are introduced to dynamically capture the identity distributions and generate high-quality pseudo-labels. Then, the dual-knowledge cooperation scheme integrates current model specialization and historical model generalization, refining noisy pseudo-labels. Through this cyclic design, reliable pseudo-labels are progressively mined to improve current-stage learning and ensure positive knowledge propagation over long-term learning. Experiments on the established Semi-LReID benchmarks show that our SPRED achieves state-of-the-art performance. Our source code is available at https://github.com/zhoujiahuan1991/ICCV2025-SPRED", "date": "2025-07-02", "categories": []}
{"id": "2507.17995", "title": "AG-VPReID.VIR: Bridging Aerial and Ground Platforms for Video-based Visible-Infrared Person Re-ID", "abstract": "Person re-identification (Re-ID) across visible and infrared modalities is crucial for 24-hour surveillance systems, but existing datasets primarily focus on ground-level perspectives. While ground-based IR systems offer nighttime capabilities, they suffer from occlusions, limited coverage, and vulnerability to obstructions--problems that aerial perspectives uniquely solve. To address these limitations, we introduce AG-VPReID.VIR, the first aerial-ground cross-modality video-based person Re-ID dataset. This dataset captures 1,837 identities across 4,861 tracklets (124,855 frames) using both UAV-mounted and fixed CCTV cameras in RGB and infrared modalities. AG-VPReID.VIR presents unique challenges including cross-viewpoint variations, modality discrepancies, and temporal dynamics. Additionally, we propose TCC-VPReID, a novel three-stream architecture designed to address the joint challenges of cross-platform and cross-modality person Re-ID. Our approach bridges the domain gaps between aerial-ground perspectives and RGB-IR modalities, through style-robust feature learning, memory-based cross-view adaptation, and intermediary-guided temporal modeling. Experiments show that AG-VPReID.VIR presents distinctive challenges compared to existing datasets, with our TCC-VPReID framework achieving significant performance gains across multiple evaluation protocols. Dataset and code are available at https://github.com/agvpreid25/AG-VPReID.VIR.", "date": "2025-07-24", "categories": []}
{"id": "2507.18627", "title": "Gait Recognition Based on Tiny ML and IMU Sensors", "abstract": "This project presents the development of a gait recognition system using Tiny Machine Learning (Tiny ML) and Inertial Measurement Unit (IMU) sensors. The system leverages the XIAO-nRF52840 Sense microcontroller and the LSM6DS3 IMU sensor to capture motion data, including acceleration and angular velocity, from four distinct activities: walking, stationary, going upstairs, and going downstairs. The data collected is processed through Edge Impulse, an edge AI platform, which enables the training of machine learning models that can be deployed directly onto the microcontroller for real-time activity classification.The data preprocessing step involves extracting relevant features from the raw sensor data using techniques such as sliding windows and data normalization, followed by training a Deep Neural Network (DNN) classifier for activity recognition. The model achieves over 80% accuracy on a test dataset, demonstrating its ability to classify the four activities effectively. Additionally, the platform enables anomaly detection, further enhancing the robustness of the system. The integration of Tiny ML ensures low-power operation, making it suitable for battery-powered or energy-harvesting devices.", "date": "2025-07-24", "categories": []}
{"id": "2507.17158", "title": "DOOMGAN:High-Fidelity Dynamic Identity Obfuscation Ocular Generative Morphing", "abstract": "Ocular biometrics in the visible spectrum have emerged as a prominent modality due to their high accuracy, resistance to spoofing, and non-invasive nature. However, morphing attacks, synthetic biometric traits created by blending features from multiple individuals, threaten biometric system integrity. While extensively studied for near-infrared iris and face biometrics, morphing in visible-spectrum ocular data remains underexplored. Simulating such attacks demands advanced generation models that handle uncontrolled conditions while preserving detailed ocular features like iris boundaries and periocular textures. To address this gap, we introduce DOOMGAN, that encompasses landmark-driven encoding of visible ocular anatomy, attention-guided generation for realistic morph synthesis, and dynamic weighting of multi-faceted losses for optimized convergence. DOOMGAN achieves over 20% higher attack success rates than baseline methods under stringent thresholds, along with 20% better elliptical iris structure generation and 30% improved gaze consistency. We also release the first comprehensive ocular morphing dataset to support further research in this domain.", "date": "2025-07-23", "categories": []}
{"id": "2507.17729", "title": "A Comprehensive Evaluation Framework for the Study of the Effects of Facial Filters on Face Recognition Accuracy", "abstract": "Facial filters are now commonplace for social media users around the world. Previous work has demonstrated that facial filters can negatively impact automated face recognition performance. However, these studies focus on small numbers of hand-picked filters in particular styles. In order to more effectively incorporate the wide ranges of filters present on various social media applications, we introduce a framework that allows for larger-scale study of the impact of facial filters on automated recognition. This framework includes a controlled dataset of face images, a principled filter selection process that selects a representative range of filters for experimentation, and a set of experiments to evaluate the filters' impact on recognition. We demonstrate our framework with a case study of filters from the American applications Instagram and Snapchat and the Chinese applications Meitu and Pitu to uncover cross-cultural differences. Finally, we show how the filtering effect in a face embedding space can easily be detected and restored to improve face recognition performance.", "date": "2025-07-23", "categories": []}
{"id": "2507.10978", "title": "Mind the Gap: Bridging Occlusion in Gait Recognition via Residual Gap Correction", "abstract": "Gait is becoming popular as a method of person re-identification because of its ability to identify people at a distance. However, most current works in gait recognition do not address the practical problem of occlusions. Among those which do, some require paired tuples of occluded and holistic sequences, which are impractical to collect in the real world. Further, these approaches work on occlusions but fail to retain performance on holistic inputs. To address these challenges, we propose RG-Gait, a method for residual correction for occluded gait recognition with holistic retention. We model the problem as a residual learning task, conceptualizing the occluded gait signature as a residual deviation from the holistic gait representation. Our proposed network adaptively integrates the learned residual, significantly improving performance on occluded gait sequences without compromising the holistic recognition accuracy. We evaluate our approach on the challenging Gait3D, GREW and BRIAR datasets and show that learning the residual can be an effective technique to tackle occluded gait recognition with holistic retention. We release our code publicly at https://github.com/Ayush-00/rg-gait.", "date": "2025-07-15", "categories": []}
{"id": "2507.16201", "title": "A Single-step Accurate Fingerprint Registration Method Based on Local Feature Matching", "abstract": "Distortion of the fingerprint images leads to a decline in fingerprint recognition performance, and fingerprint registration can mitigate this distortion issue by accurately aligning two fingerprint images. Currently, fingerprint registration methods often consist of two steps: an initial registration based on minutiae, and a dense registration based on matching points. However, when the quality of fingerprint image is low, the number of detected minutiae is reduced, leading to frequent failures in the initial registration, which ultimately causes the entire fingerprint registration process to fail. In this study, we propose an end-to-end single-step fingerprint registration algorithm that aligns two fingerprints by directly predicting the semi-dense matching points correspondences between two fingerprints. Thus, our method minimizes the risk of minutiae registration failure and also leverages global-local attentions to achieve end-to-end pixel-level alignment between the two fingerprints. Experiment results prove that our method can achieve the state-of-the-art matching performance with only single-step registration, and it can also be used in conjunction with dense registration algorithms for further performance improvements.", "date": "2025-07-22", "categories": []}
{"id": "2507.16790", "title": "Enhancing Domain Diversity in Synthetic Data Face Recognition with Dataset Fusion", "abstract": "While the accuracy of face recognition systems has improved significantly in recent years, the datasets used to train these models are often collected through web crawling without the explicit consent of users, raising ethical and privacy concerns. To address this, many recent approaches have explored the use of synthetic data for training face recognition models. However, these models typically underperform compared to those trained on real-world data. A common limitation is that a single generator model is often used to create the entire synthetic dataset, leading to model-specific artifacts that may cause overfitting to the generator's inherent biases and artifacts. In this work, we propose a solution by combining two state-of-the-art synthetic face datasets generated using architecturally distinct backbones. This fusion reduces model-specific artifacts, enhances diversity in pose, lighting, and demographics, and implicitly regularizes the face recognition model by emphasizing identity-relevant features. We evaluate the performance of models trained on this combined dataset using standard face recognition benchmarks and demonstrate that our approach achieves superior performance across many of these benchmarks.", "date": "2025-07-22", "categories": []}
{"id": "2206.01658", "title": "Identification via Retinal Vessels Combining LBP and HOG", "abstract": "With development of information technology and necessity for high security, using different identification methods has become very important. Each biometric feature has its own advantages and disadvantages and choosing each of them depends on our usage. Retinal scanning is a bio scale method for identification. The retina is composed of vessels and optical disk. The vessels distribution pattern is one the remarkable retinal identification methods. In this paper, a new approach is presented for identification via retinal images using LBP and hog methods. In the proposed method, it will be tried to separate the retinal vessels accurately via machine vision techniques which will have good sustainability in rotation and size change. HOG-based or LBP-based methods or their combination can be used for separation and also HSV color space can be used too. Having extracted the features, the similarity criteria can be used for identification. The implementation of proposed method and its comparison with one of the newly-presented methods in this area shows better performance of the proposed method.", "date": "2022-06-03", "categories": []}
{"id": "2507.15297", "title": "Minutiae-Anchored Local Dense Representation for Fingerprint Matching", "abstract": "Fingerprint matching under diverse capture conditions remains a fundamental challenge in biometric recognition. To achieve robust and accurate performance in such scenarios, we propose DMD, a minutiae-anchored local dense representation which captures both fine-grained ridge textures and discriminative minutiae features in a spatially structured manner. Specifically, descriptors are extracted from local patches centered and oriented on each detected minutia, forming a three-dimensional tensor, where two dimensions represent spatial locations on the fingerprint plane and the third encodes semantic features. This representation explicitly captures abstract features of local image patches, enabling a multi-level, fine-grained description that aggregates information from multiple minutiae and their surrounding ridge structures. Furthermore, thanks to its strong spatial correspondence with the patch image, DMD allows for the use of foreground segmentation masks to identify valid descriptor regions. During matching, comparisons are then restricted to overlapping foreground areas, improving efficiency and robustness. Extensive experiments on rolled, plain, parital, contactless, and latent fingerprint datasets demonstrate the effectiveness and generalizability of the proposed method. It achieves state-of-the-art accuracy across multiple benchmarks while maintaining high computational efficiency, showing strong potential for large-scale fingerprint recognition. Corresponding code is available at https://github.com/Yu-Yy/DMD.", "date": "2025-07-21", "categories": []}
{"id": "2507.07393", "title": "KeyRe-ID: Keypoint-Guided Person Re-Identification using Part-Aware Representation in Videos", "abstract": "We propose \\textbf{KeyRe-ID}, a keypoint-guided video-based person re-identification framework consisting of global and local branches that leverage human keypoints for enhanced spatiotemporal representation learning. The global branch captures holistic identity semantics through Transformer-based temporal aggregation, while the local branch dynamically segments body regions based on keypoints to generate fine-grained, part-aware features. Extensive experiments on MARS and iLIDS-VID benchmarks demonstrate state-of-the-art performance, achieving 91.73\\% mAP and 97.32\\% Rank-1 accuracy on MARS, and 96.00\\% Rank-1 and 100.0\\% Rank-5 accuracy on iLIDS-VID. The code for this work will be publicly available on GitHub upon publication.", "date": "2025-07-10", "categories": []}
{"id": "2507.12854", "title": "Transformer-Based Person Identification via Wi-Fi CSI Amplitude and Phase Perturbations", "abstract": "Wi-Fi sensing is gaining momentum as a non-intrusive and privacy-preserving alternative to vision-based systems for human identification. However, person identification through wireless signals, particularly without user motion, remains largely unexplored. Most prior wireless-based approaches rely on movement patterns, such as walking gait, to extract biometric cues. In contrast, we propose a transformer-based method that identifies individuals from Channel State Information (CSI) recorded while the subject remains stationary. CSI captures fine-grained amplitude and phase distortions induced by the unique interaction between the human body and the radio signal. To support evaluation, we introduce a dataset acquired with ESP32 devices in a controlled indoor environment, featuring six participants observed across multiple orientations. A tailored preprocessing pipeline, including outlier removal, smoothing, and phase calibration, enhances signal quality. Our dual-branch transformer architecture processes amplitude and phase modalities separately and achieves 99.82\\% classification accuracy, outperforming convolutional and multilayer perceptron baselines. These results demonstrate the discriminative potential of CSI perturbations, highlighting their capacity to encode biometric traits in a consistent manner. They further confirm the viability of passive, device-free person identification using low-cost commodity Wi-Fi hardware in real-world settings.", "date": "2025-07-17", "categories": []}
{"id": "2507.12873", "title": "An Investigation of Ear-EEG Signals for a Novel Biometric Authentication System", "abstract": "This work explores the feasibility of biometric authentication using EEG signals acquired through in-ear devices, commonly referred to as ear-EEG. Traditional EEG-based biometric systems, while secure, often suffer from low usability due to cumbersome scalp-based electrode setups. In this study, we propose a novel and practical framework leveraging ear-EEG signals as a user-friendly alternative for everyday biometric authentication. The system extracts an original combination of temporal and spectral features from ear-EEG signals and feeds them into a fully connected deep neural network for subject identification. Experimental results on the only currently available ear-EEG dataset suitable for different purposes, including biometric authentication, demonstrate promising performance, with an average accuracy of 82\\% in a subject identification scenario. These findings confirm the potential of ear-EEG as a viable and deployable direction for next-generation real-world biometric systems.", "date": "2025-07-17", "categories": []}
{"id": "2507.12942", "title": "Weakly Supervised Visible-Infrared Person Re-Identification via Heterogeneous Expert Collaborative Consistency Learning", "abstract": "To reduce the reliance of visible-infrared person re-identification (ReID) models on labeled cross-modal samples, this paper explores a weakly supervised cross-modal person ReID method that uses only single-modal sample identity labels, addressing scenarios where cross-modal identity labels are unavailable. To mitigate the impact of missing cross-modal labels on model performance, we propose a heterogeneous expert collaborative consistency learning framework, designed to establish robust cross-modal identity correspondences in a weakly supervised manner. This framework leverages labeled data from each modality to independently train dedicated classification experts. To associate cross-modal samples, these classification experts act as heterogeneous predictors, predicting the identities of samples from the other modality. To improve prediction accuracy, we design a cross-modal relationship fusion mechanism that effectively integrates predictions from different experts. Under the implicit supervision provided by cross-modal identity correspondences, collaborative and consistent learning among the experts is encouraged, significantly enhancing the model's ability to extract modality-invariant features and improve cross-modal identity recognition. Experimental results on two challenging datasets validate the effectiveness of the proposed method.", "date": "2025-07-17", "categories": []}
{"id": "2505.07530", "title": "FLUXSynID: A Framework for Identity-Controlled Synthetic Face Generation with Document and Live Images", "abstract": "Synthetic face datasets are increasingly used to overcome the limitations of real-world biometric data, including privacy concerns, demographic imbalance, and high collection costs. However, many existing methods lack fine-grained control over identity attributes and fail to produce paired, identity-consistent images under structured capture conditions. We introduce FLUXSynID, a framework for generating high-resolution synthetic face datasets along with a dataset of 14,889 synthetic identities. We generate synthetic faces with user-defined identity attribute distributions, offering both document-style and trusted live capture images. The dataset generated using the FLUXSynID framework shows improved alignment with real-world identity distributions and greater inter-class diversity compared to prior work. Our work is publicly released to support biometric research, including face recognition and morphing attack detection.", "date": "2025-05-12", "categories": []}
{"id": "2507.12050", "title": "IDFace: Face Template Protection for Efficient and Secure Identification", "abstract": "As face recognition systems (FRS) become more widely used, user privacy becomes more important. A key privacy issue in FRS is protecting the user's face template, as the characteristics of the user's face image can be recovered from the template. Although recent advances in cryptographic tools such as homomorphic encryption (HE) have provided opportunities for securing the FRS, HE cannot be used directly with FRS in an efficient plug-and-play manner. In particular, although HE is functionally complete for arbitrary programs, it is basically designed for algebraic operations on encrypted data of predetermined shape, such as a polynomial ring. Thus, a non-tailored combination of HE and the system can yield very inefficient performance, and many previous HE-based face template protection methods are hundreds of times slower than plain systems without protection. In this study, we propose IDFace, a new HE-based secure and efficient face identification method with template protection. IDFace is designed on the basis of two novel techniques for efficient searching on a (homomorphically encrypted) biometric database with an angular metric. The first technique is a template representation transformation that sharply reduces the unit cost for the matching test. The second is a space-efficient encoding that reduces wasted space from the encryption algorithm, thus saving the number of operations on encrypted templates. Through experiments, we show that IDFace can identify a face template from among a database of 1M encrypted templates in 126ms, showing only 2X overhead compared to the identification over plaintexts.", "date": "2025-07-16", "categories": []}
{"id": "2507.12133", "title": "HyDRA: A Hybrid Dual-Mode Network for Closed- and Open-Set RFFI with Optimized VMD", "abstract": "Device recognition is vital for security in wireless communication systems, particularly for applications like access control. Radio Frequency Fingerprint Identification (RFFI) offers a non-cryptographic solution by exploiting hardware-induced signal distortions. This paper proposes HyDRA, a Hybrid Dual-mode RF Architecture that integrates an optimized Variational Mode Decomposition (VMD) with a novel architecture based on the fusion of Convolutional Neural Networks (CNNs), Transformers, and Mamba components, designed to support both closed-set and open-set classification tasks. The optimized VMD enhances preprocessing efficiency and classification accuracy by fixing center frequencies and using closed-form solutions. HyDRA employs the Transformer Dynamic Sequence Encoder (TDSE) for global dependency modeling and the Mamba Linear Flow Encoder (MLFE) for linear-complexity processing, adapting to varying conditions. Evaluation on public datasets demonstrates state-of-the-art (SOTA) accuracy in closed-set scenarios and robust performance in our proposed open-set classification method, effectively identifying unauthorized devices. Deployed on NVIDIA Jetson Xavier NX, HyDRA achieves millisecond-level inference speed with low power consumption, providing a practical solution for real-time wireless authentication in real-world environments.", "date": "2025-07-16", "categories": []}
{"id": "2507.01504", "title": "Following the Clues: Experiments on Person Re-ID using Cross-Modal Intelligence", "abstract": "The collection and release of street-level recordings as Open Data play a vital role in advancing autonomous driving systems and AI research. However, these datasets pose significant privacy risks, particularly for pedestrians, due to the presence of Personally Identifiable Information (PII) that extends beyond biometric traits such as faces. In this paper, we present cRID, a novel cross-modal framework combining Large Vision-Language Models, Graph Attention Networks, and representation learning to detect textual describable clues of PII and enhance person re-identification (Re-ID). Our approach focuses on identifying and leveraging interpretable features, enabling the detection of semantically meaningful PII beyond low-level appearance cues. We conduct a systematic evaluation of PII presence in person image datasets. Our experiments show improved performance in practical cross-dataset Re-ID scenarios, notably from Market-1501 to CUHK03-np (detected), highlighting the framework's practical utility. Code is available at https://github.com/RAufschlaeger/cRID.", "date": "2025-07-02", "categories": []}
{"id": "2507.06858", "title": "Longitudinal Study of Facial Biometrics at the BEZ: Temporal Variance Analysis", "abstract": "This study presents findings from long-term biometric evaluations conducted at the Biometric Evaluation Center (bez). Over the course of two and a half years, our ongoing research with over 400 participants representing diverse ethnicities, genders, and age groups were regularly assessed using a variety of biometric tools and techniques at the controlled testing facilities. Our findings are based on the General Data Protection Regulation-compliant local bez database with more than 238.000 biometric data sets categorized into multiple biometric modalities such as face and finger. We used state-of-the-art face recognition algorithms to analyze long-term comparison scores. Our results show that these scores fluctuate more significantly between individual days than over the entire measurement period. These findings highlight the importance of testing biometric characteristics of the same individuals over a longer period of time in a controlled measurement environment and lays the groundwork for future advancements in biometric data analysis.", "date": "2025-07-09", "categories": []}
{"id": "2507.11119", "title": "Try Harder: Hard Sample Generation and Learning for Clothes-Changing Person Re-ID", "abstract": "Hard samples pose a significant challenge in person re-identification (ReID) tasks, particularly in clothing-changing person Re-ID (CC-ReID). Their inherent ambiguity or similarity, coupled with the lack of explicit definitions, makes them a fundamental bottleneck. These issues not only limit the design of targeted learning strategies but also diminish the model's robustness under clothing or viewpoint changes. In this paper, we propose a novel multimodal-guided Hard Sample Generation and Learning (HSGL) framework, which is the first effort to unify textual and visual modalities to explicitly define, generate, and optimize hard samples within a unified paradigm. HSGL comprises two core components: (1) Dual-Granularity Hard Sample Generation (DGHSG), which leverages multimodal cues to synthesize semantically consistent samples, including both coarse- and fine-grained hard positives and negatives for effectively increasing the hardness and diversity of the training data. (2) Hard Sample Adaptive Learning (HSAL), which introduces a hardness-aware optimization strategy that adjusts feature distances based on textual semantic labels, encouraging the separation of hard positives and drawing hard negatives closer in the embedding space to enhance the model's discriminative capability and robustness to hard samples. Extensive experiments on multiple CC-ReID benchmarks demonstrate the effectiveness of our approach and highlight the potential of multimodal-guided hard sample generation and learning for robust CC-ReID. Notably, HSAL significantly accelerates the convergence of the targeted learning procedure and achieves state-of-the-art performance on both PRCC and LTCC datasets. The code is available at https://github.com/undooo/TryHarder-ACMMM25.", "date": "2025-07-15", "categories": []}
{"id": "2507.11372", "title": "Attributes Shape the Embedding Space of Face Recognition Models", "abstract": "Face Recognition (FR) tasks have made significant progress with the advent of Deep Neural Networks, particularly through margin-based triplet losses that embed facial images into high-dimensional feature spaces. During training, these contrastive losses focus exclusively on identity information as labels. However, we observe a multiscale geometric structure emerging in the embedding space, influenced by interpretable facial (e.g., hair color) and image attributes (e.g., contrast). We propose a geometric approach to describe the dependence or invariance of FR models to these attributes and introduce a physics-inspired alignment metric. We evaluate the proposed metric on controlled, simplified models and widely used FR models fine-tuned with synthetic data for targeted attribute augmentation. Our findings reveal that the models exhibit varying degrees of invariance across different attributes, providing insight into their strengths and weaknesses and enabling deeper interpretability. Code available here: https://github.com/mantonios107/attrs-fr-embs}{https://github.com/mantonios107/attrs-fr-embs", "date": "2025-07-15", "categories": []}
{"id": "2507.08329", "title": "Cross-Domain Identity Representation for Skull to Face Matching with Benchmark DataSet", "abstract": "Craniofacial reconstruction in forensic science is crucial for the identification of the victims of crimes and disasters. The objective is to map a given skull to its corresponding face in a corpus of faces with known identities using recent advancements in computer vision, such as deep learning. In this paper, we presented a framework for the identification of a person given the X-ray image of a skull using convolutional Siamese networks for cross-domain identity representation. Siamese networks are twin networks that share the same architecture and can be trained to discover a feature space where nearby observations that are similar are grouped and dissimilar observations are moved apart. To do this, the network is exposed to two sets of comparable and different data. The Euclidean distance is then minimized between similar pairs and maximized between dissimilar ones. Since getting pairs of skull and face images are difficult, we prepared our own dataset of 40 volunteers whose front and side skull X-ray images and optical face images were collected. Experiments were conducted on the collected cross-domain dataset to train and validate the Siamese networks. The experimental results provide satisfactory results on the identification of a person from the given skull.", "date": "2025-07-11", "categories": []}
{"id": "2507.08520", "title": "Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation for Occluded Person Re-Identification", "abstract": "Occluded person re-identification aims to retrieve holistic images based on occluded ones. Existing methods often rely on aligning visible body parts, applying occlusion augmentation, or complementing missing semantics using holistic images. However, they face challenges in handling diverse occlusion scenarios not seen during training and the issue of feature contamination from holistic images. To address these limitations, we propose Occlusion-Guided Feature Purification Learning via Reinforced Knowledge Distillation (OGFR), which simultaneously mitigates these challenges. OGFR adopts a teacher-student distillation architecture that effectively incorporates diverse occlusion patterns into feature representation while transferring the purified discriminative holistic knowledge from the holistic to the occluded branch through reinforced knowledge distillation. Specifically, an Occlusion-Aware Vision Transformer is designed to leverage learnable occlusion pattern embeddings to explicitly model such diverse occlusion types, thereby guiding occlusion-aware robust feature representation. Moreover, we devise a Feature Erasing and Purification Module within the holistic branch, in which an agent is employed to identify low-quality patch tokens of holistic images that contain noisy negative information via deep reinforcement learning, and substitute these patch tokens with learnable embedding tokens to avoid feature contamination and further excavate identity-related discriminative clues. Afterward, with the assistance of knowledge distillation, the student branch effectively absorbs the purified holistic knowledge to precisely learn robust representation regardless of the interference of occlusions.", "date": "2025-07-11", "categories": []}
{"id": "2507.07230", "title": "Colors See Colors Ignore: Clothes Changing ReID with Color Disentanglement", "abstract": "Clothes-Changing Re-Identification (CC-ReID) aims to recognize individuals across different locations and times, irrespective of clothing. Existing methods often rely on additional models or annotations to learn robust, clothing-invariant features, making them resource-intensive. In contrast, we explore the use of color - specifically foreground and background colors - as a lightweight, annotation-free proxy for mitigating appearance bias in ReID models. We propose Colors See, Colors Ignore (CSCI), an RGB-only method that leverages color information directly from raw images or video frames. CSCI efficiently captures color-related appearance bias ('Color See') while disentangling it from identity-relevant ReID features ('Color Ignore'). To achieve this, we introduce S2A self-attention, a novel self-attention to prevent information leak between color and identity cues within the feature space. Our analysis shows a strong correspondence between learned color embeddings and clothing attributes, validating color as an effective proxy when explicit clothing labels are unavailable. We demonstrate the effectiveness of CSCI on both image and video ReID with extensive experiments on four CC-ReID datasets. We improve the baseline by Top-1 2.9% on LTCC and 5.0% on PRCC for image-based ReID, and 1.0% on CCVID and 2.5% on MeVID for video-based ReID without relying on additional supervision. Our results highlight the potential of color as a cost-effective solution for addressing appearance bias in CC-ReID. Github: https://github.com/ppriyank/ICCV-CSCI-Person-ReID.", "date": "2025-07-09", "categories": []}
{"id": "2507.07262", "title": "DisenQ: Disentangling Q-Former for Activity-Biometrics", "abstract": "In this work, we address activity-biometrics, which involves identifying individuals across diverse set of activities. Unlike traditional person identification, this setting introduces additional challenges as identity cues become entangled with motion dynamics and appearance variations, making biometrics feature learning more complex. While additional visual data like pose and/or silhouette help, they often struggle from extraction inaccuracies. To overcome this, we propose a multimodal language-guided framework that replaces reliance on additional visual data with structured textual supervision. At its core, we introduce \\textbf{DisenQ} (\\textbf{Disen}tangling \\textbf{Q}-Former), a unified querying transformer that disentangles biometrics, motion, and non-biometrics features by leveraging structured language guidance. This ensures identity cues remain independent of appearance and motion variations, preventing misidentifications. We evaluate our approach on three activity-based video benchmarks, achieving state-of-the-art performance. Additionally, we demonstrate strong generalization to complex real-world scenario with competitive performance on a traditional video-based identification benchmark, showing the effectiveness of our framework.", "date": "2025-07-09", "categories": []}
{"id": "2408.05582", "title": "Non-Negative Reduced Biquaternion Matrix Factorization with Applications in Color Face Recognition", "abstract": "Reduced biquaternion (RB), as a four-dimensional algebra highly suitable for representing color pixels, has recently garnered significant attention from numerous scholars. In this paper, for color image processing problems, we introduce a concept of the non-negative RB matrix and then use the multiplication properties of RB to propose a non-negative RB matrix factorization (NRBMF) model. The NRBMF model is introduced to address the challenge of reasonably establishing a non-negative quaternion matrix factorization model, which is primarily hindered by the multiplication properties of traditional quaternions. Furthermore, this paper transforms the problem of solving the NRBMF model into an RB alternating non-negative least squares (RB-ANNLS) problem. Then, by introducing a method to compute the gradient of the real function with RB matrix variables, we solve the RB-ANNLS optimization problem using the RB projected gradient algorithm and conduct a convergence analysis of the algorithm. Finally, we validate the effectiveness and superiority of the proposed NRBMF model in color face recognition.", "date": "2024-08-10", "categories": []}
{"id": "2412.03349", "title": "Fairer Analysis and Demographically Balanced Face Generation for Fairer Face Verification", "abstract": "Face recognition and verification are two computer vision tasks whose performances have advanced with the introduction of deep representations. However, ethical, legal, and technical challenges due to the sensitive nature of face data and biases in real-world training datasets hinder their development. Generative AI addresses privacy by creating fictitious identities, but fairness problems remain. Using the existing DCFace SOTA framework, we introduce a new controlled generation pipeline that improves fairness. Through classical fairness metrics and a proposed in-depth statistical analysis based on logit models and ANOVA, we show that our generation pipeline improves fairness more than other bias mitigation approaches while slightly improving raw performance.", "date": "2024-12-04", "categories": []}
{"id": "2507.02414", "title": "Privacy-preserving Preselection for Face Identification Based on Packing", "abstract": "Face identification systems operating in the ciphertext domain have garnered significant attention due to increasing privacy concerns and the potential recovery of original facial data. However, as the size of ciphertext template libraries grows, the face retrieval process becomes progressively more time-intensive. To address this challenge, we propose a novel and efficient scheme for face retrieval in the ciphertext domain, termed Privacy-Preserving Preselection for Face Identification Based on Packing (PFIP). PFIP incorporates an innovative preselection mechanism to reduce computational overhead and a packing module to enhance the flexibility of biometric systems during the enrollment stage. Extensive experiments conducted on the LFW and CASIA datasets demonstrate that PFIP preserves the accuracy of the original face recognition model, achieving a 100% hit rate while retrieving 1,000 ciphertext face templates within 300 milliseconds. Compared to existing approaches, PFIP achieves a nearly 50x improvement in retrieval efficiency.", "date": "2025-07-03", "categories": []}
{"id": "2507.00506", "title": "SCING:Towards More Efficient and Robust Person Re-Identification through Selective Cross-modal Prompt Tuning", "abstract": "Recent advancements in adapting vision-language pre-training models like CLIP for person re-identification (ReID) tasks often rely on complex adapter design or modality-specific tuning while neglecting cross-modal interaction, leading to high computational costs or suboptimal alignment. To address these limitations, we propose a simple yet effective framework named Selective Cross-modal Prompt Tuning (SCING) that enhances cross-modal alignment and robustness against real-world perturbations. Our method introduces two key innovations: Firstly, we proposed Selective Visual Prompt Fusion (SVIP), a lightweight module that dynamically injects discriminative visual features into text prompts via a cross-modal gating mechanism. Moreover, the proposed Perturbation-Driven Consistency Alignment (PDCA) is a dual-path training strategy that enforces invariant feature alignment under random image perturbations by regularizing consistency between original and augmented cross-modal embeddings. Extensive experiments are conducted on several popular benchmarks covering Market1501, DukeMTMC-ReID, Occluded-Duke, Occluded-REID, and P-DukeMTMC, which demonstrate the impressive performance of the proposed method. Notably, our framework eliminates heavy adapters while maintaining efficient inference, achieving an optimal trade-off between performance and computational overhead. The code will be released upon acceptance.", "date": "2025-07-01", "categories": []}
{"id": "2409.20003", "title": "Multibiometrics Using a Single Face Image", "abstract": "Multibiometrics, which uses multiple biometric traits to improve recognition performance instead of using only one biometric trait to authenticate individuals, has been investigated. Previous studies have combined individually acquired biometric traits or have not fully considered the convenience of the system. Focusing on a single face image, we propose a novel multibiometric method that combines five biometric traits, i.e., face, iris, periocular, nose, eyebrow, that can be extracted from a single face image. The proposed method does not sacrifice the convenience of biometrics since only a single face image is used as input. Through a variety of experiments using the CASIA Iris Distance database, we demonstrate the effectiveness of the proposed multibiometrics method.", "date": "2024-09-30", "categories": []}
{"id": "2503.03501", "title": "CarGait: Cross-Attention based Re-ranking for Gait recognition", "abstract": "Gait recognition is a computer vision task that identifies individuals based on their walking patterns. Gait recognition performance is commonly evaluated by ranking a gallery of candidates and measuring the accuracy at the top Rank-$K$. Existing models are typically single-staged, i.e. searching for the probe's nearest neighbors in a gallery using a single global feature representation. Although these models typically excel at retrieving the correct identity within the top-$K$ predictions, they struggle when hard negatives appear in the top short-list, leading to relatively low performance at the highest ranks (e.g., Rank-1). In this paper, we introduce CarGait, a Cross-Attention Re-ranking method for gait recognition, that involves re-ordering the top-$K$ list leveraging the fine-grained correlations between pairs of gait sequences through cross-attention between gait strips. This re-ranking scheme can be adapted to existing single-stage models to enhance their final results. We demonstrate the capabilities of CarGait by extensive experiments on three common gait datasets, Gait3D, GREW, and OU-MVLP, and seven different gait models, showing consistent improvements in Rank-1,5 accuracy, superior results over existing re-ranking methods, and strong baselines.", "date": "2025-03-05", "categories": []}
{"id": "2506.22843", "title": "AG-VPReID 2025: Aerial-Ground Video-based Person Re-identification Challenge Results", "abstract": "Person re-identification (ReID) across aerial and ground vantage points has become crucial for large-scale surveillance and public safety applications. Although significant progress has been made in ground-only scenarios, bridging the aerial-ground domain gap remains a formidable challenge due to extreme viewpoint differences, scale variations, and occlusions. Building upon the achievements of the AG-ReID 2023 Challenge, this paper introduces the AG-VPReID 2025 Challenge - the first large-scale video-based competition focused on high-altitude (80-120m) aerial-ground ReID. Constructed on the new AG-VPReID dataset with 3,027 identities, over 13,500 tracklets, and approximately 3.7 million frames captured from UAVs, CCTV, and wearable cameras, the challenge featured four international teams. These teams developed solutions ranging from multi-stream architectures to transformer-based temporal reasoning and physics-informed modeling. The leading approach, X-TFCLIP from UAM, attained 72.28% Rank-1 accuracy in the aerial-to-ground ReID setting and 70.77% in the ground-to-aerial ReID setting, surpassing existing baselines while highlighting the dataset's complexity. For additional details, please refer to the official website at https://agvpreid25.github.io.", "date": "2025-06-28", "categories": []}
{"id": "2506.22347", "title": "Closing the Performance Gap in Biometric Cryptosystems: A Deeper Analysis on Unlinkable Fuzzy Vaults", "abstract": "This paper analyses and addresses the performance gap in the fuzzy vault-based \\ac{BCS}. We identify unstable error correction capabilities, which are caused by variable feature set sizes and their influence on similarity thresholds, as a key source of performance degradation. This issue is further compounded by information loss introduced through feature type transformations. To address both problems, we propose a novel feature quantization method based on \\it{equal frequent intervals}. This method guarantees fixed feature set sizes and supports training-free adaptation to any number of intervals. The proposed approach significantly reduces the performance gap introduced by template protection. Additionally, it integrates seamlessly with existing systems to minimize the negative effects of feature transformation. Experiments on state-of-the-art face, fingerprint, and iris recognition systems confirm that only minimal performance degradation remains, demonstrating the effectiveness of the method across major biometric modalities.", "date": "2025-06-27", "categories": []}
{"id": "2503.09749", "title": "A Siamese Network to Detect If Two Iris Images Are Monozygotic", "abstract": "This study presents the first automated classifier designed to determine whether a pair of iris images originates from monozygotic individuals, addressing a previously untackled problem in biometric recognition. In Daugman-style iris recognition, the textures of the left and right irises of the same person are traditionally considered as being as different as the irises of two unrelated persons. However, previous research indicates that humans can detect that two iris images are from different eyes of the same person, or eyes of monozygotic twins, with an accuracy of about 80%. In this work, we employ a Siamese network architecture and contrastive learning to categorize a pair of iris images as coming from monozygotic or non-monozygotic irises. This could potentially be applied, for example, as a fast, noninvasive test to determine if twins are monozygotic or non-monozygotic. We construct a dataset comprising both synthetic monozygotic pairs (images of different irises of the same individual) and natural monozygotic pairs (images of different images from persons who are identical twins), in addition to non-monozygotic pairs from unrelated individuals, ensuring a comprehensive evaluation of the model's capabilities. To gain deeper insights into the learned representations, we train and analyze three variants of the model using (1) the original input images, (2) iris-only images (masking everything but the iris region), and (3) non-iris-only images (masking the iris region). This comparison reveals that both iris texture and surrounding ocular structure contain information useful for the model to classify the image pairs as monozygotic or non-monozygotic. Our approach achieves accuracy levels using the full iris image that exceed those previously reported for human classification of monozygotic iris pairs.", "date": "2025-03-12", "categories": []}
{"id": "2506.20312", "title": "On the Burstiness of Faces in Set", "abstract": "Burstiness, a phenomenon observed in text and image retrieval, refers to that particular elements appear more times in a set than a statistically independent model assumes. We argue that in the context of set-based face recognition (SFR), burstiness exists widely and degrades the performance in two aspects: Firstly, the bursty faces, where faces with particular attributes %exist frequently in a face set, dominate the training instances and dominate the training face sets and lead to poor generalization ability to unconstrained scenarios. Secondly, the bursty faces %dominating the evaluation sets interfere with the similarity comparison in set verification and identification when evaluation. To detect the bursty faces in a set, we propose three strategies based on Quickshift++, feature self-similarity, and generalized max-pooling (GMP). We apply the burst detection results on training and evaluation stages to enhance the sampling ratios or contributions of the infrequent faces. When evaluation, we additionally propose the quality-aware GMP that enables awareness of the face quality and robustness to the low-quality faces for the original GMP. We give illustrations and extensive experiments on the SFR benchmarks to demonstrate that burstiness is widespread and suppressing burstiness considerably improves the recognition performance.", "date": "2025-06-25", "categories": []}
{"id": "2501.06533", "title": "DivTrackee versus DynTracker: Promoting Diversity in Anti-Facial Recognition against Dynamic FR Strategy", "abstract": "The widespread adoption of facial recognition (FR) models raises serious concerns about their potential misuse, motivating the development of anti-facial recognition (AFR) to protect user facial privacy. In this paper, we argue that the static FR strategy, predominantly adopted in prior literature for evaluating AFR efficacy, cannot faithfully characterize the actual capabilities of determined trackers who aim to track a specific target identity. In particular, we introduce DynTracker, a dynamic FR strategy where the model's gallery database is iteratively updated with newly recognized target identity images. Surprisingly, such a simple approach renders all the existing AFR protections ineffective. To mitigate the privacy threats posed by DynTracker, we advocate for explicitly promoting diversity in the AFR-protected images. We hypothesize that the lack of diversity is the primary cause of the failure of existing AFR methods. Specifically, we develop DivTrackee, a novel method for crafting diverse AFR protections that builds upon a text-guided image generation framework and diversity-promoting adversarial losses. Through comprehensive experiments on various image benchmarks and feature extractors, we demonstrate DynTracker's strength in breaking existing AFR methods and the superiority of DivTrackee in preventing user facial images from being identified by dynamic FR strategies. We believe our work can act as an important initial step towards developing more effective AFR methods for protecting user facial privacy against determined trackers.", "date": "2025-01-11", "categories": []}
{"id": "2502.20803", "title": "Two-Stream Spatial-Temporal Transformer Framework for Person Identification via Natural Conversational Keypoints", "abstract": "In the age of AI-driven generative technologies, traditional biometric recognition systems face unprecedented challenges, particularly from sophisticated deepfake and face reenactment techniques. In this study, we propose a Two-Stream Spatial-Temporal Transformer Framework for person identification using upper body keypoints visible during online conversations, which we term conversational keypoints. Our framework processes both spatial relationships between keypoints and their temporal evolution through two specialized branches: a Spatial Transformer (STR) that learns distinctive structural patterns in keypoint configurations, and a Temporal Transformer (TTR) that captures sequential motion patterns. Using the state-of-the-art Sapiens pose estimator, we extract 133 keypoints (based on COCO-WholeBody format) representing facial features, head pose, and hand positions. The framework was evaluated on a dataset of 114 individuals engaged in natural conversations, achieving recognition accuracies of 80.12% for the spatial stream, 63.61% for the temporal stream. We then explored two fusion strategies: a shared loss function approach achieving 82.22% accuracy, and a feature-level fusion method that concatenates feature maps from both streams, significantly improving performance to 94.86%. By jointly modeling both static anatomical relationships and dynamic movement patterns, our approach learns comprehensive identity signatures that are more robust to spoofing than traditional appearance-based methods.", "date": "2025-02-28", "categories": []}
{"id": "2506.19533", "title": "Identifying Physically Realizable Triggers for Backdoored Face Recognition Networks", "abstract": "Backdoor attacks embed a hidden functionality into deep neural networks, causing the network to display anomalous behavior when activated by a predetermined pattern in the input Trigger, while behaving well otherwise on public test data. Recent works have shown that backdoored face recognition (FR) systems can respond to natural-looking triggers like a particular pair of sunglasses. Such attacks pose a serious threat to the applicability of FR systems in high-security applications. We propose a novel technique to (1) detect whether an FR network is compromised with a natural, physically realizable trigger, and (2) identify such triggers given a compromised network. We demonstrate the effectiveness of our methods with a compromised FR network, where we are able to identify the trigger (e.g., green sunglasses or red hat) with a top-5 accuracy of 74%, whereas a naive brute force baseline achieves 56% accuracy.", "date": "2025-06-24", "categories": []}
{"id": "2506.18731", "title": "Deep CNN Face Matchers Inherently Support Revocable Biometric Templates", "abstract": "One common critique of biometric authentication is that if an individual's biometric is compromised, then the individual has no recourse. The concept of revocable biometrics was developed to address this concern. A biometric scheme is revocable if an individual can have their current enrollment in the scheme revoked, so that the compromised biometric template becomes worthless, and the individual can re-enroll with a new template that has similar recognition power. We show that modern deep CNN face matchers inherently allow for a robust revocable biometric scheme. For a given state-of-the-art deep CNN backbone and training set, it is possible to generate an unlimited number of distinct face matcher models that have both (1) equivalent recognition power, and (2) strongly incompatible biometric templates. The equivalent recognition power extends to the point of generating impostor and genuine distributions that have the same shape and placement on the similarity dimension, meaning that the models can share a similarity threshold for a 1-in-10,000 false match rate. The biometric templates from different model instances are so strongly incompatible that the cross-instance similarity score for images of the same person is typically lower than the same-instance similarity score for images of different persons. That is, a stolen biometric template that is revoked is of less value in attempting to match the re-enrolled identity than the average impostor template. We also explore the feasibility of using a Vision Transformer (ViT) backbone-based face matcher in the revocable biometric system proposed in this work and demonstrate that it is less suitable compared to typical ResNet-based deep CNN backbones.", "date": "2025-06-23", "categories": []}
{"id": "2506.15852", "title": "Assessing the impact of Binarization for Writer Identification in Greek Papyrus", "abstract": "This paper tackles the task of writer identification for Greek papyri. A common preprocessing step in writer identification pipelines is image binarization, which prevents the model from learning background features. This is challenging in historical documents, in our case Greek papyri, as background is often non-uniform, fragmented, and discolored with visible fiber structures. We compare traditional binarization methods to state-of-the-art Deep Learning (DL) models, evaluating the impact of binarization quality on subsequent writer identification performance. DL models are trained with and without a custom data augmentation technique, as well as different model selection criteria are applied. The performance of these binarization methods, is then systematically evaluated on the DIBCO 2019 dataset. The impact of binarization on writer identification is subsequently evaluated using a state-of-the-art approach for writer identification. The results of this analysis highlight the influence of data augmentation for DL methods. Furthermore, findings indicate a strong correlation between binarization effectiveness on papyri documents of DIBCO 2019 and downstream writer identification performance.", "date": "2025-06-18", "categories": []}
{"id": "2504.04818", "title": "SUEDE:Shared Unified Experts for Physical-Digital Face Attack Detection Enhancement", "abstract": "Face recognition systems are vulnerable to physical attacks (e.g., printed photos) and digital threats (e.g., DeepFake), which are currently being studied as independent visual tasks, such as Face Anti-Spoofing and Forgery Detection. The inherent differences among various attack types present significant challenges in identifying a common feature space, making it difficult to develop a unified framework for detecting data from both attack modalities simultaneously. Inspired by the efficacy of Mixture-of-Experts (MoE) in learning across diverse domains, we explore utilizing multiple experts to learn the distinct features of various attack types. However, the feature distributions of physical and digital attacks overlap and differ. This suggests that relying solely on distinct experts to learn the unique features of each attack type may overlook shared knowledge between them. To address these issues, we propose SUEDE, the Shared Unified Experts for Physical-Digital Face Attack Detection Enhancement. SUEDE combines a shared expert (always activated) to capture common features for both attack types and multiple routed experts (selectively activated) for specific attack types. Further, we integrate CLIP as the base network to ensure the shared expert benefits from prior visual knowledge and align visual-text representations in a unified space. Extensive results demonstrate SUEDE achieves superior performance compared to state-of-the-art unified detection methods.", "date": "2025-04-07", "categories": []}
{"id": "2506.14511", "title": "MOL: Joint Estimation of Micro-Expression, Optical Flow, and Landmark via Transformer-Graph-Style Convolution", "abstract": "Facial micro-expression recognition (MER) is a challenging problem, due to transient and subtle micro-expression (ME) actions. Most existing methods depend on hand-crafted features, key frames like onset, apex, and offset frames, or deep networks limited by small-scale and low-diversity datasets. In this paper, we propose an end-to-end micro-action-aware deep learning framework with advantages from transformer, graph convolution, and vanilla convolution. In particular, we propose a novel F5C block composed of fully-connected convolution and channel correspondence convolution to directly extract local-global features from a sequence of raw frames, without the prior knowledge of key frames. The transformer-style fully-connected convolution is proposed to extract local features while maintaining global receptive fields, and the graph-style channel correspondence convolution is introduced to model the correlations among feature patterns. Moreover, MER, optical flow estimation, and facial landmark detection are jointly trained by sharing the local-global features. The two latter tasks contribute to capturing facial subtle action information for MER, which can alleviate the impact of insufficient training data. Extensive experiments demonstrate that our framework (i) outperforms the state-of-the-art MER methods on CASME II, SAMM, and SMIC benchmarks, (ii) works well for optical flow estimation and facial landmark detection, and (iii) can capture facial subtle muscle actions in local regions associated with MEs. The code is available at https://github.com/CYF-cuber/MOL.", "date": "2025-06-17", "categories": []}
{"id": "2405.18302", "title": "Deep Network Pruning: A Comparative Study on CNNs in Face Recognition", "abstract": "The widespread use of mobile devices for all kinds of transactions makes necessary reliable and real-time identity authentication, leading to the adoption of face recognition (FR) via the cameras embedded in such devices. Progress of deep Convolutional Neural Networks (CNNs) has provided substantial advances in FR. Nonetheless, the size of state-of-the-art architectures is unsuitable for mobile deployment, since they often encompass hundreds of megabytes and millions of parameters. We address this by studying methods for deep network compression applied to FR. In particular, we apply network pruning based on Taylor scores, where less important filters are removed iteratively. The method is tested on three networks based on the small SqueezeNet (1.24M parameters) and the popular MobileNetv2 (3.5M) and ResNet50 (23.5M) architectures. These have been selected to showcase the method on CNNs with different complexities and sizes. We observe that a substantial percentage of filters can be removed with minimal performance loss. Also, filters with the highest amount of output channels tend to be removed first, suggesting that high-dimensional spaces within popular CNNs are over-dimensioned.", "date": "2024-05-28", "categories": []}
{"id": "2506.12413", "title": "Domain Generalization for Person Re-identification: A Survey Towards Domain-Agnostic Person Matching", "abstract": "Person Re-identification (ReID) aims to retrieve images of the same individual captured across non-overlapping camera views, making it a critical component of intelligent surveillance systems. Traditional ReID methods assume that the training and test domains share similar characteristics and primarily focus on learning discriminative features within a given domain. However, they often fail to generalize to unseen domains due to domain shifts caused by variations in viewpoint, background, and lighting conditions. To address this issue, Domain-Adaptive ReID (DA-ReID) methods have been proposed. These approaches incorporate unlabeled target domain data during training and improve performance by aligning feature distributions between source and target domains. Domain-Generalizable ReID (DG-ReID) tackles a more realistic and challenging setting by aiming to learn domain-invariant features without relying on any target domain data. Recent methods have explored various strategies to enhance generalization across diverse environments, but the field remains relatively underexplored. In this paper, we present a comprehensive survey of DG-ReID. We first review the architectural components of DG-ReID including the overall setting, commonly used backbone networks and multi-source input configurations. Then, we categorize and analyze domain generalization modules that explicitly aim to learn domain-invariant and identity-discriminative representations. To examine the broader applicability of these techniques, we further conduct a case study on a related task that also involves distribution shifts. Finally, we discuss recent trends, open challenges, and promising directions for future research in DG-ReID. To the best of our knowledge, this is the first systematic survey dedicated to DG-ReID.", "date": "2025-06-14", "categories": []}
{"id": "2506.13039", "title": "Evolution of ReID: From Early Methods to LLM Integration", "abstract": "Person re-identification (ReID) has evolved from handcrafted feature-based methods to deep learning approaches and, more recently, to models incorporating large language models (LLMs). Early methods struggled with variations in lighting, pose, and viewpoint, but deep learning addressed these issues by learning robust visual features. Building on this, LLMs now enable ReID systems to integrate semantic and contextual information through natural language. This survey traces that full evolution and offers one of the first comprehensive reviews of ReID approaches that leverage LLMs, where textual descriptions are used as privileged information to improve visual matching. A key contribution is the use of dynamic, identity-specific prompts generated by GPT-4o, which enhance the alignment between images and text in vision-language ReID systems. Experimental results show that these descriptions improve accuracy, especially in complex or ambiguous cases. To support further research, we release a large set of GPT-4o-generated descriptions for standard ReID datasets. By bridging computer vision and natural language processing, this survey offers a unified perspective on the field's development and outlines key future directions such as better prompt design, cross-modal transfer learning, and real-world adaptability.", "date": "2025-06-16", "categories": []}
{"id": "2506.09777", "title": "Inverting Black-Box Face Recognition Systems via Zero-Order Optimization in Eigenface Space", "abstract": "Reconstructing facial images from black-box recognition models poses a significant privacy threat. While many methods require access to embeddings, we address the more challenging scenario of model inversion using only similarity scores. This paper introduces DarkerBB, a novel approach that reconstructs color faces by performing zero-order optimization within a PCA-derived eigenface space. Despite this highly limited information, experiments on LFW, AgeDB-30, and CFP-FP benchmarks demonstrate that DarkerBB achieves state-of-the-art verification accuracies in the similarity-only setting, with competitive query efficiency.", "date": "2025-06-11", "categories": []}
{"id": "2506.06759", "title": "LitMAS: A Lightweight and Generalized Multi-Modal Anti-Spoofing Framework for Biometric Security", "abstract": "Biometric authentication systems are increasingly being deployed in critical applications, but they remain susceptible to spoofing. Since most of the research efforts focus on modality-specific anti-spoofing techniques, building a unified, resource-efficient solution across multiple biometric modalities remains a challenge. To address this, we propose LitMAS, a $\\textbf{Li}$gh$\\textbf{t}$ weight and generalizable $\\textbf{M}$ulti-modal $\\textbf{A}$nti-$\\textbf{S}$poofing framework designed to detect spoofing attacks in speech, face, iris, and fingerprint-based biometric systems. At the core of LitMAS is a Modality-Aligned Concentration Loss, which enhances inter-class separability while preserving cross-modal consistency and enabling robust spoof detection across diverse biometric traits. With just 6M parameters, LitMAS surpasses state-of-the-art methods by $1.36\\%$ in average EER across seven datasets, demonstrating high efficiency, strong generalizability, and suitability for edge deployment. Code and trained models are available at https://github.com/IAB-IITJ/LitMAS.", "date": "2025-06-07", "categories": []}
{"id": "2506.05383", "title": "Can Vision Transformers with ResNet's Global Features Fairly Authenticate Demographic Faces?", "abstract": "Biometric face authentication is crucial in computer vision, but ensuring fairness and generalization across demographic groups remains a big challenge. Therefore, we investigated whether Vision Transformer (ViT) and ResNet, leveraging pre-trained global features, can fairly authenticate different demographic faces while relying minimally on local features. In this investigation, we used three pre-trained state-of-the-art (SOTA) ViT foundation models from Facebook, Google, and Microsoft for global features as well as ResNet-18. We concatenated the features from ViT and ResNet, passed them through two fully connected layers, and trained on customized face image datasets to capture the local features. Then, we designed a novel few-shot prototype network with backbone features embedding. We also developed new demographic face image support and query datasets for this empirical study. The network's testing was conducted on this dataset in one-shot, three-shot, and five-shot scenarios to assess how performance improves as the size of the support set increases. We observed results across datasets with varying races/ethnicities, genders, and age groups. The Microsoft Swin Transformer backbone performed better among the three SOTA ViT for this task. The code and data are available at: https://github.com/Sufianlab/FairVitBio.", "date": "2025-06-03", "categories": []}
{"id": "2409.00372", "title": "First Competition on Presentation Attack Detection on ID Card", "abstract": "This paper summarises the Competition on Presentation Attack Detection on ID Cards (PAD-IDCard) held at the 2024 International Joint Conference on Biometrics (IJCB2024). The competition attracted a total of ten registered teams, both from academia and industry. In the end, the participating teams submitted five valid submissions, with eight models to be evaluated by the organisers. The competition presented an independent assessment of current state-of-the-art algorithms. Today, no independent evaluation on cross-dataset is available; therefore, this work determined the state-of-the-art on ID cards. To reach this goal, a sequestered test set and baseline algorithms were used to evaluate and compare all the proposals. The sequestered test dataset contains ID cards from four different countries. In summary, a team that chose to be \"Anonymous\" reached the best average ranking results of 74.80%, followed very closely by the \"IDVC\" team with 77.65%.", "date": "2024-08-31", "categories": []}
{"id": "2506.05263", "title": "Can Foundation Models Generalise the Presentation Attack Detection Capabilities on ID Cards?", "abstract": "Nowadays, one of the main challenges in presentation attack detection (PAD) on ID cards is obtaining generalisation capabilities for a diversity of countries that are issuing ID cards. Most PAD systems are trained on one, two, or three ID documents because of privacy protection concerns. As a result, they do not obtain competitive results for commercial purposes when tested in an unknown new ID card country. In this scenario, Foundation Models (FM) trained on huge datasets can help to improve generalisation capabilities. This work intends to improve and benchmark the capabilities of FM and how to use them to adapt the generalisation on PAD of ID Documents. Different test protocols were used, considering zero-shot and fine-tuning and two different ID card datasets. One private dataset based on Chilean IDs and one open-set based on three ID countries: Finland, Spain, and Slovakia. Our findings indicate that bona fide images are the key to generalisation.", "date": "2025-06-05", "categories": []}
{"id": "2506.03635", "title": "FingerVeinSyn-5M: A Million-Scale Dataset and Benchmark for Finger Vein Recognition", "abstract": "A major challenge in finger vein recognition is the lack of large-scale public datasets. Existing datasets contain few identities and limited samples per finger, restricting the advancement of deep learning-based methods. To address this, we introduce FVeinSyn, a synthetic generator capable of producing diverse finger vein patterns with rich intra-class variations. Using FVeinSyn, we created FingerVeinSyn-5M -- the largest available finger vein dataset -- containing 5 million samples from 50,000 unique fingers, each with 100 variations including shift, rotation, scale, roll, varying exposure levels, skin scattering blur, optical blur, and motion blur. FingerVeinSyn-5M is also the first to offer fully annotated finger vein images, supporting deep learning applications in this field. Models pretrained on FingerVeinSyn-5M and fine-tuned with minimal real data achieve an average 53.91\\% performance gain across multiple benchmarks. The dataset is publicly available at: https://github.com/EvanWang98/FingerVeinSyn-5M.", "date": "2025-06-04", "categories": []}
{"id": "2506.02062", "title": "Predicting Blood Type: Assessing Model Performance with ROC Analysis", "abstract": "Introduction: Personal identification is a critical aspect of forensic sciences, security, and healthcare. While conventional biometrics systems such as DNA profiling and iris scanning offer high accuracy, they are time-consuming and costly. Objectives: This study investigates the relationship between fingerprint patterns and ABO blood group classification to explore potential correlations between these two traits. Methods: The study analyzed 200 individuals, categorizing their fingerprints into three types: loops, whorls, and arches. Blood group classification was also recorded. Statistical analysis, including chi-square and Pearson correlation tests, was used to assess associations between fingerprint patterns and blood groups. Results: Loops were the most common fingerprint pattern, while blood group O+ was the most prevalent among the participants. Statistical analysis revealed no significant correlation between fingerprint patterns and blood groups (p > 0.05), suggesting that these traits are independent. Conclusions: Although the study showed limited correlation between fingerprint patterns and ABO blood groups, it highlights the importance of future research using larger and more diverse populations, incorporating machine learning approaches, and integrating multiple biometric signals. This study contributes to forensic science by emphasizing the need for rigorous protocols and comprehensive investigations in personal identification.", "date": "2025-06-01", "categories": []}
{"id": "2506.02439", "title": "Video-Level Language-Driven Video-Based Visible-Infrared Person Re-Identification", "abstract": "Video-based Visible-Infrared Person Re-Identification (VVI-ReID) aims to match pedestrian sequences across modalities by extracting modality-invariant sequence-level features. As a high-level semantic representation, language provides a consistent description of pedestrian characteristics in both infrared and visible modalities. Leveraging the Contrastive Language-Image Pre-training (CLIP) model to generate video-level language prompts and guide the learning of modality-invariant sequence-level features is theoretically feasible. However, the challenge of generating and utilizing modality-shared video-level language prompts to address modality gaps remains a critical problem. To address this problem, we propose a simple yet powerful framework, video-level language-driven VVI-ReID (VLD), which consists of two core modules: invariant-modality language prompting (IMLP) and spatial-temporal prompting (STP). IMLP employs a joint fine-tuning strategy for the visual encoder and the prompt learner to effectively generate modality-shared text prompts and align them with visual features from different modalities in CLIP's multimodal space, thereby mitigating modality differences. Additionally, STP models spatiotemporal information through two submodules, the spatial-temporal hub (STH) and spatial-temporal aggregation (STA), which further enhance IMLP by incorporating spatiotemporal information into text prompts. The STH aggregates and diffuses spatiotemporal information into the [CLS] token of each frame across the vision transformer (ViT) layers, whereas STA introduces dedicated identity-level loss and specialized multihead attention to ensure that the STH focuses on identity-relevant spatiotemporal feature aggregation. The VLD framework achieves state-of-the-art results on two VVI-ReID benchmarks. The code will be released at https://github.com/Visuang/VLD.", "date": "2025-06-03", "categories": []}
{"id": "2506.02626", "title": "Synthetic Iris Image Databases and Identity Leakage: Risks and Mitigation Strategies", "abstract": "This paper presents a comprehensive overview of iris image synthesis methods, which can alleviate the issues associated with gathering large, diverse datasets of biometric data from living individuals, which are considered pivotal for biometric methods development. These methods for synthesizing iris data range from traditional, hand crafted image processing-based techniques, through various iterations of GAN-based image generators, variational autoencoders (VAEs), as well as diffusion models. The potential and fidelity in iris image generation of each method is discussed and examples of inferred predictions are provided. Furthermore, the risks of individual biometric features leakage from the training sets are considered, together with possible strategies for preventing them, which have to be implemented should these generative methods be considered a valid replacement of real-world biometric datasets.", "date": "2025-06-03", "categories": []}
{"id": "2405.09138", "title": "OpenGait: A Comprehensive Benchmark Study for Gait Recognition towards Better Practicality", "abstract": "Gait recognition, a rapidly advancing vision technology for person identification from a distance, has made significant strides in indoor settings. However, evidence suggests that existing methods often yield unsatisfactory results when applied to newly released real-world gait datasets. Furthermore, conclusions drawn from indoor gait datasets may not easily generalize to outdoor ones. Therefore, the primary goal of this paper is to present a comprehensive benchmark study aimed at improving practicality rather than solely focusing on enhancing performance. To this end, we developed OpenGait, a flexible and efficient gait recognition platform. Using OpenGait, we conducted in-depth ablation experiments to revisit recent developments in gait recognition. Surprisingly, we detected some imperfect parts of some prior methods and thereby uncovered several critical yet previously neglected insights. These findings led us to develop three structurally simple yet empirically powerful and practically robust baseline models: DeepGaitV2, SkeletonGait, and SkeletonGait++, which represent the appearance-based, model-based, and multi-modal methodologies for gait pattern description, respectively. In addition to achieving state-of-the-art performance, our careful exploration provides new perspectives on the modeling experience of deep gait models and the representational capacity of typical gait modalities. In the end, we discuss the key trends and challenges in current gait recognition, aiming to inspire further advancements towards better practicality. The code is available at https://github.com/ShiqiYu/OpenGait.", "date": "2024-05-15", "categories": []}
{"id": "2502.19958", "title": "ChatReID: Open-ended Interactive Person Retrieval via Hierarchical Progressive Tuning for Vision Language Models", "abstract": "Person re-identification (Re-ID) is a crucial task in computer vision, aiming to recognize individuals across non-overlapping camera views. While recent advanced vision-language models (VLMs) excel in logical reasoning and multi-task generalization, their applications in Re-ID tasks remain limited. They either struggle to perform accurate matching based on identity-relevant features or assist image-dominated branches as auxiliary semantics. In this paper, we propose a novel framework ChatReID, that shifts the focus towards a text-side-dominated retrieval paradigm, enabling flexible and interactive re-identification. To integrate the reasoning abilities of language models into Re-ID pipelines, We first present a large-scale instruction dataset, which contains more than 8 million prompts to promote the model fine-tuning. Next. we introduce a hierarchical progressive tuning strategy, which endows Re-ID ability through three stages of tuning, i.e., from person attribute understanding to fine-grained image retrieval and to multi-modal task reasoning. Extensive experiments across ten popular benchmarks demonstrate that ChatReID outperforms existing methods, achieving state-of-the-art performance in all Re-ID tasks. More experiments demonstrate that ChatReID not only has the ability to recognize fine-grained details but also to integrate them into a coherent reasoning process.", "date": "2025-02-27", "categories": []}
{"id": "2506.01069", "title": "Revolutionizing Blood Banks: AI-Driven Fingerprint-Blood Group Correlation for Enhanced Safety", "abstract": "Identification of a person is central in forensic science, security, and healthcare. Methods such as iris scanning and genomic profiling are more accurate but expensive, time-consuming, and more difficult to implement. This study focuses on the relationship between the fingerprint patterns and the ABO blood group as a biometric identification tool. A total of 200 subjects were included in the study, and fingerprint types (loops, whorls, and arches) and blood groups were compared. Associations were evaluated with statistical tests, including chi-square and Pearson correlation. The study found that the loops were the most common fingerprint pattern and the O+ blood group was the most prevalent. Even though there was some associative pattern, there was no statistically significant difference in the fingerprint patterns of different blood groups. Overall, the results indicate that blood group data do not significantly improve personal identification when used in conjunction with fingerprinting. Although the study shows weak correlation, it may emphasize the efforts of multi-modal based biometric systems in enhancing the current biometric systems. Future studies may focus on larger and more diverse samples, and possibly machine learning and additional biometrics to improve identification methods. This study addresses an element of the ever-changing nature of the fields of forensic science and biometric identification, highlighting the importance of resilient analytical methods for personal identification.", "date": "2025-06-01", "categories": []}
{"id": "2505.24214", "title": "Benchmarking Foundation Models for Zero-Shot Biometric Tasks", "abstract": "The advent of foundation models, particularly Vision-Language Models (VLMs) and Multi-modal Large Language Models (MLLMs), has redefined the frontiers of artificial intelligence, enabling remarkable generalization across diverse tasks with minimal or no supervision. Yet, their potential in biometric recognition and analysis remains relatively underexplored. In this work, we introduce a comprehensive benchmark that evaluates the zero-shot and few-shot performance of state-of-the-art publicly available VLMs and MLLMs across six biometric tasks spanning the face and iris modalities: face verification, soft biometric attribute prediction (gender and race), iris recognition, presentation attack detection (PAD), and face manipulation detection (morphs and deepfakes). A total of 41 VLMs were used in this evaluation. Experiments show that embeddings from these foundation models can be used for diverse biometric tasks with varying degrees of success. For example, in the case of face verification, a True Match Rate (TMR) of 96.77 percent was obtained at a False Match Rate (FMR) of 1 percent on the Labeled Face in the Wild (LFW) dataset, without any fine-tuning. In the case of iris recognition, the TMR at 1 percent FMR on the IITD-R-Full dataset was 97.55 percent without any fine-tuning. Further, we show that applying a simple classifier head to these embeddings can help perform DeepFake detection for faces, Presentation Attack Detection (PAD) for irides, and extract soft biometric attributes like gender and ethnicity from faces with reasonably high accuracy. This work reiterates the potential of pretrained models in achieving the long-term vision of Artificial General Intelligence.", "date": "2025-05-30", "categories": []}
{"id": "2505.20540", "title": "Causality and \"In-the-Wild\" Video-Based Person Re-ID: A Survey", "abstract": "Video-based person re-identification (Re-ID) remains brittle in real-world deployments despite impressive benchmark performance. Most existing models rely on superficial correlations such as clothing, background, or lighting that fail to generalize across domains, viewpoints, and temporal variations. This survey examines the emerging role of causal reasoning as a principled alternative to traditional correlation-based approaches in video-based Re-ID. We provide a structured and critical analysis of methods that leverage structural causal models, interventions, and counterfactual reasoning to isolate identity-specific features from confounding factors. The survey is organized around a novel taxonomy of causal Re-ID methods that spans generative disentanglement, domain-invariant modeling, and causal transformers. We review current evaluation metrics and introduce causal-specific robustness measures. In addition, we assess practical challenges of scalability, fairness, interpretability, and privacy that must be addressed for real-world adoption. Finally, we identify open problems and outline future research directions that integrate causal modeling with efficient architectures and self-supervised learning. This survey aims to establish a coherent foundation for causal video-based person Re-ID and to catalyze the next phase of research in this rapidly evolving domain.", "date": "2025-05-26", "categories": []}
{"id": "2504.10174", "title": "LLaVA-ReID: Selective Multi-image Questioner for Interactive Person Re-Identification", "abstract": "Traditional text-based person ReID assumes that person descriptions from witnesses are complete and provided at once. However, in real-world scenarios, such descriptions are often partial or vague. To address this limitation, we introduce a new task called interactive person re-identification (Inter-ReID). Inter-ReID is a dialogue-based retrieval task that iteratively refines initial descriptions through ongoing interactions with the witnesses. To facilitate the study of this new task, we construct a dialogue dataset that incorporates multiple types of questions by decomposing fine-grained attributes of individuals. We further propose LLaVA-ReID, a question model that generates targeted questions based on visual and textual contexts to elicit additional details about the target person. Leveraging a looking-forward strategy, we prioritize the most informative questions as supervision during training. Experimental results on both Inter-ReID and text-based ReID benchmarks demonstrate that LLaVA-ReID significantly outperforms baselines.", "date": "2025-04-14", "categories": []}
{"id": "2505.18337", "title": "DART$^3$: Leveraging Distance for Test Time Adaptation in Person Re-Identification", "abstract": "Person re-identification (ReID) models are known to suffer from camera bias, where learned representations cluster according to camera viewpoints rather than identity, leading to significant performance degradation under (inter-camera) domain shifts in real-world surveillance systems when new cameras are added to camera networks. State-of-the-art test-time adaptation (TTA) methods, largely designed for classification tasks, rely on classification entropy-based objectives that fail to generalize well to ReID, thus making them unsuitable for tackling camera bias. In this paper, we introduce DART$^3$, a TTA framework specifically designed to mitigate camera-induced domain shifts in person ReID. DART$^3$ (Distance-Aware Retrieval Tuning at Test Time) leverages a distance-based objective that aligns better with image retrieval tasks like ReID by exploiting the correlation between nearest-neighbor distance and prediction error. Unlike prior ReID-specific domain adaptation methods, DART$^3$ requires no source data, architectural modifications, or retraining, and can be deployed in both fully black-box and hybrid settings. Empirical evaluations on multiple ReID benchmarks indicate that DART$^3$ and DART$^3$ LITE, a lightweight alternative to the approach, consistently outperforms state-of-the-art TTA baselines, making for a viable option to online learning to mitigate the adverse effects of camera bias.", "date": "2025-05-23", "categories": []}
{"id": "2505.19920", "title": "A Responsible Face Recognition Approach for Small and Mid-Scale Systems Through Personalized Neural Networks", "abstract": "Traditional face recognition systems rely on extracting fixed face representations, known as templates, to store and verify identities. These representations are typically generated by neural networks that often lack explainability and raise concerns regarding fairness and privacy. In this work, we propose a novel model-template (MOTE) approach that replaces vector-based face templates with small personalized neural networks. This design enables more responsible face recognition for small and medium-scale systems. During enrollment, MOTE creates a dedicated binary classifier for each identity, trained to determine whether an input face matches the enrolled identity. Each classifier is trained using only a single reference sample, along with synthetically balanced samples to allow adjusting fairness at the level of a single individual during enrollment. Extensive experiments across multiple datasets and recognition systems demonstrate substantial improvements in fairness and particularly in privacy. Although the method increases inference time and storage requirements, it presents a strong solution for small- and mid-scale applications where fairness and privacy are critical.", "date": "2025-05-26", "categories": []}
{"id": "2505.17445", "title": "PawPrint: Whose Footprints Are These? Identifying Animal Individuals by Their Footprints", "abstract": "In the United States, as of 2023, pet ownership has reached 66% of households and continues to rise annually. This trend underscores the critical need for effective pet identification and monitoring methods, particularly as nearly 10 million cats and dogs are reported stolen or lost each year. However, traditional methods for finding lost animals like GPS tags or ID photos have limitations-they can be removed, face signal issues, and depend on someone finding and reporting the pet. To address these limitations, we introduce PawPrint and PawPrint+, the first publicly available datasets focused on individual-level footprint identification for dogs and cats. Through comprehensive benchmarking of both modern deep neural networks (e.g., CNN, Transformers) and classical local features, we observe varying advantages and drawbacks depending on substrate complexity and data availability. These insights suggest future directions for combining learned global representations with local descriptors to enhance reliability across diverse, real-world conditions. As this approach provides a non-invasive alternative to traditional ID tags, we anticipate promising applications in ethical pet management and wildlife conservation efforts.", "date": "2025-05-23", "categories": []}
{"id": "2402.10847", "title": "Enhancement-Driven Pretraining for Robust Fingerprint Representation Learning", "abstract": "Fingerprint recognition stands as a pivotal component of biometric technology, with diverse applications from identity verification to advanced search tools. In this paper, we propose a unique method for deriving robust fingerprint representations by leveraging enhancement-based pre-training. Building on the achievements of U-Net-based fingerprint enhancement, our method employs a specialized encoder to derive representations from fingerprint images in a self-supervised manner. We further refine these representations, aiming to enhance the verification capabilities. Our experimental results, tested on publicly available fingerprint datasets, reveal a marked improvement in verification performance against established self-supervised training techniques. Our findings not only highlight the effectiveness of our method but also pave the way for potential advancements. Crucially, our research indicates that it is feasible to extract meaningful fingerprint representations from degraded images without relying on enhanced samples.", "date": "2024-02-16", "categories": []}
{"id": "2505.15825", "title": "Multilinear subspace learning for person re-identification based fusion of high order tensor features", "abstract": "Video surveillance image analysis and processing is a challenging field in computer vision, with one of its most difficult tasks being Person Re-Identification (PRe-ID). PRe-ID aims to identify and track target individuals who have already been detected in a network of cameras, using a robust description of their pedestrian images. The success of recent research in person PRe-ID is largely due to effective feature extraction and representation, as well as the powerful learning of these features to reliably discriminate between pedestrian images. To this end, two powerful features, Convolutional Neural Networks (CNN) and Local Maximal Occurrence (LOMO), are modeled on multidimensional data using the proposed method, High-Dimensional Feature Fusion (HDFF). Specifically, a new tensor fusion scheme is introduced to leverage and combine these two types of features in a single tensor, even though their dimensions are not identical. To enhance the system's accuracy, we employ Tensor Cross-View Quadratic Analysis (TXQDA) for multilinear subspace learning, followed by cosine similarity for matching. TXQDA efficiently facilitates learning while reducing the high dimensionality inherent in high-order tensor data. The effectiveness of our approach is verified through experiments on three widely-used PRe-ID datasets: VIPeR, GRID, and PRID450S. Extensive experiments demonstrate that our approach outperforms recent state-of-the-art methods.", "date": "2025-05-09", "categories": []}
{"id": "2505.16412", "title": "Pose-invariant face recognition via feature-space pose frontalization", "abstract": "Pose-invariant face recognition has become a challenging problem for modern AI-based face recognition systems. It aims at matching a profile face captured in the wild with a frontal face registered in a database. Existing methods perform face frontalization via either generative models or learning a pose robust feature representation. In this paper, a new method is presented to perform face frontalization and recognition within the feature space. First, a novel feature space pose frontalization module (FSPFM) is proposed to transform profile images with arbitrary angles into frontal counterparts. Second, a new training paradigm is proposed to maximize the potential of FSPFM and boost its performance. The latter consists of a pre-training and an attention-guided fine-tuning stage. Moreover, extensive experiments have been conducted on five popular face recognition benchmarks. Results show that not only our method outperforms the state-of-the-art in the pose-invariant face recognition task but also maintains superior performance in other standard scenarios.", "date": "2025-05-22", "categories": []}
{"id": "2505.15272", "title": "DiffProb: Data Pruning for Face Recognition", "abstract": "Face recognition models have made substantial progress due to advances in deep learning and the availability of large-scale datasets. However, reliance on massive annotated datasets introduces challenges related to training computational cost and data storage, as well as potential privacy concerns regarding managing large face datasets. This paper presents DiffProb, the first data pruning approach for the application of face recognition. DiffProb assesses the prediction probabilities of training samples within each identity and prunes the ones with identical or close prediction probability values, as they are likely reinforcing the same decision boundaries, and thus contribute minimally with new information. We further enhance this process with an auxiliary cleaning mechanism to eliminate mislabeled and label-flipped samples, boosting data quality with minimal loss. Extensive experiments on CASIA-WebFace with different pruning ratios and multiple benchmarks, including LFW, CFP-FP, and IJB-C, demonstrate that DiffProb can prune up to 50% of the dataset while maintaining or even, in some settings, improving the verification accuracies. Additionally, we demonstrate DiffProb's robustness across different architectures and loss functions. Our method significantly reduces training cost and data volume, enabling efficient face recognition training and reducing the reliance on massive datasets and their demanding management.", "date": "2025-05-21", "categories": []}
{"id": "2505.15332", "title": "Towards Zero-Shot Differential Morphing Attack Detection with Multimodal Large Language Models", "abstract": "Leveraging the power of multimodal large language models (LLMs) offers a promising approach to enhancing the accuracy and interpretability of morphing attack detection (MAD), especially in real-world biometric applications. This work introduces the use of LLMs for differential morphing attack detection (D-MAD). To the best of our knowledge, this is the first study to employ multimodal LLMs to D-MAD using real biometric data. To effectively utilize these models, we design Chain-of-Thought (CoT)-based prompts to reduce failure-to-answer rates and enhance the reasoning behind decisions. Our contributions include: (1) the first application of multimodal LLMs for D-MAD using real data subjects, (2) CoT-based prompt engineering to improve response reliability and explainability, (3) comprehensive qualitative and quantitative benchmarking of LLM performance using data from 54 individuals captured in passport enrollment scenarios, and (4) comparative analysis of two multimodal LLMs: ChatGPT-4o and Gemini providing insights into their morphing attack detection accuracy and decision transparency. Experimental results show that ChatGPT-4o outperforms Gemini in detection accuracy, especially against GAN-based morphs, though both models struggle under challenging conditions. While Gemini offers more consistent explanations, ChatGPT-4o is more resilient but prone to a higher failure-to-answer rate.", "date": "2025-05-21", "categories": []}
{"id": "2505.15476", "title": "Pura: An Efficient Privacy-Preserving Solution for Face Recognition", "abstract": "Face recognition is an effective technology for identifying a target person by facial images. However, sensitive facial images raises privacy concerns. Although privacy-preserving face recognition is one of potential solutions, this solution neither fully addresses the privacy concerns nor is efficient enough. To this end, we propose an efficient privacy-preserving solution for face recognition, named Pura, which sufficiently protects facial privacy and supports face recognition over encrypted data efficiently. Specifically, we propose a privacy-preserving and non-interactive architecture for face recognition through the threshold Paillier cryptosystem. Additionally, we carefully design a suite of underlying secure computing protocols to enable efficient operations of face recognition over encrypted data directly. Furthermore, we introduce a parallel computing mechanism to enhance the performance of the proposed secure computing protocols. Privacy analysis demonstrates that Pura fully safeguards personal facial privacy. Experimental evaluations demonstrate that Pura achieves recognition speeds up to 16 times faster than the state-of-the-art.", "date": "2025-05-21", "categories": []}
{"id": "2505.14320", "title": "Accuracy and Fairness of Facial Recognition Technology in Low-Quality Police Images: An Experiment With Synthetic Faces", "abstract": "Facial recognition technology (FRT) is increasingly used in criminal investigations, yet most evaluations of its accuracy rely on high-quality images, unlike those often encountered by law enforcement. This study examines how five common forms of image degradation--contrast, brightness, motion blur, pose shift, and resolution--affect FRT accuracy and fairness across demographic groups. Using synthetic faces generated by StyleGAN3 and labeled with FairFace, we simulate degraded images and evaluate performance using Deepface with ArcFace loss in 1:n identification tasks. We perform an experiment and find that false positive rates peak near baseline image quality, while false negatives increase as degradation intensifies--especially with blur and low resolution. Error rates are consistently higher for women and Black individuals, with Black females most affected. These disparities raise concerns about fairness and reliability when FRT is used in real-world investigative contexts. Nevertheless, even under the most challenging conditions and for the most affected subgroups, FRT accuracy remains substantially higher than that of many traditional forensic methods. This suggests that, if appropriately validated and regulated, FRT should be considered a valuable investigative tool. However, algorithmic accuracy alone is not sufficient: we must also evaluate how FRT is used in practice, including user-driven data manipulation. Such cases underscore the need for transparency and oversight in FRT deployment to ensure both fairness and forensic validity.", "date": "2025-05-20", "categories": []}
{"id": "2412.01345", "title": "See What You Seek: Semantic Contextual Integration for Cloth-Changing Person Re-Identification", "abstract": "Cloth-changing person re-identification (CC-ReID) aims to match individuals across surveillance cameras despite variations in clothing. Existing methods typically mitigate the impact of clothing changes or enhance identity (ID)-relevant features, but they often struggle to capture complex semantic information. In this paper, we propose a novel prompt learning framework Semantic Contextual Integration (SCI), which leverages the visual-textual representation capabilities of CLIP to reduce clothing-induced discrepancies and strengthen ID cues. Specifically, we introduce the Semantic Separation Enhancement (SSE) module, which employs dual learnable text tokens to disentangle clothing-related semantics from confounding factors, thereby isolating ID-relevant features. Furthermore, we develop a Semantic-Guided Interaction Module (SIM) that uses orthogonalized text features to guide visual representations, sharpening the focus of the model on distinctive ID characteristics. This semantic integration improves the discriminative power of the model and enriches the visual context with high-dimensional insights. Extensive experiments on three CC-ReID datasets demonstrate that our method outperforms state-of-the-art techniques. The code will be released at https://github.com/hxy-499/CCREID-SCI.", "date": "2024-12-02", "categories": []}
{"id": "2505.10497", "title": "MorphGuard: Morph Specific Margin Loss for Enhancing Robustness to Face Morphing Attacks", "abstract": "Face recognition has evolved significantly with the advancement of deep learning techniques, enabling its widespread adoption in various applications requiring secure authentication. However, this progress has also increased its exposure to presentation attacks, including face morphing, which poses a serious security threat by allowing one identity to impersonate another. Therefore, modern face recognition systems must be robust against such attacks.\n  In this work, we propose a novel approach for training deep networks for face recognition with enhanced robustness to face morphing attacks. Our method modifies the classification task by introducing a dual-branch classification strategy that effectively handles the ambiguity in the labeling of face morphs. This adaptation allows the model to incorporate morph images into the training process, improving its ability to distinguish them from bona fide samples.\n  Our strategy has been validated on public benchmarks, demonstrating its effectiveness in enhancing robustness against face morphing attacks. Furthermore, our approach is universally applicable and can be integrated into existing face recognition training pipelines to improve classification-based recognition methods.", "date": "2025-05-15", "categories": []}
{"id": "2505.08233", "title": "G-MSGINet: A Grouped Multi-Scale Graph-Involution Network for Contactless Fingerprint Recognition", "abstract": "This paper presents G-MSGINet, a unified and efficient framework for robust contactless fingerprint recognition that jointly performs minutiae localization and identity embedding directly from raw input images. Existing approaches rely on multi-branch architectures, orientation labels, or complex preprocessing steps, which limit scalability and generalization across real-world acquisition scenarios. In contrast, the proposed architecture introduces the GMSGI layer, a novel computational module that integrates grouped pixel-level involution, dynamic multi-scale kernel generation, and graph-based relational modelling into a single processing unit. Stacked GMSGI layers progressively refine both local minutiae-sensitive features and global topological representations through end-to-end optimization. The architecture eliminates explicit orientation supervision and adapts graph connectivity directly from learned kernel descriptors, thereby capturing meaningful structural relationships among fingerprint regions without fixed heuristics. Extensive experiments on three benchmark datasets, namely PolyU, CFPose, and Benchmark 2D/3D, demonstrate that G-MSGINet consistently achieves minutiae F1-scores in the range of $0.83\\pm0.02$ and Rank-1 identification accuracies between 97.0% and 99.1%, while maintaining an Equal Error Rate (EER) as low as 0.5%. These results correspond to improvements of up to 4.8% in F1-score and 1.4% in Rank-1 accuracy when compared to prior methods, using only 0.38 million parameters and 6.63 giga floating-point operations, which represents up to ten times fewer parameters than competitive baselines. This highlights the scalability and effectiveness of G-MSGINet in real-world contactless biometric recognition scenarios.", "date": "2025-05-13", "categories": []}
{"id": "2505.08801", "title": "OptiGait-LGBM: An Efficient Approach of Gait-based Person Re-identification in Non-Overlapping Regions", "abstract": "Gait recognition, known for its ability to identify individuals from a distance, has gained significant attention in recent times due to its non-intrusive verification. While video-based gait identification systems perform well on large public datasets, their performance drops when applied to real-world, unconstrained gait data due to various factors. Among these, uncontrolled outdoor environments, non-overlapping camera views, varying illumination, and computational efficiency are core challenges in gait-based authentication. Currently, no dataset addresses all these challenges simultaneously. In this paper, we propose an OptiGait-LGBM model capable of recognizing person re-identification under these constraints using a skeletal model approach, which helps mitigate inconsistencies in a person's appearance. The model constructs a dataset from landmark positions, minimizing memory usage by using non-sequential data. A benchmark dataset, RUET-GAIT, is introduced to represent uncontrolled gait sequences in complex outdoor environments. The process involves extracting skeletal joint landmarks, generating numerical datasets, and developing an OptiGait-LGBM gait classification model. Our aim is to address the aforementioned challenges with minimal computational cost compared to existing methods. A comparative analysis with ensemble techniques such as Random Forest and CatBoost demonstrates that the proposed approach outperforms them in terms of accuracy, memory usage, and training time. This method provides a novel, low-cost, and memory-efficient video-based gait recognition solution for real-world scenarios.", "date": "2025-05-10", "categories": []}
{"id": "2505.09073", "title": "2D-3D Attention and Entropy for Pose Robust 2D Facial Recognition", "abstract": "Despite recent advances in facial recognition, there remains a fundamental issue concerning degradations in performance due to substantial perspective (pose) differences between enrollment and query (probe) imagery. Therefore, we propose a novel domain adaptive framework to facilitate improved performances across large discrepancies in pose by enabling image-based (2D) representations to infer properties of inherently pose invariant point cloud (3D) representations. Specifically, our proposed framework achieves better pose invariance by using (1) a shared (joint) attention mapping to emphasize common patterns that are most correlated between 2D facial images and 3D facial data and (2) a joint entropy regularizing loss to promote better consistency$\\unicode{x2014}$enhancing correlations among the intersecting 2D and 3D representations$\\unicode{x2014}$by leveraging both attention maps. This framework is evaluated on FaceScape and ARL-VTF datasets, where it outperforms competitive methods by achieving profile (90$\\unicode{x00b0}$$\\unicode{x002b}$) TAR @ 1$\\unicode{x0025}$ FAR improvements of at least 7.1$\\unicode{x0025}$ and 1.57$\\unicode{x0025}$, respectively.", "date": "2025-05-14", "categories": []}
{"id": "2505.09256", "title": "Test-Time Augmentation for Pose-invariant Face Recognition", "abstract": "The goal of this paper is to enhance face recognition performance by augmenting head poses during the testing phase. Existing methods often rely on training on frontalised images or learning pose-invariant representations, yet both approaches typically require re-training and testing for each dataset, involving a substantial amount of effort. In contrast, this study proposes Pose-TTA, a novel approach that aligns faces at inference time without additional training. To achieve this, we employ a portrait animator that transfers the source image identity into the pose of a driving image. Instead of frontalising a side-profile face -- which can introduce distortion -- Pose-TTA generates matching side-profile images for comparison, thereby reducing identity information loss. Furthermore, we propose a weighted feature aggregation strategy to address any distortions or biases arising from the synthetic data, thus enhancing the reliability of the augmented images. Extensive experiments on diverse datasets and with various pre-trained face recognition models demonstrate that Pose-TTA consistently improves inference performance. Moreover, our method is straightforward to integrate into existing face recognition pipelines, as it requires no retraining or fine-tuning of the underlying recognition models.", "date": "2025-05-14", "categories": []}
{"id": "2504.21415", "title": "Optimizing Mouse Dynamics for User Authentication by Machine Learning: Addressing Data Sufficiency, Accuracy-Practicality Trade-off, and Model Performance Challenges", "abstract": "User authentication is essential to ensure secure access to computer systems, yet traditional methods face limitations in usability, cost, and security. Mouse dynamics authentication, based on the analysis of users' natural interaction behaviors with mouse devices, offers a cost-effective, non-intrusive, and adaptable solution. However, challenges remain in determining the optimal data volume, balancing accuracy and practicality, and effectively capturing temporal behavioral patterns. In this study, we propose a statistical method using Gaussian kernel density estimate (KDE) and Kullback-Leibler (KL) divergence to estimate the sufficient data volume for training authentication models. We introduce the Mouse Authentication Unit (MAU), leveraging Approximate Entropy (ApEn) to optimize segment length for efficient and accurate behavioral representation. Furthermore, we design the Local-Time Mouse Authentication (LT-AMouse) framework, integrating 1D-ResNet for local feature extraction and GRU for modeling long-term temporal dependencies. Taking the Balabit and DFL datasets as examples, we significantly reduced the data scale, particularly by a factor of 10 for the DFL dataset, greatly alleviating the training burden. Additionally, we determined the optimal input recognition unit length for the user authentication system on different datasets based on the slope of Approximate Entropy. Training with imbalanced samples, our model achieved a successful defense AUC 98.52% for blind attack on the DFL dataset and 94.65% on the Balabit dataset, surpassing the current sota performance.", "date": "2025-04-30", "categories": []}
{"id": "2505.07540", "title": "SynID: Passport Synthetic Dataset for Presentation Attack Detection", "abstract": "The demand for Presentation Attack Detection (PAD) to identify fraudulent ID documents in remote verification systems has significantly risen in recent years. This increase is driven by several factors, including the rise of remote work, online purchasing, migration, and advancements in synthetic images. Additionally, we have noticed a surge in the number of attacks aimed at the enrolment process. Training a PAD to detect fake ID documents is very challenging because of the limited number of ID documents available due to privacy concerns. This work proposes a new passport dataset generated from a hybrid method that combines synthetic data and open-access information using the ICAO requirement to obtain realistic training and testing images.", "date": "2025-05-12", "categories": []}
{"id": "2505.04793", "title": "DetReIDX: A Stress-Test Dataset for Real-World UAV-Based Person\n  Recognition", "abstract": "  Person reidentification (ReID) technology has been considered to perform\nrelatively well under controlled, ground-level conditions, but it breaks down\nwhen deployed in challenging real-world settings. Evidently, this is due to\nextreme data variability factors such as resolution, viewpoint changes, scale\nvariations, occlusions, and appearance shifts from clothing or session drifts.\nMoreover, the publicly available data sets do not realistically incorporate\nsuch kinds and magnitudes of variability, which limits the progress of this\ntechnology. This paper introduces DetReIDX, a large-scale aerial-ground person\ndataset, that was explicitly designed as a stress test to ReID under real-world\nconditions. DetReIDX is a multi-session set that includes over 13 million\nbounding boxes from 509 identities, collected in seven university campuses from\nthree continents, with drone altitudes between 5.8 and 120 meters. More\nimportant, as a key novelty, DetReIDX subjects were recorded in (at least) two\nsessions on different days, with changes in clothing, daylight and location,\nmaking it suitable to actually evaluate long-term person ReID. Plus, data were\nannotated from 16 soft biometric attributes and multitask labels for detection,\ntracking, ReID, and action recognition. In order to provide empirical evidence\nof DetReIDX usefulness, we considered the specific tasks of human detection and\nReID, where SOTA methods catastrophically degrade performance (up to 80% in\ndetection accuracy and over 70% in Rank-1 ReID) when exposed to DetReIDXs\nconditions. The dataset, annotations, and official evaluation protocols are\npublicly available at https://www.it.ubi.pt/DetReIDX/\n", "date": "2025-05-07", "categories": []}
{"id": "2505.04922", "title": "Canny2Palm: Realistic and Controllable Palmprint Generation for\n  Large-scale Pre-training", "abstract": "  Palmprint recognition is a secure and privacy-friendly method of biometric\nidentification. One of the major challenges to improve palmprint recognition\naccuracy is the scarcity of palmprint data. Recently, a popular line of\nresearch revolves around the synthesis of virtual palmprints for large-scale\npre-training purposes. In this paper, we propose a novel synthesis method named\nCanny2Palm that extracts palm textures with Canny edge detector and uses them\nto condition a Pix2Pix network for realistic palmprint generation. By\nre-assembling palmprint textures from different identities, we are able to\ncreate new identities by seeding the generator with new assemblies. Canny2Palm\nnot only synthesizes realistic data following the distribution of real\npalmprints but also enables controllable diversity to generate large-scale new\nidentities. On open-set palmprint recognition benchmarks, models pre-trained\nwith Canny2Palm synthetic data outperform the state-of-the-art with up to 7.2%\nhigher identification accuracy. Moreover, the performance of models pre-trained\nwith Canny2Palm continues to improve given 10,000 synthetic IDs while those\nwith existing methods already saturate, demonstrating the potential of our\nmethod for large-scale pre-training.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.05015", "title": "An Agent-Based Modeling Approach to Free-Text Keyboard Dynamics for\n  Continuous Authentication", "abstract": "  Continuous authentication systems leveraging free-text keyboard dynamics\noffer a promising additional layer of security in a multifactor authentication\nsetup that can be used in a transparent way with no impact on user experience.\nThis study investigates the efficacy of behavioral biometrics by employing an\nAgent-Based Model (ABM) to simulate diverse typing profiles across mechanical\nand membrane keyboards. Specifically, we generated synthetic keystroke data\nfrom five unique agents, capturing features related to dwell time, flight time,\nand error rates within sliding 5-second windows updated every second. Two\nmachine learning approaches, One-Class Support Vector Machine (OC-SVM) and\nRandom Forest (RF), were evaluated for user verification. Results revealed a\nstark contrast in performance: while One-Class SVM failed to differentiate\nindividual users within each group, Random Forest achieved robust\nintra-keyboard user recognition (Accuracy > 0.7) but struggled to generalize\nacross keyboards for the same user, highlighting the significant impact of\nkeyboard hardware on typing behavior. These findings suggest that: (1)\nkeyboard-specific user profiles may be necessary for reliable authentication,\nand (2) ensemble methods like RF outperform One-Class SVM in capturing\nfine-grained user-specific patterns.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.05374", "title": "OcularAge: A Comparative Study of Iris and Periocular Images for\n  Pediatric Age Estimation", "abstract": "  Estimating a child's age from ocular biometric images is challenging due to\nsubtle physiological changes and the limited availability of longitudinal\ndatasets. Although most biometric age estimation studies have focused on facial\nfeatures and adult subjects, pediatric-specific analysis, particularly of the\niris and periocular regions, remains relatively unexplored. This study presents\na comparative evaluation of iris and periocular images for estimating the ages\nof children aged between 4 and 16 years. We utilized a longitudinal dataset\ncomprising more than 21,000 near-infrared (NIR) images, collected from 288\npediatric subjects over eight years using two different imaging sensors. A\nmulti-task deep learning framework was employed to jointly perform age\nprediction and age-group classification, enabling a systematic exploration of\nhow different convolutional neural network (CNN) architectures, particularly\nthose adapted for non-square ocular inputs, capture the complex variability\ninherent in pediatric eye images. The results show that periocular models\nconsistently outperform iris-based models, achieving a mean absolute error\n(MAE) of 1.33 years and an age-group classification accuracy of 83.82%. These\nresults mark the first demonstration that reliable age estimation is feasible\nfrom children's ocular images, enabling privacy-preserving age checks in\nchild-centric applications. This work establishes the first longitudinal\nbenchmark for pediatric ocular age estimation, providing a foundation for\ndesigning robust, child-focused biometric systems. The developed models proved\nresilient across different imaging sensors, confirming their potential for\nreal-world deployment. They also achieved inference speeds of less than 10\nmilliseconds per image on resource-constrained VR headsets, demonstrating their\nsuitability for real-time applications.\n", "date": "2025-05-08", "categories": []}
{"id": "2505.03837", "title": "Explainable Face Recognition via Improved Localization", "abstract": "  Biometric authentication has become one of the most widely used tools in the\ncurrent technological era to authenticate users and to distinguish between\ngenuine users and imposters. Face is the most common form of biometric modality\nthat has proven effective. Deep learning-based face recognition systems are now\ncommonly used across different domains. However, these systems usually operate\nlike black-box models that do not provide necessary explanations or\njustifications for their decisions. This is a major disadvantage because users\ncannot trust such artificial intelligence-based biometric systems and may not\nfeel comfortable using them when clear explanations or justifications are not\nprovided. This paper addresses this problem by applying an efficient method for\nexplainable face recognition systems. We use a Class Activation Mapping\n(CAM)-based discriminative localization (very narrow/specific localization)\ntechnique called Scaled Directed Divergence (SDD) to visually explain the\nresults of deep learning-based face recognition systems. We perform fine\nlocalization of the face features relevant to the deep learning model for its\nprediction/decision. Our experiments show that the SDD Class Activation Map\n(CAM) highlights the relevant face features very specifically compared to the\ntraditional CAM and very accurately. The provided visual explanations with\nnarrow localization of relevant features can ensure much-needed transparency\nand trust for deep learning-based face recognition systems.\n", "date": "2025-05-04", "categories": []}
{"id": "2505.04306", "title": "MoDE: Mixture of Diffusion Experts for Any Occluded Face Recognition", "abstract": "  With the continuous impact of epidemics, people have become accustomed to\nwearing masks. However, most current occluded face recognition (OFR) algorithms\nlack prior knowledge of occlusions, resulting in poor performance when dealing\nwith occluded faces of varying types and severity in reality. Recognizing\noccluded faces is still a significant challenge, which greatly affects the\nconvenience of people's daily lives. In this paper, we propose an\nidentity-gated mixture of diffusion experts (MoDE) for OFR. Each\ndiffusion-based generative expert estimates one possible complete image for\noccluded faces. Considering the random sampling process of the diffusion model,\nwhich introduces inevitable differences and variations between the inpainted\nfaces and the real ones. To ensemble effective information from\nmulti-reconstructed faces, we introduce an identity-gating network to evaluate\nthe contribution of each reconstructed face to the identity and adaptively\nintegrate the predictions in the decision space. Moreover, our MoDE is a\nplug-and-play module for most existing face recognition models. Extensive\nexperiments on three public face datasets and two datasets in the wild validate\nour advanced performance for various occlusions in comparison with the\ncompeting methods.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.04502", "title": "Leveraging Simultaneous Usage of Edge GPU Hardware Engines for Video\n  Face Detection and Recognition", "abstract": "  Video face detection and recognition in public places at the edge is required\nin several applications, such as security reinforcement and contactless access\nto authorized venues. This paper aims to maximize the simultaneous usage of\nhardware engines available in edge GPUs nowadays by leveraging the concurrency\nand pipelining of tasks required for face detection and recognition. This also\nincludes the video decoding task, which is required in most face monitoring\napplications as the video streams are usually carried via Gbps Ethernet\nnetwork. This constitutes an improvement over previous works where the tasks\nare usually allocated to a single engine due to the lack of a unified and\nautomated framework that simultaneously explores all hardware engines. In\naddition, previously, the input faces were usually embedded in still images or\nwithin raw video streams that overlook the burst delay caused by the decoding\nstage. The results on real-life video streams suggest that simultaneously using\nall the hardware engines available in the recent NVIDIA edge Orin GPU, higher\nthroughput, and a slight saving of power consumption of around 300 mW,\naccounting for around 5%, have been achieved while satisfying the real-time\nperformance constraint. The performance gets even higher by considering several\nvideo streams simultaneously. Further performance improvement could have been\nobtained if the number of shuffle layers that were created by the tensor RT\nframework for the face recognition task was lower. Thus, the paper suggests\nsome hardware improvements to the existing edge GPU processors to enhance their\nperformance even higher.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.04616", "title": "Person Recognition at Altitude and Range: Fusion of Face, Body Shape and\n  Gait", "abstract": "  We address the problem of whole-body person recognition in unconstrained\nenvironments. This problem arises in surveillance scenarios such as those in\nthe IARPA Biometric Recognition and Identification at Altitude and Range\n(BRIAR) program, where biometric data is captured at long standoff distances,\nelevated viewing angles, and under adverse atmospheric conditions (e.g.,\nturbulence and high wind velocity). To this end, we propose FarSight, a unified\nend-to-end system for person recognition that integrates complementary\nbiometric cues across face, gait, and body shape modalities. FarSight\nincorporates novel algorithms across four core modules: multi-subject detection\nand tracking, recognition-aware video restoration, modality-specific biometric\nfeature encoding, and quality-guided multi-modal fusion. These components are\ndesigned to work cohesively under degraded image conditions, large pose and\nscale variations, and cross-domain gaps. Extensive experiments on the BRIAR\ndataset, one of the most comprehensive benchmarks for long-range, multi-modal\nbiometric recognition, demonstrate the effectiveness of FarSight. Compared to\nour preliminary system, this system achieves a 34.1% absolute gain in 1:1\nverification accuracy (TAR@0.1% FAR), a 17.8% increase in closed-set\nidentification (Rank-20), and a 34.3% reduction in open-set identification\nerrors (FNIR@1% FPIR). Furthermore, FarSight was evaluated in the 2025 NIST RTE\nFace in Video Evaluation (FIVE), which conducts standardized face recognition\ntesting on the BRIAR dataset. These results establish FarSight as a\nstate-of-the-art solution for operational biometric recognition in challenging\nreal-world conditions.\n", "date": "2025-05-07", "categories": []}
{"id": "2505.03012", "title": "GIF: Generative Inspiration for Face Recognition at Scale", "abstract": "  Aiming to reduce the computational cost of Softmax in massive label space of\nFace Recognition (FR) benchmarks, recent studies estimate the output using a\nsubset of identities. Although promising, the association between the\ncomputation cost and the number of identities in the dataset remains linear\nonly with a reduced ratio. A shared characteristic among available FR methods\nis the employment of atomic scalar labels during training. Consequently, the\ninput to label matching is through a dot product between the feature vector of\nthe input and the Softmax centroids. Inspired by generative modeling, we\npresent a simple yet effective method that substitutes scalar labels with\nstructured identity code, i.e., a sequence of integers. Specifically, we\npropose a tokenization scheme that transforms atomic scalar labels into\nstructured identity codes. Then, we train an FR backbone to predict the code\nfor each input instead of its scalar label. As a result, the associated\ncomputational cost becomes logarithmic w.r.t. number of identities. We\ndemonstrate the benefits of the proposed method by conducting experiments. In\nparticular, our method outperforms its competitors by 1.52%, and 0.6% at\nTAR@FAR$=1e-4$ on IJB-B and IJB-C, respectively, while transforming the\nassociation between computational cost and the number of identities from linear\nto logarithmic. See code at https://github.com/msed-Ebrahimi/GIF\n", "date": "2025-05-05", "categories": []}
{"id": "2405.01101", "title": "Enhancing person re-identification via Uncertainty Feature Fusion Method\n  and Auto-weighted Measure Combination", "abstract": "  Person re-identification (Re-ID) is a challenging task that involves\nidentifying the same person across different camera views in surveillance\nsystems. Current methods usually rely on features from single-camera views,\nwhich can be limiting when dealing with multiple cameras and challenges such as\nchanging viewpoints and occlusions. In this paper, a new approach is introduced\nthat enhances the capability of ReID models through the Uncertain Feature\nFusion Method (UFFM) and Auto-weighted Measure Combination (AMC). UFFM\ngenerates multi-view features using features extracted independently from\nmultiple images to mitigate view bias. However, relying only on similarity\nbased on multi-view features is limited because these features ignore the\ndetails represented in single-view features. Therefore, we propose the AMC\nmethod to generate a more robust similarity measure by combining various\nmeasures. Our method significantly improves Rank@1 accuracy and Mean Average\nPrecision (mAP) when evaluated on person re-identification datasets. Combined\nwith the BoT Baseline on challenging datasets, we achieve impressive results,\nwith a 7.9% improvement in Rank@1 and a 12.1% improvement in mAP on the MSMT17\ndataset. On the Occluded-DukeMTMC dataset, our method increases Rank@1 by 22.0%\nand mAP by 18.4%. Code is available:\nhttps://github.com/chequanghuy/Enhancing-Person-Re-Identification-via-UFFM-and-AMC\n", "date": "2024-05-02", "categories": []}
{"id": "2408.13431", "title": "Face Clustering via Early Stopping and Edge Recall", "abstract": "  Large-scale face clustering has achieved significant progress, with many\nefforts dedicated to learning to cluster large-scale faces with\nsupervised-learning. However, complex model design and tedious clustering\nprocesses are typical in existing methods. Such limitations result in\ninfeasible clustering in real-world applications. Reasonable and efficient\nmodel design and training need to be taken into account. Besides, developing\nunsupervised face clustering algorithms is crucial, which are more realistic in\nreal-world applications. In this paper, we propose a novel unsupervised face\nclustering algorithm FC-ES and a novel supervised face clustering algorithm\nFC-ESER to address these issues. An efficient and effective neighbor-based edge\nprobability and a novel early stopping strategy are proposed in FC-ES,\nguaranteeing the accuracy and recall of large-scale face clustering\nsimultaneously. Furthermore, to take advantage of supervised learning, a novel\nedge recall strategy is proposed in FC-ESER to further recall the edge\nconnections that are not connected in FC-ES. Extensive experiments on multiple\nbenchmarks for face, person, and vehicle clustering show that our proposed\nFC-ES and FC-ESER significantly outperform previous state-of-the-art methods.\nOur code will be available at https://github.com/jumptoliujj/FC-ESER.\n", "date": "2024-08-24", "categories": []}
{"id": "2505.02815", "title": "Database-Agnostic Gait Enrollment using SetTransformers", "abstract": "  Gait recognition has emerged as a powerful tool for unobtrusive and\nlong-range identity analysis, with growing relevance in surveillance and\nmonitoring applications. Although recent advances in deep learning and\nlarge-scale datasets have enabled highly accurate recognition under closed-set\nconditions, real-world deployment demands open-set gait enrollment, which means\ndetermining whether a new gait sample corresponds to a known identity or\nrepresents a previously unseen individual. In this work, we introduce a\ntransformer-based framework for open-set gait enrollment that is both\ndataset-agnostic and recognition-architecture-agnostic. Our method leverages a\nSetTransformer to make enrollment decisions based on the embedding of a probe\nsample and a context set drawn from the gallery, without requiring\ntask-specific thresholds or retraining for new environments. By decoupling\nenrollment from the main recognition pipeline, our model is generalized across\ndifferent datasets, gallery sizes, and identity distributions. We propose an\nevaluation protocol that uses existing datasets in different ratios of\nidentities and walks per identity. We instantiate our method using\nskeleton-based gait representations and evaluate it on two benchmark datasets\n(CASIA-B and PsyMo), using embeddings from three state-of-the-art recognition\nmodels (GaitGraph, GaitFormer, and GaitPT). We show that our method is\nflexible, is able to accurately perform enrollment in different scenarios, and\nscales better with data compared to traditional approaches. We will make the\ncode and dataset scenarios publicly available.\n", "date": "2025-05-05", "categories": []}
{"id": "2312.12028", "title": "EyePreserve: Identity-Preserving Iris Synthesis", "abstract": "  Synthesis of same-identity biometric iris images, both for existing and\nnon-existing identities while preserving the identity across a wide range of\npupil sizes, is complex due to the intricate iris muscle constriction\nmechanism, requiring a precise model of iris non-linear texture deformations to\nbe embedded into the synthesis pipeline. This paper presents the first method\nof fully data-driven, identity-preserving, pupil size-varying synthesis of iris\nimages. This approach is capable of synthesizing images of irises with\ndifferent pupil sizes representing non-existing identities, as well as\nnon-linearly deforming the texture of iris images of existing subjects given\nthe segmentation mask of the target iris image. Iris recognition experiments\nsuggest that the proposed deformation model both preserves the identity when\nchanging the pupil size, and offers better similarity between same-identity\niris samples with significant differences in pupil size, compared to\nstate-of-the-art linear and non-linear (bio-mechanical-based) iris deformation\nmodels. Two immediate applications of the proposed approach are: (a) synthesis\nof, or enhancement of the existing biometric datasets for iris recognition,\nmimicking those acquired with iris sensors, and (b) helping forensic human\nexperts examine iris image pairs with significant differences in pupil\ndilation. Images considered in this work conform to selected ISO/IEC 29794-6\nquality metrics to make them applicable in biometric systems. The source codes\nand model weights are offered with this paper.\n", "date": "2023-12-19", "categories": []}
{"id": "2505.00380", "title": "The Invisible Threat: Evaluating the Vulnerability of Cross-Spectral\n  Face Recognition to Presentation Attacks", "abstract": "  Cross-spectral face recognition systems are designed to enhance the\nperformance of facial recognition systems by enabling cross-modal matching\nunder challenging operational conditions. A particularly relevant application\nis the matching of near-infrared (NIR) images to visible-spectrum (VIS) images,\nenabling the verification of individuals by comparing NIR facial captures\nacquired with VIS reference images. The use of NIR imaging offers several\nadvantages, including greater robustness to illumination variations, better\nvisibility through glasses and glare, and greater resistance to presentation\nattacks. Despite these claimed benefits, the robustness of NIR-based systems\nagainst presentation attacks has not been systematically studied in the\nliterature. In this work, we conduct a comprehensive evaluation into the\nvulnerability of NIR-VIS cross-spectral face recognition systems to\npresentation attacks. Our empirical findings indicate that, although these\nsystems exhibit a certain degree of reliability, they remain vulnerable to\nspecific attacks, emphasizing the need for further research in this area.\n", "date": "2025-05-01", "categories": []}
{"id": "2306.07520", "title": "Instruct-ReID: A Multi-purpose Person Re-identification Task with\n  Instructions", "abstract": "  Human intelligence can retrieve any person according to both visual and\nlanguage descriptions. However, the current computer vision community studies\nspecific person re-identification (ReID) tasks in different scenarios\nseparately, which limits the applications in the real world. This paper strives\nto resolve this problem by proposing a new instruct-ReID task that requires the\nmodel to retrieve images according to the given image or language instructions.\nOur instruct-ReID is a more general ReID setting, where existing 6 ReID tasks\ncan be viewed as special cases by designing different instructions. We propose\na large-scale OmniReID benchmark and an adaptive triplet loss as a baseline\nmethod to facilitate research in this new setting. Experimental results show\nthat the proposed multi-purpose ReID model, trained on our OmniReID benchmark\nwithout fine-tuning, can improve +0.5%, +0.6%, +7.7% mAP on Market1501, MSMT17,\nCUHK03 for traditional ReID, +6.4%, +7.1%, +11.2% mAP on PRCC, VC-Clothes, LTCC\nfor clothes-changing ReID, +11.7% mAP on COCAS+ real2 for clothes template\nbased clothes-changing ReID when using only RGB images, +24.9% mAP on COCAS+\nreal2 for our newly defined language-instructed ReID, +4.3% on LLCM for\nvisible-infrared ReID, +2.6% on CUHK-PEDES for text-to-image ReID. The\ndatasets, the model, and code will be available at\nhttps://github.com/hwz-zju/Instruct-ReID.\n", "date": "2023-06-13", "categories": []}
{"id": "2405.17790", "title": "Instruct-ReID++: Towards Universal Purpose Instruction-Guided Person\n  Re-identification", "abstract": "  Human intelligence can retrieve any person according to both visual and\nlanguage descriptions. However, the current computer vision community studies\nspecific person re-identification (ReID) tasks in different scenarios\nseparately, which limits the applications in the real world. This paper strives\nto resolve this problem by proposing a novel instruct-ReID task that requires\nthe model to retrieve images according to the given image or language\ninstructions. Instruct-ReID is the first exploration of a general ReID setting,\nwhere existing 6 ReID tasks can be viewed as special cases by assigning\ndifferent instructions. To facilitate research in this new instruct-ReID task,\nwe propose a large-scale OmniReID++ benchmark equipped with diverse data and\ncomprehensive evaluation methods e.g., task specific and task-free evaluation\nsettings. In the task-specific evaluation setting, gallery sets are categorized\naccording to specific ReID tasks. We propose a novel baseline model, IRM, with\nan adaptive triplet loss to handle various retrieval tasks within a unified\nframework. For task-free evaluation setting, where target person images are\nretrieved from task-agnostic gallery sets, we further propose a new method\ncalled IRM++ with novel memory bank-assisted learning. Extensive evaluations of\nIRM and IRM++ on OmniReID++ benchmark demonstrate the superiority of our\nproposed methods, achieving state-of-the-art performance on 10 test sets. The\ndatasets, the model, and the code will be available at\nhttps://github.com/hwz-zju/Instruct-ReID\n", "date": "2024-05-28", "categories": []}
{"id": "2504.20104", "title": "An on-production high-resolution longitudinal neonatal fingerprint\n  database in Brazil", "abstract": "  The neonatal period is critical for survival, requiring accurate and early\nidentification to enable timely interventions such as vaccinations, HIV\ntreatment, and nutrition programs. Biometric solutions offer potential for\nchild protection by helping to prevent baby swaps, locate missing children, and\nsupport national identity systems. However, developing effective biometric\nidentification systems for newborns remains a major challenge due to the\nphysiological variability caused by finger growth, weight changes, and skin\ntexture alterations during early development. Current literature has attempted\nto address these issues by applying scaling factors to emulate growth-induced\ndistortions in minutiae maps, but such approaches fail to capture the complex\nand non-linear growth patterns of infants. A key barrier to progress in this\ndomain is the lack of comprehensive, longitudinal biometric datasets capturing\nthe evolution of neonatal fingerprints over time. This study addresses this gap\nby focusing on designing and developing a high-quality biometric database of\nneonatal fingerprints, acquired at multiple early life stages. The dataset is\nintended to support the training and evaluation of machine learning models\naimed at emulating the effects of growth on biometric features. We hypothesize\nthat such a dataset will enable the development of more robust and accurate\nDeep Learning-based models, capable of predicting changes in the minutiae map\nwith higher fidelity than conventional scaling-based methods. Ultimately, this\neffort lays the groundwork for more reliable biometric identification systems\ntailored to the unique developmental trajectory of newborns.\n", "date": "2025-04-27", "categories": []}
{"id": "2504.18910", "title": "Kinship Verification through a Forest Neural Network", "abstract": "  Early methods used face representations in kinship verification, which are\nless accurate than joint representations of parents' and children's facial\nimages learned from scratch. We propose an approach featuring graph neural\nnetwork concepts to utilize face representations and have comparable results to\njoint representation algorithms. Moreover, we designed the structure of the\nclassification module and introduced a new combination of losses to engage the\ncenter loss gradually in training our network. Additionally, we conducted\nexperiments on KinFaceW-I and II, demonstrating the effectiveness of our\napproach. We achieved the best result on KinFaceW-II, an average improvement of\nnearly 1.6 for all kinship types, and we were near the best on KinFaceW-I. The\ncode is available at https://github.com/ali-nazari/Kinship-Verification\n", "date": "2025-04-26", "categories": []}
{"id": "2504.19646", "title": "xEdgeFace: Efficient Cross-Spectral Face Recognition for Edge Devices", "abstract": "  Heterogeneous Face Recognition (HFR) addresses the challenge of matching face\nimages across different sensing modalities, such as thermal to visible or\nnear-infrared to visible, expanding the applicability of face recognition\nsystems in real-world, unconstrained environments. While recent HFR methods\nhave shown promising results, many rely on computation-intensive architectures,\nlimiting their practicality for deployment on resource-constrained edge\ndevices. In this work, we present a lightweight yet effective HFR framework by\nadapting a hybrid CNN-Transformer architecture originally designed for face\nrecognition. Our approach enables efficient end-to-end training with minimal\npaired heterogeneous data while preserving strong performance on standard RGB\nface recognition tasks. This makes it a compelling solution for both\nhomogeneous and heterogeneous scenarios. Extensive experiments across multiple\nchallenging HFR and face recognition benchmarks demonstrate that our method\nconsistently outperforms state-of-the-art approaches while maintaining a low\ncomputational overhead.\n", "date": "2025-04-28", "categories": []}
{"id": "2412.07199", "title": "A Parametric Approach to Adversarial Augmentation for Cross-Domain Iris\n  Presentation Attack Detection", "abstract": "  Iris-based biometric systems are vulnerable to presentation attacks (PAs),\nwhere adversaries present physical artifacts (e.g., printed iris images,\ntextured contact lenses) to defeat the system. This has led to the development\nof various presentation attack detection (PAD) algorithms, which typically\nperform well in intra-domain settings. However, they often struggle to\ngeneralize effectively in cross-domain scenarios, where training and testing\nemploy different sensors, PA instruments, and datasets. In this work, we use\nadversarial training samples of both bonafide irides and PAs to improve the\ncross-domain performance of a PAD classifier. The novelty of our approach lies\nin leveraging transformation parameters from classical data augmentation\nschemes (e.g., translation, rotation) to generate adversarial samples. We\nachieve this through a convolutional autoencoder, ADV-GEN, that inputs original\ntraining samples along with a set of geometric and photometric transformations.\nThe transformation parameters act as regularization variables, guiding ADV-GEN\nto generate adversarial samples in a constrained search space. Experiments\nconducted on the LivDet-Iris 2017 database, comprising four datasets, and the\nLivDet-Iris 2020 dataset, demonstrate the efficacy of our proposed method. The\ncode is available at https://github.com/iPRoBe-lab/ADV-GEN-IrisPAD.\n", "date": "2024-12-10", "categories": []}
{"id": "2504.18025", "title": "ShapeSpeak: Body Shape-Aware Textual Alignment for Visible-Infrared\n  Person Re-Identification", "abstract": "  Visible-Infrared Person Re-identification (VIReID) aims to match visible and\ninfrared pedestrian images, but the modality differences and the complexity of\nidentity features make it challenging. Existing methods rely solely on identity\nlabel supervision, which makes it difficult to fully extract high-level\nsemantic information. Recently, vision-language pre-trained models have been\nintroduced to VIReID, enhancing semantic information modeling by generating\ntextual descriptions. However, such methods do not explicitly model body shape\nfeatures, which are crucial for cross-modal matching. To address this, we\npropose an effective Body Shape-aware Textual Alignment (BSaTa) framework that\nexplicitly models and utilizes body shape information to improve VIReID\nperformance. Specifically, we design a Body Shape Textual Alignment (BSTA)\nmodule that extracts body shape information using a human parsing model and\nconverts it into structured text representations via CLIP. We also design a\nText-Visual Consistency Regularizer (TVCR) to ensure alignment between body\nshape textual representations and visual body shape features. Furthermore, we\nintroduce a Shape-aware Representation Learning (SRL) mechanism that combines\nMulti-text Supervision and Distribution Consistency Constraints to guide the\nvisual encoder to learn modality-invariant and discriminative identity\nfeatures, thus enhancing modality invariance. Experimental results demonstrate\nthat our method achieves superior performance on the SYSU-MM01 and RegDB\ndatasets, validating its effectiveness.\n", "date": "2025-04-25", "categories": []}
{"id": "2501.04444", "title": "A novel Facial Recognition technique with Focusing on Masked Faces", "abstract": "  Recognizing the same faces with and without masks is important for ensuring\nconsistent identification in security, access control, and public safety. This\ncapability is crucial in scenarios like law enforcement, healthcare, and\nsurveillance, where accurate recognition must be maintained despite facial\nocclusion. This research focuses on the challenge of recognizing the same faces\nwith and without masks by employing cosine similarity as the primary technique.\nWith the increased use of masks, traditional facial recognition systems face\nsignificant accuracy issues, making it crucial to develop methods that can\nreliably identify individuals in masked conditions. For that reason, this study\nproposed Masked-Unmasked Face Matching Model (MUFM). This model employs\ntransfer learning using the Visual Geometry Group (VGG16) model to extract\nsignificant facial features, which are subsequently classified utilizing the\nK-Nearest Neighbors (K-NN) algorithm. The cosine similarity metric is employed\nto compare masked and unmasked faces of the same individuals. This approach\nrepresents a novel contribution, as the task of recognizing the same individual\nwith and without a mask using cosine similarity has not been previously\naddressed. By integrating these advanced methodologies, the research\ndemonstrates effective identification of individuals despite the presence of\nmasks, addressing a significant limitation in traditional systems. Using data\nis another essential part of this work, by collecting and preparing an image\ndataset from three different sources especially some of those data are real\nprovided a comprehensive power of this research. The image dataset used were\nalready collected in three different datasets of masked and unmasked for the\nsame faces.\n", "date": "2025-01-08", "categories": []}
{"id": "2504.15823", "title": "Human-Imperceptible Physical Adversarial Attack for NIR Face Recognition\n  Models", "abstract": "  Near-infrared (NIR) face recognition systems, which can operate effectively\nin low-light conditions or in the presence of makeup, exhibit vulnerabilities\nwhen subjected to physical adversarial attacks. To further demonstrate the\npotential risks in real-world applications, we design a novel, stealthy, and\npractical adversarial patch to attack NIR face recognition systems in a\nblack-box setting. We achieved this by utilizing human-imperceptible\ninfrared-absorbing ink to generate multiple patches with digitally optimized\nshapes and positions for infrared images. To address the optimization mismatch\nbetween digital and real-world NIR imaging, we develop a light reflection model\nfor human skin to minimize pixel-level discrepancies by simulating NIR light\nreflection.\n  Compared to state-of-the-art (SOTA) physical attacks on NIR face recognition\nsystems, the experimental results show that our method improves the attack\nsuccess rate in both digital and physical domains, particularly maintaining\neffectiveness across various face postures. Notably, the proposed approach\noutperforms SOTA methods, achieving an average attack success rate of 82.46% in\nthe physical domain across different models, compared to 64.18% for existing\nmethods. The artifact is available at\nhttps://anonymous.4open.science/r/Human-imperceptible-adversarial-patch-0703/.\n", "date": "2025-04-22", "categories": []}
{"id": "2504.14253", "title": "ColorVein: Colorful Cancelable Vein Biometrics", "abstract": "  Vein recognition technologies have become one of the primary solutions for\nhigh-security identification systems. However, the issue of biometric\ninformation leakage can still pose a serious threat to user privacy and\nanonymity. Currently, there is no cancelable biometric template generation\nscheme specifically designed for vein biometrics. Therefore, this paper\nproposes an innovative cancelable vein biometric generation scheme: ColorVein.\nUnlike previous cancelable template generation schemes, ColorVein does not\ndestroy the original biometric features and introduces additional color\ninformation to grayscale vein images. This method significantly enhances the\ninformation density of vein images by transforming static grayscale information\ninto dynamically controllable color representations through interactive\ncolorization. ColorVein allows users/administrators to define a controllable\npseudo-random color space for grayscale vein images by editing the position,\nnumber, and color of hint points, thereby generating protected cancelable\ntemplates. Additionally, we propose a new secure center loss to optimize the\ntraining process of the protected feature extraction model, effectively\nincreasing the feature distance between enrolled users and any potential\nimpostors. Finally, we evaluate ColorVein's performance on all types of vein\nbiometrics, including recognition performance, unlinkability, irreversibility,\nand revocability, and conduct security and privacy analyses. ColorVein achieves\ncompetitive performance compared with state-of-the-art methods.\n", "date": "2025-04-19", "categories": []}
{"id": "2504.15105", "title": "A triple-branch network for latent fingerprint enhancement guided by\n  orientation fields and minutiae", "abstract": "  Latent fingerprint enhancement is a critical step in the process of latent\nfingerprint identification. Existing deep learning-based enhancement methods\nstill fall short of practical application requirements, particularly in\nrestoring low-quality fingerprint regions. Recognizing that different regions\nof latent fingerprints require distinct enhancement strategies, we propose a\nTriple Branch Spatial Fusion Network (TBSFNet), which simultaneously enhances\ndifferent regions of the image using tailored strategies. Furthermore, to\nimprove the generalization capability of the network, we integrate orientation\nfield and minutiae-related modules into TBSFNet and introduce a Multi-Level\nFeature Guidance Network (MLFGNet). Experimental results on the MOLF and MUST\ndatasets demonstrate that MLFGNet outperforms existing enhancement algorithms.\n", "date": "2025-04-21", "categories": []}
{"id": "2412.06235", "title": "VariFace: Fair and Diverse Synthetic Dataset Generation for Face\n  Recognition", "abstract": "  The use of large-scale, web-scraped datasets to train face recognition models\nhas raised significant privacy and bias concerns. Synthetic methods mitigate\nthese concerns and provide scalable and controllable face generation to enable\nfair and accurate face recognition. However, existing synthetic datasets\ndisplay limited intraclass and interclass diversity and do not match the face\nrecognition performance obtained using real datasets. Here, we propose\nVariFace, a two-stage diffusion-based pipeline to create fair and diverse\nsynthetic face datasets to train face recognition models. Specifically, we\nintroduce three methods: Face Recognition Consistency to refine demographic\nlabels, Face Vendi Score Guidance to improve interclass diversity, and\nDivergence Score Conditioning to balance the identity preservation-intraclass\ndiversity trade-off. When constrained to the same dataset size, VariFace\nconsiderably outperforms previous synthetic datasets (0.9200 $\\rightarrow$\n0.9405) and achieves comparable performance to face recognition models trained\nwith real data (Real Gap = -0.0065). In an unconstrained setting, VariFace not\nonly consistently achieves better performance compared to previous synthetic\nmethods across dataset sizes but also, for the first time, outperforms the real\ndataset (CASIA-WebFace) across six evaluation datasets. This sets a new\nstate-of-the-art performance with an average face verification accuracy of\n0.9567 (Real Gap = +0.0097) across LFW, CFP-FP, CPLFW, AgeDB, and CALFW\ndatasets and 0.9366 (Real Gap = +0.0380) on the RFW dataset.\n", "date": "2024-12-09", "categories": []}
{"id": "2504.11798", "title": "Neighbor-Based Feature and Index Enhancement for Person\n  Re-Identification", "abstract": "  Person re-identification (Re-ID) aims to match the same pedestrian in a large\ngallery with different cameras and views. Enhancing the robustness of the\nextracted feature representations is a main challenge in Re-ID. Existing\nmethods usually improve feature representation by improving model architecture,\nbut most methods ignore the potential contextual information, which limits the\neffectiveness of feature representation and retrieval performance. Neighborhood\ninformation, especially the potential information of multi-order neighborhoods,\ncan effectively enrich feature expression and improve retrieval accuracy, but\nthis has not been fully explored in existing research. Therefore, we propose a\nnovel model DMON-ARO that leverages latent neighborhood information to enhance\nboth feature representation and index performance. Our approach is built on two\ncomplementary modules: Dynamic Multi-Order Neighbor Modeling (DMON) and\nAsymmetric Relationship Optimization (ARO). The DMON module dynamically\naggregates multi-order neighbor relationships, allowing it to capture richer\ncontextual information and enhance feature representation through adaptive\nneighborhood modeling. Meanwhile, ARO refines the distance matrix by optimizing\nquery-to-gallery relationships, improving the index accuracy. Extensive\nexperiments on three benchmark datasets demonstrate that our approach achieves\nperformance improvements against baseline models, which illustrate the\neffectiveness of our model. Specifically, our model demonstrates improvements\nin Rank-1 accuracy and mAP. Moreover, this method can also be directly extended\nto other re-identification tasks.\n", "date": "2025-04-16", "categories": []}
{"id": "2504.11306", "title": "Context-Aware Palmprint Recognition via a Relative Similarity Metric", "abstract": "  We propose a new approach to matching mechanism for palmprint recognition by\nintroducing a Relative Similarity Metric (RSM) that enhances the robustness and\ndiscriminability of existing matching frameworks. While conventional systems\nrely on direct pairwise similarity measures, such as cosine or Euclidean\ndistances, these metrics fail to capture how a pairwise similarity compares\nwithin the context of the entire dataset. Our method addresses this by\nevaluating the relative consistency of similarity scores across up to all\nidentities, allowing for better suppression of false positives and negatives.\nApplied atop the CCNet architecture, our method achieves a new state-of-the-art\n0.000036% Equal Error Rate (EER) on the Tongji dataset, outperforming previous\nmethods and demonstrating the efficacy of incorporating relational structure\ninto the palmprint matching process.\n", "date": "2025-04-15", "categories": []}
{"id": "2503.04707", "title": "Iris Style Transfer: Enhancing Iris Recognition with Style Features and\n  Privacy Preservation through Neural Style Transfer", "abstract": "  Iris texture is widely regarded as a gold standard biometric modality for\nauthentication and identification. The demand for robust iris recognition\nmethods, coupled with growing security and privacy concerns regarding iris\nattacks, has escalated recently. Inspired by neural style transfer, an advanced\ntechnique that leverages neural networks to separate content and style\nfeatures, we hypothesize that iris texture's style features provide a reliable\nfoundation for recognition and are more resilient to variations like rotation\nand perspective shifts than traditional approaches. Our experimental results\nsupport this hypothesis, showing a significantly higher classification accuracy\ncompared to conventional features. Further, we propose using neural style\ntransfer to obfuscate the identifiable iris style features, ensuring the\nprotection of sensitive biometric information while maintaining the utility of\neye images for tasks like eye segmentation and gaze estimation. This work opens\nnew avenues for iris-oriented, secure, and privacy-aware biometric systems.\n", "date": "2025-03-06", "categories": []}
{"id": "2408.09752", "title": "A Unified Framework for Iris Anti-Spoofing: Introducing Iris\n  Anti-Spoofing Cross-Domain-Testing Protocol and Masked-MoE Method", "abstract": "  Iris recognition is widely used in high-security scenarios due to its\nstability and distinctiveness. However, iris images captured by different\ndevices exhibit certain and device-related consistent differences, which has a\ngreater impact on the classification algorithm for anti-spoofing. The iris of\nvarious races would also affect the classification, causing the risk of\nidentity theft. So it is necessary to improve the cross-domain capabilities of\nthe iris anti-spoofing (IAS) methods to enable it more robust in facing\ndifferent races and devices. However, there is no existing protocol that is\ncomprehensively available. To address this gap, we propose an Iris\nAnti-Spoofing Cross-Domain-Testing (IAS-CDT) Protocol, which involves 10\ndatasets, belonging to 7 databases, published by 4 institutions, and collected\nwith 6 different devices. It contains three sub-protocols hierarchically, aimed\nat evaluating average performance, cross-racial generalization, and\ncross-device generalization of IAS models. Moreover, to address the\ncross-device generalization challenge brought by the IAS-CDT Protocol, we\nemploy multiple model parameter sets to learn from the multiple sub-datasets.\nSpecifically, we utilize the Mixture of Experts (MoE) to fit complex data\ndistributions using multiple sub-neural networks. To further enhance the\ngeneralization capabilities, we propose a novel method Masked-MoE (MMoE), which\nrandomly masks a portion of tokens for some experts and requires their outputs\nto be similar to the unmasked experts, which can effectively mitigate the\noverfitting issue of MoE. For the evaluation, we selected ResNet50, VIT-B/16,\nCLIP, and FLIP as representative models and benchmarked them under the proposed\nIAS-CDT Protocol.\n", "date": "2024-08-19", "categories": []}
{"id": "2503.06965", "title": "SeCap: Self-Calibrating and Adaptive Prompts for Cross-view Person\n  Re-Identification in Aerial-Ground Networks", "abstract": "  When discussing the Aerial-Ground Person Re-identification (AGPReID) task, we\nface the main challenge of the significant appearance variations caused by\ndifferent viewpoints, making identity matching difficult. To address this\nissue, previous methods attempt to reduce the differences between viewpoints by\ncritical attributes and decoupling the viewpoints. While these methods can\nmitigate viewpoint differences to some extent, they still face two main issues:\n(1) difficulty in handling viewpoint diversity and (2) neglect of the\ncontribution of local features. To effectively address these challenges, we\ndesign and implement the Self-Calibrating and Adaptive Prompt (SeCap) method\nfor the AGPReID task. The core of this framework relies on the Prompt\nRe-calibration Module (PRM), which adaptively re-calibrates prompts based on\nthe input. Combined with the Local Feature Refinement Module (LFRM), SeCap can\nextract view-invariant features from local features for AGPReID. Meanwhile,\ngiven the current scarcity of datasets in the AGPReID field, we further\ncontribute two real-world Large-scale Aerial-Ground Person Re-Identification\ndatasets, LAGPeR and G2APS-ReID. The former is collected and annotated by us\nindependently, covering $4,231$ unique identities and containing $63,841$\nhigh-quality images; the latter is reconstructed from the person search dataset\nG2APS. Through extensive experiments on AGPReID datasets, we demonstrate that\nSeCap is a feasible and effective solution for the AGPReID task. The datasets\nand source code available on https://github.com/wangshining681/SeCap-AGPReID.\n", "date": "2025-03-10", "categories": []}
{"id": "2311.11882", "title": "Multi-Task Faces (MTF) Data Set: A Legally and Ethically Compliant\n  Collection of Face Images for Various Classification Tasks", "abstract": "  Human facial data offers valuable potential for tackling classification\nproblems, including face recognition, age estimation, gender identification,\nemotion analysis, and race classification. However, recent privacy regulations,\nparticularly the EU General Data Protection Regulation, have restricted the\ncollection and usage of human images in research. As a result, several\npreviously published face data sets have been removed from the internet due to\ninadequate data collection methods and privacy concerns. While synthetic data\nsets have been suggested as an alternative, they fall short of accurately\nrepresenting the real data distribution. Additionally, most existing data sets\nare labeled for just a single task, which limits their versatility. To address\nthese limitations, we introduce the Multi-Task Face (MTF) data set, designed\nfor various tasks, including face recognition and classification by race,\ngender, and age, as well as for aiding in training generative networks. The MTF\ndata set comes in two versions: a non-curated set containing 132,816 images of\n640 individuals and a manually curated set with 5,246 images of 240\nindividuals, meticulously selected to maximize their classification quality.\nBoth data sets were ethically sourced, using publicly available celebrity\nimages in full compliance with copyright regulations. Along with providing\ndetailed descriptions of data collection and processing, we evaluated the\neffectiveness of the MTF data set in training five deep learning models across\nthe aforementioned classification tasks, achieving up to 98.88\\% accuracy for\ngender classification, 95.77\\% for race classification, 97.60\\% for age\nclassification, and 79.87\\% for face recognition with the ConvNeXT model. Both\nMTF data sets can be accessed through the following link.\nhttps://github.com/RamiHaf/MTF_data_set\n", "date": "2023-11-20", "categories": []}
{"id": "2504.01035", "title": "Novel sparse PCA method via Runge Kutta numerical method(s) for face\n  recognition", "abstract": "  Face recognition is a crucial topic in data science and biometric security,\nwith applications spanning military, finance, and retail industries. This paper\nexplores the implementation of sparse Principal Component Analysis (PCA) using\nthe Proximal Gradient method (also known as ISTA) and the Runge-Kutta numerical\nmethods. To address the face recognition problem, we integrate sparse PCA with\neither the k-nearest neighbor method or the kernel ridge regression method.\nExperimental results demonstrate that combining sparse PCA-solved via the\nProximal Gradient method or the Runge-Kutta numerical approach-with a\nclassification system yields higher accuracy compared to standard PCA.\nAdditionally, we observe that the Runge-Kutta-based sparse PCA computation\nconsistently outperforms the Proximal Gradient method in terms of speed.\n", "date": "2025-03-30", "categories": []}
{"id": "2504.01047", "title": "Predicting Movie Production Years through Facial Recognition of Actors\n  with Machine Learning", "abstract": "  This study used machine learning algorithms to identify actors and extract\nthe age of actors from images taken randomly from movies. The use of images\ntaken from Arab movies includes challenges such as non-uniform lighting,\ndifferent and multiple poses for the actors and multiple elements with the\nactor or a group of actors. Additionally, the use of make-up, wigs, beards, and\nwearing different accessories and costumes made it difficult for the system to\nidentify the personality of the same actor. The Arab Actors Dataset-AAD\ncomprises 574 images sourced from various movies, encompassing both black and\nwhite as well as color compositions. The images depict complete scenes or\nfragments thereof. Multiple models were employed for feature extraction, and\ndiverse machine learning algorithms were utilized during the classification and\nprediction stages to determine the most effective algorithm for handling such\nimage types. The study demonstrated the effectiveness of the Logistic\nRegression model exhibited the best performance compared to other models in the\ntraining phase, as evidenced by its AUC, precision, CA and F1score values of\n99%, 86%, 85.5% and 84.2% respectively. The findings of this study can be used\nto improve the precision and reliability of facial recognition technology for\nvarious uses as with movies search services, movie suggestion algorithms, and\ngenre classification of movies.\n", "date": "2025-04-01", "categories": []}
{"id": "2504.01213", "title": "GRU-AUNet: A Domain Adaptation Framework for Contactless Fingerprint\n  Presentation Attack Detection", "abstract": "  Although contactless fingerprints offer user comfort, they are more\nvulnerable to spoofing. The current solution for anti-spoofing in the area of\ncontactless fingerprints relies on domain adaptation learning, limiting their\ngeneralization and scalability. To address these limitations, we introduce\nGRU-AUNet, a domain adaptation approach that integrates a Swin\nTransformer-based UNet architecture with GRU-enhanced attention mechanisms, a\nDynamic Filter Network in the bottleneck, and a combined Focal and Contrastive\nLoss function. Trained in both genuine and spoof fingerprint images, GRU-AUNet\ndemonstrates robust resilience against presentation attacks, achieving an\naverage BPCER of 0.09\\% and APCER of 1.2\\% in the CLARKSON, COLFISPOOF, and\nIIITD datasets, outperforming state-of-the-art domain adaptation methods.\n", "date": "2025-04-01", "categories": []}
{"id": "2311.14395", "title": "MSCMNet: Multi-scale Semantic Correlation Mining for Visible-Infrared\n  Person Re-Identification", "abstract": "  The main challenge in the Visible-Infrared Person Re-Identification (VI-ReID)\ntask lies in how to extract discriminative features from different modalities\nfor matching purposes. While the existing well works primarily focus on\nminimizing the modal discrepancies, the modality information can not thoroughly\nbe leveraged. To solve this problem, a Multi-scale Semantic Correlation Mining\nnetwork (MSCMNet) is proposed to comprehensively exploit semantic features at\nmultiple scales and simultaneously reduce modality information loss as small as\npossible in feature extraction. The proposed network contains three novel\ncomponents. Firstly, after taking into account the effective utilization of\nmodality information, the Multi-scale Information Correlation Mining Block\n(MIMB) is designed to explore semantic correlations across multiple scales.\nSecondly, in order to enrich the semantic information that MIMB can utilize, a\nquadruple-stream feature extractor (QFE) with non-shared parameters is\nspecifically designed to extract information from different dimensions of the\ndataset. Finally, the Quadruple Center Triplet Loss (QCT) is further proposed\nto address the information discrepancy in the comprehensive features. Extensive\nexperiments on the SYSU-MM01, RegDB, and LLCM datasets demonstrate that the\nproposed MSCMNet achieves the greatest accuracy.\n", "date": "2023-11-24", "categories": []}
{"id": "2504.00430", "title": "Data Synthesis with Diverse Styles for Face Recognition via 3DMM-Guided\n  Diffusion", "abstract": "  Identity-preserving face synthesis aims to generate synthetic face images of\nvirtual subjects that can substitute real-world data for training face\nrecognition models. While prior arts strive to create images with consistent\nidentities and diverse styles, they face a trade-off between them. Identifying\ntheir limitation of treating style variation as subject-agnostic and observing\nthat real-world persons actually have distinct, subject-specific styles, this\npaper introduces MorphFace, a diffusion-based face generator. The generator\nlearns fine-grained facial styles, e.g., shape, pose and expression, from the\nrenderings of a 3D morphable model (3DMM). It also learns identities from an\noff-the-shelf recognition model. To create virtual faces, the generator is\nconditioned on novel identities of unlabeled synthetic faces, and novel styles\nthat are statistically sampled from a real-world prior distribution. The\nsampling especially accounts for both intra-subject variation and subject\ndistinctiveness. A context blending strategy is employed to enhance the\ngenerator's responsiveness to identity and style conditions. Extensive\nexperiments show that MorphFace outperforms the best prior arts in face\nrecognition efficacy.\n", "date": "2025-04-01", "categories": []}
{"id": "2410.13567", "title": "CCUP: A Controllable Synthetic Data Generation Pipeline for Pretraining\n  Cloth-Changing Person Re-Identification Models", "abstract": "  Cloth-changing person re-identification (CC-ReID), also known as Long-Term\nPerson Re-Identification (LT-ReID) is a critical and challenging research topic\nin computer vision that has recently garnered significant attention. However,\ndue to the high cost of constructing CC-ReID data, the existing data-driven\nmodels are hard to train efficiently on limited data, causing overfitting\nissue. To address this challenge, we propose a low-cost and efficient pipeline\nfor generating controllable and high-quality synthetic data simulating the\nsurveillance of real scenarios specific to the CC-ReID task. Particularly, we\nconstruct a new self-annotated CC-ReID dataset named Cloth-Changing Unreal\nPerson (CCUP), containing 6,000 IDs, 1,179,976 images, 100 cameras, and 26.5\noutfits per individual. Based on this large-scale dataset, we introduce an\neffective and scalable pretrain-finetune framework for enhancing the\ngeneralization capabilities of the traditional CC-ReID models. The extensive\nexperiments demonstrate that two typical models namely TransReID and FIRe^2,\nwhen integrated into our framework, outperform other state-of-the-art models\nafter pretraining on CCUP and finetuning on the benchmarks such as PRCC,\nVC-Clothes and NKUP. The CCUP is available at:\nhttps://github.com/yjzhao1019/CCUP.\n", "date": "2024-10-17", "categories": []}
{"id": "2503.22912", "title": "DIFFER: Disentangling Identity Features via Semantic Cues for\n  Clothes-Changing Person Re-ID", "abstract": "  Clothes-changing person re-identification (CC-ReID) aims to recognize\nindividuals under different clothing scenarios. Current CC-ReID approaches\neither concentrate on modeling body shape using additional modalities including\nsilhouette, pose, and body mesh, potentially causing the model to overlook\nother critical biometric traits such as gender, age, and style, or they\nincorporate supervision through additional labels that the model tries to\ndisregard or emphasize, such as clothing or personal attributes. However, these\nannotations are discrete in nature and do not capture comprehensive\ndescriptions.\n  In this work, we propose DIFFER: Disentangle Identity Features From Entangled\nRepresentations, a novel adversarial learning method that leverages textual\ndescriptions to disentangle identity features. Recognizing that image features\ninherently mix inseparable information, DIFFER introduces NBDetach, a mechanism\ndesigned for feature disentanglement by leveraging the separable nature of text\ndescriptions as supervision. It partitions the feature space into distinct\nsubspaces and, through gradient reversal layers, effectively separates\nidentity-related features from non-biometric features. We evaluate DIFFER on 4\ndifferent benchmark datasets (LTCC, PRCC, CelebreID-Light, and CCVID) to\ndemonstrate its effectiveness and provide state-of-the-art performance across\nall the benchmarks. DIFFER consistently outperforms the baseline method, with\nimprovements in top-1 accuracy of 3.6% on LTCC, 3.4% on PRCC, 2.5% on\nCelebReID-Light, and 1% on CCVID. Our code can be found here.\n", "date": "2025-03-28", "categories": []}
{"id": "2503.23275", "title": "Improved Ear Verification with Vision Transformers and Overlapping\n  Patches", "abstract": "  Ear recognition has emerged as a promising biometric modality due to the\nrelative stability in appearance during adulthood. Although Vision Transformers\n(ViTs) have been widely used in image recognition tasks, their efficiency in\near recognition has been hampered by a lack of attention to overlapping\npatches, which is crucial for capturing intricate ear features. In this study,\nwe evaluate ViT-Tiny (ViT-T), ViT-Small (ViT-S), ViT-Base (ViT-B) and ViT-Large\n(ViT-L) configurations on a diverse set of datasets (OPIB, AWE, WPUT, and\nEarVN1.0), using an overlapping patch selection strategy. Results demonstrate\nthe critical importance of overlapping patches, yielding superior performance\nin 44 of 48 experiments in a structured study. Moreover, upon comparing the\nresults of the overlapping patches with the non-overlapping configurations, the\nincrease is significant, reaching up to 10% for the EarVN1.0 dataset. In terms\nof model performance, the ViT-T model consistently outperformed the ViT-S,\nViT-B, and ViT-L models on the AWE, WPUT, and EarVN1.0 datasets. The highest\nscores were achieved in a configuration with a patch size of 28x28 and a stride\nof 14 pixels. This patch-stride configuration represents 25% of the normalized\nimage area (112x112 pixels) for the patch size and 12.5% of the row or column\nsize for the stride. This study confirms that transformer architectures with\noverlapping patch selection can serve as an efficient and high-performing\noption for ear-based biometric recognition tasks in verification scenarios.\n", "date": "2025-03-30", "categories": []}
{"id": "2503.23930", "title": "Exploring Reliable PPG Authentication on Smartwatches in Daily Scenarios", "abstract": "  Photoplethysmography (PPG) Sensors, widely deployed in smartwatches, offer a\nsimple and non-invasive authentication approach for daily use. However, PPG\nauthentication faces reliability issues due to motion artifacts from physical\nactivity and physiological variability over time. To address these challenges,\nwe propose MTL-RAPID, an efficient and reliable PPG authentication model, that\nemploys a multitask joint training strategy, simultaneously assessing signal\nquality and verifying user identity. The joint optimization of these two tasks\nin MTL-RAPID results in a structure that outperforms models trained on\nindividual tasks separately, achieving stronger performance with fewer\nparameters. In our comprehensive user studies regarding motion artifacts (N =\n30), time variations (N = 32), and user preferences (N = 16), MTL-RAPID\nachieves a best AUC of 99.2\\% and an EER of 3.5\\%, outperforming existing\nbaselines. We opensource our PPG authentication dataset along with the\nMTL-RAPID model to facilitate future research on GitHub.\n", "date": "2025-03-31", "categories": []}
{"id": "2503.23949", "title": "AMB-FHE: Adaptive Multi-biometric Fusion with Fully Homomorphic\n  Encryption", "abstract": "  Biometric systems strive to balance security and usability. The use of\nmulti-biometric systems combining multiple biometric modalities is usually\nrecommended for high-security applications. However, the presentation of\nmultiple biometric modalities can impair the user-friendliness of the overall\nsystem and might not be necessary in all cases. In this work, we present a\nsimple but flexible approach to increase the privacy protection of\nhomomorphically encrypted multi-biometric reference templates while enabling\nadaptation to security requirements at run-time: An adaptive multi-biometric\nfusion with fully homomorphic encryption (AMB-FHE). AMB-FHE is benchmarked\nagainst a bimodal biometric database consisting of the CASIA iris and MCYT\nfingerprint datasets using deep neural networks for feature extraction. Our\ncontribution is easy to implement and increases the flexibility of biometric\nauthentication while offering increased privacy protection through joint\nencryption of templates from multiple modalities.\n", "date": "2025-03-31", "categories": []}
{"id": "2412.02479", "title": "OODFace: Benchmarking Robustness of Face Recognition under Common\n  Corruptions and Appearance Variations", "abstract": "  With the rise of deep learning, facial recognition technology has seen\nextensive research and rapid development. Although facial recognition is\nconsidered a mature technology, we find that existing open-source models and\ncommercial algorithms lack robustness in certain complex Out-of-Distribution\n(OOD) scenarios, raising concerns about the reliability of these systems. In\nthis paper, we introduce OODFace, which explores the OOD challenges faced by\nfacial recognition models from two perspectives: common corruptions and\nappearance variations. We systematically design 30 OOD scenarios across 9 major\ncategories tailored for facial recognition. By simulating these challenges on\npublic datasets, we establish three robustness benchmarks: LFW-C/V, CFP-FP-C/V,\nand YTF-C/V. We then conduct extensive experiments on 19 facial recognition\nmodels and 3 commercial APIs, along with extended physical experiments on face\nmasks to assess their robustness. Next, we explore potential solutions from two\nperspectives: defense strategies and Vision-Language Models (VLMs). Based on\nthe results, we draw several key insights, highlighting the vulnerability of\nfacial recognition systems to OOD data and suggesting possible solutions.\nAdditionally, we offer a unified toolkit that includes all corruption and\nvariation types, easily extendable to other datasets. We hope that our\nbenchmarks and findings can provide guidance for future improvements in facial\nrecognition model robustness.\n", "date": "2024-12-03", "categories": []}
{"id": "2412.08885", "title": "Residual Channel Boosts Contrastive Learning for Radio Frequency\n  Fingerprint Identification", "abstract": "  In order to address the issue of limited data samples for the deployment of\npre-trained models in unseen environments, this paper proposes a residual\nchannel-based data augmentation strategy for Radio Frequency Fingerprint\nIdentification (RFFI), coupled with a lightweight SimSiam contrastive learning\nframework. By applying least square (LS) and minimum mean square error (MMSE)\nchannel estimations followed by equalization, signals with different residual\nchannel effects are generated. These residual channels enable the model to\nlearn more effective representations. Then the pre-trained model is fine-tuned\nwith 1% samples in a novel environment for RFFI. Experimental results\ndemonstrate that our method significantly enhances both feature extraction\nability and generalization while requiring fewer samples and less time, making\nit suitable for practical wireless security applications.\n", "date": "2024-12-12", "categories": []}
{"id": "2503.20108", "title": "Peepers & Pixels: Human Recognition Accuracy on Low Resolution Faces", "abstract": "  Automated one-to-many (1:N) face recognition is a powerful investigative tool\ncommonly used by law enforcement agencies. In this context, potential matches\nresulting from automated 1:N recognition are reviewed by human examiners prior\nto possible use as investigative leads. While automated 1:N recognition can\nachieve near-perfect accuracy under ideal imaging conditions, operational\nscenarios may necessitate the use of surveillance imagery, which is often\ndegraded in various quality dimensions. One important quality dimension is\nimage resolution, typically quantified by the number of pixels on the face. The\ncommon metric for this is inter-pupillary distance (IPD), which measures the\nnumber of pixels between the pupils. Low IPD is known to degrade the accuracy\nof automated face recognition. However, the threshold IPD for reliability in\nhuman face recognition remains undefined. This study aims to explore the\nboundaries of human recognition accuracy by systematically testing accuracy\nacross a range of IPD values. We find that at low IPDs (10px, 5px), human\naccuracy is at or below chance levels (50.7%, 35.9%), even as confidence in\ndecision-making remains relatively high (77%, 70.7%). Our findings indicate\nthat, for low IPD images, human recognition ability could be a limiting factor\nto overall system accuracy.\n", "date": "2025-03-25", "categories": []}
{"id": "2503.19478", "title": "TeLL Me what you cant see", "abstract": "  During criminal investigations, images of persons of interest directly\ninfluence the success of identification procedures. However, law enforcement\nagencies often face challenges related to the scarcity of high-quality images\nor their obsolescence, which can affect the accuracy and success of people\nsearching processes. This paper introduces a novel forensic mugshot\naugmentation framework aimed at addressing these limitations. Our approach\nenhances the identification probability of individuals by generating\nadditional, high-quality images through customizable data augmentation\ntechniques, while maintaining the biometric integrity and consistency of the\noriginal data. Several experimental results show that our method significantly\nimproves identification accuracy and robustness across various forensic\nscenarios, demonstrating its effectiveness as a trustworthy tool law\nenforcement applications. Index Terms: Digital Forensics, Person\nre-identification, Feature extraction, Data augmentation, Visual-Language\nmodels.\n", "date": "2025-03-25", "categories": []}
{"id": "2409.05595", "title": "SynMorph: Generating Synthetic Face Morphing Dataset with Mated Samples", "abstract": "  Face morphing attack detection (MAD) algorithms have become essential to\novercome the vulnerability of face recognition systems. To solve the lack of\nlarge-scale and public-available datasets due to privacy concerns and\nrestrictions, in this work we propose a new method to generate a synthetic face\nmorphing dataset with 2450 identities and more than 100k morphs. The proposed\nsynthetic face morphing dataset is unique for its high-quality samples,\ndifferent types of morphing algorithms, and the generalization for both single\nand differential morphing attack detection algorithms. For experiments, we\napply face image quality assessment and vulnerability analysis to evaluate the\nproposed synthetic face morphing dataset from the perspective of biometric\nsample quality and morphing attack potential on face recognition systems. The\nresults are benchmarked with an existing SOTA synthetic dataset and a\nrepresentative non-synthetic and indicate improvement compared with the SOTA.\nAdditionally, we design different protocols and study the applicability of\nusing the proposed synthetic dataset on training morphing attack detection\nalgorithms.\n", "date": "2024-09-09", "categories": []}
{"id": "2503.17096", "title": "Multi-modal Multi-platform Person Re-Identification: Benchmark and\n  Method", "abstract": "  Conventional person re-identification (ReID) research is often limited to\nsingle-modality sensor data from static cameras, which fails to address the\ncomplexities of real-world scenarios where multi-modal signals are increasingly\nprevalent. For instance, consider an urban ReID system integrating stationary\nRGB cameras, nighttime infrared sensors, and UAVs equipped with dynamic\ntracking capabilities. Such systems face significant challenges due to\nvariations in camera perspectives, lighting conditions, and sensor modalities,\nhindering effective person ReID. To address these challenges, we introduce the\nMP-ReID benchmark, a novel dataset designed specifically for multi-modality and\nmulti-platform ReID. This benchmark uniquely compiles data from 1,930\nidentities across diverse modalities, including RGB, infrared, and thermal\nimaging, captured by both UAVs and ground-based cameras in indoor and outdoor\nenvironments. Building on this benchmark, we introduce Uni-Prompt ReID, a\nframework with specific-designed prompts, tailored for cross-modality and\ncross-platform scenarios. Our method consistently outperforms state-of-the-art\napproaches, establishing a robust foundation for future research in complex and\ndynamic ReID environments. Our dataset are available\nat:https://mp-reid.github.io/.\n", "date": "2025-03-21", "categories": []}
{"id": "2503.18652", "title": "Robust face recognition based on the wing loss and the $\\ell_1$\n  regularization", "abstract": "  In recent years, sparse sampling techniques based on regression analysis have\nwitnessed extensive applications in face recognition research. Presently,\nnumerous sparse sampling models based on regression analysis have been explored\nby various researchers. Nevertheless, the recognition rates of the majority of\nthese models would be significantly decreased when confronted with highly\noccluded and highly damaged face images. In this paper, a new wing-constrained\nsparse coding model(WCSC) and its weighted version(WWCSC) are introduced, so as\nto deal with the face recognition problem in complex circumstances, where the\nalternating direction method of multipliers (ADMM) algorithm is employed to\nsolve the corresponding minimization problems. In addition, performances of the\nproposed method are examined based on the four well-known facial databases,\nnamely the ORL facial database, the Yale facial database, the AR facial\ndatabase and the FERET facial database. Also, compared to the other methods in\nthe literatures, the WWCSC has a very high recognition rate even in complex\nsituations where face images have high occlusion or high damage, which\nillustrates the robustness of the WWCSC method in facial recognition.\n", "date": "2025-03-24", "categories": []}
{"id": "2308.02738", "title": "Exploring Part-Informed Visual-Language Learning for Person\n  Re-Identification", "abstract": "  Recently, visual-language learning (VLL) has shown great potential in\nenhancing visual-based person re-identification (ReID). Existing VLL-based ReID\nmethods typically focus on image-text feature alignment at the whole-body\nlevel, while neglecting supervision on fine-grained part features, thus lacking\nconstraints for local feature semantic consistency. To this end, we propose\nPart-Informed Visual-language Learning ($\\pi$-VL) to enhance fine-grained\nvisual features with part-informed language supervisions for ReID tasks.\nSpecifically, $\\pi$-VL introduces a human parsing-guided prompt tuning strategy\nand a hierarchical visual-language alignment paradigm to ensure within-part\nfeature semantic consistency. The former combines both identity labels and\nhuman parsing maps to constitute pixel-level text prompts, and the latter fuses\nmulti-scale visual features with a light-weight auxiliary head to perform\nfine-grained image-text alignment. As a plug-and-play and inference-free\nsolution, our $\\pi$-VL achieves performance comparable to or better than\nstate-of-the-art methods on four commonly used ReID benchmarks. Notably, it\nreports 91.0% Rank-1 and 76.9% mAP on the challenging MSMT17 database, without\nbells and whistles.\n", "date": "2023-08-04", "categories": []}
{"id": "2401.10090", "title": "Cross-Modality Perturbation Synergy Attack for Person Re-identification", "abstract": "  In recent years, there has been significant research focusing on addressing\nsecurity concerns in single-modal person re-identification (ReID) systems that\nare based on RGB images. However, the safety of cross-modality scenarios, which\nare more commonly encountered in practical applications involving images\ncaptured by infrared cameras, has not received adequate attention. The main\nchallenge in cross-modality ReID lies in effectively dealing with visual\ndifferences between different modalities. For instance, infrared images are\ntypically grayscale, unlike visible images that contain color information.\nExisting attack methods have primarily focused on the characteristics of the\nvisible image modality, overlooking the features of other modalities and the\nvariations in data distribution among different modalities. This oversight can\npotentially undermine the effectiveness of these methods in image retrieval\nacross diverse modalities. This study represents the first exploration into the\nsecurity of cross-modality ReID models and proposes a universal perturbation\nattack specifically designed for cross-modality ReID. This attack optimizes\nperturbations by leveraging gradients from diverse modality data, thereby\ndisrupting the discriminator and reinforcing the differences between\nmodalities. We conducted experiments on three widely used cross-modality\ndatasets, namely RegDB, SYSU, and LLCM. The results not only demonstrate the\neffectiveness of our method but also provide insights for future improvements\nin the robustness of cross-modality ReID systems. The code will be available at\nhttps://github.com/finger-monkey/cmps__attack.\n", "date": "2024-01-18", "categories": []}
{"id": "2503.15648", "title": "Cancelable Biometric Template Generation Using Random Feature Vector\n  Transformations", "abstract": "  Cancelable biometric schemes are designed to extract an identity-preserving,\nnon-invertible as well as revocable pseudo-identifier from biometric data.\nRecognition systems need to store only this pseudo-identifier, to avoid\ntampering and/or stealing of original biometric data during the recognition\nprocess. State-of-the-art cancelable schemes generate pseudo-identifiers by\ntransforming the original template using either user-specific salting or\nmany-to-one transformations. In addition to the performance concerns, most of\nsuch schemes are modality-specific and prone to reconstruction attacks as there\nare chances for unauthorized access to security-critical transformation keys. A\nnovel, modality-independent cancelable biometric scheme is proposed to overcome\nthese limitations. In this scheme, a cancelable template (pseudo identifier) is\ngenerated as a distance vector between multiple random transformations of the\nbiometric feature vector. These transformations were done by grouping feature\nvector components based on a set of user-specific random vectors. The proposed\nscheme nullifies the possibility of template reconstruction as the generated\ncancelable template contains only the distance values between the different\nrandom transformations of the feature vector and it does not store any details\nof the biometric template. The recognition performance of the proposed scheme\nis evaluated for face and fingerprint modalities. Equal Error Rate (EER) of 1.5\nis obtained for face and 1.7 is obtained for the fingerprint in the worst case.\n", "date": "2025-03-19", "categories": []}
{"id": "2407.18112", "title": "Keypoint Promptable Re-Identification", "abstract": "  Occluded Person Re-Identification (ReID) is a metric learning task that\ninvolves matching occluded individuals based on their appearance. While many\nstudies have tackled occlusions caused by objects, multi-person occlusions\nremain less explored. In this work, we identify and address a critical\nchallenge overlooked by previous occluded ReID methods: the Multi-Person\nAmbiguity (MPA) arising when multiple individuals are visible in the same\nbounding box, making it impossible to determine the intended ReID target among\nthe candidates. Inspired by recent work on prompting in vision, we introduce\nKeypoint Promptable ReID (KPR), a novel formulation of the ReID problem that\nexplicitly complements the input bounding box with a set of semantic keypoints\nindicating the intended target. Since promptable re-identification is an\nunexplored paradigm, existing ReID datasets lack the pixel-level annotations\nnecessary for prompting. To bridge this gap and foster further research on this\ntopic, we introduce Occluded-PoseTrack ReID, a novel ReID dataset with\nkeypoints labels, that features strong inter-person occlusions. Furthermore, we\nrelease custom keypoint labels for four popular ReID benchmarks. Experiments on\nperson retrieval, but also on pose tracking, demonstrate that our method\nsystematically surpasses previous state-of-the-art approaches on various\noccluded scenarios. Our code, dataset and annotations are available at\nhttps://github.com/VlSomers/keypoint_promptable_reidentification.\n", "date": "2024-07-25", "categories": []}
{"id": "2503.13495", "title": "TransECG: Leveraging Transformers for Explainable ECG Re-identification\n  Risk Analysis", "abstract": "  Electrocardiogram (ECG) signals are widely shared across multiple clinical\napplications for diagnosis, health monitoring, and biometric authentication.\nWhile valuable for healthcare, they also carry unique biometric identifiers\nthat pose privacy risks, especially when ECG data shared across multiple\nentities. These risks are amplified in shared environments, where\nre-identification threats can compromise patient privacy. Existing deep\nlearning re-identification models prioritize accuracy but lack explainability,\nmaking it challenging to understand how the unique biometric characteristics\nencoded within ECG signals are recognized and utilized for identification.\nWithout these insights, despite high accuracy, developing secure and trustable\nECG data-sharing frameworks remains difficult, especially in diverse,\nmulti-source environments. In this work, we introduce TransECG, a Vision\nTransformer (ViT)-based method that uses attention mechanisms to pinpoint\ncritical ECG segments associated with re-identification tasks like gender, age,\nand participant ID. Our approach demonstrates high accuracy (89.9% for gender,\n89.9% for age, and 88.6% for ID re-identification) across four real-world\ndatasets with 87 participants. Importantly, we provide key insights into ECG\ncomponents such as the R-wave, QRS complex, and P-Q interval in\nre-identification. For example, in the gender classification, the R wave\ncontributed 58.29% to the model's attention, while in the age classification,\nthe P-R interval contributed 46.29%. By combining high predictive performance\nwith enhanced explainability, TransECG provides a robust solution for\nprivacy-conscious ECG data sharing, supporting the development of secure and\ntrusted healthcare data environment.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.13573", "title": "Online Signature Verification based on the Lagrange formulation with 2D\n  and 3D robotic models", "abstract": "  Online Signature Verification commonly relies on function-based features,\nsuch as time-sampled horizontal and vertical coordinates, as well as the\npressure exerted by the writer, obtained through a digitizer. Although\ninferring additional information about the writers arm pose, kinematics, and\ndynamics based on digitizer data can be useful, it constitutes a challenge. In\nthis paper, we tackle this challenge by proposing a new set of features based\non the dynamics of online signatures. These new features are inferred through a\nLagrangian formulation, obtaining the sequences of generalized coordinates and\ntorques for 2D and 3D robotic arm models. By combining kinematic and dynamic\nrobotic features, our results demonstrate their significant effectiveness for\nonline automatic signature verification and achieving state-of-the-art results\nwhen integrated into deep learning models.\n", "date": "2025-03-17", "categories": []}
{"id": "2503.14138", "title": "Exploring Disparity-Accuracy Trade-offs in Face Recognition Systems: The\n  Role of Datasets, Architectures, and Loss Functions", "abstract": "  Automated Face Recognition Systems (FRSs), developed using deep learning\nmodels, are deployed worldwide for identity verification and facial attribute\nanalysis. The performance of these models is determined by a complex\ninterdependence among the model architecture, optimization/loss function and\ndatasets. Although FRSs have surpassed human-level accuracy, they continue to\nbe disparate against certain demographics. Due to the ubiquity of applications,\nit is extremely important to understand the impact of the three components --\nmodel architecture, loss function and face image dataset on the\naccuracy-disparity trade-off to design better, unbiased platforms. In this\nwork, we perform an in-depth analysis of three FRSs for the task of gender\nprediction, with various architectural modifications resulting in ten\ndeep-learning models coupled with four loss functions and benchmark them on\nseven face datasets across 266 evaluation configurations. Our results show that\nall three components have an individual as well as a combined impact on both\naccuracy and disparity. We identify that datasets have an inherent property\nthat causes them to perform similarly across models, independent of the choice\nof loss functions. Moreover, the choice of dataset determines the model's\nperceived bias -- the same model reports bias in opposite directions for three\ngender-balanced datasets of ``in-the-wild'' face images of popular individuals.\nStudying the facial embeddings shows that the models are unable to generalize a\nuniform definition of what constitutes a ``female face'' as opposed to a ``male\nface'', due to dataset diversity. We provide recommendations to model\ndevelopers on using our study as a blueprint for model development and\nsubsequent deployment.\n", "date": "2025-03-18", "categories": []}
{"id": "2411.16832", "title": "Edit Away and My Face Will not Stay: Personal Biometric Defense against\n  Malicious Generative Editing", "abstract": "  Recent advancements in diffusion models have made generative image editing\nmore accessible, enabling creative edits but raising ethical concerns,\nparticularly regarding malicious edits to human portraits that threaten privacy\nand identity security. Existing protection methods primarily rely on\nadversarial perturbations to nullify edits but often fail against diverse\nediting requests. We propose FaceLock, a novel approach to portrait protection\nthat optimizes adversarial perturbations to destroy or significantly alter\nbiometric information, rendering edited outputs biometrically unrecognizable.\nFaceLock integrates facial recognition and visual perception into perturbation\noptimization to provide robust protection against various editing attempts. We\nalso highlight flaws in commonly used evaluation metrics and reveal how they\ncan be manipulated, emphasizing the need for reliable assessments of\nprotection. Experiments show FaceLock outperforms baselines in defending\nagainst malicious edits and is robust against purification techniques. Ablation\nstudies confirm its stability and broad applicability across diffusion-based\nediting algorithms. Our work advances biometric defense and sets the foundation\nfor privacy-preserving practices in image editing. The code is available at:\nhttps://github.com/taco-group/FaceLock.\n", "date": "2024-11-25", "categories": []}
{"id": "2503.08121", "title": "AG-VPReID: A Challenging Large-Scale Benchmark for Aerial-Ground\n  Video-based Person Re-Identification", "abstract": "  We introduce AG-VPReID, a new large-scale dataset for aerial-ground\nvideo-based person re-identification (ReID) that comprises 6,632 subjects,\n32,321 tracklets and over 9.6 million frames captured by drones (altitudes\nranging from 15-120m), CCTV, and wearable cameras. This dataset offers a\nreal-world benchmark for evaluating the robustness to significant viewpoint\nchanges, scale variations, and resolution differences in cross-platform\naerial-ground settings. In addition, to address these challenges, we propose\nAG-VPReID-Net, an end-to-end framework composed of three complementary streams:\n(1) an Adapted Temporal-Spatial Stream addressing motion pattern\ninconsistencies and facilitating temporal feature learning, (2) a Normalized\nAppearance Stream leveraging physics-informed techniques to tackle resolution\nand appearance changes, and (3) a Multi-Scale Attention Stream handling scale\nvariations across drone altitudes. We integrate visual-semantic cues from all\nstreams to form a robust, viewpoint-invariant whole-body representation.\nExtensive experiments demonstrate that AG-VPReID-Net outperforms\nstate-of-the-art approaches on both our new dataset and existing video-based\nReID benchmarks, showcasing its effectiveness and generalizability.\nNevertheless, the performance gap observed on AG-VPReID across all methods\nunderscores the dataset's challenging nature. The dataset, code and trained\nmodels are available at https://github.com/agvpreid25/AG-VPReID-Net.\n", "date": "2025-03-11", "categories": []}
{"id": "2503.11930", "title": "Generating a Biometrically Unique and Realistic Iris Database", "abstract": "  The use of the iris as a biometric identifier has increased dramatically over\nthe last 30 years, prompting privacy and security concerns about the use of\niris images in research. It can be difficult to acquire iris image databases\ndue to ethical concerns, and this can be a barrier for those performing\nbiometrics research. In this paper, we describe and show how to create a\ndatabase of realistic, biometrically unidentifiable colored iris images by\ntraining a diffusion model within an open-source diffusion framework. Not only\nwere we able to verify that our model is capable of creating iris textures that\nare biometrically unique from the training data, but we were also able to\nverify that our model output creates a full distribution of realistic iris\npigmentations. We highlight the fact that the utility of diffusion networks to\nachieve these criteria with relative ease, warrants additional research in its\nuse within the context of iris database generation and presentation attack\nsecurity.\n", "date": "2025-03-15", "categories": []}
{"id": "2503.12472", "title": "Diffusion-based Synthetic Data Generation for Visible-Infrared Person\n  Re-Identification", "abstract": "  The performance of models is intricately linked to the abundance of training\ndata. In Visible-Infrared person Re-IDentification (VI-ReID) tasks, collecting\nand annotating large-scale images of each individual under various cameras and\nmodalities is tedious, time-expensive, costly and must comply with data\nprotection laws, posing a severe challenge in meeting dataset requirements.\nCurrent research investigates the generation of synthetic data as an efficient\nand privacy-ensuring alternative to collecting real data in the field. However,\na specific data synthesis technique tailored for VI-ReID models has yet to be\nexplored. In this paper, we present a novel data generation framework, dubbed\nDiffusion-based VI-ReID data Expansion (DiVE), that automatically obtain\nmassive RGB-IR paired images with identity preserving by decoupling identity\nand modality to improve the performance of VI-ReID models. Specifically,\nidentity representation is acquired from a set of samples sharing the same ID,\nwhereas the modality of images is learned by fine-tuning the Stable Diffusion\n(SD) on modality-specific data. DiVE extend the text-driven image synthesis to\nidentity-preserving RGB-IR multimodal image synthesis. This approach\nsignificantly reduces data collection and annotation costs by directly\nincorporating synthetic data into ReID model training. Experiments have\ndemonstrated that VI-ReID models trained on synthetic data produced by DiVE\nconsistently exhibit notable enhancements. In particular, the state-of-the-art\nmethod, CAJ, trained with synthetic images, achieves an improvement of about\n$9\\%$ in mAP over the baseline on the LLCM dataset. Code:\nhttps://github.com/BorgDiven/DiVE\n", "date": "2025-03-16", "categories": []}
{"id": "2503.12786", "title": "Privacy-Preserving Biometric Verification with Handwritten Random Digit\n  String", "abstract": "  Handwriting verification has stood as a steadfast identity authentication\nmethod for decades. However, this technique risks potential privacy breaches\ndue to the inclusion of personal information in handwritten biometrics such as\nsignatures. To address this concern, we propose using the Random Digit String\n(RDS) for privacy-preserving handwriting verification. This approach allows\nusers to authenticate themselves by writing an arbitrary digit sequence,\neffectively ensuring privacy protection. To evaluate the effectiveness of RDS,\nwe construct a new HRDS4BV dataset composed of online naturally handwritten\nRDS. Unlike conventional handwriting, RDS encompasses unconstrained and\nvariable content, posing significant challenges for modeling consistent\npersonal writing style. To surmount this, we propose the Pattern Attentive\nVErification Network (PAVENet), along with a Discriminative Pattern Mining\n(DPM) module. DPM adaptively enhances the recognition of consistent and\ndiscriminative writing patterns, thus refining handwriting style\nrepresentation. Through comprehensive evaluations, we scrutinize the\napplicability of online RDS verification and showcase a pronounced\noutperformance of our model over existing methods. Furthermore, we discover a\nnoteworthy forgery phenomenon that deviates from prior findings and discuss its\npositive impact in countering malicious impostor attacks. Substantially, our\nwork underscores the feasibility of privacy-preserving biometric verification\nand propels the prospects of its broader acceptance and application.\n", "date": "2025-03-17", "categories": []}
{"id": "2503.10931", "title": "Multi-Domain Biometric Recognition using Body Embeddings", "abstract": "  Biometric recognition becomes increasingly challenging as we move away from\nthe visible spectrum to infrared imagery, where domain discrepancies\nsignificantly impact identification performance. In this paper, we show that\nbody embeddings perform better than face embeddings for cross-spectral person\nidentification in medium-wave infrared (MWIR) and long-wave infrared (LWIR)\ndomains. Due to the lack of multi-domain datasets, previous research on\ncross-spectral body identification - also known as Visible-Infrared Person\nRe-Identification (VI-ReID) - has primarily focused on individual infrared\nbands, such as near-infrared (NIR) or LWIR, separately. We address the\nmulti-domain body recognition problem using the IARPA Janus Benchmark\nMulti-Domain Face (IJB-MDF) dataset, which enables matching of short-wave\ninfrared (SWIR), MWIR, and LWIR images against RGB (VIS) images. We leverage a\nvision transformer architecture to establish benchmark results on the IJB-MDF\ndataset and, through extensive experiments, provide valuable insights into the\ninterrelation of infrared domains, the adaptability of VIS-pretrained models,\nthe role of local semantic features in body-embeddings, and effective training\nstrategies for small datasets. Additionally, we show that finetuning a body\nmodel, pretrained exclusively on VIS data, with a simple combination of\ncross-entropy and triplet losses achieves state-of-the-art mAP scores on the\nLLCM dataset.\n", "date": "2025-03-13", "categories": []}
{"id": "2403.08277", "title": "VIGFace: Virtual Identity Generation for Privacy-Free Face Recognition", "abstract": "  Deep learning-based face recognition continues to face challenges due to its\nreliance on huge datasets obtained from web crawling, which can be costly to\ngather and raise significant real-world privacy concerns. To address this\nissue, we propose VIGFace, a novel framework capable of generating synthetic\nfacial images. Our idea originates from pre-assigning virtual identities in the\nfeature space. Initially, we train the face recognition model using a real face\ndataset and create a feature space for both real and virtual identities, where\nvirtual prototypes are orthogonal to other prototypes. Subsequently, we train\nthe diffusion model based on the established feature space, enabling it to\ngenerate authentic human face images from real prototypes and synthesize\nvirtual face images from virtual prototypes. Our proposed framework provides\ntwo significant benefits. Firstly, it shows clear separability between existing\nindividuals and virtual face images, allowing one to create synthetic images\nwith confidence and without concerns about privacy and portrait rights.\nSecondly, it ensures improved performance through data augmentation by\nincorporating real existing images. Extensive experiments demonstrate the\nsuperiority of our virtual face dataset and framework, outperforming the\nprevious state-of-the-art on various face recognition benchmarks.\n", "date": "2024-03-13", "categories": []}
{"id": "2503.00938", "title": "From Poses to Identity: Training-Free Person Re-Identification via\n  Feature Centralization", "abstract": "  Person re-identification (ReID) aims to extract accurate identity\nrepresentation features. However, during feature extraction, individual samples\nare inevitably affected by noise (background, occlusions, and model\nlimitations). Considering that features from the same identity follow a normal\ndistribution around identity centers after training, we propose a Training-Free\nFeature Centralization ReID framework (Pose2ID) by aggregating the same\nidentity features to reduce individual noise and enhance the stability of\nidentity representation, which preserves the feature's original distribution\nfor following strategies such as re-ranking. Specifically, to obtain samples of\nthe same identity, we introduce two components: Identity-Guided Pedestrian\nGeneration: by leveraging identity features to guide the generation process, we\nobtain high-quality images with diverse poses, ensuring identity consistency\neven in complex scenarios such as infrared, and occlusion. Neighbor Feature\nCentralization: it explores each sample's potential positive samples from its\nneighborhood. Experiments demonstrate that our generative model exhibits strong\ngeneralization capabilities and maintains high identity consistency. With the\nFeature Centralization framework, we achieve impressive performance even with\nan ImageNet pre-trained model without ReID training, reaching mAP/Rank-1 of\n52.81/78.92 on Market1501. Moreover, our method sets new state-of-the-art\nresults across standard, cross-modality, and occluded ReID tasks, showcasing\nstrong adaptability.\n", "date": "2025-03-02", "categories": []}
{"id": "2503.06187", "title": "MSConv: Multiplicative and Subtractive Convolution for Face Recognition", "abstract": "  In Neural Networks, there are various methods of feature fusion. Different\nstrategies can significantly affect the effectiveness of feature\nrepresentation, consequently influencing the ability of model to extract\nrepresentative and discriminative features. In the field of face recognition,\ntraditional feature fusion methods include feature concatenation and feature\naddition. Recently, various attention mechanism-based fusion strategies have\nemerged. However, we found that these methods primarily focus on the important\nfeatures in the image, referred to as salient features in this paper, while\nneglecting another equally important set of features for image recognition\ntasks, which we term differential features. This may cause the model to\noverlook critical local differences when dealing with complex facial samples.\nTherefore, in this paper, we propose an efficient convolution module called\nMSConv (Multiplicative and Subtractive Convolution), designed to balance the\nlearning of model about salient and differential features. Specifically, we\nemploy multi-scale mixed convolution to capture both local and broader\ncontextual information from face images, and then utilize Multiplication\nOperation (MO) and Subtraction Operation (SO) to extract salient and\ndifferential features, respectively. Experimental results demonstrate that by\nintegrating both salient and differential features, MSConv outperforms models\nthat only focus on salient features.\n", "date": "2025-03-08", "categories": []}
{"id": "2503.06839", "title": "AttFC: Attention Fully-Connected Layer for Large-Scale Face Recognition\n  with One GPU", "abstract": "  Nowadays, with the advancement of deep neural networks (DNNs) and the\navailability of large-scale datasets, the face recognition (FR) model has\nachieved exceptional performance. However, since the parameter magnitude of the\nfully connected (FC) layer directly depends on the number of identities in the\ndataset. If training the FR model on large-scale datasets, the size of the\nmodel parameter will be excessively huge, leading to substantial demand for\ncomputational resources, such as time and memory. This paper proposes the\nattention fully connected (AttFC) layer, which could significantly reduce\ncomputational resources. AttFC employs an attention loader to generate the\ngenerative class center (GCC), and dynamically store the class center with\nDynamic Class Container (DCC). DCC only stores a small subset of all class\ncenters in FC, thus its parameter count is substantially less than the FC\nlayer. Also, training face recognition models on large-scale datasets with one\nGPU often encounter out-of-memory (OOM) issues. AttFC overcomes this and\nachieves comparable performance to state-of-the-art methods.\n", "date": "2025-03-10", "categories": []}
{"id": "2503.04837", "title": "FedPalm: A General Federated Learning Framework for Closed- and Open-Set\n  Palmprint Verification", "abstract": "  Current deep learning (DL)-based palmprint verification models rely on\ncentralized training with large datasets, which raises significant privacy\nconcerns due to biometric data's sensitive and immutable nature. Federated\nlearning~(FL), a privacy-preserving distributed learning paradigm, offers a\ncompelling alternative by enabling collaborative model training without the\nneed for data sharing. However, FL-based palmprint verification faces critical\nchallenges, including data heterogeneity from diverse identities and the\nabsence of standardized evaluation benchmarks. This paper addresses these gaps\nby establishing a comprehensive benchmark for FL-based palmprint verification,\nwhich explicitly defines and evaluates two practical scenarios: closed-set and\nopen-set verification. We propose FedPalm, a unified FL framework that balances\nlocal adaptability with global generalization. Each client trains a\npersonalized textural expert tailored to local data and collaboratively\ncontributes to a shared global textural expert for extracting generalized\nfeatures. To further enhance verification performance, we introduce a Textural\nExpert Interaction Module that dynamically routes textural features among\nexperts to generate refined side textural features. Learnable parameters are\nemployed to model relationships between original and side features, fostering\ncross-texture-expert interaction and improving feature discrimination.\nExtensive experiments validate the effectiveness of FedPalm, demonstrating\nrobust performance across both scenarios and providing a promising foundation\nfor advancing FL-based palmprint verification research.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.05247", "title": "ColFigPhotoAttnNet: Reliable Finger Photo Presentation Attack Detection\n  Leveraging Window-Attention on Color Spaces", "abstract": "  Finger photo Presentation Attack Detection (PAD) can significantly strengthen\nsmartphone device security. However, these algorithms are trained to detect\ncertain types of attacks. Furthermore, they are designed to operate on images\nacquired by specific capture devices, leading to poor generalization and a lack\nof robustness in handling the evolving nature of mobile hardware. The proposed\ninvestigation is the first to systematically analyze the performance\ndegradation of existing deep learning PAD systems, convolutional and\ntransformers, in cross-capture device settings. In this paper, we introduce the\nColFigPhotoAttnNet architecture designed based on window attention on color\nchannels, followed by the nested residual network as the predictor to achieve a\nreliable PAD. Extensive experiments using various capture devices, including\niPhone13 Pro, GooglePixel 3, Nokia C5, and OnePlusOne, were carried out to\nevaluate the performance of proposed and existing methods on three publicly\navailable databases. The findings underscore the effectiveness of our approach.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.05514", "title": "Noise-Robust Radio Frequency Fingerprint Identification Using Denoise\n  Diffusion Model", "abstract": "  Securing Internet of Things (IoT) devices presents increasing challenges due\nto their limited computational and energy resources. Radio Frequency\nFingerprint Identification (RFFI) emerges as a promising authentication\ntechnique to identify wireless devices through hardware impairments. RFFI\nperformance under low signal-to-noise ratio (SNR) scenarios is significantly\ndegraded because the minute hardware features can be easily swamped in noise.\nIn this paper, we leveraged the diffusion model to effectively restore the RFF\nunder low SNR scenarios. Specifically, we trained a powerful noise predictor\nand tailored a noise removal algorithm to effectively reduce the noise level in\nthe received signal and restore the device fingerprints. We used Wi-Fi as a\ncase study and created a testbed involving 6 commercial off-the-shelf Wi-Fi\ndongles and a USRP N210 software-defined radio (SDR) platform. We conducted\nexperimental evaluations on various SNR scenarios. The experimental results\nshow that the proposed algorithm can improve the classification accuracy by up\nto 34.9%.\n", "date": "2025-03-07", "categories": []}
{"id": "2503.03204", "title": "Find Matching Faces Based On Face Parameters", "abstract": "  This paper presents an innovative approach that enables the user to find\nmatching faces based on the user-selected face parameters. Through gradio-based\nuser interface, the users can interactively select the face parameters they\nwant in their desired partner. These user-selected face parameters are\ntransformed into a text prompt which is used by the Text-To-Image generation\nmodel to generate a realistic face image. Further, the generated image along\nwith the images downloaded from the Jeevansathi.com are processed through face\ndetection and feature extraction model, which results in high dimensional\nvector embedding of 512 dimensions. The vector embeddings generated from the\ndownloaded images are stored into vector database. Now, the similarity search\nis carried out between the vector embedding of generated image and the stored\nvector embeddings. As a result, it displays the top five similar faces based on\nthe user-selected face parameters. This contribution holds a significant\npotential to turn into a high-quality personalized face matching tool.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03330", "title": "Automated Attendee Recognition System for Large-Scale Social Events or\n  Conference Gathering", "abstract": "  Manual attendance tracking at large-scale events, such as marriage functions\nor conferences, is often inefficient and prone to human error. To address this\nchallenge, we propose an automated, cloud-based attendance tracking system that\nuses cameras mounted at the entrance and exit gates. The mounted cameras\ncontinuously capture video and send the video data to cloud services to perform\nreal-time face detection and recognition. Unlike existing solutions, our system\naccurately identifies attendees even when they are not looking directly at the\ncamera, allowing natural movements, such as looking around or talking while\nwalking. To the best of our knowledge, this is the first system to achieve high\nrecognition rates under such dynamic conditions. Our system demonstrates\noverall 90% accuracy, with each video frame processed in 5 seconds, ensuring\nreal time operation without frame loss. In addition, notifications are sent\npromptly to security personnel within the same latency. This system achieves\n100% accuracy for individuals without facial obstructions and successfully\nrecognizes all attendees appearing within the camera's field of view, providing\na robust solution for attendee recognition in large-scale social events.\n", "date": "2025-03-05", "categories": []}
{"id": "2503.03528", "title": "AdaSin: Enhancing Hard Sample Metrics with Dual Adaptive Penalty for\n  Face Recognition", "abstract": "  In recent years, the emergence of deep convolutional neural networks has\npositioned face recognition as a prominent research focus in computer vision.\nTraditional loss functions, such as margin-based, hard-sample mining-based, and\nhybrid approaches, have achieved notable performance improvements, with some\nleveraging curriculum learning to optimize training. However, these methods\noften fall short in effectively quantifying the difficulty of hard samples. To\naddress this, we propose Adaptive Sine (AdaSin) loss function, which introduces\nthe sine of the angle between a sample's embedding feature and its ground-truth\nclass center as a novel difficulty metric. This metric enables precise and\neffective penalization of hard samples. By incorporating curriculum learning,\nthe model dynamically adjusts classification boundaries across different\ntraining stages. Unlike previous adaptive-margin loss functions, AdaSin\nintroduce a dual adaptive penalty, applied to both the positive and negative\ncosine similarities of hard samples. This design imposes stronger constraints,\nenhancing intra-class compactness and inter-class separability. The combination\nof the dual adaptive penalty and curriculum learning is guided by a\nwell-designed difficulty metric. It enables the model to focus more effectively\non hard samples in later training stages, and lead to the extraction of highly\ndiscriminative face features. Extensive experiments across eight benchmarks\ndemonstrate that AdaSin achieves superior accuracy compared to other\nstate-of-the-art methods.\n", "date": "2025-03-05", "categories": []}
{"id": "2310.07584", "title": "Centrality of the Fingerprint Core Location", "abstract": "  Fingerprints have long been recognized as a unique and reliable means of\npersonal identification. Central to the analysis and enhancement of\nfingerprints is the concept of the fingerprint core. Although the location of\nthe core is used in many applications, to the best of our knowledge, this study\nis the first to investigate the empirical distribution of the core over a\nlarge, combined dataset of rolled, as well as plain fingerprint recordings. We\nidentify and investigate the extent of incomplete rolling during the rolled\nfingerprint acquisition and investigate the centrality of the core. After\ncorrecting for the incomplete rolling, we find that the core deviates from the\nfingerprint center by 5.7% $\\pm$ 5.2% to 7.6% $\\pm$ 6.9%, depending on the\nfinger. Additionally, we find that the assumption of normal distribution of the\ncore position of plain fingerprint recordings cannot be rejected, but for\nrolled ones it can. Therefore, we use a multi-step process to find the\ndistribution of the rolled fingerprint recordings. The process consists of an\nAnderson-Darling normality test, the Bayesian Information Criterion to reduce\nthe number of possible candidate distributions and finally a Generalized Monte\nCarlo goodness-of-fit procedure to find the best fitting distribution. We find\nthe non-central Fischer distribution best describes the cores' horizontal\npositions. Finally, we investigate the correlation between mean core position\noffset and the NFIQ 2 score and find that the NFIQ 2 prefers rolled fingerprint\nrecordings where the core sits slightly below the fingerprint center.\n", "date": "2023-10-11", "categories": []}
{"id": "2503.02547", "title": "PVTree: Realistic and Controllable Palm Vein Generation for Recognition\n  Tasks", "abstract": "  Palm vein recognition is an emerging biometric technology that offers\nenhanced security and privacy. However, acquiring sufficient palm vein data for\ntraining deep learning-based recognition models is challenging due to the high\ncosts of data collection and privacy protection constraints. This has led to a\ngrowing interest in generating pseudo-palm vein data using generative models.\nExisting methods, however, often produce unrealistic palm vein patterns or\nstruggle with controlling identity and style attributes. To address these\nissues, we propose a novel palm vein generation framework named PVTree. First,\nthe palm vein identity is defined by a complex and authentic 3D palm vascular\ntree, created using an improved Constrained Constructive Optimization (CCO)\nalgorithm. Second, palm vein patterns of the same identity are generated by\nprojecting the same 3D vascular tree into 2D images from different views and\nconverting them into realistic images using a generative model. As a result,\nPVTree satisfies the need for both identity consistency and intra-class\ndiversity. Extensive experiments conducted on several publicly available\ndatasets demonstrate that our proposed palm vein generation method surpasses\nexisting methods and achieves a higher TAR@FAR=1e-4 under the 1:1 Open-set\nprotocol. To the best of our knowledge, this is the first time that the\nperformance of a recognition model trained on synthetic palm vein data exceeds\nthat of the recognition model trained on real data, which indicates that palm\nvein image generation research has a promising future.\n", "date": "2025-03-04", "categories": []}
{"id": "2209.04821", "title": "Local-Aware Global Attention Network for Person Re-Identification Based\n  on Body and Hand Images", "abstract": "  Learning representative, robust and discriminative information from images is\nessential for effective person re-identification (Re-Id). In this paper, we\npropose a compound approach for end-to-end discriminative deep feature learning\nfor person Re-Id based on both body and hand images. We carefully design the\nLocal-Aware Global Attention Network (LAGA-Net), a multi-branch deep network\narchitecture consisting of one branch for spatial attention, one branch for\nchannel attention, one branch for global feature representations and another\nbranch for local feature representations. The attention branches focus on the\nrelevant features of the image while suppressing the irrelevant backgrounds. In\norder to overcome the weakness of the attention mechanisms, equivariant to\npixel shuffling, we integrate relative positional encodings into the spatial\nattention module to capture the spatial positions of pixels. The global branch\nintends to preserve the global context or structural information. For the the\nlocal branch, which intends to capture the fine-grained information, we perform\nuniform partitioning to generate stripes on the conv-layer horizontally. We\nretrieve the parts by conducting a soft partition without explicitly\npartitioning the images or requiring external cues such as pose estimation. A\nset of ablation study shows that each component contributes to the increased\nperformance of the LAGA-Net. Extensive evaluations on four popular body-based\nperson Re-Id benchmarks and two publicly available hand datasets demonstrate\nthat our proposed method consistently outperforms existing state-of-the-art\nmethods.\n", "date": "2022-09-11", "categories": []}
{"id": "2303.15263", "title": "Joint Person Identity, Gender and Age Estimation from Hand Images using\n  Deep Multi-Task Representation Learning", "abstract": "  In this paper, we propose a multi-task representation learning framework to\njointly estimate the identity, gender and age of individuals from their hand\nimages for the purpose of criminal investigations since the hand images are\noften the only available information in cases of serious crime such as sexual\nabuse. We investigate different up-to-date deep learning architectures and\ncompare their performance for joint estimation of identity, gender and age from\nhand images of perpetrators of serious crime. To simplify the age prediction,\nwe create age groups for the age estimation. We make extensive evaluations and\ncomparisons of both convolution-based and transformer-based deep learning\narchitectures on a publicly available 11k hands dataset. Our experimental\nanalysis shows that it is possible to efficiently estimate not only identity\nbut also other attributes such as gender and age of suspects jointly from hand\nimages for criminal investigations, which is crucial in assisting international\npolice forces in the court to identify and convict abusers.\n", "date": "2023-03-27", "categories": []}
{"id": "2411.08470", "title": "HyperFace: Generating Synthetic Face Recognition Datasets by Exploring\n  Face Embedding Hypersphere", "abstract": "  Face recognition datasets are often collected by crawling Internet and\nwithout individuals' consents, raising ethical and privacy concerns. Generating\nsynthetic datasets for training face recognition models has emerged as a\npromising alternative. However, the generation of synthetic datasets remains\nchallenging as it entails adequate inter-class and intra-class variations.\nWhile advances in generative models have made it easier to increase intra-class\nvariations in face datasets (such as pose, illumination, etc.), generating\nsufficient inter-class variation is still a difficult task. In this paper, we\nformulate the dataset generation as a packing problem on the embedding space\n(represented on a hypersphere) of a face recognition model and propose a new\nsynthetic dataset generation approach, called HyperFace. We formalize our\npacking problem as an optimization problem and solve it with a gradient\ndescent-based approach. Then, we use a conditional face generator model to\nsynthesize face images from the optimized embeddings. We use our generated\ndatasets to train face recognition models and evaluate the trained models on\nseveral benchmarking real datasets. Our experimental results show that models\ntrained with HyperFace achieve state-of-the-art performance in training face\nrecognition using synthetic datasets.\n", "date": "2024-11-13", "categories": []}
{"id": "2503.01612", "title": "Robust Palm-Vein Recognition Using the MMD Filter: Improving SIFT-Based\n  Feature Matching", "abstract": "  A major challenge with palm vein images is that slight movements of the\nfingers and thumb, or variations in hand posture, can stretch the skin in\ndifferent areas and alter the vein patterns. This can result in an infinite\nnumber of variations in palm vein images for a given individual. This paper\nintroduces a novel filtering technique for SIFT-based feature matching, known\nas the Mean and Median Distance (MMD) Filter. This method evaluates the\ndifferences in keypoint coordinates and computes the mean and median in each\ndirection to eliminate incorrect matches. Experiments conducted on the 850nm\nsubset of the CASIA dataset indicate that the proposed MMD filter effectively\npreserves correct points while reducing false positives detected by other\nfiltering methods. A comparison with existing SIFT-based palm vein recognition\nsystems demonstrates that the proposed MMD filter delivers outstanding\nperformance, achieving lower Equal Error Rate (EER) values. This article\npresents an extended author's version based on our previous work, A Keypoint\nFiltering Method for SIFT based Palm-Vein Recognition.\n", "date": "2025-03-03", "categories": []}
{"id": "2502.19803", "title": "UIFace: Unleashing Inherent Model Capabilities to Enhance Intra-Class\n  Diversity in Synthetic Face Recognition", "abstract": "  Face recognition (FR) stands as one of the most crucial applications in\ncomputer vision. The accuracy of FR models has significantly improved in recent\nyears due to the availability of large-scale human face datasets. However,\ndirectly using these datasets can inevitably lead to privacy and legal\nproblems. Generating synthetic data to train FR models is a feasible solution\nto circumvent these issues. While existing synthetic-based face recognition\nmethods have made significant progress in generating identity-preserving\nimages, they are severely plagued by context overfitting, resulting in a lack\nof intra-class diversity of generated images and poor face recognition\nperformance. In this paper, we propose a framework to Unleash Inherent\ncapability of the model to enhance intra-class diversity for synthetic face\nrecognition, shortened as UIFace. Our framework first trains a diffusion model\nthat can perform sampling conditioned on either identity contexts or a\nlearnable empty context. The former generates identity-preserving images but\nlacks variations, while the latter exploits the model's intrinsic ability to\nsynthesize intra-class-diversified images but with random identities. Then we\nadopt a novel two-stage sampling strategy during inference to fully leverage\nthe strengths of both types of contexts, resulting in images that are diverse\nas well as identitypreserving. Moreover, an attention injection module is\nintroduced to further augment the intra-class variations by utilizing attention\nmaps from the empty context to guide the sampling process in ID-conditioned\ngeneration. Experiments show that our method significantly surpasses previous\napproaches with even less training data and half the size of synthetic dataset.\nThe proposed UIFace even achieves comparable performance with FR models trained\non real datasets when we further increase the number of synthetic identities.\n", "date": "2025-02-27", "categories": []}
{"id": "2502.18514", "title": "CipherFace: A Fully Homomorphic Encryption-Driven Framework for Secure\n  Cloud-Based Facial Recognition", "abstract": "  Facial recognition systems rely on embeddings to represent facial images and\ndetermine identity by verifying if the distance between embeddings is below a\npre-tuned threshold. While embeddings are not reversible to original images,\nthey still contain sensitive information, making their security critical.\nTraditional encryption methods like AES are limited in securely utilizing cloud\ncomputational power for distance calculations. Homomorphic Encryption, allowing\ncalculations on encrypted data, offers a robust alternative. This paper\nintroduces CipherFace, a homomorphic encryption-driven framework for secure\ncloud-based facial recognition, which we have open-sourced at\nhttp://github.com/serengil/cipherface. By leveraging FHE, CipherFace ensures\nthe privacy of embeddings while utilizing the cloud for efficient distance\ncomputation. Furthermore, we propose a novel encrypted distance computation\nmethod for both Euclidean and Cosine distances, addressing key challenges in\nperforming secure similarity calculations on encrypted data. We also conducted\nexperiments with different facial recognition models, various embedding sizes,\nand cryptosystem configurations, demonstrating the scalability and\neffectiveness of CipherFace in real-world applications.\n", "date": "2025-02-22", "categories": []}
{"id": "2502.15934", "title": "Dissecting Human Body Representations in Deep Networks Trained for\n  Person Identification", "abstract": "  Long-term body identification algorithms have emerged recently with the\nincreased availability of high-quality training data. We seek to fill knowledge\ngaps about these models by analyzing body image embeddings from four body\nidentification networks trained with 1.9 million images across 4,788 identities\nand 9 databases. By analyzing a diverse range of architectures (ViT, SWIN-ViT,\nCNN, and linguistically primed CNN), we first show that the face contributes to\nthe accuracy of body identification algorithms and that these algorithms can\nidentify faces to some extent -- with no explicit face training. Second, we\nshow that representations (embeddings) generated by body identification\nalgorithms encode information about gender, as well as image-based information\nincluding view (yaw) and even the dataset from which the image originated.\nThird, we demonstrate that identification accuracy can be improved without\nadditional training by operating directly and selectively on the learned\nembedding space. Leveraging principal component analysis (PCA), identity\ncomparisons were consistently more accurate in subspaces that eliminated\ndimensions that explained large amounts of variance. These three findings were\nsurprisingly consistent across architectures and test datasets. This work\nrepresents the first analysis of body representations produced by long-term\nre-identification networks trained on challenging unconstrained datasets.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.14996", "title": "A Rapid Test for Accuracy and Bias of Face Recognition Technology", "abstract": "  Measuring the accuracy of face recognition (FR) systems is essential for\nimproving performance and ensuring responsible use. Accuracy is typically\nestimated using large annotated datasets, which are costly and difficult to\nobtain. We propose a novel method for 1:1 face verification that benchmarks FR\nsystems quickly and without manual annotation, starting from approximate labels\n(e.g., from web search results). Unlike previous methods for training set label\ncleaning, ours leverages the embedding representation of the models being\nevaluated, achieving high accuracy in smaller-sized test datasets. Our approach\nreliably estimates FR accuracy and ranking, significantly reducing the time and\ncost of manual labeling. We also introduce the first public benchmark of five\nFR cloud services, revealing demographic biases, particularly lower accuracy\nfor Asian women. Our rapid test method can democratize FR testing, promoting\nscrutiny and responsible use of the technology. Our method is provided as a\npublicly accessible tool at https://github.com/caltechvisionlab/frt-rapid-test\n", "date": "2025-02-20", "categories": []}
{"id": "2502.15632", "title": "Continual Person Identification using Footstep-Induced Floor Vibrations\n  on Heterogeneous Floor Structures", "abstract": "  Person identification is important for smart buildings to provide\npersonalized services such as health monitoring, activity tracking, and\npersonnel management. However, previous person identification relies on\npre-collected data from everyone, which is impractical in many buildings and\npublic facilities in which visitors are typically expected. This calls for a\ncontinual person identification system that gradually learns people's\nidentities on the fly. Existing studies use cameras to achieve this goal, but\nthey require direct line-of-sight and also have raised privacy concerns in\npublic. Other modalities such as wearables and pressure mats are limited by the\nrequirement of device-carrying or dense deployment. Thus, prior studies\nintroduced footstep-induced structural vibration sensing, which is\nnon-intrusive and perceived as more privacy-friendly. However, this approach\nhas a significant challenge: the high variability of vibration data due to\nstructural heterogeneity and human gait variations, which makes online person\nidentification algorithms perform poorly. In this paper, we characterize the\nvariability in footstep-induced structural vibration data for accurate online\nperson identification. To achieve this, we quantify and decompose different\nsources of variability and then design a feature transformation function to\nreduce the variability within each person's data to make different people's\ndata more separable. We evaluate our approach through field experiments with 20\npeople. The results show a 70% variability reduction and a 90% accuracy for\nonline person identification.\n", "date": "2025-02-21", "categories": []}
{"id": "2502.11023", "title": "DT4ECG: A Dual-Task Learning Framework for ECG-Based Human Identity\n  Recognition and Human Activity Detection", "abstract": "  This article introduces DT4ECG, an innovative dual-task learning framework\nfor Electrocardiogram (ECG)-based human identity recognition and activity\ndetection. The framework employs a robust one-dimensional convolutional neural\nnetwork (1D-CNN) backbone integrated with residual blocks to extract\ndiscriminative ECG features. To enhance feature representation, we propose a\nnovel Sequence Channel Attention (SCA) mechanism, which combines channel-wise\nand sequential context attention to prioritize informative features across both\ntemporal and channel dimensions. Furthermore, to address gradient imbalance in\nmulti-task learning, we integrate GradNorm, a technique that dynamically\nadjusts loss weights based on gradient magnitudes, ensuring balanced training\nacross tasks. Experimental results demonstrate the superior performance of our\nmodel, achieving accuracy rates of 99.12% in ID classification and 90.11% in\nactivity classification. These findings underscore the potential of the DT4ECG\nframework in enhancing security and user experience across various applications\nsuch as fitness monitoring and personalized healthcare, thereby presenting a\ntransformative approach to integrating ECG-based biometrics in everyday\ntechnologies.\n", "date": "2025-02-16", "categories": []}
{"id": "2502.11337", "title": "A Comparison of Human and Machine Learning Errors in Face Recognition", "abstract": "  Machine learning applications in high-stakes scenarios should always operate\nunder human oversight. Developing an optimal combination of human and machine\nintelligence requires an understanding of their complementarities, particularly\nregarding the similarities and differences in the way they make mistakes. We\nperform extensive experiments in the area of face recognition and compare two\nautomated face recognition systems against human annotators through a\ndemographically balanced user study. Our research uncovers important ways in\nwhich machine learning errors and human errors differ from each other, and\nsuggests potential strategies in which human-machine collaboration can improve\naccuracy in face recognition.\n", "date": "2025-02-17", "categories": []}
{"id": "2502.10195", "title": "Exploring the Camera Bias of Person Re-identification", "abstract": "  We empirically investigate the camera bias of person re-identification (ReID)\nmodels. Previously, camera-aware methods have been proposed to address this\nissue, but they are largely confined to training domains of the models. We\nmeasure the camera bias of ReID models on unseen domains and reveal that camera\nbias becomes more pronounced under data distribution shifts. As a debiasing\nmethod for unseen domain data, we revisit feature normalization on embedding\nvectors. While the normalization has been used as a straightforward solution,\nits underlying causes and broader applicability remain unexplored. We analyze\nwhy this simple method is effective at reducing bias and show that it can be\napplied to detailed bias factors such as low-level image properties and body\nangle. Furthermore, we validate its generalizability across various models and\nbenchmarks, highlighting its potential as a simple yet effective test-time\npostprocessing method for ReID. In addition, we explore the inherent risk of\ncamera bias in unsupervised learning of ReID models. The unsupervised models\nremain highly biased towards camera labels even for seen domain data,\nindicating substantial room for improvement. Based on observations of the\nnegative impact of camera-biased pseudo labels on training, we suggest simple\ntraining strategies to mitigate the bias. By applying these strategies to\nexisting unsupervised learning algorithms, we show that significant performance\nimprovements can be achieved with minor modifications.\n", "date": "2025-02-14", "categories": []}
{"id": "2501.16065", "title": "CILP-FGDI: Exploiting Vision-Language Model for Generalizable Person\n  Re-Identification", "abstract": "  The Visual Language Model, known for its robust cross-modal capabilities, has\nbeen extensively applied in various computer vision tasks. In this paper, we\nexplore the use of CLIP (Contrastive Language-Image Pretraining), a\nvision-language model pretrained on large-scale image-text pairs to align\nvisual and textual features, for acquiring fine-grained and domain-invariant\nrepresentations in generalizable person re-identification. The adaptation of\nCLIP to the task presents two primary challenges: learning more fine-grained\nfeatures to enhance discriminative ability, and learning more domain-invariant\nfeatures to improve the model's generalization capabilities. To mitigate the\nfirst challenge thereby enhance the ability to learn fine-grained features, a\nthree-stage strategy is proposed to boost the accuracy of text descriptions.\nInitially, the image encoder is trained to effectively adapt to person\nre-identification tasks. In the second stage, the features extracted by the\nimage encoder are used to generate textual descriptions (i.e., prompts) for\neach image. Finally, the text encoder with the learned prompts is employed to\nguide the training of the final image encoder. To enhance the model's\ngeneralization capabilities to unseen domains, a bidirectional guiding method\nis introduced to learn domain-invariant image features. Specifically,\ndomain-invariant and domain-relevant prompts are generated, and both positive\n(pulling together image features and domain-invariant prompts) and negative\n(pushing apart image features and domain-relevant prompts) views are used to\ntrain the image encoder. Collectively, these strategies contribute to the\ndevelopment of an innovative CLIP-based framework for learning fine-grained\ngeneralized features in person re-identification.\n", "date": "2025-01-27", "categories": []}
{"id": "2502.07130", "title": "Unconstrained Body Recognition at Altitude and Range: Comparing Four\n  Approaches", "abstract": "  This study presents an investigation of four distinct approaches to long-term\nperson identification using body shape. Unlike short-term re-identification\nsystems that rely on temporary features (e.g., clothing), we focus on learning\npersistent body shape characteristics that remain stable over time. We\nintroduce a body identification model based on a Vision Transformer (ViT) (Body\nIdentification from Diverse Datasets, BIDDS) and on a Swin-ViT model\n(Swin-BIDDS). We also expand on previous approaches based on the Linguistic and\nNon-linguistic Core ResNet Identity Models (LCRIM and NLCRIM), but with\nimproved training. All models are trained on a large and diverse dataset of\nover 1.9 million images of approximately 5k identities across 9 databases.\nPerformance was evaluated on standard re-identification benchmark datasets\n(MARS, MSMT17, Outdoor Gait, DeepChange) and on an unconstrained dataset that\nincludes images at a distance (from close-range to 1000m), at altitude (from an\nunmanned aerial vehicle, UAV), and with clothing change. A comparative analysis\nacross these models provides insights into how different backbone architectures\nand input image sizes impact long-term body identification performance across\nreal-world conditions.\n", "date": "2025-02-10", "categories": []}
{"id": "2502.07734", "title": "EdgeEar: Efficient and Accurate Ear Recognition for Edge Devices", "abstract": "  Ear recognition is a contactless and unobtrusive biometric technique with\napplications across various domains. However, deploying high-performing ear\nrecognition models on resource-constrained devices is challenging, limiting\ntheir applicability and widespread adoption. This paper introduces EdgeEar, a\nlightweight model based on a proposed hybrid CNN-transformer architecture to\nsolve this problem. By incorporating low-rank approximations into specific\nlinear layers, EdgeEar reduces its parameter count by a factor of 50 compared\nto the current state-of-the-art, bringing it below two million while\nmaintaining competitive accuracy. Evaluation on the Unconstrained Ear\nRecognition Challenge (UERC2023) benchmark shows that EdgeEar achieves the\nlowest EER while significantly reducing computational costs. These findings\ndemonstrate the feasibility of efficient and accurate ear recognition, which we\nbelieve will contribute to the wider adoption of ear biometrics.\n", "date": "2025-02-11", "categories": []}
{"id": "2307.03240", "title": "Adaptive Generation of Privileged Intermediate Information for\n  Visible-Infrared Person Re-Identification", "abstract": "  Visible-infrared person re-identification seeks to retrieve images of the\nsame individual captured over a distributed network of RGB and IR sensors.\nSeveral V-I ReID approaches directly integrate both V and I modalities to\ndiscriminate persons within a shared representation space. However, given the\nsignificant gap in data distributions between V and I modalities, cross-modal\nV-I ReID remains challenging. Some recent approaches improve generalization by\nleveraging intermediate spaces that can bridge V and I modalities, yet\neffective methods are required to select or generate data for such informative\ndomains. In this paper, the Adaptive Generation of Privileged Intermediate\nInformation training approach is introduced to adapt and generate a virtual\ndomain that bridges discriminant information between the V and I modalities.\nThe key motivation behind AGPI^2 is to enhance the training of a deep V-I ReID\nbackbone by generating privileged images that provide additional information.\nThese privileged images capture shared discriminative features that are not\neasily accessible within the original V or I modalities alone. Towards this\ngoal, a non-linear generative module is trained with an adversarial objective,\ntranslating V images into intermediate spaces with a smaller domain shift\nw.r.t. the I domain. Meanwhile, the embedding module within AGPI^2 aims to\nproduce similar features for both V and generated images, encouraging the\nextraction of features that are common to all modalities. In addition to these\ncontributions, AGPI^2 employs adversarial objectives for adapting the\nintermediate images, which play a crucial role in creating a\nnon-modality-specific space to address the large domain shifts between V and I\ndomains. Experimental results conducted on challenging V-I ReID datasets\nindicate that AGPI^2 increases matching accuracy without extra computational\nresources during inference.\n", "date": "2023-07-06", "categories": []}
{"id": "2403.10782", "title": "Bidirectional Multi-Step Domain Generalization for Visible-Infrared\n  Person Re-Identification", "abstract": "  A key challenge in visible-infrared person re-identification (V-I ReID) is\ntraining a backbone model capable of effectively addressing the significant\ndiscrepancies across modalities. State-of-the-art methods that generate a\nsingle intermediate bridging domain are often less effective, as this generated\ndomain may not adequately capture sufficient common discriminant information.\nThis paper introduces Bidirectional Multi-step Domain Generalization (BMDG), a\nnovel approach for unifying feature representations across diverse modalities.\nBMDG creates multiple virtual intermediate domains by learning and aligning\nbody part features extracted from both I and V modalities. In particular, our\nmethod aims to minimize the cross-modal gap in two steps. First, BMDG aligns\nmodalities in the feature space by learning shared and modality-invariant body\npart prototypes from V and I images. Then, it generalizes the feature\nrepresentation by applying bidirectional multi-step learning, which\nprogressively refines feature representations in each step and incorporates\nmore prototypes from both modalities. Based on these prototypes, multiple\nbridging steps enhance the feature representation. Experiments conducted on V-I\nReID datasets indicate that our BMDG approach can outperform state-of-the-art\npart-based and intermediate generation methods, and can be integrated into\nother part-based methods to enhance their V-I ReID performance. (Our code is\navailable at:https:/alehdaghi.github.io/BMDG/ )\n", "date": "2024-03-16", "categories": []}
{"id": "2405.01088", "title": "Type2Branch: Keystroke Biometrics based on a Dual-branch Architecture\n  with Attention Mechanisms and Set2set Loss", "abstract": "  In 2021, the pioneering work TypeNet showed that keystroke dynamics\nverification could scale to hundreds of thousands of users with minimal\nperformance degradation. Recently, the KVC-onGoing competition has provided an\nopen and robust experimental protocol for evaluating keystroke dynamics\nverification systems of such scale. %, including considerations of algorithmic\nfairness. This article describes Type2Branch, the model and techniques that\nachieved the lowest error rates at the KVC-onGoing, in both desktop and mobile\ntyping scenarios. The novelty aspects of the proposed Type2Branch include: i)\nsynthesized timing features emphasizing user behavior deviation from the\ngeneral population, ii) a dual-branch architecture combining recurrent and\nconvolutional paths with various attention mechanisms, iii) a new loss function\nnamed Set2set that captures the global structure of the embedding space, and\niv) a training curriculum of increasing difficulty. Considering five enrollment\nsamples per subject of approximately 50 characters typed, the proposed\nType2Branch achieves state-of-the-art performance with mean per-subject Equal\nError Rates (EERs) of 0.77% and 1.03% on evaluation sets of respectively 15,000\nand 5,000 subjects for desktop and mobile scenarios. With a fixed global\nthreshold for all subjects, the EERs are respectively 3.25% and 3.61% for\ndesktop and mobile scenarios, outperforming previous approaches by a\nsignificant margin. The source code for dataset generation, model, and training\nprocess is publicly available.\n", "date": "2024-05-02", "categories": []}
{"id": "2405.19695", "title": "Distribution Aligned Semantics Adaption for Lifelong Person\n  Re-Identification", "abstract": "  In real-world scenarios, person Re-IDentification (Re-ID) systems need to be\nadaptable to changes in space and time. Therefore, the adaptation of Re-ID\nmodels to new domains while preserving previously acquired knowledge is\ncrucial, known as Lifelong person Re-IDentification (LReID). Advanced LReID\nmethods rely on replaying exemplars from old domains and applying knowledge\ndistillation in logits with old models. However, due to privacy concerns,\nretaining previous data is inappropriate. Additionally, the fine-grained and\nopen-set characteristics of Re-ID limit the effectiveness of the distillation\nparadigm for accumulating knowledge. We argue that a Re-ID model trained on\ndiverse and challenging pedestrian images at a large scale can acquire robust\nand general human semantic knowledge. These semantics can be readily utilized\nas shared knowledge for lifelong applications. In this paper, we identify the\nchallenges and discrepancies associated with adapting a pre-trained model to\neach application domain and introduce the Distribution Aligned Semantics\nAdaption (DASA) framework. It efficiently adjusts Batch Normalization (BN) to\nmitigate interference from data distribution discrepancy and freezes the\npre-trained convolutional layers to preserve shared knowledge. Additionally, we\npropose the lightweight Semantics Adaption (SA) module, which effectively\nadapts learned semantics to enhance pedestrian representations. Extensive\nexperiments demonstrate the remarkable superiority of our proposed framework\nover advanced LReID methods, and it exhibits significantly reduced storage\nconsumption. DASA presents a novel and cost-effective perspective on\neffectively adapting pre-trained models for LReID. The code is available at\nhttps://github.com/QizaoWang/DASA-LReID.\n", "date": "2024-05-30", "categories": []}
{"id": "2409.02979", "title": "Vec2Face: Scaling Face Dataset Generation with Loosely Constrained\n  Vectors", "abstract": "  This paper studies how to synthesize face images of non-existent persons, to\ncreate a dataset that allows effective training of face recognition (FR)\nmodels. Besides generating realistic face images, two other important goals\nare: 1) the ability to generate a large number of distinct identities\n(inter-class separation), and 2) a proper variation in appearance of the images\nfor each identity (intra-class variation). However, existing works 1) are\ntypically limited in how many well-separated identities can be generated and 2)\neither neglect or use an external model for attribute augmentation. We propose\nVec2Face, a holistic model that uses only a sampled vector as input and can\nflexibly generate and control the identity of face images and their attributes.\nComposed of a feature masked autoencoder and an image decoder, Vec2Face is\nsupervised by face image reconstruction and can be conveniently used in\ninference. Using vectors with low similarity among themselves as inputs,\nVec2Face generates well-separated identities. Randomly perturbing an input\nidentity vector within a small range allows Vec2Face to generate faces of the\nsame identity with proper variation in face attributes. It is also possible to\ngenerate images with designated attributes by adjusting vector values with a\ngradient descent method. Vec2Face has efficiently synthesized as many as 300K\nidentities, whereas 60K is the largest number of identities created in the\nprevious works. As for performance, FR models trained with the generated HSFace\ndatasets, from 10k to 300k identities, achieve state-of-the-art accuracy, from\n92% to 93.52%, on five real-world test sets (\\emph{i.e.}, LFW, CFP-FP,\nAgeDB-30, CALFW, and CPLFW). For the first time, the FR model trained using our\nsynthetic training set achieves higher accuracy than that trained using a\nsame-scale training set of real face images on the CALFW, IJBB, and IJBC test\nsets.\n", "date": "2024-09-04", "categories": []}
{"id": "2410.23831", "title": "FRoundation: Are Foundation Models Ready for Face Recognition?", "abstract": "  Foundation models are predominantly trained in an unsupervised or\nself-supervised manner on highly diverse and large-scale datasets, making them\nbroadly applicable to various downstream tasks. In this work, we investigate\nfor the first time whether such models are suitable for the specific domain of\nface recognition (FR). We further propose and demonstrate the adaptation of\nthese models for FR across different levels of data availability, including\nsynthetic data. Extensive experiments are conducted on multiple foundation\nmodels and datasets of varying scales for training and fine-tuning, with\nevaluation on a wide range of benchmarks. Our results indicate that, despite\ntheir versatility, pre-trained foundation models tend to underperform in FR in\ncomparison with similar architectures trained specifically for this task.\nHowever, fine-tuning foundation models yields promising results, often\nsurpassing models trained from scratch, particularly when training data is\nlimited. For example, after fine-tuning only on 1K identities, DINOv2 ViT-S\nachieved average verification accuracy on LFW, CALFW, CPLFW, CFP-FP, and\nAgeDB30 benchmarks of 87.10%, compared to 64.70% achieved by the same model and\nwithout fine-tuning. While training the same model architecture, ViT-S, from\nscratch on 1k identities reached 69.96%. With access to larger-scale FR\ntraining datasets, these performances reach 96.03% and 95.59% for the DINOv2\nand CLIP ViT-L models, respectively. In comparison to the ViT-based\narchitectures trained from scratch for FR, fine-tuned same architectures of\nfoundation models achieve similar performance while requiring lower training\ncomputational costs and not relying on the assumption of extensive data\navailability. We further demonstrated the use of synthetic face data, showing\nimproved performances over both pre-trained foundation and ViT models.\n", "date": "2024-10-31", "categories": []}
{"id": "2502.04361", "title": "Predicting 3D Motion from 2D Video for Behavior-Based VR Biometrics", "abstract": "  Critical VR applications in domains such as healthcare, education, and\nfinance that use traditional credentials, such as PIN, password, or\nmulti-factor authentication, stand the chance of being compromised if a\nmalicious person acquires the user credentials or if the user hands over their\ncredentials to an ally. Recently, a number of approaches on user authentication\nhave emerged that use motions of VR head-mounted displays (HMDs) and hand\ncontrollers during user interactions in VR to represent the user's behavior as\na VR biometric signature. One of the fundamental limitations of behavior-based\napproaches is that current on-device tracking for HMDs and controllers lacks\ncapability to perform tracking of full-body joint articulation, losing key\nsignature data encapsulated by the user articulation. In this paper, we propose\nan approach that uses 2D body joints, namely shoulder, elbow, wrist, hip, knee,\nand ankle, acquired from the right side of the participants using an external\n2D camera. Using a Transformer-based deep neural network, our method uses the\n2D data of body joints that are not tracked by the VR device to predict past\nand future 3D tracks of the right controller, providing the benefit of\naugmenting 3D knowledge in authentication. Our approach provides a minimum\nequal error rate (EER) of 0.025, and a maximum EER drop of 0.040 over prior\nwork that uses single-unit 3D trajectory as the input.\n", "date": "2025-02-05", "categories": []}
{"id": "2502.04680", "title": "Performance Evaluation of Image Enhancement Techniques on Transfer\n  Learning for Touchless Fingerprint Recognition", "abstract": "  Fingerprint recognition remains one of the most reliable biometric\ntechnologies due to its high accuracy and uniqueness. Traditional systems rely\non contact-based scanners, which are prone to issues such as image degradation\nfrom surface contamination and inconsistent user interaction. To address these\nlimitations, contactless fingerprint recognition has emerged as a promising\nalternative, providing non-intrusive and hygienic authentication. This study\nevaluates the impact of image enhancement tech-niques on the performance of\npre-trained deep learning models using transfer learning for touchless\nfingerprint recognition. The IIT-Bombay Touchless and Touch-Based Fingerprint\nDatabase, containing data from 200 subjects, was employed to test the\nper-formance of deep learning architectures such as VGG-16, VGG-19,\nInception-V3, and ResNet-50. Experimental results reveal that transfer learning\nmethods with fingerprint image enhance-ment (indirect method) significantly\noutperform those without enhancement (direct method). Specifically, VGG-16\nachieved an accuracy of 98% in training and 93% in testing when using the\nenhanced images, demonstrating superior performance compared to the direct\nmethod.\n  This paper provides a detailed comparison of the effectiveness of image\nenhancement in improving the accuracy of transfer learning models for touchless\nfingerprint recognition, offering key insights for developing more efficient\nbiometric systems.\n", "date": "2025-02-07", "categories": []}
{"id": "2502.04973", "title": "DE-PADA: Personalized Augmentation and Domain Adaptation for ECG\n  Biometrics Across Physiological States", "abstract": "  Electrocardiogram (ECG)-based biometrics offer a promising method for user\nidentification, combining intrinsic liveness detection with morphological\nuniqueness. However, elevated heart rates introduce significant physiological\nvariability, posing challenges to pattern recognition systems and leading to a\nnotable performance gap between resting and post-exercise conditions.\nAddressing this gap is critical for advancing ECG-based biometric systems for\nreal-world applications. We propose DE-PADA, a Dual Expert model with\nPersonalized Augmentation and Domain Adaptation, designed to enhance robustness\nacross diverse physiological states. The model is trained primarily on\nresting-state data from the evaluation dataset, without direct exposure to\ntheir exercise data. To address variability, DE-PADA incorporates ECG-specific\ninnovations, including heartbeat segmentation into the PQRS interval, known for\nits relative temporal consistency, and the heart rate-sensitive ST interval,\nenabling targeted feature extraction tailored to each region's unique\ncharacteristics. Personalized augmentation simulates subject-specific T-wave\nvariability across heart rates using individual T-wave peak predictions to\nadapt augmentation ranges. Domain adaptation further improves generalization by\nleveraging auxiliary data from supplementary subjects used exclusively for\ntraining, including both resting and exercise conditions. Experiments on the\nUniversity of Toronto ECG Database demonstrate the model's effectiveness.\nDE-PADA achieves relative improvements in post-exercise identification rates of\n26.75% in the initial recovery phase and 11.72% in the late recovery phase,\nwhile maintaining a 98.12% identification rate in the sitting position. These\nresults highlight DE-PADA's ability to address intra-subject variability and\nenhance the robustness of ECG-based biometric systems across diverse\nphysiological states.\n", "date": "2025-02-07", "categories": []}
{"id": "2501.14070", "title": "Expanding on the BRIAR Dataset: A Comprehensive Whole Body Biometric\n  Recognition Resource at Extreme Distances and Real-World Scenarios\n  (Collections 1-4)", "abstract": "  The state-of-the-art in biometric recognition algorithms and operational\nsystems has advanced quickly in recent years providing high accuracy and\nrobustness in more challenging collection environments and consumer\napplications. However, the technology still suffers greatly when applied to\nnon-conventional settings such as those seen when performing identification at\nextreme distances or from elevated cameras on buildings or mounted to UAVs.\nThis paper summarizes an extension to the largest dataset currently focused on\naddressing these operational challenges, and describes its composition as well\nas methodologies of collection, curation, and annotation.\n", "date": "2025-01-23", "categories": []}
{"id": "2402.02108", "title": "Synthetic-To-Real Video Person Re-ID", "abstract": "  Person re-identification (Re-ID) is an important task and has significant\napplications for public security and information forensics, which has\nprogressed rapidly with the development of deep learning. In this work, we\ninvestigate a novel and challenging setting of Re-ID, i.e., cross-domain\nvideo-based person Re-ID. Specifically, we utilize synthetic video datasets as\nthe source domain for training and real-world videos for testing, notably\nreducing the reliance on expensive real data acquisition and annotation. To\nharness the potential of synthetic data, we first propose a self-supervised\ndomain-invariant feature learning strategy for both static and dynamic\n(temporal) features. Additionally, to enhance person identification accuracy in\nthe target domain, we propose a mean-teacher scheme incorporating a\nself-supervised ID consistency loss. Experimental results across five real\ndatasets validate the rationale behind cross-synthetic-real domain adaptation\nand demonstrate the efficacy of our method. Notably, the discovery that\nsynthetic data outperforms real data in the cross-domain scenario is a\nsurprising outcome. The code and data are publicly available at\nhttps://github.com/XiangqunZhang/UDA_Video_ReID.\n", "date": "2024-02-03", "categories": []}
{"id": "2502.00665", "title": "Cross-Modal Synergies: Unveiling the Potential of Motion-Aware Fusion\n  Networks in Handling Dynamic and Static ReID Scenarios", "abstract": "  Navigating the complexities of person re-identification (ReID) in varied\nsurveillance scenarios, particularly when occlusions occur, poses significant\nchallenges. We introduce an innovative Motion-Aware Fusion (MOTAR-FUSE) network\nthat utilizes motion cues derived from static imagery to significantly enhance\nReID capabilities. This network incorporates a dual-input visual adapter\ncapable of processing both images and videos, thereby facilitating more\neffective feature extraction. A unique aspect of our approach is the\nintegration of a motion consistency task, which empowers the motion-aware\ntransformer to adeptly capture the dynamics of human motion. This technique\nsubstantially improves the recognition of features in scenarios where\nocclusions are prevalent, thereby advancing the ReID process. Our comprehensive\nevaluations across multiple ReID benchmarks, including holistic, occluded, and\nvideo-based scenarios, demonstrate that our MOTAR-FUSE network achieves\nsuperior performance compared to existing approaches.\n", "date": "2025-02-02", "categories": []}
{"id": "2501.09229", "title": "Tessellated Linear Model for Age Prediction from Voice", "abstract": "  Voice biometric tasks, such as age estimation require modeling the often\ncomplex relationship between voice features and the biometric variable. While\ndeep learning models can handle such complexity, they typically require large\namounts of accurately labeled data to perform well. Such data are often scarce\nfor biometric tasks such as voice-based age prediction. On the other hand,\nsimpler models like linear regression can work with smaller datasets but often\nfail to generalize to the underlying non-linear patterns present in the data.\nIn this paper we propose the Tessellated Linear Model (TLM), a piecewise linear\napproach that combines the simplicity of linear models with the capacity of\nnon-linear functions. TLM tessellates the feature space into convex regions and\nfits a linear model within each region. We optimize the tessellation and the\nlinear models using a hierarchical greedy partitioning. We evaluated TLM on the\nTIMIT dataset on the task of age prediction from voice, where it outperformed\nstate-of-the-art deep learning models.\n", "date": "2025-01-16", "categories": []}
{"id": "2501.16811", "title": "Not Every Patch is Needed: Towards a More Efficient and Effective\n  Backbone for Video-based Person Re-identification", "abstract": "  This paper proposes a new effective and efficient plug-and-play backbone for\nvideo-based person re-identification (ReID). Conventional video-based ReID\nmethods typically use CNN or transformer backbones to extract deep features for\nevery position in every sampled video frame. Here, we argue that this\nexhaustive feature extraction could be unnecessary, since we find that\ndifferent frames in a ReID video often exhibit small differences and contain\nmany similar regions due to the relatively slight movements of human beings.\nInspired by this, a more selective, efficient paradigm is explored in this\npaper. Specifically, we introduce a patch selection mechanism to reduce\ncomputational cost by choosing only the crucial and non-repetitive patches for\nfeature extraction. Additionally, we present a novel network structure that\ngenerates and utilizes pseudo frame global context to address the issue of\nincomplete views resulting from sparse inputs. By incorporating these new\ndesigns, our backbone can achieve both high performance and low computational\ncost. Extensive experiments on multiple datasets show that our approach reduces\nthe computational cost by 74\\% compared to ViT-B and 28\\% compared to ResNet50,\nwhile the accuracy is on par with ViT-B and outperforms ResNet50 significantly.\n", "date": "2025-01-28", "categories": []}
{"id": "2501.16896", "title": "Frequency Matters: Explaining Biases of Face Recognition in the\n  Frequency Domain", "abstract": "  Face recognition (FR) models are vulnerable to performance variations across\ndemographic groups. The causes for these performance differences are unclear\ndue to the highly complex deep learning-based structure of face recognition\nmodels. Several works aimed at exploring possible roots of gender and ethnicity\nbias, identifying semantic reasons such as hairstyle, make-up, or facial hair\nas possible sources. Motivated by recent discoveries of the importance of\nfrequency patterns in convolutional neural networks, we explain bias in face\nrecognition using state-of-the-art frequency-based explanations. Our extensive\nresults show that different frequencies are important to FR models depending on\nthe ethnicity of the samples.\n", "date": "2025-01-28", "categories": []}
{"id": "2501.15666", "title": "MimicGait: A Model Agnostic approach for Occluded Gait Recognition using\n  Correlational Knowledge Distillation", "abstract": "  Gait recognition is an important biometric technique over large distances.\nState-of-the-art gait recognition systems perform very well in controlled\nenvironments at close range. Recently, there has been an increased interest in\ngait recognition in the wild prompted by the collection of outdoor, more\nchallenging datasets containing variations in terms of illumination, pitch\nangles, and distances. An important problem in these environments is that of\nocclusion, where the subject is partially blocked from camera view. While\nimportant, this problem has received little attention. Thus, we propose\nMimicGait, a model-agnostic approach for gait recognition in the presence of\nocclusions. We train the network using a multi-instance correlational\ndistillation loss to capture both inter-sequence and intra-sequence\ncorrelations in the occluded gait patterns of a subject, utilizing an auxiliary\nVisibility Estimation Network to guide the training of the proposed mimic\nnetwork. We demonstrate the effectiveness of our approach on challenging\nreal-world datasets like GREW, Gait3D and BRIAR. We release the code in\nhttps://github.com/Ayush-00/mimicgait.\n", "date": "2025-01-26", "categories": []}
{"id": "2501.13889", "title": "Generating Realistic Forehead-Creases for User Verification via\n  Conditioned Piecewise Polynomial Curves", "abstract": "  We propose a trait-specific image generation method that models forehead\ncreases geometrically using B-spline and B\\'ezier curves. This approach ensures\nthe realistic generation of both principal creases and non-prominent crease\npatterns, effectively constructing detailed and authentic forehead-crease\nimages. These geometrically rendered images serve as visual prompts for a\ndiffusion-based Edge-to-Image translation model, which generates corresponding\nmated samples. The resulting novel synthetic identities are then used to train\na forehead-crease verification network. To enhance intra-subject diversity in\nthe generated samples, we employ two strategies: (a) perturbing the control\npoints of B-splines under defined constraints to maintain label consistency,\nand (b) applying image-level augmentations to the geometric visual prompts,\nsuch as dropout and elastic transformations, specifically tailored to crease\npatterns. By integrating the proposed synthetic dataset with real-world data,\nour method significantly improves the performance of forehead-crease\nverification systems under a cross-database verification protocol.\n", "date": "2025-01-23", "categories": []}
{"id": "2305.00352", "title": "Embedding Aggregation for Forensic Facial Comparison", "abstract": "  In forensic facial comparison, questioned-source images are usually captured\nin uncontrolled environments, with non-uniform lighting, and from\nnon-cooperative subjects. The poor quality of such material usually compromises\ntheir value as evidence in legal matters. On the other hand, in forensic\ncasework, multiple images of the person of interest are usually available. In\nthis paper, we propose to aggregate deep neural network embeddings from various\nimages of the same person to improve performance in facial verification. We\nobserve significant performance improvements, especially for very low-quality\nimages. Further improvements are obtained by aggregating embeddings of more\nimages and by applying quality-weighted aggregation. We demonstrate the\nbenefits of this approach in forensic evaluation settings with the development\nand validation of score-based likelihood ratio systems and report improvements\nin Cllr of up to 95% (from 0.249 to 0.012) for CCTV images and of up to 96%\n(from 0.083 to 0.003) for social media images.\n", "date": "2023-04-29", "categories": []}
{"id": "2410.24031", "title": "A Multi-Modal Approach for Face Anti-Spoofing in Non-Calibrated Systems\n  using Disparity Maps", "abstract": "  Face recognition technologies are increasingly used in various applications,\nyet they are vulnerable to face spoofing attacks. These spoofing attacks often\ninvolve unique 3D structures, such as printed papers or mobile device screens.\nAlthough stereo-depth cameras can detect such attacks effectively, their\nhigh-cost limits their widespread adoption. Conversely, two-sensor systems\nwithout extrinsic calibration offer a cost-effective alternative but are unable\nto calculate depth using stereo techniques. In this work, we propose a method\nto overcome this challenge by leveraging facial attributes to derive disparity\ninformation and estimate relative depth for anti-spoofing purposes, using\nnon-calibrated systems. We introduce a multi-modal anti-spoofing model, coined\nDisparity Model, that incorporates created disparity maps as a third modality\nalongside the two original sensor modalities. We demonstrate the effectiveness\nof the Disparity Model in countering various spoof attacks using a\ncomprehensive dataset collected from the Intel RealSense ID Solution F455. Our\nmethod outperformed existing methods in the literature, achieving an Equal\nError Rate (EER) of 1.71% and a False Negative Rate (FNR) of 2.77% at a False\nPositive Rate (FPR) of 1%. These errors are lower by 2.45% and 7.94% than the\nerrors of the best comparison method, respectively. Additionally, we introduce\na model ensemble that addresses 3D spoof attacks as well, achieving an EER of\n2.04% and an FNR of 3.83% at an FPR of 1%. Overall, our work provides a\nstate-of-the-art solution for the challenging task of anti-spoofing in\nnon-calibrated systems that lack depth information.\n", "date": "2024-10-31", "categories": []}
{"id": "2501.09048", "title": "Anthropomorphic Features for On-Line Signatures", "abstract": "  Many features have been proposed in on-line signature verification.\nGenerally, these features rely on the position of the on-line signature samples\nand their dynamic properties, as recorded by a tablet. This paper proposes a\nnovel feature space to describe efficiently on-line signatures. Since producing\na signature requires a skeletal arm system and its associated muscles, the new\nfeature space is based on characterizing the movement of the shoulder, the\nelbow and the wrist joints when signing. As this motion is not directly\nobtained from a digital tablet, the new features are calculated by means of a\nvirtual skeletal arm (VSA) model, which simulates the architecture of a real\narm and forearm. Specifically, the VSA motion is described by its 3D joint\nposition and its joint angles. These anthropomorphic features are worked out\nfrom both pen position and orientation through the VSA forward and direct\nkinematic model. The anthropomorphic features' robustness is proved by\nachieving state-of-the-art performance with several verifiers and multiple\nbenchmarks on third party signature databases, which were collected with\ndifferent devices and in different languages and scripts.\n", "date": "2025-01-15", "categories": []}
{"id": "2411.00330", "title": "Multiple Information Prompt Learning for Cloth-Changing Person\n  Re-Identification", "abstract": "  Cloth-changing person re-identification is a subject closer to the real\nworld, which focuses on solving the problem of person re-identification after\npedestrians change clothes. The primary challenge in this field is to overcome\nthe complex interplay between intra-class and inter-class variations and to\nidentify features that remain unaffected by changes in appearance. Sufficient\ndata collection for model training would significantly aid in addressing this\nproblem. However, it is challenging to gather diverse datasets in practice.\nCurrent methods focus on implicitly learning identity information from the\noriginal image or introducing additional auxiliary models, which are largely\nlimited by the quality of the image and the performance of the additional\nmodel. To address these issues, inspired by prompt learning, we propose a novel\nmultiple information prompt learning (MIPL) scheme for cloth-changing person\nReID, which learns identity robust features through the common prompt guidance\nof multiple messages. Specifically, the clothing information stripping (CIS)\nmodule is designed to decouple the clothing information from the original RGB\nimage features to counteract the influence of clothing appearance. The\nBio-guided attention (BGA) module is proposed to increase the learning\nintensity of the model for key information. A dual-length hybrid patch (DHP)\nmodule is employed to make the features have diverse coverage to minimize the\nimpact of feature bias. Extensive experiments demonstrate that the proposed\nmethod outperforms all state-of-the-art methods on the LTCC, Celeb-reID,\nCeleb-reID-light, and CSCC datasets, achieving rank-1 scores of 74.8%, 73.3%,\n66.0%, and 88.1%, respectively. When compared to AIM (CVPR23), ACID (TIP23),\nand SCNet (MM23), MIPL achieves rank-1 improvements of 11.3%, 13.8%, and 7.9%,\nrespectively, on the PRCC dataset.\n", "date": "2024-11-01", "categories": []}
{"id": "2501.08440", "title": "FARE: A Deep Learning-Based Framework for Radar-based Face Recognition\n  and Out-of-distribution Detection", "abstract": "  In this work, we propose a novel pipeline for face recognition and\nout-of-distribution (OOD) detection using short-range FMCW radar. The proposed\nsystem utilizes Range-Doppler and micro Range-Doppler Images. The architecture\nfeatures a primary path (PP) responsible for the classification of\nin-distribution (ID) faces, complemented by intermediate paths (IPs) dedicated\nto OOD detection. The network is trained in two stages: first, the PP is\ntrained using triplet loss to optimize ID face classification. In the second\nstage, the PP is frozen, and the IPs-comprising simple linear autoencoder\nnetworks-are trained specifically for OOD detection. Using our dataset\ngenerated with a 60 GHz FMCW radar, our method achieves an ID classification\naccuracy of 99.30% and an OOD detection AUROC of 96.91%.\n", "date": "2025-01-14", "categories": []}
{"id": "2501.08910", "title": "Lights, Camera, Matching: The Role of Image Illumination in Fair Face\n  Recognition", "abstract": "  Facial brightness is a key image quality factor impacting face recognition\naccuracy differentials across demographic groups. In this work, we aim to\ndecrease the accuracy gap between the similarity score distributions for\nCaucasian and African American female mated image pairs, as measured by d'\nbetween distributions. To balance brightness across demographic groups, we\nconduct three experiments, interpreting brightness in the face skin region\neither as median pixel value or as the distribution of pixel values. Balancing\nbased on median brightness alone yields up to a 46.8% decrease in d', while\nbalancing based on brightness distribution yields up to a 57.6% decrease. In\nall three cases, the similarity scores of the individual distributions improve,\nwith mean scores maximally improving 5.9% for Caucasian females and 3.7% for\nAfrican American females.\n", "date": "2025-01-15", "categories": []}
{"id": "2411.02188", "title": "Digi2Real: Bridging the Realism Gap in Synthetic Data Face Recognition\n  via Foundation Models", "abstract": "  The accuracy of face recognition systems has improved significantly in the\npast few years, thanks to the large amount of data collected and advancements\nin neural network architectures. However, these large-scale datasets are often\ncollected without explicit consent, raising ethical and privacy concerns. To\naddress this, there have been proposals to use synthetic datasets for training\nface recognition models. Yet, such models still rely on real data to train the\ngenerative models and generally exhibit inferior performance compared to those\ntrained on real datasets. One of these datasets, DigiFace, uses a graphics\npipeline to generate different identities and intra-class variations without\nusing real data in model training. However, the performance of this approach is\npoor on face recognition benchmarks, possibly due to the lack of realism in the\nimages generated by the graphics pipeline. In this work, we introduce a novel\nframework for realism transfer aimed at enhancing the realism of synthetically\ngenerated face images. Our method leverages the large-scale face foundation\nmodel, and we adapt the pipeline for realism enhancement. By integrating the\ncontrollable aspects of the graphics pipeline with our realism enhancement\ntechnique, we generate a large amount of realistic variations, combining the\nadvantages of both approaches. Our empirical evaluations demonstrate that\nmodels trained using our enhanced dataset significantly improve the performance\nof face recognition systems over the baseline. The source code and dataset will\nbe publicly accessible at the following link:\nhttps://www.idiap.ch/paper/digi2real\n", "date": "2024-11-04", "categories": []}
{"id": "2501.07600", "title": "Impact of Data Breadth and Depth on Performance of Siamese Neural\n  Network Model: Experiments with Three Keystroke Dynamic Datasets", "abstract": "  Deep learning models, such as the Siamese Neural Networks (SNN), have shown\ngreat potential in capturing the intricate patterns in behavioral data.\nHowever, the impacts of dataset breadth (i.e., the number of subjects) and\ndepth (e.g., the amount of training samples per subject) on the performance of\nthese models is often informally assumed, and remains under-explored. To this\nend, we have conducted extensive experiments using the concepts of \"feature\nspace\" and \"density\" to guide and gain deeper understanding on the impact of\ndataset breadth and depth on three publicly available keystroke datasets\n(Aalto, CMU and Clarkson II). Through varying the number of training subjects,\nnumber of samples per subject, amount of data in each sample, and number of\ntriplets used in training, we found that when feasible, increasing dataset\nbreadth enables the training of a well-trained model that effectively captures\nmore inter-subject variability. In contrast, we find that the extent of depth's\nimpact from a dataset depends on the nature of the dataset. Free-text datasets\nare influenced by all three depth-wise factors; inadequate samples per subject,\nsequence length, training triplets and gallery sample size, which may all lead\nto an under-trained model. Fixed-text datasets are less affected by these\nfactors, and as such make it easier to create a well-trained model. These\nfindings shed light on the importance of dataset breadth and depth in training\ndeep learning models for behavioral biometrics and provide valuable insights\nfor designing more effective authentication systems.\n", "date": "2025-01-10", "categories": []}
{"id": "2409.14432", "title": "EM-DARTS: Hierarchical Differentiable Architecture Search for Eye\n  Movement Recognition", "abstract": "  Eye movement biometrics has received increasing attention thanks to its\nhighly secure identification. Although deep learning (DL) models have shown\nsuccess in eye movement recognition, their architectures largely rely on human\nprior knowledge. Differentiable Neural Architecture Search (DARTS) automates\nthe manual process of architecture design with high search efficiency. However,\nDARTS typically stacks multiple cells to form a convolutional network, which\nlimits the diversity of architecture. Furthermore, DARTS generally searches for\narchitectures using shallower networks than those used in the evaluation,\ncreating a significant disparity in architecture depth between the search and\nevaluation phases. To address this issue, we propose EM-DARTS, a hierarchical\ndifferentiable architecture search algorithm to automatically design the DL\narchitecture for eye movement recognition. First, we define a supernet and\npropose a global and local alternate Neural Architecture Search method to\nsearch the optimal architecture alternately with a differentiable neural\narchitecture search. The local search strategy aims to find an optimal\narchitecture for different cells while the global search strategy is\nresponsible for optimizing the architecture of the target network. To minimize\nredundancy, transfer entropy is proposed to compute the information amount of\neach layer, thereby further simplifying the network search process.\nExperimental results on three public datasets demonstrate that the proposed\nEM-DARTS is capable of producing an optimal architecture that leads to\nstate-of-the-art recognition performance, {Specifically, the recognition models\ndeveloped using EM-DARTS achieved the lowest EERs of 0.0453 on the GazeBase\ndataset, 0.0377 on the JuDo1000 dataset, and 0.1385 on the EMglasses dataset.\n", "date": "2024-09-22", "categories": []}
{"id": "2501.07296", "title": "Event-based Video Person Re-identification via Cross-Modality and\n  Temporal Collaboration", "abstract": "  Video-based person re-identification (ReID) has become increasingly important\ndue to its applications in video surveillance applications. By employing events\nin video-based person ReID, more motion information can be provided between\ncontinuous frames to improve recognition accuracy. Previous approaches have\nassisted by introducing event data into the video person ReID task, but they\nstill cannot avoid the privacy leakage problem caused by RGB images. In order\nto avoid privacy attacks and to take advantage of the benefits of event data,\nwe consider using only event data. To make full use of the information in the\nevent stream, we propose a Cross-Modality and Temporal Collaboration (CMTC)\nnetwork for event-based video person ReID. First, we design an event transform\nnetwork to obtain corresponding auxiliary information from the input of raw\nevents. Additionally, we propose a differential modality collaboration module\nto balance the roles of events and auxiliaries to achieve complementary\neffects. Furthermore, we introduce a temporal collaboration module to exploit\nmotion information and appearance cues. Experimental results demonstrate that\nour method outperforms others in the task of event-based video person ReID.\n", "date": "2025-01-13", "categories": []}
{"id": "2501.05851", "title": "Identity-aware Feature Decoupling Learning for Clothing-change Person\n  Re-identification", "abstract": "  Clothing-change person re-identification (CC Re-ID) has attracted increasing\nattention in recent years due to its application prospect. Most existing works\nstruggle to adequately extract the ID-related information from the original RGB\nimages. In this paper, we propose an Identity-aware Feature Decoupling (IFD)\nlearning framework to mine identity-related features. Particularly, IFD\nexploits a dual stream architecture that consists of a main stream and an\nattention stream. The attention stream takes the clothing-masked images as\ninputs and derives the identity attention weights for effectively transferring\nthe spatial knowledge to the main stream and highlighting the regions with\nabundant identity-related information. To eliminate the semantic gap between\nthe inputs of two streams, we propose a clothing bias diminishing module\nspecific to the main stream to regularize the features of clothing-relevant\nregions. Extensive experimental results demonstrate that our framework\noutperforms other baseline models on several widely-used CC Re-ID datasets.\n", "date": "2025-01-10", "categories": []}
{"id": "2501.05076", "title": "TipSegNet: Fingertip Segmentation in Contactless Fingerprint Imaging", "abstract": "  Contactless fingerprint recognition systems offer a hygienic, user-friendly,\nand efficient alternative to traditional contact-based methods. However, their\naccuracy heavily relies on precise fingertip detection and segmentation,\nparticularly under challenging background conditions. This paper introduces\nTipSegNet, a novel deep learning model that achieves state-of-the-art\nperformance in segmenting fingertips directly from grayscale hand images.\nTipSegNet leverages a ResNeXt-101 backbone for robust feature extraction,\ncombined with a Feature Pyramid Network (FPN) for multi-scale representation,\nenabling accurate segmentation across varying finger poses and image qualities.\nFurthermore, we employ an extensive data augmentation strategy to enhance the\nmodel's generalizability and robustness. TipSegNet outperforms existing\nmethods, achieving a mean Intersection over Union (mIoU) of 0.987 and an\naccuracy of 0.999, representing a significant advancement in contactless\nfingerprint segmentation. This enhanced accuracy has the potential to\nsubstantially improve the reliability and effectiveness of contactless\nbiometric systems in real-world applications.\n", "date": "2025-01-09", "categories": []}
{"id": "2312.16246", "title": "Nighttime Person Re-Identification via Collaborative Enhancement Network\n  with Multi-domain Learning", "abstract": "  Prevalent nighttime person re-identification (ReID) methods typically combine\nimage relighting and ReID networks in a sequential manner. However, their\nperformance (recognition accuracy) is limited by the quality of relighting\nimages and insufficient collaboration between image relighting and ReID tasks.\nTo handle these problems, we propose a novel Collaborative Enhancement Network\ncalled CENet, which performs the multilevel feature interactions in a parallel\nframework, for nighttime person ReID. In particular, the designed parallel\nstructure of CENet can not only avoid the impact of the quality of relighting\nimages on ReID performance, but also allow us to mine the collaborative\nrelations between image relighting and person ReID tasks. To this end, we\nintegrate the multilevel feature interactions in CENet, where we first share\nthe Transformer encoder to build the low-level feature interaction, and then\nperform the feature distillation that transfers the high-level features from\nimage relighting to ReID, thereby alleviating the severe image degradation\nissue caused by the nighttime scenario while avoiding the impact of relighting\nimages. In addition, the sizes of existing real-world nighttime person ReID\ndatasets are limited, and large-scale synthetic ones exhibit substantial domain\ngaps with real-world data. To leverage both small-scale real-world and\nlarge-scale synthetic training data, we develop a multi-domain learning\nalgorithm, which alternately utilizes both kinds of data to reduce the\ninter-domain difference in training procedure. Extensive experiments on two\nreal nighttime datasets, \\textit{Night600} and \\textit{RGBNT201$_{rgb}$}, and a\nsynthetic nighttime ReID dataset are conducted to validate the effectiveness of\nCENet. We release the code and synthetic dataset at:\n\\hyperlink{https://github.com/Alexadlu/CENet}{\\color{red}\nhttps://github.com/Alexadlu/CENet}.\n", "date": "2023-12-25", "categories": []}
{"id": "2501.01983", "title": "ECG-guided individual identification via PPG", "abstract": "  Photoplethsmography (PPG)-based individual identification aiming at\nrecognizing humans via intrinsic cardiovascular activities has raised extensive\nattention due to its high security and resistance to mimicry. However, this\nkind of technology witnesses unpromising results due to the limitation of low\ninformation density. To this end, electrocardiogram (ECG) signals have been\nintroduced as a novel modality to enhance the density of input information.\nSpecifically, a novel cross-modal knowledge distillation framework is\nimplemented to propagate discriminate knowledge from ECG modality to PPG\nmodality without incurring additional computational demands at the inference\nphase. Furthermore, to ensure efficient knowledge propagation, Contrastive\nLanguage-Image Pre-training (CLIP)-based knowledge alignment and\ncross-knowledge assessment modules are proposed respectively. Comprehensive\nexperiments are conducted and results show our framework outperforms the\nbaseline model with the improvement of 2.8% and 3.0% in terms of overall\naccuracy on seen- and unseen individual recognitions.\n", "date": "2024-12-30", "categories": []}
{"id": "2407.03043", "title": "SlerpFace: Face Template Protection via Spherical Linear Interpolation", "abstract": "  Contemporary face recognition systems use feature templates extracted from\nface images to identify persons. To enhance privacy, face template protection\ntechniques are widely employed to conceal sensitive identity and appearance\ninformation stored in the template. This paper identifies an emerging privacy\nattack form utilizing diffusion models that could nullify prior protection. The\nattack can synthesize high-quality, identity-preserving face images from\ntemplates, revealing persons' appearance. Based on studies of the diffusion\nmodel's generative capability, this paper proposes a defense by rotating\ntemplates to a noise-like distribution. This is achieved efficiently by\nspherically and linearly interpolating templates on their located hypersphere.\nThis paper further proposes to group-wisely divide and drop out templates'\nfeature dimensions, to enhance the irreversibility of rotated templates. The\nproposed techniques are concretized as a novel face template protection\ntechnique, SlerpFace. Extensive experiments show that SlerpFace provides\nsatisfactory recognition accuracy and comprehensive protection against\ninversion and other attack forms, superior to prior arts.\n", "date": "2024-07-03", "categories": []}
{"id": "2412.19111", "title": "Spectral Enhancement and Pseudo-Anchor Guidance for Infrared-Visible\n  Person Re-Identification", "abstract": "  The development of deep learning has facilitated the application of person\nre-identification (ReID) technology in intelligent security. Visible-infrared\nperson re-identification (VI-ReID) aims to match pedestrians across infrared\nand visible modality images enabling 24-hour surveillance. Current studies\nrelying on unsupervised modality transformations as well as inefficient\nembedding constraints to bridge the spectral differences between infrared and\nvisible images, however, limit their potential performance. To tackle the\nlimitations of the above approaches, this paper introduces a simple yet\neffective Spectral Enhancement and Pseudo-anchor Guidance Network, named\nSEPG-Net. Specifically, we propose a more homogeneous spectral enhancement\nscheme based on frequency domain information and greyscale space, which avoids\nthe information loss typically caused by inefficient modality transformations.\nFurther, a Pseudo Anchor-guided Bidirectional Aggregation (PABA) loss is\nintroduced to bridge local modality discrepancies while better preserving\ndiscriminative identity embeddings. Experimental results on two public\nbenchmark datasets demonstrate the superior performance of SEPG-Net against\nother state-of-the-art methods. The code is available at\nhttps://github.com/1024AILab/ReID-SEPG.\n", "date": "2024-12-26", "categories": []}
{"id": "2501.01375", "title": "Iris Recognition for Infants", "abstract": "  Non-invasive, efficient, physical token-less, accurate and stable\nidentification methods for newborns may prevent baby swapping at birth, limit\nbaby abductions and improve post-natal health monitoring across geographies,\nwithin the context of both the formal (i.e., hospitals) and informal (i.e.,\nhumanitarian and fragile settings) health sectors. This paper explores the\nfeasibility of application iris recognition to build biometric identifiers for\n4-6 week old infants. We (a) collected near infrared (NIR) iris images from 17\ninfants using a specially-designed NIR iris sensor; (b) evaluated six iris\nrecognition methods to assess readiness of the state-of-the-art iris\nrecognition to be applied to newborns and infants; (c) proposed a new\nsegmentation model that correctly detects iris texture within infants iris\nimages, and coupled it with several iris texture encoding approaches to offer,\nto the first of our knowledge, a fully-operational infant iris recognition\nsystem; and, (d) trained a StyleGAN-based model to synthesize iris images\nmimicking samples acquired from infants to deliver to the research community\nprivacy-safe infant iris images. The proposed system, incorporating the\nspecially-designed iris sensor and segmenter, and applied to the collected\ninfant iris samples, achieved Equal Error Rate (EER) of 3\\% and Area Under ROC\nCurve (AUC) of 99\\%, compared to EER$\\geq$20\\% and AUC$\\leq$88\\% obtained for\nstate of the art adult iris recognition systems. This suggests that it may be\nfeasible to design methods that succesfully extract biometric features from\ninfant irises.\n", "date": "2025-01-02", "categories": []}
{"id": "2401.11598", "title": "TetraLoss: Improving the Robustness of Face Recognition against Morphing\n  Attacks", "abstract": "  Face recognition systems are widely deployed in high-security applications\nsuch as for biometric verification at border controls. Despite their high\naccuracy on pristine data, it is well-known that digital manipulations, such as\nface morphing, pose a security threat to face recognition systems. Malicious\nactors can exploit the facilities offered by the identity document issuance\nprocess to obtain identity documents containing morphed images. Thus, subjects\nwho contributed to the creation of the morphed image can with high probability\nuse the identity document to bypass automated face recognition systems. In\nrecent years, no-reference (i.e., single image) and differential morphing\nattack detectors have been proposed to tackle this risk. These systems are\ntypically evaluated in isolation from the face recognition system that they\nhave to operate jointly with and do not consider the face recognition process.\nContrary to most existing works, we present a novel method for adapting deep\nlearning-based face recognition systems to be more robust against face morphing\nattacks. To this end, we introduce TetraLoss, a novel loss function that learns\nto separate morphed face images from its contributing subjects in the embedding\nspace while still achieving high biometric verification performance. In a\ncomprehensive evaluation, we show that the proposed method can significantly\nenhance the original system while also significantly outperforming other tested\nbaseline methods.\n", "date": "2024-01-21", "categories": []}
{"id": "2412.04719", "title": "Mix-Modality Person Re-Identification: A New and Practical Paradigm", "abstract": "  Current visible-infrared cross-modality person re-identification research has\nonly focused on exploring the bi-modality mutual retrieval paradigm, and we\npropose a new and more practical mix-modality retrieval paradigm. Existing\nVisible-Infrared person re-identification (VI-ReID) methods have achieved some\nresults in the bi-modality mutual retrieval paradigm by learning the\ncorrespondence between visible and infrared modalities. However, significant\nperformance degradation occurs due to the modality confusion problem when these\nmethods are applied to the new mix-modality paradigm. Therefore, this paper\nproposes a Mix-Modality person re-identification (MM-ReID) task, explores the\ninfluence of modality mixing ratio on performance, and constructs mix-modality\ntest sets for existing datasets according to the new mix-modality testing\nparadigm. To solve the modality confusion problem in MM-ReID, we propose a\nCross-Identity Discrimination Harmonization Loss (CIDHL) adjusting the\ndistribution of samples in the hyperspherical feature space, pulling the\ncenters of samples with the same identity closer, and pushing away the centers\nof samples with different identities while aggregating samples with the same\nmodality and the same identity. Furthermore, we propose a Modality Bridge\nSimilarity Optimization Strategy (MBSOS) to optimize the cross-modality\nsimilarity between the query and queried samples with the help of the similar\nbridge sample in the gallery. Extensive experiments demonstrate that compared\nto the original performance of existing cross-modality methods on MM-ReID, the\naddition of our CIDHL and MBSOS demonstrates a general improvement.\n", "date": "2024-12-06", "categories": []}
{"id": "2412.20327", "title": "Motion Transfer-Driven intra-class data augmentation for Finger Vein\n  Recognition", "abstract": "  Finger vein recognition (FVR) has emerged as a secure biometric technique\nbecause of the confidentiality of vascular bio-information. Recently, deep\nlearning-based FVR has gained increased popularity and achieved promising\nperformance. However, the limited size of public vein datasets has caused\noverfitting issues and greatly limits the recognition performance. Although\ntraditional data augmentation can partially alleviate this data shortage issue,\nit cannot capture the real finger posture variations due to the rigid\nlabel-preserving image transformations, bringing limited performance\nimprovement. To address this issue, we propose a novel motion transfer (MT)\nmodel for finger vein image data augmentation via modeling the actual finger\nposture and rotational movements. The proposed model first utilizes a key point\ndetector to extract the key point and pose map of the source and drive finger\nvein images. We then utilize a dense motion module to estimate the motion\noptical flow, which is fed to an image generation module for generating the\nimage with the target pose. Experiments conducted on three public finger vein\ndatabases demonstrate that the proposed motion transfer model can effectively\nimprove recognition accuracy. Code is available at:\nhttps://github.com/kevinhuangxf/FingerVeinRecognition.\n", "date": "2024-12-29", "categories": []}
{"id": "2412.20530", "title": "KVC-onGoing: Keystroke Verification Challenge", "abstract": "  This article presents the Keystroke Verification Challenge - onGoing\n(KVC-onGoing), on which researchers can easily benchmark their systems in a\ncommon platform using large-scale public databases, the Aalto University\nKeystroke databases, and a standard experimental protocol. The keystroke data\nconsist of tweet-long sequences of variable transcript text from over 185,000\nsubjects, acquired through desktop and mobile keyboards simulating real-life\nconditions. The results on the evaluation set of KVC-onGoing have proved the\nhigh discriminative power of keystroke dynamics, reaching values as low as\n3.33% of Equal Error Rate (EER) and 11.96% of False Non-Match Rate (FNMR) @1%\nFalse Match Rate (FMR) in the desktop scenario, and 3.61% of EER and 17.44% of\nFNMR @1% at FMR in the mobile scenario, significantly improving previous\nstate-of-the-art results. Concerning demographic fairness, the analyzed scores\nreflect the subjects' age and gender to various extents, not negligible in a\nfew cases. The framework runs on CodaLab.\n", "date": "2024-12-29", "categories": []}
{"id": "2409.06371", "title": "Distilling Generative-Discriminative Representations for Very\n  Low-Resolution Face Recognition", "abstract": "  Very low-resolution face recognition is challenging due to the serious loss\nof informative facial details in resolution degradation. In this paper, we\npropose a generative-discriminative representation distillation approach that\ncombines generative representation with cross-resolution aligned knowledge\ndistillation. This approach facilitates very low-resolution face recognition by\njointly distilling generative and discriminative models via two distillation\nmodules. Firstly, the generative representation distillation takes the encoder\nof a diffusion model pretrained for face super-resolution as the generative\nteacher to supervise the learning of the student backbone via feature\nregression, and then freezes the student backbone. After that, the\ndiscriminative representation distillation further considers a pretrained face\nrecognizer as the discriminative teacher to supervise the learning of the\nstudent head via cross-resolution relational contrastive distillation. In this\nway, the general backbone representation can be transformed into discriminative\nhead representation, leading to a robust and discriminative student model for\nvery low-resolution face recognition. Our approach improves the recovery of the\nmissing details in very low-resolution faces and achieves better knowledge\ntransfer. Extensive experiments on face datasets demonstrate that our approach\nenhances the recognition accuracy of very low-resolution faces, showcasing its\neffectiveness and adaptability.\n", "date": "2024-09-10", "categories": []}
{"id": "2412.00433", "title": "Dynamic Token Selection for Aerial-Ground Person Re-Identification", "abstract": "  Aerial-Ground Person Re-identification (AGPReID) holds significant practical\nvalue but faces unique challenges due to pronounced variations in viewing\nangles, lighting conditions, and background interference. Traditional methods,\noften involving a global analysis of the entire image, frequently lead to\ninefficiencies and susceptibility to irrelevant data. In this paper, we propose\na novel Dynamic Token Selective Transformer (DTST) tailored for AGPReID, which\ndynamically selects pivotal tokens to concentrate on pertinent regions.\nSpecifically, we segment the input image into multiple tokens, with each token\nrepresenting a unique region or feature within the image. Using a Top-k\nstrategy, we extract the k most significant tokens that contain vital\ninformation essential for identity recognition. Subsequently, an attention\nmechanism is employed to discern interrelations among diverse tokens, thereby\nenhancing the representation of identity features. Extensive experiments on\nbenchmark datasets showcases the superiority of our method over existing works.\nNotably, on the CARGO dataset, our proposed method gains 1.18% mAP improvements\nwhen compared to the second place. In addition, we comprehensively analyze the\nimpact of different numbers of tokens, token insertion positions, and numbers\nof heads on model performance.\n", "date": "2024-11-30", "categories": []}
{"id": "2412.05721", "title": "Impact of Sunglasses on One-to-Many Facial Identification Accuracy", "abstract": "  One-to-many facial identification is documented to achieve high accuracy in\nthe case where both the probe and the gallery are \"mugshot quality\" images.\nHowever, an increasing number of documented instances of wrongful arrest\nfollowing one-to-many facial identification have raised questions about its\naccuracy. Probe images used in one-to-many facial identification are often\ncropped from frames of surveillance video and deviate from \"mugshot quality\" in\nvarious ways. This paper systematically explores how the accuracy of\none-to-many facial identification is degraded by the person in the probe image\nchoosing to wear dark sunglasses. We show that sunglasses degrade accuracy for\nmugshot-quality images by an amount similar to strong blur or noticeably lower\nresolution. Further, we demonstrate that the combination of sunglasses with\nblur or lower resolution results in even more pronounced loss in accuracy.\nThese results have important implications for developing objective criteria to\nqualify a probe image for the level of accuracy to be expected if it used for\none-to-many identification. To ameliorate the accuracy degradation caused by\ndark sunglasses, we show that it is possible to recover about 38% of the lost\naccuracy by synthetically adding sunglasses to all the gallery images, without\nmodel re-training. We also show that the frequency of wearing-sunglasses images\nis very low in existing training sets, and that increasing the representation\nof wearing-sunglasses images can greatly reduce the error rate. The image set\nassembled for this research is available at https://cvrl.nd.edu/projects/data/\nto support replication and further research.\n", "date": "2024-12-07", "categories": []}
{"id": "2412.19134", "title": "Extended Cross-Modality United Learning for Unsupervised\n  Visible-Infrared Person Re-identification", "abstract": "  Unsupervised learning visible-infrared person re-identification (USL-VI-ReID)\naims to learn modality-invariant features from unlabeled cross-modality\ndatasets and reduce the inter-modality gap. However, the existing methods lack\ncross-modality clustering or excessively pursue cluster-level association,\nwhich makes it difficult to perform reliable modality-invariant features\nlearning. To deal with this issue, we propose a Extended Cross-Modality United\nLearning (ECUL) framework, incorporating Extended Modality-Camera Clustering\n(EMCC) and Two-Step Memory Updating Strategy (TSMem) modules. Specifically, we\ndesign ECUL to naturally integrates intra-modality clustering, inter-modality\nclustering and inter-modality instance selection, establishing compact and\naccurate cross-modality associations while reducing the introduction of noisy\nlabels. Moreover, EMCC captures and filters the neighborhood relationships by\nextending the encoding vector, which further promotes the learning of\nmodality-invariant and camera-invariant knowledge in terms of clustering\nalgorithm. Finally, TSMem provides accurate and generalized proxy points for\ncontrastive learning by updating the memory in stages. Extensive experiments\nresults on SYSU-MM01 and RegDB datasets demonstrate that the proposed ECUL\nshows promising performance and even outperforms certain supervised methods.\n", "date": "2024-12-26", "categories": []}
{"id": "2402.01269", "title": "Spectrum-guided Feature Enhancement Network for Event Person\n  Re-Identification", "abstract": "  As a cutting-edge biosensor, the event camera holds significant potential in\nthe field of computer vision, particularly regarding privacy preservation.\nHowever, compared to traditional cameras, event streams often contain noise and\npossess extremely sparse semantics, posing a formidable challenge for\nevent-based person re-identification (event Re-ID). To address this, we\nintroduce a novel event person re-identification network: the Spectrum-guided\nFeature Enhancement Network (SFE-Net). This network consists of two innovative\ncomponents: the Multi-grain Spectrum Attention Mechanism (MSAM) and the\nConsecutive Patch Dropout Module (CPDM). MSAM employs a fourier spectrum\ntransform strategy to filter event noise, while also utilizing an event-guided\nmulti-granularity attention strategy to enhance and capture discriminative\nperson semantics. CPDM employs a consecutive patch dropout strategy to generate\nmultiple incomplete feature maps, encouraging the deep Re-ID model to equally\nperceive each effective region of the person's body and capture robust person\ndescriptors. Extensive experiments on Event Re-ID datasets demonstrate that our\nSFE-Net achieves the best performance in this task.\n", "date": "2024-02-02", "categories": []}
{"id": "2412.17239", "title": "Unity is Strength: Unifying Convolutional and Transformeral Features for\n  Better Person Re-Identification", "abstract": "  Person Re-identification (ReID) aims to retrieve the specific person across\nnon-overlapping cameras, which greatly helps intelligent transportation\nsystems. As we all know, Convolutional Neural Networks (CNNs) and Transformers\nhave the unique strengths to extract local and global features, respectively.\nConsidering this fact, we focus on the mutual fusion between them to learn more\ncomprehensive representations for persons. In particular, we utilize the\ncomplementary integration of deep features from different model structures. We\npropose a novel fusion framework called FusionReID to unify the strengths of\nCNNs and Transformers for image-based person ReID. More specifically, we first\ndeploy a Dual-branch Feature Extraction (DFE) to extract features through CNNs\nand Transformers from a single image. Moreover, we design a novel\nDual-attention Mutual Fusion (DMF) to achieve sufficient feature fusions. The\nDMF comprises Local Refinement Units (LRU) and Heterogenous Transmission\nModules (HTM). LRU utilizes depth-separable convolutions to align deep features\nin channel dimensions and spatial sizes. HTM consists of a Shared Encoding Unit\n(SEU) and two Mutual Fusion Units (MFU). Through the continuous stacking of\nHTM, deep features after LRU are repeatedly utilized to generate more\ndiscriminative features. Extensive experiments on three public ReID benchmarks\ndemonstrate that our method can attain superior performances than most\nstate-of-the-arts. The source code is available at\nhttps://github.com/924973292/FusionReID.\n", "date": "2024-12-23", "categories": []}
{"id": "2412.13099", "title": "Accuracy Limits as a Barrier to Biometric System Security", "abstract": "  Biometric systems are widely used for identity verification and\nidentification, including authentication (i.e., one-to-one matching to verify a\nclaimed identity) and identification (i.e., one-to-many matching to find a\nsubject in a database). The matching process relies on measuring similarities\nor dissimilarities between a fresh biometric template and enrolled templates.\nThe False Match Rate FMR is a key metric for assessing the accuracy and\nreliability of such systems. This paper analyzes biometric systems based on\ntheir FMR, with two main contributions. First, we explore untargeted attacks,\nwhere an adversary aims to impersonate any user within a database. We determine\nthe number of trials required for an attacker to successfully impersonate a\nuser and derive the critical population size (i.e., the maximum number of users\nin the database) required to maintain a given level of security. Furthermore,\nwe compute the critical FMR value needed to ensure resistance against\nuntargeted attacks as the database size increases. Second, we revisit the\nbiometric birthday problem to evaluate the approximate and exact probabilities\nthat two users in a database collide (i.e., can impersonate each other). Based\non this analysis, we derive both the approximate critical population size and\nthe critical FMR value needed to bound the likelihood of such collisions\noccurring with a given probability. These thresholds offer insights for\ndesigning systems that mitigate the risk of impersonation and collisions,\nparticularly in large-scale biometric databases. Our findings indicate that\ncurrent biometric systems fail to deliver sufficient accuracy to achieve an\nadequate security level against untargeted attacks, even in small-scale\ndatabases. Moreover, state-of-the-art systems face significant challenges in\naddressing the biometric birthday problem, especially as database sizes grow.\n", "date": "2024-12-17", "categories": []}
{"id": "2412.14404", "title": "Enhancing Fingerprint Recognition Systems: Comparative Analysis of\n  Biometric Authentication Algorithms and Techniques for Improved Accuracy and\n  Reliability", "abstract": "  Fingerprint recognition systems stand as pillars in the realm of biometric\nauthentication, providing indispensable security measures across various\ndomains. This study investigates integrating Convolutional Neural Networks\n(CNNs) with Gabor filters to improve fingerprint recognition accuracy and\nrobustness. Leveraging a diverse dataset sourced from the Sokoto Coventry\nFingerprint Dataset, our experiments meticulously evaluate the efficacy of\ndifferent classification algorithms. Our findings underscore the supremacy of\nCNN-based approaches, boasting an impressive overall accuracy of 94\\%.\nFurthermore, the amalgamation of Gabor filters with CNN architectures unveils\npromising strides in discerning altered fingerprints, illuminating new pathways\nfor enhancing biometric authentication systems. While the CNN-Gabor fusion\nshowcases commendable performance, our exploration of hybrid approaches\ncombining multiple classifiers reveals nuanced outcomes. Despite these mixed\nresults, our study illuminates the transformative potential of deep learning\nmethodologies in reshaping the landscape of fingerprint recognition. Through\nrigorous experimentation and insightful analysis, this research not only\ncontributes to advancing biometric authentication technologies but also sheds\nlight on the intricate interplay between traditional feature extraction methods\nand cutting-edge deep learning architectures. These findings offer actionable\ninsights for optimizing fingerprint recognition systems for real-world\ndeployment, paving the way for enhanced security and reliability in diverse\napplications.\n", "date": "2024-12-18", "categories": []}
{"id": "2404.09507", "title": "Clothes-Changing Person Re-Identification with Feasibility-Aware\n  Intermediary Matching", "abstract": "  Current clothes-changing person re-identification (re-id) approaches usually\nperform retrieval based on clothes-irrelevant features, while neglecting the\npotential of clothes-relevant features. However, we observe that relying solely\non clothes-irrelevant features for clothes-changing re-id is limited, since\nthey often lack adequate identity information and suffer from large intra-class\nvariations. On the contrary, clothes-relevant features can be used to discover\nsame-clothes intermediaries that possess informative identity clues. Based on\nthis observation, we propose a Feasibility-Aware Intermediary Matching (FAIM)\nframework to additionally utilize clothes-relevant features for retrieval.\nFirstly, an Intermediary Matching (IM) module is designed to perform an\nintermediary-assisted matching process. This process involves using\nclothes-relevant features to find informative intermediates, and then using\nclothes-irrelevant features of these intermediates to complete the matching.\nSecondly, in order to reduce the negative effect of low-quality intermediaries,\nan Intermediary-Based Feasibility Weighting (IBFW) module is designed to\nevaluate the feasibility of intermediary matching process by assessing the\nquality of intermediaries. Extensive experiments demonstrate that our method\noutperforms state-of-the-art methods on several widely-used clothes-changing\nre-id benchmarks.\n", "date": "2024-04-15", "categories": []}
{"id": "2404.10476", "title": "Enhanced Facial Feature Extraction and Recignation Using Optimal Fully\n  Dispersed Haar-like Filters", "abstract": "  Haar-like filters are renowned for their simplicity, speed, and accuracy in\nvarious computer vision tasks. This paper proposes a novel algorithm to\nidentify optimal fully dispersed Haar-like filters for enhanced facial feature\nextraction and recognation. Unlike traditional Haar-like filters, these novel\nfilters allow pixels to move freely within images, enabling more effictive\ncapture of intricate local features...\n", "date": "2024-04-16", "categories": []}
{"id": "2412.09224", "title": "DASK: Distribution Rehearsing via Adaptive Style Kernel Learning for\n  Exemplar-Free Lifelong Person Re-Identification", "abstract": "  Lifelong person re-identification (LReID) is an important but challenging\ntask that suffers from catastrophic forgetting due to significant domain gaps\nbetween training steps. Existing LReID approaches typically rely on data replay\nand knowledge distillation to mitigate this issue. However, data replay methods\ncompromise data privacy by storing historical exemplars, while knowledge\ndistillation methods suffer from limited performance due to the cumulative\nforgetting of undistilled knowledge. To overcome these challenges, we propose a\nnovel paradigm that models and rehearses the distribution of the old domains to\nenhance knowledge consolidation during the new data learning, possessing a\nstrong anti-forgetting capacity without storing any exemplars. Specifically, we\nintroduce an exemplar-free LReID method called Distribution Rehearsing via\nAdaptive Style Kernel Learning (DASK). DASK includes a Distribution Rehearser\nLearning (DRL) mechanism that learns to transform arbitrary distribution data\ninto the current data style at each learning step. To enhance the style\ntransfer capacity of DRL, an Adaptive Kernel Prediction Network (AKPNet) is\nexplored to achieve an instance-specific distribution adjustment. Additionally,\nwe design a Distribution Rehearsing-driven LReID Training (DRRT) module, which\nrehearses old distribution based on the new data via the old AKPNet model,\nachieving effective new-old knowledge accumulation under a joint knowledge\nconsolidation scheme. Experimental results show our DASK outperforms the\nexisting methods by 3.6%-6.8% and 4.5%-6.5% on anti-forgetting and\ngeneralization capacity, respectively. Our code is available at\nhttps://github.com/zhoujiahuan1991/AAAI2025-LReID-DASK\n", "date": "2024-12-12", "categories": []}
{"id": "2412.13063", "title": "Smartphone-based Iris Recognition through High-Quality Visible Spectrum\n  Iris Capture", "abstract": "  Iris recognition is widely acknowledged for its exceptional accuracy in\nbiometric authentication, traditionally relying on near-infrared (NIR) imaging.\nRecently, visible spectrum (VIS) imaging via accessible smartphone cameras has\nbeen explored for biometric capture. However, a thorough study of iris\nrecognition using smartphone-captured 'High-Quality' VIS images and\ncross-spectral matching with previously enrolled NIR images has not been\nconducted. The primary challenge lies in capturing high-quality biometrics, a\nknown limitation of smartphone cameras. This study introduces a novel Android\napplication designed to consistently capture high-quality VIS iris images\nthrough automated focus and zoom adjustments. The application integrates a\nYOLOv3-tiny model for precise eye and iris detection and a lightweight\nGhost-Attention U-Net (G-ATTU-Net) for segmentation, while adhering to ISO/IEC\n29794-6 standards for image quality. The approach was validated using\nsmartphone-captured VIS and NIR iris images from 47 subjects, achieving a True\nAcceptance Rate (TAR) of 96.57% for VIS images and 97.95% for NIR images, with\nconsistent performance across various capture distances and iris colors. This\nrobust solution is expected to significantly advance the field of iris\nbiometrics, with important implications for enhancing smartphone security.\n", "date": "2024-12-17", "categories": []}
{"id": "2401.02447", "title": "User authentication system based on human exhaled breath physics", "abstract": "  This work, in a pioneering approach, attempts to build a biometric system\nthat works purely based on the fluid mechanics governing exhaled breath. We\ntest the hypothesis that the structure of turbulence in exhaled human breath\ncan be exploited to build biometric algorithms. This work relies on the idea\nthat the extrathoracic airway is unique for every individual, making the\nexhaled breath a biomarker. Methods including classical multi-dimensional\nhypothesis testing approach and machine learning models are employed in\nbuilding user authentication algorithms, namely user confirmation and user\nidentification. A user confirmation algorithm tries to verify whether a user is\nthe person they claim to be. A user identification algorithm tries to identify\na user's identity with no prior information available. A dataset of exhaled\nbreath time series samples from 94 human subjects was used to evaluate the\nperformance of these algorithms. The user confirmation algorithms performed\nexceedingly well for the given dataset with over $97\\%$ true confirmation rate.\nThe machine learning based algorithm achieved a good true confirmation rate,\nreiterating our understanding of why machine learning based algorithms\ntypically outperform classical hypothesis test based algorithms. The user\nidentification algorithm performs reasonably well with the provided dataset\nwith over $50\\%$ of the users identified as being within two possible suspects.\nWe show surprisingly unique turbulent signatures in the exhaled breath that\nhave not been discovered before. In addition to discussions on a novel\nbiometric system, we make arguments to utilise this idea as a tool to gain\ninsights into the morphometric variation of extrathoracic airway across\nindividuals. Such tools are expected to have future potential in the area of\npersonalised medicines.\n", "date": "2024-01-02", "categories": []}
{"id": "2405.03959", "title": "Joint Identity Verification and Pose Alignment for Partial Fingerprints", "abstract": "  Currently, portable electronic devices are becoming more and more popular.\nFor lightweight considerations, their fingerprint recognition modules usually\nuse limited-size sensors. However, partial fingerprints have few matchable\nfeatures, especially when there are differences in finger pressing posture or\nimage quality, which makes partial fingerprint verification challenging. Most\nexisting methods regard fingerprint position rectification and identity\nverification as independent tasks, ignoring the coupling relationship between\nthem -- relative pose estimation typically relies on paired features as\nanchors, and authentication accuracy tends to improve with more precise pose\nalignment. In this paper, we propose a novel framework for joint identity\nverification and pose alignment of partial fingerprint pairs, aiming to\nleverage their inherent correlation to improve each other. To achieve this, we\npresent a multi-task CNN (Convolutional Neural Network)-Transformer hybrid\nnetwork, and design a pre-training task to enhance the feature extraction\ncapability. Experiments on multiple public datasets (NIST SD14, FVC2002 DB1A &\nDB3A, FVC2004 DB1A & DB2A, FVC2006 DB1A) and an in-house dataset show that our\nmethod achieves state-of-the-art performance in both partial fingerprint\nverification and relative pose estimation, while being more efficient than\nprevious methods.\n", "date": "2024-05-07", "categories": []}
{"id": "2412.12031", "title": "RepFace: Refining Closed-Set Noise with Progressive Label Correction for\n  Face Recognition", "abstract": "  Face recognition has made remarkable strides, driven by the expanding scale\nof datasets, advancements in various backbone and discriminative losses.\nHowever, face recognition performance is heavily affected by the label noise,\nespecially closed-set noise. While numerous studies have focused on handling\nlabel noise, addressing closed-set noise still poses challenges. This paper\nidentifies this challenge as training isn't robust to noise at the early-stage\ntraining, and necessitating an appropriate learning strategy for samples with\nlow confidence, which are often misclassified as closed-set noise in later\ntraining phases. To address these issues, we propose a new framework to\nstabilize the training at early stages and split the samples into clean,\nambiguous and noisy groups which are devised with separate training strategies.\nInitially, we employ generated auxiliary closed-set noisy samples to enable the\nmodel to identify noisy data at the early stages of training. Subsequently, we\nintroduce how samples are split into clean, ambiguous and noisy groups by their\nsimilarity to the positive and nearest negative centers. Then we perform label\nfusion for ambiguous samples by incorporating accumulated model predictions.\nFinally, we apply label smoothing within the closed set, adjusting the label to\na point between the nearest negative class and the initially assigned label.\nExtensive experiments validate the effectiveness of our method on mainstream\nface datasets, achieving state-of-the-art results. The code will be released\nupon acceptance.\n", "date": "2024-12-16", "categories": []}
{"id": "2407.07627", "title": "Synthetic to Authentic: Transferring Realism to 3D Face Renderings for\n  Boosting Face Recognition", "abstract": "  In this paper, we investigate the potential of image-to-image translation\n(I2I) techniques for transferring realism to 3D-rendered facial images in the\ncontext of Face Recognition (FR) systems. The primary motivation for using\n3D-rendered facial images lies in their ability to circumvent the challenges\nassociated with collecting large real face datasets for training FR systems.\nThese images are generated entirely by 3D rendering engines, facilitating the\ngeneration of synthetic identities. However, it has been observed that FR\nsystems trained on such synthetic datasets underperform when compared to those\ntrained on real datasets, on various FR benchmarks. In this work, we\ndemonstrate that by transferring the realism to 3D-rendered images (i.e.,\nmaking the 3D-rendered images look more real), we can boost the performance of\nFR systems trained on these more photorealistic images. This improvement is\nevident when these systems are evaluated against FR benchmarks utilizing\nreal-world data, thereby paving new pathways for employing synthetic data in\nreal-world applications.\n", "date": "2024-07-10", "categories": []}
{"id": "2207.11759", "title": "Spatial-Temporal Federated Learning for Lifelong Person\n  Re-identification on Distributed Edges", "abstract": "  Data drift is a thorny challenge when deploying person re-identification\n(ReID) models into real-world devices, where the data distribution is\nsignificantly different from that of the training environment and keeps\nchanging. To tackle this issue, we propose a federated spatial-temporal\nincremental learning approach, named FedSTIL, which leverages both lifelong\nlearning and federated learning to continuously optimize models deployed on\nmany distributed edge clients. Unlike previous efforts, FedSTIL aims to mine\nspatial-temporal correlations among the knowledge learnt from different edge\nclients. Specifically, the edge clients first periodically extract general\nrepresentations of drifted data to optimize their local models. Then, the\nlearnt knowledge from edge clients will be aggregated by centralized parameter\nserver, where the knowledge will be selectively and attentively distilled from\nspatial- and temporal-dimension with carefully designed mechanisms. Finally,\nthe distilled informative spatial-temporal knowledge will be sent back to\ncorrelated edge clients to further improve the recognition accuracy of each\nedge client with a lifelong learning method. Extensive experiments on a mixture\nof five real-world datasets demonstrate that our method outperforms others by\nnearly 4% in Rank-1 accuracy, while reducing communication cost by 62%. All\nimplementation codes are publicly available on\nhttps://github.com/MSNLAB/Federated-Lifelong-Person-ReID\n", "date": "2022-07-24", "categories": []}
{"id": "2404.11118", "title": "FastFace: Fast-converging Scheduler for Large-scale Face Recognition\n  Training with One GPU", "abstract": "  Computing power has evolved into a foundational and indispensable resource in\nthe area of deep learning, particularly in tasks such as Face Recognition (FR)\nmodel training on large-scale datasets, where multiple GPUs are often a\nnecessity. Recognizing this challenge, some FR methods have started exploring\nways to compress the fully-connected layer in FR models. Unlike other\napproaches, our observations reveal that without prompt scheduling of the\nlearning rate (LR) during FR model training, the loss curve tends to exhibit\nnumerous stationary subsequences. To address this issue, we introduce a novel\nLR scheduler leveraging Exponential Moving Average (EMA) and Haar Convolutional\nKernel (HCK) to eliminate stationary subsequences, resulting in a significant\nreduction in converging time. However, the proposed scheduler incurs a\nconsiderable computational overhead due to its time complexity. To overcome\nthis limitation, we propose FastFace, a fast-converging scheduler with\nnegligible time complexity, i.e. O(1) per iteration, during training. In\npractice, FastFace is able to accelerate FR model training to a quarter of its\noriginal time without sacrificing more than 1% accuracy, making large-scale FR\ntraining feasible even with just one single GPU in terms of both time and space\ncomplexity. Extensive experiments validate the efficiency and effectiveness of\nFastFace. The code is publicly available at:\nhttps://github.com/amoonfana/FastFace\n", "date": "2024-04-17", "categories": []}
{"id": "2412.08048", "title": "Surveying Facial Recognition Models for Diverse Indian Demographics: A\n  Comparative Analysis on LFW and Custom Dataset", "abstract": "  Facial recognition technology has made significant advances, yet its\neffectiveness across diverse ethnic backgrounds, particularly in specific\nIndian demographics, is less explored. This paper presents a detailed\nevaluation of both traditional and deep learning-based facial recognition\nmodels using the established LFW dataset and our newly developed IITJ Faces of\nAcademia Dataset (JFAD), which comprises images of students from IIT Jodhpur.\nThis unique dataset is designed to reflect the ethnic diversity of India,\nproviding a critical test bed for assessing model performance in a focused\nacademic environment. We analyze models ranging from holistic approaches like\nEigenfaces and SIFT to advanced hybrid models that integrate CNNs with Gabor\nfilters, Laplacian transforms, and segmentation techniques. Our findings reveal\nsignificant insights into the models' ability to adapt to the ethnic\nvariability within Indian demographics and suggest modifications to enhance\naccuracy and inclusivity in real-world applications. The JFAD not only serves\nas a valuable resource for further research but also highlights the need for\ndeveloping facial recognition systems that perform equitably across diverse\npopulations.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.08276", "title": "Local Features Meet Stochastic Anonymization: Revolutionizing\n  Privacy-Preserving Face Recognition for Black-Box Models", "abstract": "  The task of privacy-preserving face recognition (PPFR) currently faces two\nmajor unsolved challenges: (1) existing methods are typically effective only on\nspecific face recognition models and struggle to generalize to black-box face\nrecognition models; (2) current methods employ data-driven reversible\nrepresentation encoding for privacy protection, making them susceptible to\nadversarial learning and reconstruction of the original image. We observe that\nface recognition models primarily rely on local features ({e.g., face contour,\nskin texture, and so on) for identification. Thus, by disrupting global\nfeatures while enhancing local features, we achieve effective recognition even\nin black-box environments. Additionally, to prevent adversarial models from\nlearning and reversing the anonymization process, we adopt an adversarial\nlearning-based approach with irreversible stochastic injection to ensure the\nstochastic nature of the anonymization. Experimental results demonstrate that\nour method achieves an average recognition accuracy of 94.21\\% on black-box\nmodels, outperforming existing methods in both privacy protection and\nanti-reconstruction capabilities.\n", "date": "2024-12-11", "categories": []}
{"id": "2412.07082", "title": "User Authentication and Vital Signs Extraction from Low-Frame-Rate and\n  Monochrome No-contact Fingerprint Captures", "abstract": "  We present our work on leveraging low-frame-rate monochrome (blue light)\nvideos of fingertips, captured with an off-the-shelf fingerprint capture\ndevice, to extract vital signs and identify users. These videos utilize\nphotoplethysmography (PPG), commonly used to measure vital signs like heart\nrate. While prior research predominantly utilizes high-frame-rate,\nmulti-wavelength PPG sensors (e.g., infrared, red, or RGB), our preliminary\nfindings demonstrate that both user identification and vital sign extraction\nare achievable with the low-frame-rate data we collected. Preliminary results\nare promising, with low error rates for both heart rate estimation and user\nauthentication. These results indicate promise for effective biometric systems.\nWe anticipate further optimization will enhance accuracy and advance healthcare\nand security.\n", "date": "2024-12-10", "categories": []}
{"id": "2412.07771", "title": "PETALface: Parameter Efficient Transfer Learning for Low-resolution Face\n  Recognition", "abstract": "  Pre-training on large-scale datasets and utilizing margin-based loss\nfunctions have been highly successful in training models for high-resolution\nface recognition. However, these models struggle with low-resolution face\ndatasets, in which the faces lack the facial attributes necessary for\ndistinguishing different faces. Full fine-tuning on low-resolution datasets, a\nnaive method for adapting the model, yields inferior performance due to\ncatastrophic forgetting of pre-trained knowledge. Additionally the domain\ndifference between high-resolution (HR) gallery images and low-resolution (LR)\nprobe images in low resolution datasets leads to poor convergence for a single\nmodel to adapt to both gallery and probe after fine-tuning. To this end, we\npropose PETALface, a Parameter-Efficient Transfer Learning approach for\nlow-resolution face recognition. Through PETALface, we attempt to solve both\nthe aforementioned problems. (1) We solve catastrophic forgetting by leveraging\nthe power of parameter efficient fine-tuning(PEFT). (2) We introduce two\nlow-rank adaptation modules to the backbone, with weights adjusted based on the\ninput image quality to account for the difference in quality for the gallery\nand probe images. To the best of our knowledge, PETALface is the first work\nleveraging the powers of PEFT for low resolution face recognition. Extensive\nexperiments demonstrate that the proposed method outperforms full fine-tuning\non low-resolution datasets while preserving performance on high-resolution and\nmixed-quality datasets, all while using only 0.48% of the parameters. Code:\nhttps://kartik-3004.github.io/PETALface/\n", "date": "2024-12-10", "categories": []}
{"id": "2412.05660", "title": "Multimodal Biometric Authentication Using Camera-Based PPG and\n  Fingerprint Fusion", "abstract": "  Camera-based photoplethysmography (PPG) obtained from smartphones has shown\ngreat promise for personalized healthcare and secure authentication. This paper\npresents a multimodal biometric system that integrates PPG signals extracted\nfrom videos with fingerprint data to enhance the accuracy of user verification.\nThe system requires users to place their fingertip on the camera lens for a few\nseconds, allowing the capture and processing of unique biometric\ncharacteristics. Our approach employs a neural network with two structured\nstate-space model (SSM) encoders to manage the distinct modalities. Fingerprint\nimages are transformed into pixel sequences, and along with segmented PPG\nwaveforms, they are input into the encoders. A cross-modal attention mechanism\nthen extracts refined feature representations, and a distribution-oriented\ncontrastive loss function aligns these features within a unified latent space.\nExperimental results demonstrate the system's superior performance across\nvarious evaluation metrics in both single-session and dual-session\nauthentication scenarios.\n", "date": "2024-12-07", "categories": []}
{"id": "2411.16169", "title": "Local and Global Feature Attention Fusion Network for Face Recognition", "abstract": "  Recognition of low-quality face images remains a challenge due to invisible\nor deformation in partial facial regions. For low-quality images dominated by\nmissing partial facial regions, local region similarity contributes more to\nface recognition (FR). Conversely, in cases dominated by local face\ndeformation, excessive attention to local regions may lead to misjudgments,\nwhile global features exhibit better robustness. However, most of the existing\nFR methods neglect the bias in feature quality of low-quality images introduced\nby different factors. To address this issue, we propose a Local and Global\nFeature Attention Fusion (LGAF) network based on feature quality. The network\nadaptively allocates attention between local and global features according to\nfeature quality and obtains more discriminative and high-quality face features\nthrough local and global information complementarity. In addition, to\neffectively obtain fine-grained information at various scales and increase the\nseparability of facial features in high-dimensional space, we introduce a\nMulti-Head Multi-Scale Local Feature Extraction (MHMS) module. Experimental\nresults demonstrate that the LGAF achieves the best average performance on $4$\nvalidation sets (CFP-FP, CPLFW, AgeDB, and CALFW), and the performance on\nTinyFace and SCFace outperforms the state-of-the-art methods (SoTA).\n", "date": "2024-11-25", "categories": []}
{"id": "2412.01048", "title": "Cerberus: Attribute-based person re-identification using semantic IDs", "abstract": "  We introduce a new framework, dubbed Cerberus, for attribute-based person\nre-identification (reID). Our approach leverages person attribute labels to\nlearn local and global person representations that encode specific traits, such\nas gender and clothing style. To achieve this, we define semantic IDs (SIDs) by\ncombining attribute labels, and use a semantic guidance loss to align the\nperson representations with the prototypical features of corresponding SIDs,\nencouraging the representations to encode the relevant semantics.\nSimultaneously, we enforce the representations of the same person to be\nembedded closely, enabling recognizing subtle differences in appearance to\ndiscriminate persons sharing the same attribute labels. To increase the\ngeneralization ability on unseen data, we also propose a regularization method\nthat takes advantage of the relationships between SID prototypes. Our framework\nperforms individual comparisons of local and global person representations\nbetween query and gallery images for attribute-based reID. By exploiting the\nSID prototypes aligned with the corresponding representations, it can also\nperform person attribute recognition (PAR) and attribute-based person search\n(APS) without bells and whistles. Experimental results on standard benchmarks\non attribute-based person reID, Market-1501 and DukeMTMC, demonstrate the\nsuperiority of our model compared to the state of the art.\n", "date": "2024-12-02", "categories": []}
{"id": "2411.19215", "title": "Cross-Spectral Attention for Unsupervised RGB-IR Face Verification and\n  Person Re-identification", "abstract": "  Cross-spectral biometrics, such as matching imagery of faces or persons from\nvisible (RGB) and infrared (IR) bands, have rapidly advanced over the last\ndecade due to increasing sensitivity, size, quality, and ubiquity of IR focal\nplane arrays and enhanced analytics beyond the visible spectrum. Current\ntechniques for mitigating large spectral disparities between RGB and IR imagery\noften include learning a discriminative common subspace by exploiting precisely\ncurated data acquired from multiple spectra. Although there are challenges with\ndetermining robust architectures for extracting common information, a critical\nlimitation for supervised methods is poor scalability in terms of acquiring\nlabeled data. Therefore, we propose a novel unsupervised cross-spectral\nframework that combines (1) a new pseudo triplet loss with cross-spectral\nvoting, (2) a new cross-spectral attention network leveraging multiple\nsubspaces, and (3) structured sparsity to perform more discriminative\ncross-spectral clustering. We extensively compare our proposed RGB-IR biometric\nlearning framework (and its individual components) with recent and previous\nstate-of-the-art models on two challenging benchmark datasets: DEVCOM Army\nResearch Laboratory Visible-Thermal Face Dataset (ARL-VTF) and RegDB person\nre-identification dataset, and, in some cases, achieve performance superior to\ncompletely supervised methods.\n", "date": "2024-11-28", "categories": []}
{"id": "2407.04127", "title": "Biometric Authentication Based on Enhanced Remote Photoplethysmography\n  Signal Morphology", "abstract": "  Remote photoplethysmography (rPPG) is a non-contact method for measuring\ncardiac signals from facial videos, offering a convenient alternative to\ncontact photoplethysmography (cPPG) obtained from contact sensors. Recent\nstudies have shown that each individual possesses a unique cPPG signal\nmorphology that can be utilized as a biometric identifier, which has inspired\nus to utilize the morphology of rPPG signals extracted from facial videos for\nperson authentication. Since the facial appearance and rPPG are mixed in the\nfacial videos, we first de-identify facial videos to remove facial appearance\nwhile preserving the rPPG information, which protects facial privacy and\nguarantees that only rPPG is used for authentication. The de-identified videos\nare fed into an rPPG model to get the rPPG signal morphology for\nauthentication. In the first training stage, unsupervised rPPG training is\nperformed to get coarse rPPG signals. In the second training stage, an\nrPPG-cPPG hybrid training is performed by incorporating external cPPG datasets\nto achieve rPPG biometric authentication and enhance rPPG signal morphology.\nOur approach needs only de-identified facial videos with subject IDs to train\nrPPG authentication models. The experimental results demonstrate that rPPG\nsignal morphology hidden in facial videos can be used for biometric\nauthentication. The code is available at\nhttps://github.com/zhaodongsun/rppg_biometrics.\n", "date": "2024-07-04", "categories": []}
{"id": "2411.17506", "title": "Neural network modelling of kinematic and dynamic features for signature\n  verification", "abstract": "  Online signature parameters, which are based on human characteristics,\nbroaden the applicability of an automatic signature verifier. Although\nkinematic and dynamic features have previously been suggested, accurately\nmeasuring features such as arm and forearm torques remains challenging. We\npresent two approaches for estimating angular velocities, angular positions,\nand force torques. The first approach involves using a physical UR5e robotic\narm to reproduce a signature while capturing those parameters over time. The\nsecond method, a cost effective approach, uses a neural network to estimate the\nsame parameters. Our findings demonstrate that a simple neural network model\ncan extract effective parameters for signature verification. Training the\nneural network with the MCYT300 dataset and cross validating with other\ndatabases, namely, BiosecurID, Visual, Blind, OnOffSigDevanagari 75 and\nOnOffSigBengali 75 confirm the models generalization capability.\n", "date": "2024-11-26", "categories": []}
{"id": "2411.18165", "title": "KAN See Your Face", "abstract": "  With the advancement of face reconstruction (FR) systems, privacy-preserving\nface recognition (PPFR) has gained popularity for its secure face recognition,\nenhanced facial privacy protection, and robustness to various attacks. Besides,\nspecific models and algorithms are proposed for face embedding protection by\nmapping embeddings to a secure space. However, there is a lack of studies on\ninvestigating and evaluating the possibility of extracting face images from\nembeddings of those systems, especially for PPFR. In this work, we introduce\nthe first approach to exploit Kolmogorov-Arnold Network (KAN) for conducting\nembedding-to-face attacks against state-of-the-art (SOTA) FR and PPFR systems.\nFace embedding mapping (FEM) models are proposed to learn the distribution\nmapping relation between the embeddings from the initial domain and target\ndomain. In comparison with Multi-Layer Perceptrons (MLP), we provide two\nvariants, FEM-KAN and FEM-MLP, for efficient non-linear embedding-to-embedding\nmapping in order to reconstruct realistic face images from the corresponding\nface embedding. To verify our methods, we conduct extensive experiments with\nvarious PPFR and FR models. We also measure reconstructed face images with\ndifferent metrics to evaluate the image quality. Through comprehensive\nexperiments, we demonstrate the effectiveness of FEMs in accurate embedding\nmapping and face reconstruction.\n", "date": "2024-11-27", "categories": []}
{"id": "2302.06529", "title": "ECG-Based Patient Identification: A Comprehensive Evaluation Across\n  Health and Activity Conditions", "abstract": "  Over the course of the past two decades, a substantial body of research has\nsubstantiated the viability of utilising cardiac signals as a biometric\nmodality. This paper presents a novel approach for patient identification in\nhealthcare systems using electrocardiogram signals. A convolutional neural\nnetwork (CNN) is employed to classify users based on electrocardiomatrices, a\nspecific type of image derived from ECG signals. The proposed identification\nsystem is evaluated in multiple databases, achieving up to 99.84\\% accuracy on\nhealthy subjects, 97.09\\% on patients with cardiovascular diseases, and 97.89%\non mixed populations including both healthy and arrhythmic patients. The system\nalso performs robustly under varying activity conditions, achieving 91.32%\naccuracy in scenarios involving different physical activities. These consistent\nand reliable results, with low error rates such as a FAR of 0.01% and FRR of\n0.157% in the best cases, demonstrate the method's significant advancement in\nsubject identification within healthcare systems. By considering patients'\ncardiovascular conditions and activity levels, the proposed approach addresses\ngaps in the existing literature, positioning it as a strong candidate for\npractical applications in real-world healthcare settings.\n", "date": "2023-02-13", "categories": []}
{"id": "2411.07205", "title": "DLCR: A Generative Data Expansion Framework via Diffusion for\n  Clothes-Changing Person Re-ID", "abstract": "  With the recent exhibited strength of generative diffusion models, an open\nresearch question is if images generated by these models can be used to learn\nbetter visual representations. While this generative data expansion may suffice\nfor easier visual tasks, we explore its efficacy on a more difficult\ndiscriminative task: clothes-changing person re-identification (CC-ReID).\nCC-ReID aims to match people appearing in non-overlapping cameras, even when\nthey change their clothes across cameras. Not only are current CC-ReID models\nconstrained by the limited diversity of clothing in current CC-ReID datasets,\nbut generating additional data that retains important personal features for\naccurate identification is a current challenge. To address this issue we\npropose DLCR, a novel data expansion framework that leverages pre-trained\ndiffusion and large language models (LLMs) to accurately generate diverse\nimages of individuals in varied attire. We generate additional data for five\nbenchmark CC-ReID datasets (PRCC, CCVID, LaST, VC-Clothes, and LTCC) and\nincrease their clothing diversity by 10X, totaling over 2.1M images generated.\nDLCR employs diffusion-based text-guided inpainting, conditioned on clothing\nprompts constructed using LLMs, to generate synthetic data that only modifies a\nsubject's clothes while preserving their personally identifiable features. With\nthis massive increase in data, we introduce two novel strategies - progressive\nlearning and test-time prediction refinement - that respectively reduce\ntraining time and further boosts CC-ReID performance. On the PRCC dataset, we\nobtain a large top-1 accuracy improvement of 11.3% by training CAL, a previous\nstate of the art (SOTA) method, with DLCR-generated data. We publicly release\nour code and generated data for each dataset here:\nhttps://github.com/CroitoruAlin/dlcr.\n", "date": "2024-11-11", "categories": []}
{"id": "2411.17305", "title": "in-Car Biometrics (iCarB) Datasets for Driver Recognition: Face,\n  Fingerprint, and Voice", "abstract": "  We present three biometric datasets (iCarB-Face, iCarB-Fingerprint,\niCarB-Voice) containing face videos, fingerprint images, and voice samples,\ncollected inside a car from 200 consenting volunteers. The data was acquired\nusing a near-infrared camera, two fingerprint scanners, and two microphones,\nwhile the volunteers were seated in the driver's seat of the car. The data\ncollection took place while the car was parked both indoors and outdoors, and\ndifferent \"noises\" were added to simulate non-ideal biometric data capture that\nmay be encountered in real-life driver recognition. Although the datasets are\nspecifically tailored to in-vehicle biometric recognition, their utility is not\nlimited to the automotive environment. The iCarB datasets, which are available\nto the research community, can be used to: (i) evaluate and benchmark face,\nfingerprint, and voice recognition systems (we provide several evaluation\nprotocols); (ii) create multimodal pseudo-identities, to train/test multimodal\nfusion algorithms; (iii) create Presentation Attacks from the biometric data,\nto evaluate Presentation Attack Detection algorithms; (iv) investigate\ndemographic and environmental biases in biometric systems, using the provided\nmetadata. To the best of our knowledge, ours are the largest and most diverse\npublicly available in-vehicle biometric datasets. Most other datasets contain\nonly one biometric modality (usually face), while our datasets consist of three\nmodalities, all acquired in the same automotive environment. Moreover,\niCarB-Fingerprint seems to be the first publicly available in-vehicle\nfingerprint dataset. Finally, the iCarB datasets boast a rare level of\ndemographic diversity among the 200 data subjects, including a 50/50 gender\nsplit, skin colours across the whole Fitzpatrick-scale spectrum, and a wide age\nrange (18-60+). So, these datasets will be valuable for advancing biometrics\nresearch.\n", "date": "2024-11-26", "categories": []}
{"id": "2411.13886", "title": "CLFace: A Scalable and Resource-Efficient Continual Learning Framework\n  for Lifelong Face Recognition", "abstract": "  An important aspect of deploying face recognition (FR) algorithms in\nreal-world applications is their ability to learn new face identities from a\ncontinuous data stream. However, the online training of existing deep neural\nnetwork-based FR algorithms, which are pre-trained offline on large-scale\nstationary datasets, encounter two major challenges: (I) catastrophic\nforgetting of previously learned identities, and (II) the need to store past\ndata for complete retraining from scratch, leading to significant storage\nconstraints and privacy concerns. In this paper, we introduce CLFace, a\ncontinual learning framework designed to preserve and incrementally extend the\nlearned knowledge. CLFace eliminates the classification layer, resulting in a\nresource-efficient FR model that remains fixed throughout lifelong learning and\nprovides label-free supervision to a student model, making it suitable for\nopen-set face recognition during incremental steps. We introduce an objective\nfunction that employs feature-level distillation to reduce drift between\nfeature maps of the student and teacher models across multiple stages.\nAdditionally, it incorporates a geometry-preserving distillation scheme to\nmaintain the orientation of the teacher model's feature embedding. Furthermore,\na contrastive knowledge distillation is incorporated to continually enhance the\ndiscriminative power of the feature representation by matching similarities\nbetween new identities. Experiments on several benchmark FR datasets\ndemonstrate that CLFace outperforms baseline approaches and state-of-the-art\nmethods on unseen identities using both in-domain and out-of-domain datasets.\n", "date": "2024-11-21", "categories": []}
{"id": "2411.12319", "title": "CLIP Unreasonable Potential in Single-Shot Face Recognition", "abstract": "  Face recognition is a core task in computer vision designed to identify and\nauthenticate individuals by analyzing facial patterns and features. This field\nintersects with artificial intelligence image processing and machine learning\nwith applications in security authentication and personalization. Traditional\napproaches in facial recognition focus on capturing facial features like the\neyes, nose and mouth and matching these against a database to verify\nidentities. However challenges such as high false positive rates have persisted\noften due to the similarity among individuals facial features. Recently\nContrastive Language Image Pretraining (CLIP) a model developed by OpenAI has\nshown promising advancements by linking natural language processing with vision\ntasks allowing it to generalize across modalities. Using CLIP's vision language\ncorrespondence and single-shot finetuning the model can achieve lower false\npositive rates upon deployment without the need of mass facial features\nextraction. This integration demonstrating CLIP's potential to address\npersistent issues in face recognition model performance without complicating\nour training paradigm.\n", "date": "2024-11-19", "categories": []}
{"id": "2411.10489", "title": "Biometrics in Extended Reality: A Review", "abstract": "  In the domain of Extended Reality (XR), particularly Virtual Reality (VR),\nextensive research has been devoted to harnessing this transformative\ntechnology in various real-world applications. However, a critical challenge\nthat must be addressed before unleashing the full potential of XR in practical\nscenarios is to ensure robust security and safeguard user privacy. This paper\npresents a systematic survey of the utility of biometric characteristics\napplied in the XR environment. To this end, we present a comprehensive overview\nof the different types of biometric modalities used for authentication and\nrepresentation of users in a virtual environment. We discuss different\nbiometric vulnerability gateways in general XR systems for the first time in\nthe literature along with taxonomy. A comprehensive discussion on generating\nand authenticating biometric-based photorealistic avatars in XR environments is\npresented with a stringent taxonomy. We also discuss the availability of\ndifferent datasets that are widely employed in evaluating biometric\nauthentication in XR environments together with performance evaluation metrics.\nFinally, we discuss the open challenges and potential future work that need to\nbe addressed in the field of biometrics in XR.\n", "date": "2024-11-14", "categories": []}
{"id": "2411.11471", "title": "Generalizable Person Re-identification via Balancing Alignment and\n  Uniformity", "abstract": "  Domain generalizable person re-identification (DG re-ID) aims to learn\ndiscriminative representations that are robust to distributional shifts. While\ndata augmentation is a straightforward solution to improve generalization,\ncertain augmentations exhibit a polarized effect in this task, enhancing\nin-distribution performance while deteriorating out-of-distribution\nperformance. In this paper, we investigate this phenomenon and reveal that it\nleads to sparse representation spaces with reduced uniformity. To address this\nissue, we propose a novel framework, Balancing Alignment and Uniformity (BAU),\nwhich effectively mitigates this effect by maintaining a balance between\nalignment and uniformity. Specifically, BAU incorporates alignment and\nuniformity losses applied to both original and augmented images and integrates\na weighting strategy to assess the reliability of augmented samples, further\nimproving the alignment loss. Additionally, we introduce a domain-specific\nuniformity loss that promotes uniformity within each source domain, thereby\nenhancing the learning of domain-invariant features. Extensive experimental\nresults demonstrate that BAU effectively exploits the advantages of data\naugmentation, which previous studies could not fully utilize, and achieves\nstate-of-the-art performance without requiring complex training procedures. The\ncode is available at \\url{https://github.com/yoonkicho/BAU}.\n", "date": "2024-11-18", "categories": []}
{"id": "2408.01233", "title": "CLIP4Sketch: Enhancing Sketch to Mugshot Matching through Dataset\n  Augmentation using Diffusion Models", "abstract": "  Forensic sketch-to-mugshot matching is a challenging task in face\nrecognition, primarily hindered by the scarcity of annotated forensic sketches\nand the modality gap between sketches and photographs. To address this, we\npropose CLIP4Sketch, a novel approach that leverages diffusion models to\ngenerate a large and diverse set of sketch images, which helps in enhancing the\nperformance of face recognition systems in sketch-to-mugshot matching. Our\nmethod utilizes Denoising Diffusion Probabilistic Models (DDPMs) to generate\nsketches with explicit control over identity and style. We combine CLIP and\nAdaface embeddings of a reference mugshot, along with textual descriptions of\nstyle, as the conditions to the diffusion model. We demonstrate the efficacy of\nour approach by generating a comprehensive dataset of sketches corresponding to\nmugshots and training a face recognition model on our synthetic data. Our\nresults show significant improvements in sketch-to-mugshot matching accuracy\nover training on an existing, limited amount of real face sketch data,\nvalidating the potential of diffusion models in enhancing the performance of\nface recognition systems across modalities. We also compare our dataset with\ndatasets generated using GAN-based methods to show its superiority.\n", "date": "2024-08-02", "categories": []}
{"id": "2411.09538", "title": "Marker-free Human Gait Analysis using a Smart Edge Sensor System", "abstract": "  The human gait is a complex interplay between the neuronal and the muscular\nsystems, reflecting an individual's neurological and physiological condition.\nThis makes gait analysis a valuable tool for biomechanics and medical experts.\nTraditional observational gait analysis is cost-effective but lacks reliability\nand accuracy, while instrumented gait analysis, particularly using marker-based\noptical systems, provides accurate data but is expensive and time-consuming. In\nthis paper, we introduce a novel markerless approach for gait analysis using a\nmulti-camera setup with smart edge sensors to estimate 3D body poses without\nfiducial markers. We propose a Siamese embedding network with triplet loss\ncalculation to identify individuals by their gait pattern. This network\neffectively maps gait sequences to an embedding space that enables clustering\nsequences from the same individual or activity closely together while\nseparating those of different ones. Our results demonstrate the potential of\nthe proposed system for efficient automated gait analysis in diverse real-world\nenvironments, facilitating a wide range of applications.\n", "date": "2024-11-14", "categories": []}
{"id": "2411.08195", "title": "An Explainable Machine Learning Approach for Age and Gender Estimation\n  in Living Individuals Using Dental Biometrics", "abstract": "  Objectives: Age and gender estimation is crucial for various applications,\nincluding forensic investigations and anthropological studies. This research\naims to develop a predictive system for age and gender estimation in living\nindividuals, leveraging dental measurements such as Coronal Height (CH),\nCoronal Pulp Cavity Height (CPCH), and Tooth Coronal Index (TCI). Methods:\nMachine learning models were employed in our study, including Cat Boost\nClassifier (Catboost), Gradient Boosting Machine (GBM), Ada Boost Classifier\n(AdaBoost), Random Forest (RF), eXtreme Gradient Boosting (XGB), Light Gradient\nBoosting Machine (LGB), and Extra Trees Classifier (ETC), to analyze dental\ndata from 862 living individuals (459 males and 403 females). Specifically,\nperiapical radiographs from six teeth per individual were utilized, including\npremolars and molars from both maxillary and mandibular. A novel ensemble\nlearning technique was developed, which uses multiple models each tailored to\ndistinct dental metrics, to estimate age and gender accurately. Furthermore, an\nexplainable AI model has been created utilizing SHAP, enabling dental experts\nto make judicious decisions based on comprehensible insight. Results: The RF\nand XGB models were particularly effective, yielding the highest F1 score for\nage and gender estimation. Notably, the XGB model showed a slightly better\nperformance in age estimation, achieving an F1 score of 73.26%. A similar trend\nfor the RF model was also observed in gender estimation, achieving a F1 score\nof 77.53%. Conclusions: This study marks a significant advancement in dental\nforensic methods, showcasing the potential of machine learning to automate age\nand gender estimation processes with improved accuracy.\n", "date": "2024-11-12", "categories": []}
{"id": "2411.08490", "title": "Impact of Iris Pigmentation on Performance Bias in Visible Iris\n  Verification Systems: A Comparative Study", "abstract": "  Iris recognition technology plays a critical role in biometric identification\nsystems, but their performance can be affected by variations in iris\npigmentation. In this work, we investigate the impact of iris pigmentation on\nthe efficacy of biometric recognition systems, focusing on a comparative\nanalysis of blue and dark irises. Data sets were collected using multiple\ndevices, including P1, P2, and P3 smartphones [4], to assess the robustness of\nthe systems in different capture environments [19]. Both traditional machine\nlearning techniques and deep learning models were used, namely Open-Iris,\nViT-b, and ResNet50, to evaluate performance metrics such as Equal Error Rate\n(EER) and True Match Rate (TMR). Our results indicate that iris recognition\nsystems generally exhibit higher accuracy for blue irises compared to dark\nirises. Furthermore, we examined the generalization capabilities of these\nsystems across different iris colors and devices, finding that while training\non diverse datasets enhances recognition performance, the degree of improvement\nis contingent on the specific model and device used. Our analysis also\nidentifies inherent biases in recognition performance related to iris color and\ncross-device variability. These findings underscore the need for more inclusive\ndataset collection and model refinement to reduce bias and promote equitable\nbiometric recognition across varying iris pigmentation and device\nconfigurations.\n", "date": "2024-11-13", "categories": []}
{"id": "2404.06025", "title": "Greedy-DiM: Greedy Algorithms for Unreasonably Effective Face Morphs", "abstract": "  Morphing attacks are an emerging threat to state-of-the-art Face Recognition\n(FR) systems, which aim to create a single image that contains the biometric\ninformation of multiple identities. Diffusion Morphs (DiM) are a recently\nproposed morphing attack that has achieved state-of-the-art performance for\nrepresentation-based morphing attacks. However, none of the existing research\non DiMs have leveraged the iterative nature of DiMs and left the DiM model as a\nblack box, treating it no differently than one would a Generative Adversarial\nNetwork (GAN) or Varational AutoEncoder (VAE). We propose a greedy strategy on\nthe iterative sampling process of DiM models which searches for an optimal step\nguided by an identity-based heuristic function. We compare our proposed\nalgorithm against ten other state-of-the-art morphing algorithms using the\nopen-source SYN-MAD 2022 competition dataset. We find that our proposed\nalgorithm is unreasonably effective, fooling all of the tested FR systems with\nan MMPMR of 100%, outperforming all other morphing algorithms compared.\n", "date": "2024-04-09", "categories": []}
{"id": "2404.06559", "title": "The Impact of Print-Scanning in Heterogeneous Morph Evaluation Scenarios", "abstract": "  Face morphing attacks pose an increasing threat to face recognition (FR)\nsystems. A morphed photo contains biometric information from two different\nsubjects to take advantage of vulnerabilities in FRs. These systems are\nparticularly susceptible to attacks when the morphs are subjected to\nprint-scanning to mask the artifacts generated during the morphing process. We\ninvestigate the impact of print-scanning on morphing attack detection through a\nseries of evaluations on heterogeneous morphing attack scenarios. Our\nexperiments show that we can increase the Mated Morph Presentation Match Rate\n(MMPMR) by up to 8.48%. Furthermore, when a Single-image Morphing Attack\nDetection (S-MAD) algorithm is not trained to detect print-scanned morphs the\nMorphing Attack Classification Error Rate (MACER) can increase by up to 96.12%,\nindicating significant vulnerability.\n", "date": "2024-04-09", "categories": []}
{"id": "2411.06023", "title": "Dynamic Textual Prompt For Rehearsal-free Lifelong Person\n  Re-identification", "abstract": "  Lifelong person re-identification attempts to recognize people across cameras\nand integrate new knowledge from continuous data streams. Key challenges\ninvolve addressing catastrophic forgetting caused by parameter updating and\ndomain shift, and maintaining performance in seen and unseen domains. Many\nprevious works rely on data memories to retain prior samples. However, the\namount of retained data increases linearly with the number of training domains,\nleading to continually increasing memory consumption. Additionally, these\nmethods may suffer significant performance degradation when data preservation\nis prohibited due to privacy concerns. To address these limitations, we propose\nusing textual descriptions as guidance to encourage the ReID model to learn\ncross-domain invariant features without retaining samples. The key insight is\nthat natural language can describe pedestrian instances with an invariant\nstyle, suggesting a shared textual space for any pedestrian images. By\nleveraging this shared textual space as an anchor, we can prompt the ReID model\nto embed images from various domains into a unified semantic space, thereby\nalleviating catastrophic forgetting caused by domain shifts. To achieve this,\nwe introduce a task-driven dynamic textual prompt framework in this paper. This\nmodel features a dynamic prompt fusion module, which adaptively constructs and\nfuses two different textual prompts as anchors. This effectively guides the\nReID model to embed images into a unified semantic space. Additionally, we\ndesign a text-visual feature alignment module to learn a more precise mapping\nbetween fine-grained visual and textual features. We also developed a learnable\nknowledge distillation module that allows our model to dynamically balance\nretaining existing knowledge with acquiring new knowledge. Extensive\nexperiments demonstrate that our method outperforms SOTAs under various\nsettings.\n", "date": "2024-11-09", "categories": []}
{"id": "2411.06510", "title": "Offline Handwritten Signature Verification Using a Stream-Based Approach", "abstract": "  Handwritten Signature Verification (HSV) systems distinguish between genuine\nand forged signatures. Traditional HSV development involves a static batch\nconfiguration, constraining the system's ability to model signatures to the\nlimited data available. Signatures exhibit high intra-class variability and are\nsensitive to various factors, including time and external influences, imparting\nthem a dynamic nature. This paper investigates the signature learning process\nwithin a data stream context. We propose a novel HSV approach with an adaptive\nsystem that receives an infinite sequence of signatures and is updated over\ntime. Experiments were carried out on GPDS Synthetic, CEDAR, and MCYT datasets.\nResults demonstrate the superior performance of the proposed method compared to\nstandard approaches that use a Support Vector Machine as a classifier.\nImplementation of the method is available at\nhttps://github.com/kdMoura/stream_hsv.\n", "date": "2024-11-10", "categories": []}
{"id": "2411.07224", "title": "TempCharBERT: Keystroke Dynamics for Continuous Access Control Based on\n  Pre-trained Language Models", "abstract": "  With the widespread of digital environments, reliable authentication and\ncontinuous access control has become crucial. It can minimize cyber attacks and\nprevent frauds, specially those associated with identity theft. A particular\ninterest lies on keystroke dynamics (KD), which refers to the task of\nrecognizing individuals' identity based on their unique typing style. In this\nwork, we propose the use of pre-trained language models (PLMs) to recognize\nsuch patterns. Although PLMs have shown high performance on multiple NLP\nbenchmarks, the use of these models on specific tasks requires customization.\nBERT and RoBERTa, for instance, rely on subword tokenization, and they cannot\nbe directly applied to KD, which requires temporal-character information to\nrecognize users. Recent character-aware PLMs are able to process both subwords\nand character-level information and can be an alternative solution.\nNotwithstanding, they are still not suitable to be directly fine-tuned for KD\nas they are not optimized to account for user's temporal typing information\n(e.g., hold time and flight time). To overcome this limitation, we propose\nTempCharBERT, an architecture that incorporates temporal-character information\nin the embedding layer of CharBERT. This allows modeling keystroke dynamics for\nthe purpose of user identification and authentication. Our results show a\nsignificant improvement with this customization. We also showed the feasibility\nof training TempCharBERT on a federated learning settings in order to foster\ndata privacy.\n", "date": "2024-11-11", "categories": []}
{"id": "2411.05395", "title": "AuthFormer: Adaptive Multimodal biometric authentication transformer for\n  middle-aged and elderly people", "abstract": "  Multimodal biometric authentication methods address the limitations of\nunimodal biometric technologies in security, robustness, and user adaptability.\nHowever, most existing methods depend on fixed combinations and numbers of\nbiometric modalities, which restricts flexibility and adaptability in\nreal-world applications. To overcome these challenges, we propose an adaptive\nmultimodal biometric authentication model, AuthFormer, tailored for elderly\nusers. AuthFormer is trained on the LUTBIO multimodal biometric database,\ncontaining biometric data from elderly individuals. By incorporating a\ncross-attention mechanism and a Gated Residual Network (GRN), the model\nimproves adaptability to physiological variations in elderly users. Experiments\nshow that AuthFormer achieves an accuracy of 99.73%. Additionally, its encoder\nrequires only two layers to perform optimally, reducing complexity compared to\ntraditional Transformer-based models.\n", "date": "2024-11-08", "categories": []}
{"id": "2411.04128", "title": "On the analysis of saturated pressure to detect fatigue", "abstract": "  This paper examines the saturation of pressure signals during various\nhandwriting tasks, including drawings, cursive text, capital words text, and\nsignature, under different levels of fatigue. Experimental results demonstrate\na significant rise in the proportion of saturated samples following strenuous\nexercise in tasks performed without resting wrist. The analysis of saturation\nhighlights significant differences when comparing the results to the baseline\nsituation and strenuous fatigue.\n", "date": "2024-10-22", "categories": []}
{"id": "2411.03831", "title": "An Enhancement of Haar Cascade Algorithm Applied to Face Recognition for\n  Gate Pass Security", "abstract": "  This study is focused on enhancing the Haar Cascade Algorithm to decrease the\nfalse positive and false negative rate in face matching and face detection to\nincrease the accuracy rate even under challenging conditions. The face\nrecognition library was implemented with Haar Cascade Algorithm in which the\n128-dimensional vectors representing the unique features of a face are encoded.\nA subprocess was applied where the grayscale image from Haar Cascade was\nconverted to RGB to improve the face encoding. Logical process and face\nfiltering are also used to decrease non-face detection. The Enhanced Haar\nCascade Algorithm produced a 98.39% accuracy rate (21.39% increase), 63.59%\nprecision rate, 98.30% recall rate, and 72.23% in F1 Score. In comparison, the\nHaar Cascade Algorithm achieved a 46.70% to 77.00% accuracy rate, 44.15%\nprecision rate, 98.61% recall rate, and 47.01% in F1 Score. Both algorithms\nused the Confusion Matrix Test with 301,950 comparisons using the same dataset\nof 550 images. The 98.39% accuracy rate shows a significant decrease in false\npositive and false negative rates in facial recognition. Face matching and face\ndetection are more accurate in images with complex backgrounds, lighting\nvariations, and occlusions, or even those with similar attributes.\n", "date": "2024-11-06", "categories": []}
{"id": "2411.03960", "title": "Face Reconstruction from Face Embeddings using Adapter to a Face\n  Foundation Model", "abstract": "  Face recognition systems extract embedding vectors from face images and use\nthese embeddings to verify or identify individuals. Face reconstruction attack\n(also known as template inversion) refers to reconstructing face images from\nface embeddings and using the reconstructed face image to enter a face\nrecognition system. In this paper, we propose to use a face foundation model to\nreconstruct face images from the embeddings of a blackbox face recognition\nmodel. The foundation model is trained with 42M images to generate face images\nfrom the facial embeddings of a fixed face recognition model. We propose to use\nan adapter to translate target embeddings into the embedding space of the\nfoundation model. The generated images are evaluated on different face\nrecognition models and different datasets, demonstrating the effectiveness of\nour method to translate embeddings of different face recognition models. We\nalso evaluate the transferability of reconstructed face images when attacking\ndifferent face recognition models. Our experimental results show that our\nreconstructed face images outperform previous reconstruction attacks against\nface recognition models.\n", "date": "2024-11-06", "categories": []}
{"id": "2305.12673", "title": "Efficient Bilateral Cross-Modality Cluster Matching for Unsupervised\n  Visible-Infrared Person ReID", "abstract": "  Unsupervised visible-infrared person re-identification (USL-VI-ReID) aims to\nmatch pedestrian images of the same identity from different modalities without\nannotations. Existing works mainly focus on alleviating the modality gap by\naligning instance-level features of the unlabeled samples. However, the\nrelationships between cross-modality clusters are not well explored. To this\nend, we propose a novel bilateral cluster matching-based learning framework to\nreduce the modality gap by matching cross-modality clusters. Specifically, we\ndesign a Many-to-many Bilateral Cross-Modality Cluster Matching (MBCCM)\nalgorithm through optimizing the maximum matching problem in a bipartite graph.\nThen, the matched pairwise clusters utilize shared visible and infrared\npseudo-labels during the model training. Under such a supervisory signal, a\nModality-Specific and Modality-Agnostic (MSMA) contrastive learning framework\nis proposed to align features jointly at a cluster-level. Meanwhile, the\ncross-modality Consistency Constraint (CC) is proposed to explicitly reduce the\nlarge modality discrepancy. Extensive experiments on the public SYSU-MM01 and\nRegDB datasets demonstrate the effectiveness of the proposed method, surpassing\nstate-of-the-art approaches by a large margin of 8.76% mAP on average.\n", "date": "2023-05-22", "categories": []}
{"id": "2305.12711", "title": "Unsupervised Visible-Infrared Person ReID by Collaborative Learning with\n  Neighbor-Guided Label Refinement", "abstract": "  Unsupervised learning visible-infrared person re-identification (USL-VI-ReID)\naims at learning modality-invariant features from unlabeled cross-modality\ndataset, which is crucial for practical applications in video surveillance\nsystems. The key to essentially address the USL-VI-ReID task is to solve the\ncross-modality data association problem for further heterogeneous joint\nlearning. To address this issue, we propose a Dual Optimal Transport Label\nAssignment (DOTLA) framework to simultaneously assign the generated labels from\none modality to its counterpart modality. The proposed DOTLA mechanism\nformulates a mutual reinforcement and efficient solution to cross-modality data\nassociation, which could effectively reduce the side-effects of some\ninsufficient and noisy label associations. Besides, we further propose a\ncross-modality neighbor consistency guided label refinement and regularization\nmodule, to eliminate the negative effects brought by the inaccurate supervised\nsignals, under the assumption that the prediction or label distribution of each\nexample should be similar to its nearest neighbors. Extensive experimental\nresults on the public SYSU-MM01 and RegDB datasets demonstrate the\neffectiveness of the proposed method, surpassing existing state-of-the-art\napproach by a large margin of 7.76% mAP on average, which even surpasses some\nsupervised VI-ReID methods.\n", "date": "2023-05-22", "categories": []}
{"id": "2409.00562", "title": "Comparative Analysis of Modality Fusion Approaches for Audio-Visual\n  Person Identification and Verification", "abstract": "  Multimodal learning involves integrating information from various modalities\nto enhance learning and comprehension. We compare three modality fusion\nstrategies in person identification and verification by processing two\nmodalities: voice and face. In this paper, a one-dimensional convolutional\nneural network is employed for x-vector extraction from voice, while the\npre-trained VGGFace2 network and transfer learning are utilized for face\nmodality. In addition, gammatonegram is used as speech representation in\nengagement with the Darknet19 pre-trained network. The proposed systems are\nevaluated using the K-fold cross-validation technique on the 118 speakers of\nthe test set of the VoxCeleb2 dataset. The comparative evaluations are done for\nsingle-modality and three proposed multimodal strategies in equal situations.\nResults demonstrate that the feature fusion strategy of gammatonegram and\nfacial features achieves the highest performance, with an accuracy of 98.37% in\nthe person identification task. However, concatenating facial features with the\nx-vector reaches 0.62% for EER in verification tasks.\n", "date": "2024-08-31", "categories": []}
{"id": "2411.01225", "title": "RLE: A Unified Perspective of Data Augmentation for Cross-Spectral\n  Re-identification", "abstract": "  This paper makes a step towards modeling the modality discrepancy in the\ncross-spectral re-identification task. Based on the Lambertain model, we\nobserve that the non-linear modality discrepancy mainly comes from diverse\nlinear transformations acting on the surface of different materials. From this\nview, we unify all data augmentation strategies for cross-spectral\nre-identification by mimicking such local linear transformations and\ncategorizing them into moderate transformation and radical transformation. By\nextending the observation, we propose a Random Linear Enhancement (RLE)\nstrategy which includes Moderate Random Linear Enhancement (MRLE) and Radical\nRandom Linear Enhancement (RRLE) to push the boundaries of both types of\ntransformation. Moderate Random Linear Enhancement is designed to provide\ndiverse image transformations that satisfy the original linear correlations\nunder constrained conditions, whereas Radical Random Linear Enhancement seeks\nto generate local linear transformations directly without relying on external\ninformation. The experimental results not only demonstrate the superiority and\neffectiveness of RLE but also confirm its great potential as a general-purpose\ndata augmentation for cross-spectral re-identification. The code is available\nat \\textcolor{magenta}{\\url{https://github.com/stone96123/RLE}}.\n", "date": "2024-11-02", "categories": []}
{"id": "2411.02112", "title": "Multi-modal biometric authentication: Leveraging shared layer\n  architectures for enhanced security", "abstract": "  In this study, we introduce a novel multi-modal biometric authentication\nsystem that integrates facial, vocal, and signature data to enhance security\nmeasures. Utilizing a combination of Convolutional Neural Networks (CNNs) and\nRecurrent Neural Networks (RNNs), our model architecture uniquely incorporates\ndual shared layers alongside modality-specific enhancements for comprehensive\nfeature extraction. The system undergoes rigorous training with a joint loss\nfunction, optimizing for accuracy across diverse biometric inputs.\nFeature-level fusion via Principal Component Analysis (PCA) and classification\nthrough Gradient Boosting Machines (GBM) further refine the authentication\nprocess. Our approach demonstrates significant improvements in authentication\naccuracy and robustness, paving the way for advanced secure identity\nverification solutions.\n", "date": "2024-11-04", "categories": []}
{"id": "2404.17159", "title": "Phase-aggregated Dual-branch Network for Efficient Fingerprint Dense\n  Registration", "abstract": "  Fingerprint dense registration aims to finely align fingerprint pairs at the\npixel level, thereby reducing intra-class differences caused by distortion.\nUnfortunately, traditional methods exhibited subpar performance when dealing\nwith low-quality fingerprints while suffering from slow inference speed.\nAlthough deep learning based approaches shows significant improvement in these\naspects, their registration accuracy is still unsatisfactory. In this paper, we\npropose a Phase-aggregated Dual-branch Registration Network (PDRNet) to\naggregate the advantages of both types of methods. A dual-branch structure with\nmulti-stage interactions is introduced between correlation information at high\nresolution and texture feature at low resolution, to perceive local fine\ndifferences while ensuring global stability. Extensive experiments are\nconducted on more comprehensive databases compared to previous works.\nExperimental results demonstrate that our method reaches the state-of-the-art\nregistration performance in terms of accuracy and robustness, while maintaining\nconsiderable competitiveness in efficiency.\n", "date": "2024-04-26", "categories": []}
{"id": "2409.18876", "title": "CemiFace: Center-based Semi-hard Synthetic Face Generation for Face\n  Recognition", "abstract": "  Privacy issue is a main concern in developing face recognition techniques.\nAlthough synthetic face images can partially mitigate potential legal risks\nwhile maintaining effective face recognition (FR) performance, FR models\ntrained by face images synthesized by existing generative approaches frequently\nsuffer from performance degradation problems due to the insufficient\ndiscriminative quality of these synthesized samples. In this paper, we\nsystematically investigate what contributes to solid face recognition model\ntraining, and reveal that face images with certain degree of similarities to\ntheir identity centers show great effectiveness in the performance of trained\nFR models. Inspired by this, we propose a novel diffusion-based approach\n(namely Center-based Semi-hard Synthetic Face Generation (CemiFace)) which\nproduces facial samples with various levels of similarity to the subject\ncenter, thus allowing to generate face datasets containing effective\ndiscriminative samples for training face recognition. Experimental results show\nthat with a modest degree of similarity, training on the generated dataset can\nproduce competitive performance compared to previous generation methods.\n", "date": "2024-09-27", "categories": []}
{"id": "2410.21663", "title": "Discriminative Pedestrian Features and Gated Channel Attention for\n  Clothes-Changing Person Re-Identification", "abstract": "  In public safety and social life, the task of Clothes-Changing Person\nRe-Identification (CC-ReID) has become increasingly significant. However, this\ntask faces considerable challenges due to appearance changes caused by clothing\nalterations. Addressing this issue, this paper proposes an innovative method\nfor disentangled feature extraction, effectively extracting discriminative\nfeatures from pedestrian images that are invariant to clothing. This method\nleverages pedestrian parsing techniques to identify and retain features closely\nassociated with individual identity while disregarding the variable nature of\nclothing attributes. Furthermore, this study introduces a gated channel\nattention mechanism, which, by adjusting the network's focus, aids the model in\nmore effectively learning and emphasizing features critical for pedestrian\nidentity recognition. Extensive experiments conducted on two standard CC-ReID\ndatasets validate the effectiveness of the proposed approach, with performance\nsurpassing current leading solutions. The Top-1 accuracy under clothing change\nscenarios on the PRCC and VC-Clothes datasets reached 64.8% and 83.7%,\nrespectively.\n", "date": "2024-10-29", "categories": []}
{"id": "2410.21938", "title": "ReMix: Training Generalized Person Re-identification on a Mixture of\n  Data", "abstract": "  Modern person re-identification (Re-ID) methods have a weak generalization\nability and experience a major accuracy drop when capturing environments\nchange. This is because existing multi-camera Re-ID datasets are limited in\nsize and diversity, since such data is difficult to obtain. At the same time,\nenormous volumes of unlabeled single-camera records are available. Such data\ncan be easily collected, and therefore, it is more diverse. Currently,\nsingle-camera data is used only for self-supervised pre-training of Re-ID\nmethods. However, the diversity of single-camera data is suppressed by\nfine-tuning on limited multi-camera data after pre-training. In this paper, we\npropose ReMix, a generalized Re-ID method jointly trained on a mixture of\nlimited labeled multi-camera and large unlabeled single-camera data. Effective\ntraining of our method is achieved through a novel data sampling strategy and\nnew loss functions that are adapted for joint use with both types of data.\nExperiments show that ReMix has a high generalization ability and outperforms\nstate-of-the-art methods in generalizable person Re-ID. To the best of our\nknowledge, this is the first work that explores joint training on a mixture of\nmulti-camera and single-camera data in person Re-ID.\n", "date": "2024-10-29", "categories": []}
{"id": "2410.21993", "title": "A Machine Learning-Based Secure Face Verification Scheme and Its\n  Applications to Digital Surveillance", "abstract": "  Face verification is a well-known image analysis application and is widely\nused to recognize individuals in contemporary society. However, most real-world\nrecognition systems ignore the importance of protecting the identity-sensitive\nfacial images that are used for verification. To address this problem, we\ninvestigate how to implement a secure face verification system that protects\nthe facial images from being imitated. In our work, we use the DeepID2\nconvolutional neural network to extract the features of a facial image and an\nEM algorithm to solve the facial verification problem. To maintain the privacy\nof facial images, we apply homomorphic encryption schemes to encrypt the facial\ndata and compute the EM algorithm in the ciphertext domain. We develop three\nface verification systems for surveillance (or entrance) control of a local\ncommunity based on three levels of privacy concerns. The associated timing\nperformances are presented to demonstrate their feasibility for practical\nimplementation.\n", "date": "2024-10-29", "categories": []}
{"id": "2207.04434", "title": "Hiding Your Signals: A Security Analysis of PPG-based Biometric\n  Authentication", "abstract": "  Recently, physiological signal-based biometric systems have received wide\nattention. Unlike traditional biometric features, physiological signals can not\nbe easily compromised (usually unobservable to human eyes).\nPhotoplethysmography (PPG) signal is easy to measure, making it more attractive\nthan many other physiological signals for biometric authentication. However,\nwith the advent of remote PPG (rPPG), unobservability has been challenged when\nthe attacker can remotely steal the rPPG signals by monitoring the victim's\nface, subsequently posing a threat to PPG-based biometrics. In PPG-based\nbiometric authentication, current attack approaches mandate the victim's PPG\nsignal, making rPPG-based attacks neglected. In this paper, we firstly analyze\nthe security of PPG-based biometrics, including user authentication and\ncommunication protocols. We evaluate the signal waveforms, heart rate and\ninter-pulse-interval information extracted by five rPPG methods, including four\ntraditional optical computing methods (CHROM, POS, LGI, PCA) and one deep\nlearning method (CL_rPPG). We conducted experiments on five datasets (PURE,\nUBFC_rPPG, UBFC_Phys, LGI_PPGI, and COHFACE) to collect a comprehensive set of\nresults. Our empirical studies show that rPPG poses a serious threat to the\nauthentication system. The success rate of the rPPG signal spoofing attack in\nthe user authentication system reached 0.35. The bit hit rate is 0.6 in\ninter-pulse-interval-based security protocols. Further, we propose an active\ndefence strategy to hide the physiological signals of the face to resist the\nattack. It reduces the success rate of rPPG spoofing attacks in user\nauthentication to 0.05. The bit hit rate was reduced to 0.5, which is at the\nlevel of a random guess. Our strategy effectively prevents the exposure of PPG\nsignals to protect users' sensitive physiological data.\n", "date": "2022-07-10", "categories": []}
{"id": "2302.07517", "title": "Versatile User Identification in Extended Reality using Pretrained\n  Similarity-Learning", "abstract": "  Various machine learning approaches have proven to be useful for user\nverification and identification based on motion data in eXtended Reality (XR).\nHowever, their real-world application still faces significant challenges\nconcerning versatility, i.e., in terms of extensibility and generalization\ncapability. This article presents a solution that is both extensible to new\nusers without expensive retraining, and that generalizes well across different\nsessions, devices, and user tasks. To this end, we developed a\nsimilarity-learning model and pretrained it on the \"Who Is Alyx?\" dataset. This\ndataset features a wide array of tasks and hence motions from users playing the\nVR game \"Half-Life: Alyx\". In contrast to previous works, we used a dedicated\nset of users for model validation and final evaluation. Furthermore, we\nextended this evaluation using an independent dataset that features completely\ndifferent users, tasks, and three different XR devices. In comparison with a\ntraditional classification-learning baseline, our model shows superior\nperformance, especially in scenarios with limited enrollment data. The\npretraining process allows immediate deployment in a diverse range of XR\napplications while maintaining high versatility. Looking ahead, our approach\npaves the way for easy integration of pretrained motion-based identification\nmodels in production XR systems.\n", "date": "2023-02-15", "categories": []}
{"id": "2311.17074", "title": "VILLS -- Video-Image Learning to Learn Semantics for Person\n  Re-Identification", "abstract": "  Person Re-identification is a research area with significant real world\napplications. Despite recent progress, existing methods face challenges in\nrobust re-identification in the wild, e.g., by focusing only on a particular\nmodality and on unreliable patterns such as clothing. A generalized method is\nhighly desired, but remains elusive to achieve due to issues such as the\ntrade-off between spatial and temporal resolution and imperfect feature\nextraction. We propose VILLS (Video-Image Learning to Learn Semantics), a\nself-supervised method that jointly learns spatial and temporal features from\nimages and videos. VILLS first designs a local semantic extraction module that\nadaptively extracts semantically consistent and robust spatial features. Then,\nVILLS designs a unified feature learning and adaptation module to represent\nimage and video modalities in a consistent feature space. By Leveraging\nself-supervised, large-scale pre-training, VILLS establishes a new\nState-of-The-Art that significantly outperforms existing image and video-based\nmethods.\n", "date": "2023-11-27", "categories": []}
{"id": "2303.11332", "title": "Deep Learning for Video-based Person Re-Identification: A Survey", "abstract": "  Video-based person re-identification (video re-ID) has lately fascinated\ngrowing attention due to its broad practical applications in various areas,\nsuch as surveillance, smart city, and public safety. Nevertheless, video re-ID\nis quite difficult and is an ongoing stage due to numerous uncertain challenges\nsuch as viewpoint, occlusion, pose variation, and uncertain video sequence,\netc. In the last couple of years, deep learning on video re-ID has continuously\nachieved surprising results on public datasets, with various approaches being\ndeveloped to handle diverse problems in video re-ID. Compared to image-based\nre-ID, video re-ID is much more challenging and complex. To encourage future\nresearch and challenges, this first comprehensive paper introduces a review of\nup-to-date advancements in deep learning approaches for video re-ID. It broadly\ncovers three important aspects, including brief video re-ID methods with their\nlimitations, major milestones with technical challenges, and architectural\ndesign. It offers comparative performance analysis on various available\ndatasets, guidance to improve video re-ID with valuable thoughts, and exciting\nresearch directions.\n", "date": "2023-03-21", "categories": []}
{"id": "2402.19026", "title": "Learning Commonality, Divergence and Variety for Unsupervised\n  Visible-Infrared Person Re-identification", "abstract": "  Unsupervised visible-infrared person re-identification (USVI-ReID) aims to\nmatch specified people in infrared images to visible images without\nannotations, and vice versa. USVI-ReID is a challenging yet under-explored\ntask. Most existing methods address the USVI-ReID using cluster-based\ncontrastive learning, which simply employs the cluster center as a\nrepresentation of a person. However, the cluster center primarily focuses on\ncommonality, overlooking divergence and variety. To address the problem, we\npropose a Progressive Contrastive Learning with Hard and Dynamic Prototypes\nmethod for USVI-ReID. In brief, we generate the hard prototype by selecting the\nsample with the maximum distance from the cluster center. We theoretically show\nthat the hard prototype is used in the contrastive loss to emphasize\ndivergence. Additionally, instead of rigidly aligning query images to a\nspecific prototype, we generate the dynamic prototype by randomly picking\nsamples within a cluster. The dynamic prototype is used to encourage the\nvariety. Finally, we introduce a progressive learning strategy to gradually\nshift the model's attention towards divergence and variety, avoiding cluster\ndeterioration. Extensive experiments conducted on the publicly available\nSYSU-MM01 and RegDB datasets validate the effectiveness of the proposed method.\n", "date": "2024-02-29", "categories": []}
{"id": "2406.16592", "title": "Toward Fairer Face Recognition Datasets", "abstract": "  Face recognition and verification are two computer vision tasks whose\nperformance has progressed with the introduction of deep representations.\nHowever, ethical, legal, and technical challenges due to the sensitive\ncharacter of face data and biases in real training datasets hinder their\ndevelopment. Generative AI addresses privacy by creating fictitious identities,\nbut fairness problems persist. We promote fairness by introducing a demographic\nattributes balancing mechanism in generated training datasets. We experiment\nwith an existing real dataset, three generated training datasets, and the\nbalanced versions of a diffusion-based dataset. We propose a comprehensive\nevaluation that considers accuracy and fairness equally and includes a rigorous\nregression-based statistical analysis of attributes. The analysis shows that\nbalancing reduces demographic unfairness. Also, a performance gap persists\ndespite generation becoming more accurate with time. The proposed balancing\nmethod and comprehensive verification evaluation promote fairer and transparent\nface recognition and verification.\n", "date": "2024-06-24", "categories": []}
{"id": "2410.15613", "title": "Exploring Stronger Transformer Representation Learning for Occluded\n  Person Re-Identification", "abstract": "  Due to some complex factors (e.g., occlusion, pose variation and diverse\ncamera perspectives), extracting stronger feature representation in person\nre-identification remains a challenging task. In this paper, we proposed a\nnovel self-supervision and supervision combining transformer-based person\nre-identification framework, namely SSSC-TransReID. Different from the general\ntransformer-based person re-identification models, we designed a\nself-supervised contrastive learning branch, which can enhance the feature\nrepresentation for person re-identification without negative samples or\nadditional pre-training. In order to train the contrastive learning branch, we\nalso proposed a novel random rectangle mask strategy to simulate the occlusion\nin real scenes, so as to enhance the feature representation for occlusion.\nFinally, we utilized the joint-training loss function to integrate the\nadvantages of supervised learning with ID tags and self-supervised contrastive\nlearning without negative samples, which can reinforce the ability of our model\nto excavate stronger discriminative features, especially for occlusion.\nExtensive experimental results on several benchmark datasets show our proposed\nmodel obtains superior Re-ID performance consistently and outperforms the\nstate-of-the-art ReID methods by large margins on the mean average accuracy\n(mAP) and Rank-1 accuracy.\n", "date": "2024-10-21", "categories": []}
{"id": "2406.06932", "title": "Synthetic Face Ageing: Evaluation, Analysis and Facilitation of\n  Age-Robust Facial Recognition Algorithms", "abstract": "  The ability to accurately recognize an individual's face with respect to\nhuman aging factor holds significant importance for various private as well as\ngovernment sectors such as customs and public security bureaus, passport\noffice, and national database systems. Therefore, developing a robust\nage-invariant face recognition system is of crucial importance to address the\nchallenges posed by ageing and maintain the reliability and accuracy of facial\nrecognition technology. In this research work, the focus is to explore the\nfeasibility of utilizing synthetic ageing data to improve the robustness of\nface recognition models that can eventually help in recognizing people at\nbroader age intervals. To achieve this, we first design set of experiments to\nevaluate state-of-the-art synthetic ageing methods. In the next stage we\nexplore the effect of age intervals on a current deep learning-based face\nrecognition algorithm by using synthetic ageing data as well as real ageing\ndata to perform rigorous training and validation. Moreover, these synthetic age\ndata have been used in facilitating face recognition algorithms. Experimental\nresults show that the recognition rate of the model trained on synthetic ageing\nimages is 3.33% higher than the results of the baseline model when tested on\nimages with an age gap of 40 years, which prove the potential of synthetic age\ndata which has been quantified to enhance the performance of age-invariant face\nrecognition systems.\n", "date": "2024-06-10", "categories": []}
{"id": "2305.18216", "title": "Towards minimizing efforts for Morphing Attacks -- Deep embeddings for\n  morphing pair selection and improved Morphing Attack Detection", "abstract": "  Face Morphing Attacks pose a threat to the security of identity documents,\nespecially with respect to a subsequent access control process, because it\nenables both individuals involved to exploit the same document. In this study,\nface embeddings serve two purposes: pre-selecting images for large-scale\nMorphing Attack generation and detecting potential Morphing Attacks. We build\nupon previous embedding studies in both use cases using the MagFace model. For\nthe first objective, we employ an pre-selection algorithm that pairs\nindividuals based on face embedding similarity. We quantify the attack\npotential of differently morphed face images to compare the usability of\npre-selection in automatically generating numerous successful Morphing Attacks.\nRegarding the second objective, we compare embeddings from two state-of-the-art\nface recognition systems in terms of their ability to detect Morphing Attacks.\nOur findings demonstrate that ArcFace and MagFace provide valuable face\nembeddings for image pre-selection. Both open-source and COTS face recognition\nsystems are susceptible to generated attacks, particularly when pre-selection\nis based on embeddings rather than random pairing which was only constrained by\nsoft biometrics. More accurate face recognition systems exhibit greater\nvulnerability to attacks, with COTS systems being the most susceptible.\nAdditionally, MagFace embeddings serve as a robust alternative for detecting\nmorphed face images compared to the previously used ArcFace embeddings. The\nresults endorse the advantages of face embeddings in more effective image\npre-selection for face morphing and accurate detection of morphed face images.\nThis is supported by extensive analysis of various designed attacks. The\nMagFace model proves to be a powerful alternative to the commonly used ArcFace\nmodel for both objectives, pre-selection and attack detection.\n", "date": "2023-05-29", "categories": []}
{"id": "2410.11176", "title": "Improving Bias in Facial Attribute Classification: A Combined Impact of\n  KL Divergence induced Loss Function and Dual Attention", "abstract": "  Ensuring that AI-based facial recognition systems produce fair predictions\nand work equally well across all demographic groups is crucial. Earlier systems\noften exhibited demographic bias, particularly in gender and racial\nclassification, with lower accuracy for women and individuals with darker skin\ntones. To tackle this issue and promote fairness in facial recognition,\nresearchers have introduced several bias-mitigation techniques for gender\nclassification and related algorithms. However, many challenges remain, such as\ndata diversity, balancing fairness with accuracy, disparity, and bias\nmeasurement. This paper presents a method using a dual attention mechanism with\na pre-trained Inception-ResNet V1 model, enhanced by KL-divergence\nregularization and a cross-entropy loss function. This approach reduces bias\nwhile improving accuracy and computational efficiency through transfer\nlearning. The experimental results show significant improvements in both\nfairness and classification accuracy, providing promising advances in\naddressing bias and enhancing the reliability of facial recognition systems.\n", "date": "2024-10-15", "categories": []}
{"id": "2410.11255", "title": "CLIP-DFGS: A Hard Sample Mining Method for CLIP in Generalizable Person\n  Re-Identification", "abstract": "  Recent advancements in pre-trained vision-language models like CLIP have\nshown promise in person re-identification (ReID) applications. However, their\nperformance in generalizable person re-identification tasks remains suboptimal.\nThe large-scale and diverse image-text pairs used in CLIP's pre-training may\nlead to a lack or insufficiency of certain fine-grained features. In light of\nthese challenges, we propose a hard sample mining method called DFGS\n(Depth-First Graph Sampler), based on depth-first search, designed to offer\nsufficiently challenging samples to enhance CLIP's ability to extract\nfine-grained features. DFGS can be applied to both the image encoder and the\ntext encoder in CLIP. By leveraging the powerful cross-modal learning\ncapabilities of CLIP, we aim to apply our DFGS method to extract challenging\nsamples and form mini-batches with high discriminative difficulty, providing\nthe image model with more efficient and challenging samples that are difficult\nto distinguish, thereby enhancing the model's ability to differentiate between\nindividuals. Our results demonstrate significant improvements over other\nmethods, confirming the effectiveness of DFGS in providing challenging samples\nthat enhance CLIP's performance in generalizable person re-identification.\n", "date": "2024-10-15", "categories": []}
{"id": "2405.12721", "title": "StarLKNet: Star Mixup with Large Kernel Networks for Palm Vein\n  Identification", "abstract": "  As a representative of a new generation of biometrics, vein identification\ntechnology offers a high level of security and convenience.Convolutional neural\nnetworks (CNNs), a prominent class of deep learning architectures, have been\nextensively utilized for vein identification. Since their performance and\nrobustness are limited by small \\emph{Effective Receptive Fields} (\\emph{e.g.},\n3$\\times$3 kernels) and insufficient training samples, however, they are unable\nto extract global feature representations from vein images effectively. To\naddress these issues, we propose \\textbf{StarLKNet}, a large kernel\nconvolution-based palm-vein identification network, with the Mixup approach.Our\nStarMix learns effectively the distribution of vein features to expand samples.\nTo enable CNNs to capture comprehensive feature representations from palm-vein\nimages, we explored the effect of convolutional kernel size on the performance\nof palm-vein identification networks and designed LaKNet, a network leveraging\nlarge kernel convolution and gating mechanism. In light of the current state of\nknowledge, this represents an inaugural instance of the deployment of a CNN\nwith large kernels in the domain of vein identification. Extensive experiments\nwere conducted to validate the performance of StarLKNet on two public palm-vein\ndatasets. The results demonstrated that \\textbf{StarMix} provided superior\naugmentation, and \\textbf{LakNet} exhibited more stable performance gains\ncompared to mainstream approaches, resulting in the highest identification\naccuracy and lowest identification error.\n", "date": "2024-05-21", "categories": []}
{"id": "2408.06167", "title": "Blind-Match: Efficient Homomorphic Encryption-Based 1:N Matching for\n  Privacy-Preserving Biometric Identification", "abstract": "  We present Blind-Match, a novel biometric identification system that\nleverages homomorphic encryption (HE) for efficient and privacy-preserving 1:N\nmatching. Blind-Match introduces a HE-optimized cosine similarity computation\nmethod, where the key idea is to divide the feature vector into smaller parts\nfor processing rather than computing the entire vector at once. By optimizing\nthe number of these parts, Blind-Match minimizes execution time while ensuring\ndata privacy through HE. Blind-Match achieves superior performance compared to\nstate-of-the-art methods across various biometric datasets. On the LFW face\ndataset, Blind-Match attains a 99.63% Rank-1 accuracy with a 128-dimensional\nfeature vector, demonstrating its robustness in face recognition tasks. For\nfingerprint identification, Blind-Match achieves a remarkable 99.55% Rank-1\naccuracy on the PolyU dataset, even with a compact 16-dimensional feature\nvector, significantly outperforming the state-of-the-art method, Blind-Touch,\nwhich achieves only 59.17%. Furthermore, Blind-Match showcases practical\nefficiency in large-scale biometric identification scenarios, such as Naver\nCloud's FaceSign, by processing 6,144 biometric samples in 0.74 seconds using a\n128-dimensional feature vector.\n", "date": "2024-08-12", "categories": []}
{"id": "2410.03977", "title": "Learning to Balance: Diverse Normalization for Cloth-Changing Person\n  Re-Identification", "abstract": "  Cloth-Changing Person Re-Identification (CC-ReID) involves recognizing\nindividuals in images regardless of clothing status. In this paper, we\nempirically and experimentally demonstrate that completely eliminating or fully\nretaining clothing features is detrimental to the task. Existing work, either\nrelying on clothing labels, silhouettes, or other auxiliary data, fundamentally\naim to balance the learning of clothing and identity features. However, we\npractically find that achieving this balance is challenging and nuanced. In\nthis study, we introduce a novel module called Diverse Norm, which expands\npersonal features into orthogonal spaces and employs channel attention to\nseparate clothing and identity features. A sample re-weighting optimization\nstrategy is also introduced to guarantee the opposite optimization direction.\nDiverse Norm presents a simple yet effective approach that does not require\nadditional data. Furthermore, Diverse Norm can be seamlessly integrated\nResNet50 and significantly outperforms the state-of-the-art methods.\n", "date": "2024-10-04", "categories": []}
{"id": "2410.09382", "title": "CLIP-SCGI: Synthesized Caption-Guided Inversion for Person\n  Re-Identification", "abstract": "  Person re-identification (ReID) has recently benefited from large pretrained\nvision-language models such as Contrastive Language-Image Pre-Training (CLIP).\nHowever, the absence of concrete descriptions necessitates the use of implicit\ntext embeddings, which demand complicated and inefficient training strategies.\nTo address this issue, we first propose one straightforward solution by\nleveraging existing image captioning models to generate pseudo captions for\nperson images, and thereby boost person re-identification with large vision\nlanguage models. Using models like the Large Language and Vision Assistant\n(LLAVA), we generate high-quality captions based on fixed templates that\ncapture key semantic attributes such as gender, clothing, and age. By\naugmenting ReID training sets from uni-modality (image) to bi-modality (image\nand text), we introduce CLIP-SCGI, a simple yet effective framework that\nleverages synthesized captions to guide the learning of discriminative and\nrobust representations. Built on CLIP, CLIP-SCGI fuses image and text\nembeddings through two modules to enhance the training process. To address\nquality issues in generated captions, we introduce a caption-guided inversion\nmodule that captures semantic attributes from images by converting relevant\nvisual information into pseudo-word tokens based on the descriptions. This\napproach helps the model better capture key information and focus on relevant\nregions. The extracted features are then utilized in a cross-modal fusion\nmodule, guiding the model to focus on regions semantically consistent with the\ncaption, thereby facilitating the optimization of the visual encoder to extract\ndiscriminative and robust representations. Extensive experiments on four\npopular ReID benchmarks demonstrate that CLIP-SCGI outperforms the\nstate-of-the-art by a significant margin.\n", "date": "2024-10-12", "categories": []}
{"id": "2410.09582", "title": "Improving 3D Finger Traits Recognition via Generalizable Neural\n  Rendering", "abstract": "  3D biometric techniques on finger traits have become a new trend and have\ndemonstrated a powerful ability for recognition and anti-counterfeiting.\nExisting methods follow an explicit 3D pipeline that reconstructs the models\nfirst and then extracts features from 3D models. However, these explicit 3D\nmethods suffer from the following problems: 1) Inevitable information dropping\nduring 3D reconstruction; 2) Tight coupling between specific hardware and\nalgorithm for 3D reconstruction. It leads us to a question: Is it indispensable\nto reconstruct 3D information explicitly in recognition tasks? Hence, we\nconsider this problem in an implicit manner, leaving the nerve-wracking 3D\nreconstruction problem for learnable neural networks with the help of neural\nradiance fields (NeRFs). We propose FingerNeRF, a novel generalizable NeRF for\n3D finger biometrics. To handle the shape-radiance ambiguity problem that may\nresult in incorrect 3D geometry, we aim to involve extra geometric priors based\non the correspondence of binary finger traits like fingerprints or finger\nveins. First, we propose a novel Trait Guided Transformer (TGT) module to\nenhance the feature correspondence with the guidance of finger traits. Second,\nwe involve extra geometric constraints on the volume rendering loss with the\nproposed Depth Distillation Loss and Trait Guided Rendering Loss. To evaluate\nthe performance of the proposed method on different modalities, we collect two\nnew datasets: SCUT-Finger-3D with finger images and SCUT-FingerVein-3D with\nfinger vein images. Moreover, we also utilize the UNSW-3D dataset with\nfingerprint images for evaluation. In experiments, our FingerNeRF can achieve\n4.37% EER on SCUT-Finger-3D dataset, 8.12% EER on SCUT-FingerVein-3D dataset,\nand 2.90% EER on UNSW-3D dataset, showing the superiority of the proposed\nimplicit method in 3D finger biometrics.\n", "date": "2024-10-12", "categories": []}
{"id": "2410.09842", "title": "Fusion Based Hand Geometry Recognition Using Dempster-Shafer Theory", "abstract": "  This paper presents a new technique for person recognition based on the\nfusion of hand geometric features of both the hands without any pose\nrestrictions. All the features are extracted from normalized left and right\nhand images. Fusion is applied at feature level and also at decision level. Two\nprobability based algorithms are proposed for classification. The first\nalgorithm computes the maximum probability for nearest three neighbors. The\nsecond algorithm determines the maximum probability of the number of matched\nfeatures with respect to a thresholding on distances. Based on these two\nhighest probabilities initial decisions are made. The final decision is\nconsidered according to the highest probability as calculated by the\nDempster-Shafer theory of evidence. Depending on the various combinations of\nthe initial decisions, three schemes are experimented with 201 subjects for\nidentification and verification. The correct identification rate found to be\n99.5%, and the False Acceptance Rate (FAR) of 0.625% has been found during\nverification.\n", "date": "2024-10-13", "categories": []}
{"id": "2410.09856", "title": "Human Identification using Selected Features from Finger Geometric\n  Profiles", "abstract": "  A finger biometric system at an unconstrained environment is presented in\nthis paper. A technique for hand image normalization is implemented at the\npreprocessing stage that decomposes the main hand contour into finger-level\nshape representation. This normalization technique follows subtraction of\ntransformed binary image from binary hand contour image to generate the left\nside of finger profiles (LSFP). Then, XOR is applied to LSFP image and hand\ncontour image to produce the right side of finger profiles (RSFP). During\nfeature extraction, initially, thirty geometric features are computed from\nevery normalized finger. The rank-based forward-backward greedy algorithm is\nfollowed to select relevant features and to enhance classification accuracy.\nTwo different subsets of features containing nine and twelve discriminative\nfeatures per finger are selected for two separate experimentations those use\nthe kNN and the Random Forest (RF) for classification on the Bosphorus hand\ndatabase. The experiments with the selected features of four fingers except the\nthumb have obtained improved performances compared to features extracted from\nfive fingers and also other existing methods evaluated on the Bosphorus\ndatabase. The best identification accuracies of 96.56% and 95.92% using the RF\nclassifier have been achieved for the right- and left-hand images of 638\nsub-jects, respectively. An equal error rate of 0.078 is obtained for both\ntypes of the hand images.\n", "date": "2024-10-13", "categories": []}
{"id": "2410.09866", "title": "Two-Stage Human Verification using HandCAPTCHA and Anti-Spoofed Finger\n  Biometrics with Feature Selection", "abstract": "  This paper presents a human verification scheme in two independent stages to\novercome the vulnerabilities of attacks and to enhance security. At the first\nstage, a hand image-based CAPTCHA (HandCAPTCHA) is tested to avert automated\nbot-attacks on the subsequent biometric stage. In the next stage, finger\nbiometric verification of a legitimate user is performed with presentation\nattack detection (PAD) using the real hand images of the person who has passed\na random HandCAPTCHA challenge. The electronic screen-based PAD is tested using\nimage quality metrics. After this spoofing detection, geometric features are\nextracted from the four fingers (excluding the thumb) of real users. A modified\nforward-backward (M-FoBa) algorithm is devised to select relevant features for\nbiometric authentication. The experiments are performed on the Bogazici\nUniversity (BU) and the IIT-Delhi (IITD) hand databases using the k-nearest\nneighbor and random forest classifiers. The average accuracy of the correct\nHandCAPTCHA solution is 98.5%, and the false accept rate of a bot is 1.23%. The\nPAD is tested on 255 subjects of BU, and the best average error is 0%. The\nfinger biometric identification accuracy of 98% and an equal error rate (EER)\nof 6.5% have been achieved for 500 subjects of the BU. For 200 subjects of the\nIITD, 99.5% identification accuracy, and 5.18% EER are obtained.\n", "date": "2024-10-13", "categories": []}
{"id": "2410.09875", "title": "ViFi-ReID: A Two-Stream Vision-WiFi Multimodal Approach for Person\n  Re-identification", "abstract": "  Person re-identification(ReID), as a crucial technology in the field of\nsecurity, plays a vital role in safety inspections, personnel counting, and\nmore. Most current ReID approaches primarily extract features from images,\nwhich are easily affected by objective conditions such as clothing changes and\nocclusions. In addition to cameras, we leverage widely available routers as\nsensing devices by capturing gait information from pedestrians through the\nChannel State Information (CSI) in WiFi signals and contribute a multimodal\ndataset. We employ a two-stream network to separately process video\nunderstanding and signal analysis tasks, and conduct multi-modal fusion and\ncontrastive learning on pedestrian video and WiFi data. Extensive experiments\nin real-world scenarios demonstrate that our method effectively uncovers the\ncorrelations between heterogeneous data, bridges the gap between visual and\nsignal modalities, significantly expands the sensing range, and improves ReID\naccuracy across multiple sensors.\n", "date": "2024-10-13", "categories": []}
{"id": "2410.06041", "title": "Block Induced Signature Generative Adversarial Network (BISGAN):\n  Signature Spoofing Using GANs and Their Evaluation", "abstract": "  Deep learning is actively being used in biometrics to develop efficient\nidentification and verification systems. Handwritten signatures are a common\nsubset of biometric data for authentication purposes. Generative adversarial\nnetworks (GANs) learn from original and forged signatures to generate forged\nsignatures. While most GAN techniques create a strong signature verifier, which\nis the discriminator, there is a need to focus more on the quality of forgeries\ngenerated by the generator model. This work focuses on creating a generator\nthat produces forged samples that achieve a benchmark in spoofing signature\nverification systems. We use CycleGANs infused with Inception model-like blocks\nwith attention heads as the generator and a variation of the SigCNN model as\nthe base Discriminator. We train our model with a new technique that results in\n80% to 100% success in signature spoofing. Additionally, we create a custom\nevaluation technique to act as a goodness measure of the generated forgeries.\nOur work advocates generator-focused GAN architectures for spoofing data\nquality that aid in a better understanding of biometric data generation and\nevaluation.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.08456", "title": "A Unified Deep Semantic Expansion Framework for Domain-Generalized\n  Person Re-identification", "abstract": "  Supervised Person Re-identification (Person ReID) methods have achieved\nexcellent performance when training and testing within one camera network.\nHowever, they usually suffer from considerable performance degradation when\napplied to different camera systems. In recent years, many Domain Adaptation\nPerson ReID methods have been proposed, achieving impressive performance\nwithout requiring labeled data from the target domain. However, these\napproaches still need the unlabeled data of the target domain during the\ntraining process, making them impractical in many real-world scenarios. Our\nwork focuses on the more practical Domain Generalized Person Re-identification\n(DG-ReID) problem. Given one or more source domains, it aims to learn a\ngeneralized model that can be applied to unseen target domains. One promising\nresearch direction in DG-ReID is the use of implicit deep semantic feature\nexpansion, and our previous method, Domain Embedding Expansion (DEX), is one\nsuch example that achieves powerful results in DG-ReID. However, in this work\nwe show that DEX and other similar implicit deep semantic feature expansion\nmethods, due to limitations in their proposed loss function, fail to reach\ntheir full potential on large evaluation benchmarks as they have a tendency to\nsaturate too early. Leveraging on this analysis, we propose Unified Deep\nSemantic Expansion, our novel framework that unifies implicit and explicit\nsemantic feature expansion techniques in a single framework to mitigate this\nearly over-fitting and achieve a new state-of-the-art (SOTA) in all DG-ReID\nbenchmarks. Further, we apply our method on more general image retrieval tasks,\nalso surpassing the current SOTA in all of these benchmarks by wide margins.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.08460", "title": "Diverse Deep Feature Ensemble Learning for Omni-Domain Generalized\n  Person Re-identification", "abstract": "  Person Re-identification (Person ReID) has progressed to a level where\nsingle-domain supervised Person ReID performance has saturated. However, such\nmethods experience a significant drop in performance when trained and tested\nacross different datasets, motivating the development of domain generalization\ntechniques. However, our research reveals that domain generalization methods\nsignificantly underperform single-domain supervised methods on single dataset\nbenchmarks. An ideal Person ReID method should be effective regardless of the\nnumber of domains involved, and when test domain data is available for training\nit should perform as well as state-of-the-art (SOTA) fully supervised methods.\nThis is a paradigm that we call Omni-Domain Generalization Person ReID\n(ODG-ReID). We propose a way to achieve ODG-ReID by creating deep feature\ndiversity with self-ensembles. Our method, Diverse Deep Feature Ensemble\nLearning (D2FEL), deploys unique instance normalization patterns that generate\nmultiple diverse views and recombines these views into a compact encoding. To\nthe best of our knowledge, our work is one of few to consider omni-domain\ngeneralization in Person ReID, and we advance the study of applying feature\nensembles in Person ReID. D2FEL significantly improves and matches the SOTA\nperformance for major domain generalization and single-domain supervised\nbenchmarks.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.08466", "title": "Aligned Divergent Pathways for Omni-Domain Generalized Person\n  Re-Identification", "abstract": "  Person Re-identification (Person ReID) has advanced significantly in fully\nsupervised and domain generalized Person R e ID. However, methods developed for\none task domain transfer poorly to the other. An ideal Person ReID method\nshould be effective regardless of the number of domains involved in training or\ntesting. Furthermore, given training data from the target domain, it should\nperform at least as well as state-of-the-art (SOTA) fully supervised Person\nReID methods. We call this paradigm Omni-Domain Generalization Person ReID,\nreferred to as ODG-ReID, and propose a way to achieve this by expanding\ncompatible backbone architectures into multiple diverse pathways. Our method,\nAligned Divergent Pathways (ADP), first converts a base architecture into a\nmulti-branch structure by copying the tail of the original backbone. We design\nour module Dynamic Max-Deviance Adaptive Instance Normalization (DyMAIN) that\nencourages learning of generalized features that are robust to omni-domain\ndirections and apply DyMAIN to the branches of ADP. Our proposed Phased\nMixture-of-Cosines (PMoC) coordinates a mix of stable and turbulent learning\nrate schedules among branches for further diversified learning. Finally, we\nrealign the feature space between branches with our proposed Dimensional\nConsistency Metric Loss (DCML). ADP outperforms the state-of-the-art (SOTA)\nresults for multi-source domain generalization and supervised ReID within the\nsame domain. Furthermore, our method demonstrates improvement on a wide range\nof single-source domain generalization benchmarks, achieving Omni-Domain\nGeneralization over Person ReID tasks.\n", "date": "2024-10-11", "categories": []}
{"id": "2410.05808", "title": "Vision Transformer based Random Walk for Group Re-Identification", "abstract": "  Group re-identification (re-ID) aims to match groups with the same people\nunder different cameras, mainly involves the challenges of group members and\nlayout changes well. Most existing methods usually use the k-nearest neighbor\nalgorithm to update node features to consider changes in group membership, but\nthese methods cannot solve the problem of group layout changes. To this end, we\npropose a novel vision transformer based random walk framework for group re-ID.\nSpecifically, we design a vision transformer based on a monocular depth\nestimation algorithm to construct a graph through the average depth value of\npedestrian features to fully consider the impact of camera distance on group\nmembers relationships. In addition, we propose a random walk module to\nreconstruct the graph by calculating affinity scores between target and gallery\nimages to remove pedestrians who do not belong to the current group.\nExperimental results show that our framework is superior to most methods.\n", "date": "2024-10-08", "categories": []}
{"id": "2410.06600", "title": "DDRN:a Data Distribution Reconstruction Network for Occluded Person\n  Re-Identification", "abstract": "  In occluded person re-identification(ReID), severe occlusions lead to a\nsignificant amount of irrelevant information that hinders the accurate\nidentification of individuals. These irrelevant cues primarily stem from\nbackground interference and occluding interference, adversely affecting the\nfinal retrieval results. Traditional discriminative models, which rely on the\nspecific content and positions of the images, often misclassify in cases of\nocclusion. To address these limitations, we propose the Data Distribution\nReconstruction Network (DDRN), a generative model that leverages data\ndistribution to filter out irrelevant details, enhancing overall feature\nperception ability and reducing irrelevant feature interference. Additionally,\nsevere occlusions lead to the complexity of the feature space. To effectively\nhandle this, we design a multi-center approach through the proposed\nHierarchical SubcenterArcface (HS-Arcface) loss function, which can better\napproximate complex feature spaces. On the Occluded-Duke dataset, we achieved a\nmAP of 62.4\\% (+1.1\\%) and a rank-1 accuracy of 71.3\\% (+0.6\\%), surpassing the\nlatest state-of-the-art methods(FRT) significantly.\n", "date": "2024-10-09", "categories": []}
{"id": "2407.19472", "title": "Combined CNN and ViT features off-the-shelf: Another astounding baseline\n  for recognition", "abstract": "  We apply pre-trained architectures, originally developed for the ImageNet\nLarge Scale Visual Recognition Challenge, for periocular recognition. These\narchitectures have demonstrated significant success in various computer vision\ntasks beyond the ones for which they were designed. This work builds on our\nprevious study using off-the-shelf Convolutional Neural Network (CNN) and\nextends it to include the more recently proposed Vision Transformers (ViT).\nDespite being trained for generic object classification, middle-layer features\nfrom CNNs and ViTs are a suitable way to recognize individuals based on\nperiocular images. We also demonstrate that CNNs and ViTs are highly\ncomplementary since their combination results in boosted accuracy. In addition,\nwe show that a small portion of these pre-trained models can achieve good\naccuracy, resulting in thinner models with fewer parameters, suitable for\nresource-limited environments such as mobiles. This efficiency improves if\ntraditional handcrafted features are added as well.\n", "date": "2024-07-28", "categories": []}
{"id": "2211.08007", "title": "Evidence-based Match-status-Aware Gait Recognition for Out-of-Gallery\n  Gait Identification", "abstract": "  Existing gait recognition methods typically identify individuals based on the\nsimilarity between probe and gallery samples. However, these methods often\nneglect the fact that the gallery may not contain identities corresponding to\nthe probes, leading to incorrect recognition.To identify Out-of-Gallery (OOG)\ngait queries, we propose an Evidence-based Match-status-Aware Gait Recognition\n(EMA-GR) framework. Inspired by Evidential Deep Learning (EDL), EMA-GR is\ndesigned to quantify the uncertainty associated with the match status of\nrecognition. Thus, EMA-GR identifies whether the probe has a counterpart in the\ngallery. Specifically, we adopt an evidence collector to gather match status\nevidence from a recognition result pair and parameterize a Dirichlet\ndistribution over the gathered evidence, following the Dempster-Shafer Theory\nof Evidence (DST). We measure the uncertainty and predict the match status of\nthe recognition results, and thus determine whether the probe is an OOG\nquery.To the best of our knowledge, our method is the first attempt to tackle\nOOG queries in gait recognition. Moreover, EMA-GR is agnostic against gait\nrecognition methods and improves the robustness against OOG queries. Extensive\nexperiments demonstrate that our method achieves state-of-the-art performance\non datasets with OOG queries, and can also generalize well to other\nidentity-retrieval tasks. Importantly, our method surpasses existing\nstate-of-the-art methods by a substantial margin, achieving a 51.26%\nimprovement when the OOG query rate is around 50% on OUMVLP.\n", "date": "2022-11-15", "categories": []}
{"id": "2409.19992", "title": "A large-scale operational study of fingerprint quality and demographics", "abstract": "  Even though a few initial works have shown on small sets of data some level\nof bias in the performance of fingerprint recognition technology with respect\nto certain demographic groups, there is still not sufficient evidence to\nunderstand the impact that certain factors such as gender, age or finger-type\nmay have on fingerprint quality and, in turn, also on fingerprint matching\naccuracy. The present work addresses this still under researched topic, on a\nlarge-scale database of operational data containing 10-print impressions of\nalmost 16,000 subjects. The results reached provide further insight into the\ndependency of fingerprint quality and demographics, and show that there in fact\nexists a certain degree of performance variability in fingerprint-based\nrecognition systems for different segments of the population. Based on the\nexperimental evaluation, the work points out new observations based on\ndata-driven evidence, provides plausible hypotheses to explain such\nobservations, and concludes with potential follow-up actions that can help to\nreduce the observed fingerprint quality differences. This way, the current\npaper can be considered as a contribution to further increase the algorithmic\nfairness and equality of biometric technology.\n", "date": "2024-09-30", "categories": []}
{"id": "2410.01498", "title": "Quo Vadis RankList-based System in Face Recognition?", "abstract": "  Face recognition in the wild has gained a lot of focus in the last few years,\nand many face recognition models are designed to verify faces in medium-quality\nimages. Especially due to the availability of large training datasets with\nsimilar conditions, deep face recognition models perform exceptionally well in\nsuch tasks. However, in other tasks where substantially less training data is\navailable, such methods struggle, especially when required to compare\nhigh-quality enrollment images with low-quality probes. On the other hand,\ntraditional RankList-based methods have been developed that compare faces\nindirectly by comparing to cohort faces with similar conditions. In this paper,\nwe revisit these RankList methods and extend them to use the logits of the\nstate-of-the-art DaliFace network, instead of an external cohort. We show that\nthrough a reasonable Logit-Cohort Selection (LoCoS) the performance of\nRankList-based functions can be improved drastically. Experiments on two\nchallenging face recognition datasets not only demonstrate the enhanced\nperformance of our proposed method but also set the stage for future\nadvancements in handling diverse image qualities.\n", "date": "2024-10-02", "categories": []}
{"id": "2211.12371", "title": "Gait Recognition in Large-scale Free Environment via Single LiDAR", "abstract": "  Human gait recognition is crucial in multimedia, enabling identification\nthrough walking patterns without direct interaction, enhancing the integration\nacross various media forms in real-world applications like smart homes,\nhealthcare and non-intrusive security. LiDAR's ability to capture depth makes\nit pivotal for robotic perception and holds promise for real-world gait\nrecognition. In this paper, based on a single LiDAR, we present the\nHierarchical Multi-representation Feature Interaction Network (HMRNet) for\nrobust gait recognition. Prevailing LiDAR-based gait datasets primarily derive\nfrom controlled settings with predefined trajectory, remaining a gap with\nreal-world scenarios. To facilitate LiDAR-based gait recognition research, we\nintroduce FreeGait, a comprehensive gait dataset from large-scale,\nunconstrained settings, enriched with multi-modal and varied 2D/3D data.\nNotably, our approach achieves state-of-the-art performance on prior dataset\n(SUSTech1K) and on FreeGait.\n", "date": "2022-11-22", "categories": []}
{"id": "2409.19563", "title": "CLIP-based Camera-Agnostic Feature Learning for Intra-camera Person\n  Re-Identification", "abstract": "  Contrastive Language-Image Pre-Training (CLIP) model excels in traditional\nperson re-identification (ReID) tasks due to its inherent advantage in\ngenerating textual descriptions for pedestrian images. However, applying CLIP\ndirectly to intra-camera supervised person re-identification (ICS ReID)\npresents challenges. ICS ReID requires independent identity labeling within\neach camera, without associations across cameras. This limits the effectiveness\nof text-based enhancements. To address this, we propose a novel framework\ncalled CLIP-based Camera-Agnostic Feature Learning (CCAFL) for ICS ReID.\nAccordingly, two custom modules are designed to guide the model to actively\nlearn camera-agnostic pedestrian features: Intra-Camera Discriminative Learning\n(ICDL) and Inter-Camera Adversarial Learning (ICAL). Specifically, we first\nestablish learnable textual prompts for intra-camera pedestrian images to\nobtain crucial semantic supervision signals for subsequent intra- and\ninter-camera learning. Then, we design ICDL to increase inter-class variation\nby considering the hard positive and hard negative samples within each camera,\nthereby learning intra-camera finer-grained pedestrian features. Additionally,\nwe propose ICAL to reduce inter-camera pedestrian feature discrepancies by\npenalizing the model's ability to predict the camera from which a pedestrian\nimage originates, thus enhancing the model's capability to recognize\npedestrians from different viewpoints. Extensive experiments on popular ReID\ndatasets demonstrate the effectiveness of our approach. Especially, on the\nchallenging MSMT17 dataset, we arrive at 58.9\\% in terms of mAP accuracy,\nsurpassing state-of-the-art methods by 7.6\\%. Code will be available at:\nhttps://github.com/Trangle12/CCAFL.\n", "date": "2024-09-29", "categories": []}
{"id": "2409.20081", "title": "ProFD: Prompt-Guided Feature Disentangling for Occluded Person\n  Re-Identification", "abstract": "  To address the occlusion issues in person Re-Identification (ReID) tasks,\nmany methods have been proposed to extract part features by introducing\nexternal spatial information. However, due to missing part appearance\ninformation caused by occlusion and noisy spatial information from external\nmodel, these purely vision-based approaches fail to correctly learn the\nfeatures of human body parts from limited training data and struggle in\naccurately locating body parts, ultimately leading to misaligned part features.\nTo tackle these challenges, we propose a Prompt-guided Feature Disentangling\nmethod (ProFD), which leverages the rich pre-trained knowledge in the textual\nmodality facilitate model to generate well-aligned part features. ProFD first\ndesigns part-specific prompts and utilizes noisy segmentation mask to\npreliminarily align visual and textual embedding, enabling the textual prompts\nto have spatial awareness. Furthermore, to alleviate the noise from external\nmasks, ProFD adopts a hybrid-attention decoder, ensuring spatial and semantic\nconsistency during the decoding process to minimize noise impact. Additionally,\nto avoid catastrophic forgetting, we employ a self-distillation strategy,\nretaining pre-trained knowledge of CLIP to mitigate over-fitting. Evaluation\nresults on the Market1501, DukeMTMC-ReID, Occluded-Duke, Occluded-ReID, and\nP-DukeMTMC datasets demonstrate that ProFD achieves state-of-the-art results.\nOur project is available at: https://github.com/Cuixxx/ProFD.\n", "date": "2024-09-30", "categories": []}
{"id": "2409.18298", "title": "Causality-based Subject and Task Fingerprints using fMRI Time-series\n  Data", "abstract": "  Recently, there has been a revived interest in system neuroscience causation\nmodels due to their unique capability to unravel complex relationships in\nmulti-scale brain networks. In this paper, our goal is to verify the\nfeasibility and effectiveness of using a causality-based approach for fMRI\nfingerprinting. Specifically, we propose an innovative method that utilizes the\ncausal dynamics activities of the brain to identify the unique cognitive\npatterns of individuals (e.g., subject fingerprint) and fMRI tasks (e.g., task\nfingerprint). The key novelty of our approach stems from the development of a\ntwo-timescale linear state-space model to extract 'spatio-temporal' (aka\ncausal) signatures from an individual's fMRI time series data. To the best of\nour knowledge, we pioneer and subsequently quantify, in this paper, the concept\nof 'causal fingerprint.' Our method is well-separated from other fingerprint\nstudies as we quantify fingerprints from a cause-and-effect perspective, which\nare then incorporated with a modal decomposition and projection method to\nperform subject identification and a GNN-based (Graph Neural Network) model to\nperform task identification. Finally, we show that the experimental results and\ncomparisons with non-causality-based methods demonstrate the effectiveness of\nthe proposed methods. We visualize the obtained causal signatures and discuss\ntheir biological relevance in light of the existing understanding of brain\nfunctionalities. Collectively, our work paves the way for further studies on\ncausal fingerprints with potential applications in both healthy controls and\nneurodegenerative diseases.\n", "date": "2024-09-26", "categories": []}
{"id": "2409.18569", "title": "Cross-video Identity Correlating for Person Re-identification\n  Pre-training", "abstract": "  Recent researches have proven that pre-training on large-scale person images\nextracted from internet videos is an effective way in learning better\nrepresentations for person re-identification. However, these researches are\nmostly confined to pre-training at the instance-level or single-video\ntracklet-level. They ignore the identity-invariance in images of the same\nperson across different videos, which is a key focus in person\nre-identification. To address this issue, we propose a Cross-video\nIdentity-cOrrelating pre-traiNing (CION) framework. Defining a noise concept\nthat comprehensively considers both intra-identity consistency and\ninter-identity discrimination, CION seeks the identity correlation from\ncross-video images by modeling it as a progressive multi-level denoising\nproblem. Furthermore, an identity-guided self-distillation loss is proposed to\nimplement better large-scale pre-training by mining the identity-invariance\nwithin person images. We conduct extensive experiments to verify the\nsuperiority of our CION in terms of efficiency and performance. CION achieves\nsignificantly leading performance with even fewer training samples. For\nexample, compared with the previous state-of-the-art~\\cite{ISR}, CION with the\nsame ResNet50-IBN achieves higher mAP of 93.3\\% and 74.3\\% on Market1501 and\nMSMT17, while only utilizing 8\\% training samples. Finally, with CION\ndemonstrating superior model-agnostic ability, we contribute a model zoo named\nReIDZoo to meet diverse research and application needs in this field. It\ncontains a series of CION pre-trained models with spanning structures and\nparameters, totaling 32 models with 10 different structures, including\nGhostNet, ConvNext, RepViT, FastViT and so on. The code and models will be made\npublicly available at https://github.com/Zplusdragon/CION_ReIDZoo.\n", "date": "2024-09-27", "categories": []}
{"id": "2409.18636", "title": "Unsupervised Fingerphoto Presentation Attack Detection With Diffusion\n  Models", "abstract": "  Smartphone-based contactless fingerphoto authentication has become a reliable\nalternative to traditional contact-based fingerprint biometric systems owing to\nrapid advances in smartphone camera technology. Despite its convenience,\nfingerprint authentication through fingerphotos is more vulnerable to\npresentation attacks, which has motivated recent research efforts towards\ndeveloping fingerphoto Presentation Attack Detection (PAD) techniques. However,\nprior PAD approaches utilized supervised learning methods that require labeled\ntraining data for both bona fide and attack samples. This can suffer from two\nkey issues, namely (i) generalization:the detection of novel presentation\nattack instruments (PAIs) unseen in the training data, and (ii) scalability:the\ncollection of a large dataset of attack samples using different PAIs. To\naddress these challenges, we propose a novel unsupervised approach based on a\nstate-of-the-art deep-learning-based diffusion model, the Denoising Diffusion\nProbabilistic Model (DDPM), which is trained solely on bona fide samples. The\nproposed approach detects Presentation Attacks (PA) by calculating the\nreconstruction similarity between the input and output pairs of the DDPM. We\npresent extensive experiments across three PAI datasets to test the accuracy\nand generalization capability of our approach. The results show that the\nproposed DDPM-based PAD method achieves significantly better detection error\nrates on several PAI classes compared to other baseline unsupervised\napproaches.\n", "date": "2024-09-27", "categories": []}
{"id": "2311.18576", "title": "Fixed-length Dense Descriptor for Efficient Fingerprint Matching", "abstract": "  In fingerprint matching, fixed-length descriptors generally offer greater\nefficiency compared to minutiae set, but the recognition accuracy is not as\ngood as that of the latter. Although much progress has been made in deep\nlearning based fixed-length descriptors recently, they often fall short when\ndealing with incomplete or partial fingerprints, diverse fingerprint poses, and\nsignificant background noise. In this paper, we propose a three-dimensional\nrepresentation called Fixed-length Dense Descriptor (FDD) for efficient\nfingerprint matching. FDD features great spatial properties, enabling it to\ncapture the spatial relationships of the original fingerprints, thereby\nenhancing interpretability and robustness. Our experiments on various\nfingerprint datasets reveal that FDD outperforms other fixed-length\ndescriptors, especially in matching fingerprints of different areas,\ncross-modal fingerprint matching, and fingerprint matching with background\nnoise.\n", "date": "2023-11-30", "categories": []}
{"id": "2409.16056", "title": "Adversarial Watermarking for Face Recognition", "abstract": "  Watermarking is an essential technique for embedding an identifier (i.e.,\nwatermark message) within digital images to assert ownership and monitor\nunauthorized alterations. In face recognition systems, watermarking plays a\npivotal role in ensuring data integrity and security. However, an adversary\ncould potentially interfere with the watermarking process, significantly\nimpairing recognition performance. We explore the interaction between\nwatermarking and adversarial attacks on face recognition models. Our findings\nreveal that while watermarking or input-level perturbation alone may have a\nnegligible effect on recognition accuracy, the combined effect of watermarking\nand perturbation can result in an adversarial watermarking attack,\nsignificantly degrading recognition performance. Specifically, we introduce a\nnovel threat model, the adversarial watermarking attack, which remains stealthy\nin the absence of watermarking, allowing images to be correctly recognized\ninitially. However, once watermarking is applied, the attack is activated,\ncausing recognition failures. Our study reveals a previously unrecognized\nvulnerability: adversarial perturbations can exploit the watermark message to\nevade face recognition systems. Evaluated on the CASIA-WebFace dataset, our\nproposed adversarial watermarking attack reduces face matching accuracy by\n67.2% with an $\\ell_\\infty$ norm-measured perturbation strength of ${2}/{255}$\nand by 95.9% with a strength of ${4}/{255}$.\n", "date": "2024-09-24", "categories": []}
{"id": "2407.00261", "title": "Generative Iris Prior Embedded Transformer for Iris Restoration", "abstract": "  Iris restoration from complexly degraded iris images, aiming to improve iris\nrecognition performance, is a challenging problem. Due to the complex\ndegradation, directly training a convolutional neural network (CNN) without\nprior cannot yield satisfactory results. In this work, we propose a generative\niris prior embedded Transformer model (Gformer), in which we build a\nhierarchical encoder-decoder network employing Transformer block and generative\niris prior. First, we tame Transformer blocks to model long-range dependencies\nin target images. Second, we pretrain an iris generative adversarial network\n(GAN) to obtain the rich iris prior, and incorporate it into the iris\nrestoration process with our iris feature modulator. Our experiments\ndemonstrate that the proposed Gformer outperforms state-of-the-art methods.\nBesides, iris recognition performance has been significantly improved after\napplying Gformer.\n", "date": "2024-06-28", "categories": []}
{"id": "2409.14774", "title": "CFVNet: An End-to-End Cancelable Finger Vein Network for Recognition", "abstract": "  Finger vein recognition technology has become one of the primary solutions\nfor high-security identification systems. However, it still has information\nleakage problems, which seriously jeopardizes users privacy and anonymity and\ncause great security risks. In addition, there is no work to consider a fully\nintegrated secure finger vein recognition system. So, different from the\nprevious systems, we integrate preprocessing and template protection into an\nintegrated deep learning model. We propose an end-to-end cancelable finger vein\nnetwork (CFVNet), which can be used to design an secure finger vein recognition\nsystem.It includes a plug-and-play BWR-ROIAlign unit, which consists of three\nsub-modules: Localization, Compression and Transformation. The localization\nmodule achieves automated localization of stable and unique finger vein ROI.\nThe compression module losslessly removes spatial and channel redundancies. The\ntransformation module uses the proposed BWR method to introduce unlinkability,\nirreversibility and revocability to the system. BWR-ROIAlign can directly plug\ninto the model to introduce the above features for DCNN-based finger vein\nrecognition systems. We perform extensive experiments on four public datasets\nto study the performance and cancelable biometric attributes of the\nCFVNet-based recognition system. The average accuracy, EERs and Dsys on the\nfour datasets are 99.82%, 0.01% and 0.025, respectively, and achieves\ncompetitive performance compared with the state-of-the-arts.\n", "date": "2024-09-23", "categories": []}
{"id": "2210.00294", "title": "Gait-based Age Group Classification with Adaptive Graph Neural Network", "abstract": "  Deep learning techniques have recently been utilized for model-free\nage-associated gait feature extraction. However, acquiring model-free gait\ndemands accurate pre-processing such as background subtraction, which is\nnon-trivial in unconstrained environments. On the other hand, model-based gait\ncan be obtained without background subtraction and is less affected by\ncovariates. For model-based gait-based age group classification problems,\npresent works rely solely on handcrafted features, where feature extraction is\ntedious and requires domain expertise. This paper proposes a deep learning\napproach to extract age-associated features from model-based gait for age group\nclassification. Specifically, we first develop an unconstrained gait dataset\ncalled Multimedia University Gait Age and Gender dataset (MMU GAG). Next, the\nbody joint coordinates are determined via pose estimation algorithms and\nrepresented as compact gait graphs via a novel part aggregation scheme. Then, a\nPart-AdaptIve Residual Graph Convolutional Neural Network (PairGCN) is designed\nfor age-associated feature learning. Experiments suggest that PairGCN features\nare far more informative than handcrafted features, yielding up to 99% accuracy\nfor classifying subjects as a child, adult, or senior in the MMU GAG dataset.\nThese results suggest the feasibility of deploying Artificial\nIntelligence-enabled solutions for access control, surveillance, and law\nenforcement in unconstrained environments.\n", "date": "2022-10-01", "categories": []}
{"id": "2409.13056", "title": "Cross-Chirality Palmprint Verification: Left is Right for the Right\n  Palmprint", "abstract": "  Palmprint recognition has emerged as a prominent biometric authentication\nmethod, owing to its high discriminative power and user-friendly nature. This\npaper introduces a novel Cross-Chirality Palmprint Verification (CCPV)\nframework that challenges the conventional wisdom in traditional palmprint\nverification systems. Unlike existing methods that typically require storing\nboth left and right palmprints, our approach enables verification using either\npalm while storing only one palmprint template. The core of our CCPV framework\nlies in a carefully designed matching rule. This rule involves flipping both\nthe gallery and query palmprints and calculating the average distance between\neach pair as the final matching distance. This approach effectively reduces\nmatching variance and enhances overall system robustness. We introduce a novel\ncross-chirality loss function to construct a discriminative and robust\ncross-chirality feature space. This loss enforces representation consistency\nacross four palmprint variants: left, right, flipped left, and flipped right.\nThe resulting compact feature space, coupled with the model's enhanced\ndiscriminative representation capability, ensures robust performance across\nvarious scenarios. We conducted extensive experiments to validate the efficacy\nof our proposed method. The evaluation encompassed multiple public datasets and\nconsidered both closed-set and open-set settings. The results demonstrate the\nCCPV framework's effectiveness and highlight its potential for real-world\napplications in palmprint authentication systems.\n", "date": "2024-09-19", "categories": []}
{"id": "2409.12318", "title": "A large-scale study of performance and equity of commercial remote\n  identity verification technologies across demographics", "abstract": "  As more types of transactions move online, there is an increasing need to\nverify someone's identity remotely. Remote identity verification (RIdV)\ntechnologies have emerged to fill this need. RIdV solutions typically use a\nsmart device to validate an identity document like a driver's license by\ncomparing a face selfie to the face photo on the document. Recent research has\nbeen focused on ensuring that biometric systems work fairly across demographic\ngroups. This study assesses five commercial RIdV solutions for equity across\nage, gender, race/ethnicity, and skin tone across 3,991 test subjects. This\npaper employs statistical methods to discern whether the RIdV result across\ndemographic groups is statistically distinguishable. Two of the RIdV solutions\nwere equitable across all demographics, while two RIdV solutions had at least\none demographic that was inequitable. For example, the results for one\ntechnology had a false negative rate of 10.5% +/- 4.5% and its performance for\neach demographic category was within the error bounds, and, hence, were\nequitable. The other technologies saw either poor overall performance or\ninequitable performance. For one of these, participants of the race\nBlack/African American (B/AA) as well as those with darker skin tones (Monk\nscale 7/8/9/10) experienced higher false rejections. Finally, one technology\ndemonstrated more favorable but inequitable performance for the Asian American\nand Pacific Islander (AAPI) demographic. This study confirms that it is\nnecessary to evaluate products across demographic groups to fully understand\nthe performance of remote identity verification technologies.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.12385", "title": "Look Through Masks: Towards Masked Face Recognition with De-Occlusion\n  Distillation", "abstract": "  Many real-world applications today like video surveillance and urban\ngovernance need to address the recognition of masked faces, where content\nreplacement by diverse masks often brings in incomplete appearance and\nambiguous representation, leading to a sharp drop in accuracy. Inspired by\nrecent progress on amodal perception, we propose to migrate the mechanism of\namodal completion for the task of masked face recognition with an end-to-end\nde-occlusion distillation framework, which consists of two modules. The\n\\textit{de-occlusion} module applies a generative adversarial network to\nperform face completion, which recovers the content under the mask and\neliminates appearance ambiguity. The \\textit{distillation} module takes a\npre-trained general face recognition model as the teacher and transfers its\nknowledge to train a student for completed faces using massive online\nsynthesized face pairs. Especially, the teacher knowledge is represented with\nstructural relations among instances in multiple orders, which serves as a\nposterior regularization to enable the adaptation. In this way, the knowledge\ncan be fully distilled and transferred to identify masked faces. Experiments on\nsynthetic and realistic datasets show the efficacy of the proposed approach.\n", "date": "2024-09-19", "categories": []}
{"id": "2409.12884", "title": "Hypersphere Secure Sketch Revisited: Probabilistic Linear Regression\n  Attack on IronMask in Multiple Usage", "abstract": "  Protection of biometric templates is a critical and urgent area of focus.\nIronMask demonstrates outstanding recognition performance while protecting\nfacial templates against existing known attacks. In high-level, IronMask can be\nconceptualized as a fuzzy commitment scheme building on the hypersphere\ndirectly. We devise an attack on IronMask targeting on the security notion of\nrenewability. Our attack, termed as Probabilistic Linear Regression Attack,\nutilizes the linearity of underlying used error correcting code. This attack is\nthe first algorithm to successfully recover the original template when getting\nmultiple protected templates in acceptable time and requirement of storage. We\nimplement experiments on IronMask applied to protect ArcFace that well verify\nthe validity of our attacks. Furthermore, we carry out experiments in noisy\nenvironments and confirm that our attacks are still applicable. Finally, we put\nforward two strategies to mitigate this type of attacks.\n", "date": "2024-09-19", "categories": []}
{"id": "2409.08345", "title": "SIG: A Synthetic Identity Generation Pipeline for Generating Evaluation\n  Datasets for Face Recognition", "abstract": "  As Artificial Intelligence applications expand, the evaluation of models\nfaces heightened scrutiny. Ensuring public readiness requires evaluation\ndatasets, which differ from training data by being disjoint and ethically\nsourced in compliance with privacy regulations. The performance and fairness of\nface recognition systems depend significantly on the quality and\nrepresentativeness of these evaluation datasets. This data is sometimes scraped\nfrom the internet without user's consent, causing ethical concerns that can\nprohibit its use without proper releases. In rare cases, data is collected in a\ncontrolled environment with consent, however, this process is time-consuming,\nexpensive, and logistically difficult to execute. This creates a barrier for\nthose unable to conjure the immense resources required to gather ethically\nsourced evaluation datasets. To address these challenges, we introduce the\nSynthetic Identity Generation pipeline, or SIG, that allows for the targeted\ncreation of ethical, balanced datasets for face recognition evaluation. Our\nproposed and demonstrated pipeline generates high-quality images of synthetic\nidentities with controllable pose, facial features, and demographic attributes,\nsuch as race, gender, and age. We also release an open-source evaluation\ndataset named ControlFace10k, consisting of 10,008 face images of 3,336 unique\nsynthetic identities balanced across race, gender, and age, generated using the\nproposed SIG pipeline. We analyze ControlFace10k along with a non-synthetic\nBUPT dataset using state-of-the-art face recognition algorithms to demonstrate\nits effectiveness as an evaluation tool. This analysis highlights the dataset's\ncharacteristics and its utility in assessing algorithmic bias across different\ndemographic groups.\n", "date": "2024-09-12", "categories": []}
{"id": "2409.11652", "title": "Relax DARTS: Relaxing the Constraints of Differentiable Architecture\n  Search for Eye Movement Recognition", "abstract": "  Eye movement biometrics is a secure and innovative identification method.\nDeep learning methods have shown good performance, but their network\narchitecture relies on manual design and combined priori knowledge. To address\nthese issues, we introduce automated network search (NAS) algorithms to the\nfield of eye movement recognition and present Relax DARTS, which is an\nimprovement of the Differentiable Architecture Search (DARTS) to realize more\nefficient network search and training. The key idea is to circumvent the issue\nof weight sharing by independently training the architecture parameters\n$\\alpha$ to achieve a more precise target architecture. Moreover, the\nintroduction of module input weights $\\beta$ allows cells the flexibility to\nselect inputs, to alleviate the overfitting phenomenon and improve the model\nperformance. Results on four public databases demonstrate that the Relax DARTS\nachieves state-of-the-art recognition performance. Notably, Relax DARTS\nexhibits adaptability to other multi-feature temporal classification tasks.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11786", "title": "Efficient Low-Resolution Face Recognition via Bridge Distillation", "abstract": "  Face recognition in the wild is now advancing towards light-weight models,\nfast inference speed and resolution-adapted capability. In this paper, we\npropose a bridge distillation approach to turn a complex face model pretrained\non private high-resolution faces into a light-weight one for low-resolution\nface recognition. In our approach, such a cross-dataset resolution-adapted\nknowledge transfer problem is solved via two-step distillation. In the first\nstep, we conduct cross-dataset distillation to transfer the prior knowledge\nfrom private high-resolution faces to public high-resolution faces and generate\ncompact and discriminative features. In the second step, the resolution-adapted\ndistillation is conducted to further transfer the prior knowledge to synthetic\nlow-resolution faces via multi-task learning. By learning low-resolution face\nrepresentations and mimicking the adapted high-resolution knowledge, a\nlight-weight student model can be constructed with high efficiency and\npromising accuracy in recognizing low-resolution faces. Experimental results\nshow that the student model performs impressively in recognizing low-resolution\nfaces with only 0.21M parameters and 0.057MB memory. Meanwhile, its speed\nreaches up to 14,705, ~934 and 763 faces per second on GPU, CPU and mobile\nphone, respectively.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11802", "title": "Latent fingerprint enhancement for accurate minutiae detection", "abstract": "  Identification of suspects based on partial and smudged fingerprints,\ncommonly referred to as fingermarks or latent fingerprints, presents a\nsignificant challenge in the field of fingerprint recognition. Although\nfixed-length embeddings have shown effectiveness in recognising rolled and slap\nfingerprints, the methods for matching latent fingerprints have primarily\ncentred around local minutiae-based embeddings, failing to fully exploit global\nrepresentations for matching purposes. Consequently, enhancing latent\nfingerprints becomes critical to ensuring robust identification for forensic\ninvestigations. Current approaches often prioritise restoring ridge patterns,\noverlooking the fine-macroeconomic details crucial for accurate fingerprint\nrecognition. To address this, we propose a novel approach that uses generative\nadversary networks (GANs) to redefine Latent Fingerprint Enhancement (LFE)\nthrough a structured approach to fingerprint generation. By directly optimising\nthe minutiae information during the generation process, the model produces\nenhanced latent fingerprints that exhibit exceptional fidelity to ground-truth\ninstances. This leads to a significant improvement in identification\nperformance. Our framework integrates minutiae locations and orientation\nfields, ensuring the preservation of both local and structural fingerprint\nfeatures. Extensive evaluations conducted on two publicly available datasets\ndemonstrate our method's dominance over existing state-of-the-art techniques,\nhighlighting its potential to significantly enhance latent fingerprint\nrecognition accuracy in forensic applications.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11816", "title": "SymFace: Additional Facial Symmetry Loss for Deep Face Recognition", "abstract": "  Over the past decade, there has been a steady advancement in enhancing face\nrecognition algorithms leveraging advanced machine learning methods. The role\nof the loss function is pivotal in addressing face verification problems and\nplaying a game-changing role. These loss functions have mainly explored\nvariations among intra-class or inter-class separation. This research examines\nthe natural phenomenon of facial symmetry in the face verification problem. The\nsymmetry between the left and right hemi faces has been widely used in many\nresearch areas in recent decades. This paper adopts this simple approach\njudiciously by splitting the face image vertically into two halves. With the\nassumption that the natural phenomena of facial symmetry can enhance face\nverification methodology, we hypothesize that the two output embedding vectors\nof split faces must project close to each other in the output embedding space.\nInspired by this concept, we penalize the network based on the disparity of\nembedding of the symmetrical pair of split faces. Symmetrical loss has the\npotential to minimize minor asymmetric features due to facial expression and\nlightning conditions, hence significantly increasing the inter-class variance\namong the classes and leading to more reliable face embedding. This loss\nfunction propels any network to outperform its baseline performance across all\nexisting network architectures and configurations, enabling us to achieve SoTA\nresults.\n", "date": "2024-09-18", "categories": []}
{"id": "2409.11224", "title": "A Human-Centered Risk Evaluation of Biometric Systems Using Conjoint\n  Analysis", "abstract": "  Biometric recognition systems, known for their convenience, are widely\nadopted across various fields. However, their security faces risks depending on\nthe authentication algorithm and deployment environment. Current risk\nassessment methods faces significant challenges in incorporating the crucial\nfactor of attacker's motivation, leading to incomplete evaluations. This paper\npresents a novel human-centered risk evaluation framework using conjoint\nanalysis to quantify the impact of risk factors, such as surveillance cameras,\non attacker's motivation. Our framework calculates risk values incorporating\nthe False Acceptance Rate (FAR) and attack probability, allowing comprehensive\ncomparisons across use cases. A survey of 600 Japanese participants\ndemonstrates our method's effectiveness, showing how security measures\ninfluence attacker's motivation. This approach helps decision-makers customize\nbiometric systems to enhance security while maintaining usability.\n", "date": "2024-09-17", "categories": []}
{"id": "2409.09274", "title": "LabellessFace: Fair Metric Learning for Face Recognition without\n  Attribute Labels", "abstract": "  Demographic bias is one of the major challenges for face recognition systems.\nThe majority of existing studies on demographic biases are heavily dependent on\nspecific demographic groups or demographic classifier, making it difficult to\naddress performance for unrecognised groups. This paper introduces\n``LabellessFace'', a novel framework that improves demographic bias in face\nrecognition without requiring demographic group labeling typically required for\nfairness considerations. We propose a novel fairness enhancement metric called\nthe class favoritism level, which assesses the extent of favoritism towards\nspecific classes across the dataset. Leveraging this metric, we introduce the\nfair class margin penalty, an extension of existing margin-based metric\nlearning. This method dynamically adjusts learning parameters based on class\nfavoritism levels, promoting fairness across all attributes. By treating each\nclass as an individual in facial recognition systems, we facilitate learning\nthat minimizes biases in authentication accuracy among individuals.\nComprehensive experiments have demonstrated that our proposed method is\neffective for enhancing fairness while maintaining authentication accuracy.\n", "date": "2024-09-14", "categories": []}
{"id": "2409.10481", "title": "Exploring 3D Face Reconstruction and Fusion Methods for Face\n  Verification: A Case-Study in Video Surveillance", "abstract": "  3D face reconstruction (3DFR) algorithms are based on specific assumptions\ntailored to distinct application scenarios. These assumptions limit their use\nwhen acquisition conditions, such as the subject's distance from the camera or\nthe camera's characteristics, are different than expected, as typically happens\nin video surveillance. Additionally, 3DFR algorithms follow various strategies\nto address the reconstruction of a 3D shape from 2D data, such as statistical\nmodel fitting, photometric stereo, or deep learning. In the present study, we\nexplore the application of three 3DFR algorithms representative of the SOTA,\nemploying each one as the template set generator for a face verification\nsystem. The scores provided by each system are combined by score-level fusion.\nWe show that the complementarity induced by different 3DFR algorithms improves\nperformance when tests are conducted at never-seen-before distances from the\ncamera and camera characteristics (cross-distance and cross-camera settings),\nthus encouraging further investigations on multiple 3DFR-based approaches.\n", "date": "2024-09-16", "categories": []}
{"id": "2407.02403", "title": "Face Reconstruction Transfer Attack as Out-of-Distribution\n  Generalization", "abstract": "  Understanding the vulnerability of face recognition systems to malicious\nattacks is of critical importance. Previous works have focused on\nreconstructing face images that can penetrate a targeted verification system.\nEven in the white-box scenario, however, naively reconstructed images\nmisrepresent the identity information, hence the attacks are easily neutralized\nonce the face system is updated or changed. In this paper, we aim to\nreconstruct face images which are capable of transferring face attacks on\nunseen encoders. We term this problem as Face Reconstruction Transfer Attack\n(FRTA) and show that it can be formulated as an out-of-distribution (OOD)\ngeneralization problem. Inspired by its OOD nature, we propose to solve FRTA by\nAveraged Latent Search and Unsupervised Validation with pseudo target (ALSUV).\nTo strengthen the reconstruction attack on OOD unseen encoders, ALSUV\nreconstructs the face by searching the latent of amortized generator StyleGAN2\nthrough multiple latent optimization, latent optimization trajectory averaging,\nand unsupervised validation with a pseudo target. We demonstrate the efficacy\nand generalization of our method on widely used face datasets, accompanying it\nwith extensive ablation studies and visually, qualitatively, and quantitatively\nanalyses. The source code will be released.\n", "date": "2024-07-02", "categories": []}
{"id": "2409.07488", "title": "Contrastive Learning-based User Identification with Limited Data on\n  Smart Textiles", "abstract": "  Pressure-sensitive smart textiles are widely applied in the fields of\nhealthcare, sports monitoring, and intelligent homes. The integration of\ndevices embedded with pressure sensing arrays is expected to enable\ncomprehensive scene coverage and multi-device integration. However, the\nimplementation of identity recognition, a fundamental function in this context,\nrelies on extensive device-specific datasets due to variations in pressure\ndistribution across different devices. To address this challenge, we propose a\nnovel user identification method based on contrastive learning. We design two\nparallel branches to facilitate user identification on both new and existing\ndevices respectively, employing supervised contrastive learning in the feature\nspace to promote domain unification. When encountering new devices, extensive\ndata collection efforts are not required; instead, user identification can be\nachieved using limited data consisting of only a few simple postures. Through\nexperimentation with two 8-subject pressure datasets (BedPressure and\nChrPressure), our proposed method demonstrates the capability to achieve user\nidentification across 12 sitting scenarios using only a dataset containing 2\npostures. Our average recognition accuracy reaches 79.05%, representing an\nimprovement of 2.62% over the best baseline model.\n", "date": "2024-09-06", "categories": []}
{"id": "2409.07128", "title": "Deep Learning Techniques for Hand Vein Biometrics: A Comprehensive\n  Review", "abstract": "  Biometric authentication has garnered significant attention as a secure and\nefficient method of identity verification. Among the various modalities, hand\nvein biometrics, including finger vein, palm vein, and dorsal hand vein\nrecognition, offer unique advantages due to their high accuracy, low\nsusceptibility to forgery, and non-intrusiveness. The vein patterns within the\nhand are highly complex and distinct for each individual, making them an ideal\nbiometric identifier. Additionally, hand vein recognition is contactless,\nenhancing user convenience and hygiene compared to other modalities such as\nfingerprint or iris recognition. Furthermore, the veins are internally located,\nrendering them less susceptible to damage or alteration, thus enhancing the\nsecurity and reliability of the biometric system. The combination of these\nfactors makes hand vein biometrics a highly effective and secure method for\nidentity verification. This review paper delves into the latest advancements in\ndeep learning techniques applied to finger vein, palm vein, and dorsal hand\nvein recognition. It encompasses all essential fundamentals of hand vein\nbiometrics, summarizes publicly available datasets, and discusses\nstate-of-the-art metrics used for evaluating the three modes. Moreover, it\nprovides a comprehensive overview of suggested approaches for finger, palm,\ndorsal, and multimodal vein techniques, offering insights into the best\nperformance achieved, data augmentation techniques, and effective transfer\nlearning methods, along with associated pretrained deep learning models.\nAdditionally, the review addresses research challenges faced and outlines\nfuture directions and perspectives, encouraging researchers to enhance existing\nmethods and propose innovative techniques.\n", "date": "2024-09-11", "categories": []}
{"id": "2409.07220", "title": "Watchlist Challenge: 3rd Open-set Face Detection and Identification", "abstract": "  In the current landscape of biometrics and surveillance, the ability to\naccurately recognize faces in uncontrolled settings is paramount. The Watchlist\nChallenge addresses this critical need by focusing on face detection and\nopen-set identification in real-world surveillance scenarios. This paper\npresents a comprehensive evaluation of participating algorithms, using the\nenhanced UnConstrained College Students (UCCS) dataset with new evaluation\nprotocols. In total, four participants submitted four face detection and nine\nopen-set face recognition systems. The evaluation demonstrates that while\ndetection capabilities are generally robust, closed-set identification\nperformance varies significantly, with models pre-trained on large-scale\ndatasets showing superior performance. However, open-set scenarios require\nfurther improvement, especially at higher true positive identification rates,\ni.e., lower thresholds.\n", "date": "2024-09-11", "categories": []}
{"id": "2408.05743", "title": "Neural Architecture Search based Global-local Vision Mamba for Palm-Vein\n  Recognition", "abstract": "  Due to the advantages such as high security, high privacy, and liveness\nrecognition, vein recognition has been received more and more attention in past\nyears. Recently, deep learning models, e.g., Mamba has shown robust feature\nrepresentation with linear computational complexity and successfully applied\nfor visual tasks. However, vision Manba can capture long-distance feature\ndependencies but unfortunately deteriorate local feature details. Besides,\nmanually designing a Mamba architecture based on human priori knowledge is very\ntime-consuming and error-prone. In this paper, first, we propose a hybrid\nnetwork structure named Global-local Vision Mamba (GLVM), to learn the local\ncorrelations in images explicitly and global dependencies among tokens for vein\nfeature representation. Secondly, we design a Multi-head Mamba to learn the\ndependencies along different directions, so as to improve the feature\nrepresentation ability of vision Mamba. Thirdly, to learn the complementary\nfeatures, we propose a ConvMamba block consisting of three branches, named\nMulti-head Mamba branch (MHMamba), Feature Iteration Unit branch (FIU), and\nConvolutional Neural Network (CNN) branch, where the Feature Iteration Unit\nbranch aims to fuse convolutional local features with Mamba-based global\nrepresentations. Finally, a Globallocal Alternate Neural Architecture Search\n(GLNAS) method is proposed to search the optimal architecture of GLVM\nalternately with the evolutionary algorithm, thereby improving the recognition\nperformance for vein recognition tasks. We conduct rigorous experiments on\nthree public palm-vein databases to estimate the performance. The experimental\nresults demonstrate that the proposed method outperforms the representative\napproaches and achieves state-of-the-art recognition accuracy.\n", "date": "2024-08-11", "categories": []}
{"id": "2409.05277", "title": "Disentangled Representations for Short-Term and Long-Term Person\n  Re-Identification", "abstract": "  We address the problem of person re-identification (reID), that is,\nretrieving person images from a large dataset, given a query image of the\nperson of interest. A key challenge is to learn person representations robust\nto intra-class variations, as different persons could have the same attribute,\nand persons' appearances look different, e.g., with viewpoint changes. Recent\nreID methods focus on learning person features discriminative only for a\nparticular factor of variations (e.g., human pose), which also requires\ncorresponding supervisory signals (e.g., pose annotations). To tackle this\nproblem, we propose to factorize person images into identity-related and\nunrelated features. Identity-related features contain information useful for\nspecifying a particular person (e.g., clothing), while identity-unrelated ones\nhold other factors (e.g., human pose). To this end, we propose a new generative\nadversarial network, dubbed identity shuffle GAN (IS-GAN). It disentangles\nidentity-related and unrelated features from person images through an\nidentity-shuffling technique that exploits identification labels alone without\nany auxiliary supervisory signals. We restrict the distribution of\nidentity-unrelated features or encourage the identity-related and unrelated\nfeatures to be uncorrelated, facilitating the disentanglement process.\nExperimental results validate the effectiveness of IS-GAN, showing\nstate-of-the-art performance on standard reID benchmarks, including\nMarket-1501, CUHK03, and DukeMTMC-reID. We further demonstrate the advantages\nof disentangling person representations on a long-term reID task, setting a new\nstate of the art on a Celeb-reID dataset.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.05642", "title": "Prototype-Driven Multi-Feature Generation for Visible-Infrared Person\n  Re-identification", "abstract": "  The primary challenges in visible-infrared person re-identification arise\nfrom the differences between visible (vis) and infrared (ir) images, including\ninter-modal and intra-modal variations. These challenges are further\ncomplicated by varying viewpoints and irregular movements. Existing methods\noften rely on horizontal partitioning to align part-level features, which can\nintroduce inaccuracies and have limited effectiveness in reducing modality\ndiscrepancies. In this paper, we propose a novel Prototype-Driven Multi-feature\ngeneration framework (PDM) aimed at mitigating cross-modal discrepancies by\nconstructing diversified features and mining latent semantically similar\nfeatures for modal alignment. PDM comprises two key components: Multi-Feature\nGeneration Module (MFGM) and Prototype Learning Module (PLM). The MFGM\ngenerates diversity features closely distributed from modality-shared features\nto represent pedestrians. Additionally, the PLM utilizes learnable prototypes\nto excavate latent semantic similarities among local features between visible\nand infrared modalities, thereby facilitating cross-modal instance-level\nalignment. We introduce the cosine heterogeneity loss to enhance prototype\ndiversity for extracting rich local features. Extensive experiments conducted\non the SYSU-MM01 and LLCM datasets demonstrate that our approach achieves\nstate-of-the-art performance. Our codes are available at\nhttps://github.com/mmunhappy/ICASSP2025-PDM.\n", "date": "2024-09-09", "categories": []}
{"id": "2409.02867", "title": "The Impact of Balancing Real and Synthetic Data on Accuracy and Fairness\n  in Face Recognition", "abstract": "  Over the recent years, the advancements in deep face recognition have fueled\nan increasing demand for large and diverse datasets. Nevertheless, the\nauthentic data acquired to create those datasets is typically sourced from the\nweb, which, in many cases, can lead to significant privacy issues due to the\nlack of explicit user consent. Furthermore, obtaining a demographically\nbalanced, large dataset is even more difficult because of the natural imbalance\nin the distribution of images from different demographic groups. In this paper,\nwe investigate the impact of demographically balanced authentic and synthetic\ndata, both individually and in combination, on the accuracy and fairness of\nface recognition models. Initially, several generative methods were used to\nbalance the demographic representations of the corresponding synthetic\ndatasets. Then a state-of-the-art face encoder was trained and evaluated using\n(combinations of) synthetic and authentic images. Our findings emphasized two\nmain points: (i) the increased effectiveness of training data generated by\ndiffusion-based models in enhancing accuracy, whether used alone or combined\nwith subsets of authentic data, and (ii) the minimal impact of incorporating\nbalanced data from pre-trained generative methods on fairness (in nearly all\ntested scenarios using combined datasets, fairness scores remained either\nunchanged or worsened, even when compared to unbalanced authentic datasets).\nSource code and data are available at \\url{https://cutt.ly/AeQy1K5G} for\nreproducibility.\n", "date": "2024-09-04", "categories": []}
{"id": "2405.15434", "title": "Biometrics and Behavior Analysis for Detecting Distractions in\n  e-Learning", "abstract": "  In this article, we explore computer vision approaches to detect abnormal\nhead pose during e-learning sessions and we introduce a study on the effects of\nmobile phone usage during these sessions. We utilize behavioral data collected\nfrom 120 learners monitored while participating in a MOOC learning sessions.\nOur study focuses on the influence of phone-usage events on behavior and\nphysiological responses, specifically attention, heart rate, and meditation,\nbefore, during, and after phone usage. Additionally, we propose an approach for\nestimating head pose events using images taken by the webcam during the MOOC\nlearning sessions to detect phone-usage events. Our hypothesis suggests that\nhead posture undergoes significant changes when learners interact with a mobile\nphone, contrasting with the typical behavior seen when learners face a computer\nduring e-learning sessions. We propose an approach designed to detect\ndeviations in head posture from the average observed during a learner's\nsession, operating as a semi-supervised method. This system flags events\nindicating alterations in head posture for subsequent human review and\nselection of mobile phone usage occurrences with a sensitivity over 90%.\n", "date": "2024-05-24", "categories": []}
{"id": "2408.07500", "title": "Cross-Platform Video Person ReID: A New Benchmark Dataset and Adaptation\n  Approach", "abstract": "  In this paper, we construct a large-scale benchmark dataset for\nGround-to-Aerial Video-based person Re-Identification, named G2A-VReID, which\ncomprises 185,907 images and 5,576 tracklets, featuring 2,788 distinct\nidentities. To our knowledge, this is the first dataset for video ReID under\nGround-to-Aerial scenarios. G2A-VReID dataset has the following\ncharacteristics: 1) Drastic view changes; 2) Large number of annotated\nidentities; 3) Rich outdoor scenarios; 4) Huge difference in resolution.\nAdditionally, we propose a new benchmark approach for cross-platform ReID by\ntransforming the cross-platform visual alignment problem into visual-semantic\nalignment through vision-language model (i.e., CLIP) and applying a\nparameter-efficient Video Set-Level-Adapter module to adapt image-based\nfoundation model to video ReID tasks, termed VSLA-CLIP. Besides, to further\nreduce the great discrepancy across the platforms, we also devise the\nplatform-bridge prompts for efficient visual feature alignment. Extensive\nexperiments demonstrate the superiority of the proposed method on all existing\nvideo ReID datasets and our proposed G2A-VReID dataset.\n", "date": "2024-08-14", "categories": []}
{"id": "2409.00779", "title": "Unbalanced Fingerprint Classification for Hybrid Fingerprint Orientation\n  Maps", "abstract": "  This paper introduces a novel fingerprint classification technique based on a\nmulti-layered fuzzy logic classifier. We target the cause of missed detection\nby identifying the fingerprints at an early stage among dry, standard, and wet.\nScanned images are classified based on clarity correlated with the proposed\nfeature points. We also propose a novel adaptive algorithm based on eigenvector\nspace for generating new samples to overcome the multiclass imbalance. Proposed\nmethods improve the performance of ensemble learners. It was also found that\nthe new approach performs better than the neural-network based classification\nmethods. Early-stage improvements give a suitable dataset for fingerprint\ndetection models. Leveraging the novel classifier, the best set of `standard'\nlabelled fingerprints is used to generate a unique hybrid fingerprint\norientation map (HFOM). We introduce a novel min-rotate max-flow optimization\nmethod inspired by the min-cut max-flow algorithm. The unique properties of\nHFOM generation introduce a new use case for biometric data protection by using\nHFOM as a virtual proxy of fingerprints.\n", "date": "2024-09-01", "categories": []}
{"id": "2409.01514", "title": "From Data to Insights: A Covariate Analysis of the IARPA BRIAR Dataset\n  for Multimodal Biometric Recognition Algorithms at Altitude and Range", "abstract": "  This paper examines covariate effects on fused whole body biometrics\nperformance in the IARPA BRIAR dataset, specifically focusing on UAV platforms,\nelevated positions, and distances up to 1000 meters. The dataset includes\noutdoor videos compared with indoor images and controlled gait recordings.\nNormalized raw fusion scores relate directly to predicted false accept rates\n(FAR), offering an intuitive means for interpreting model results. A linear\nmodel is developed to predict biometric algorithm scores, analyzing their\nperformance to identify the most influential covariates on accuracy at altitude\nand range. Weather factors like temperature, wind speed, solar loading, and\nturbulence are also investigated in this analysis. The study found that\nresolution and camera distance best predicted accuracy and findings can guide\nfuture research and development efforts in long-range/elevated/UAV biometrics\nand support the creation of more reliable and robust systems for national\nsecurity and other critical domains.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.01540", "title": "Long-Range Biometric Identification in Real World Scenarios: A\n  Comprehensive Evaluation Framework Based on Missions", "abstract": "  The considerable body of data available for evaluating biometric recognition\nsystems in Research and Development (R\\&D) environments has contributed to the\nincreasingly common problem of target performance mismatch. Biometric\nalgorithms are frequently tested against data that may not reflect the real\nworld applications they target. From a Testing and Evaluation (T\\&E)\nstandpoint, this domain mismatch causes difficulty assessing when improvements\nin State-of-the-Art (SOTA) research actually translate to improved applied\noutcomes. This problem can be addressed with thoughtful preparation of data and\nexperimental methods to reflect specific use-cases and scenarios.\n  To that end, this paper evaluates research solutions for identifying\nindividuals at ranges and altitudes, which could support various application\nareas such as counterterrorism, protection of critical infrastructure\nfacilities, military force protection, and border security. We address\nchallenges including image quality issues and reliance on face recognition as\nthe sole biometric modality. By fusing face and body features, we propose\ndeveloping robust biometric systems for effective long-range identification\nfrom both the ground and steep pitch angles. Preliminary results show promising\nprogress in whole-body recognition. This paper presents these early findings\nand discusses potential future directions for advancing long-range biometric\nidentification systems based on mission-driven metrics.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.01928", "title": "Comprehensive Equity Index (CEI): Definition and Application to Bias\n  Evaluation in Biometrics", "abstract": "  We present a novel metric designed, among other applications, to quantify\nbiased behaviors of machine learning models. As its core, the metric consists\nof a new similarity metric between score distributions that balances both their\ngeneral shapes and tails' probabilities. In that sense, our proposed metric may\nbe useful in many application areas. Here we focus on and apply it to the\noperational evaluation of face recognition systems, with special attention to\nquantifying demographic biases; an application where our metric is especially\nuseful. The topic of demographic bias and fairness in biometric recognition\nsystems has gained major attention in recent years. The usage of these systems\nhas spread in society, raising concerns about the extent to which these systems\ntreat different population groups. A relevant step to prevent and mitigate\ndemographic biases is first to detect and quantify them. Traditionally, two\napproaches have been studied to quantify differences between population groups\nin machine learning literature: 1) measuring differences in error rates, and 2)\nmeasuring differences in recognition score distributions. Our proposed\nComprehensive Equity Index (CEI) trade-offs both approaches combining both\nerrors from distribution tails and general distribution shapes. This new metric\nis well suited to real-world scenarios, as measured on NIST FRVT evaluations,\ninvolving high-performance systems and realistic face databases including a\nwide range of covariates and demographic groups. We first show the limitations\nof existing metrics to correctly assess the presence of biases in realistic\nsetups and then propose our new metric to tackle these limitations. We tested\nthe proposed metric with two state-of-the-art models and four widely used\ndatabases, showing its capacity to overcome the main flaws of previous bias\nmetrics.\n", "date": "2024-09-03", "categories": []}
{"id": "2409.02049", "title": "Low-Resolution Face Recognition via Adaptable Instance-Relation\n  Distillation", "abstract": "  Low-resolution face recognition is a challenging task due to the missing of\ninformative details. Recent approaches based on knowledge distillation have\nproven that high-resolution clues can well guide low-resolution face\nrecognition via proper knowledge transfer. However, due to the distribution\ndifference between training and testing faces, the learned models often suffer\nfrom poor adaptability. To address that, we split the knowledge transfer\nprocess into distillation and adaptation steps, and propose an adaptable\ninstance-relation distillation approach to facilitate low-resolution face\nrecognition. In the approach, the student distills knowledge from\nhigh-resolution teacher in both instance level and relation level, providing\nsufficient cross-resolution knowledge transfer. Then, the learned student can\nbe adaptable to recognize low-resolution faces with adaptive batch\nnormalization in inference. In this manner, the capability of recovering\nmissing details of familiar low-resolution faces can be effectively enhanced,\nleading to a better knowledge transfer. Extensive experiments on low-resolution\nface recognition clearly demonstrate the effectiveness and adaptability of our\napproach.\n", "date": "2024-09-03", "categories": []}
{"id": "2408.17399", "title": "How Knowledge Distillation Mitigates the Synthetic Gap in Fair Face\n  Recognition", "abstract": "  Leveraging the capabilities of Knowledge Distillation (KD) strategies, we\ndevise a strategy to fight the recent retraction of face recognition datasets.\nGiven a pretrained Teacher model trained on a real dataset, we show that\ncarefully utilising synthetic datasets, or a mix between real and synthetic\ndatasets to distil knowledge from this teacher to smaller students can yield\nsurprising results. In this sense, we trained 33 different models with and\nwithout KD, on different datasets, with different architectures and losses. And\nour findings are consistent, using KD leads to performance gains across all\nethnicities and decreased bias. In addition, it helps to mitigate the\nperformance gap between real and synthetic datasets. This approach addresses\nthe limitations of synthetic data training, improving both the accuracy and\nfairness of face recognition models.\n", "date": "2024-08-30", "categories": []}
{"id": "2408.16003", "title": "Meta-Learning for Federated Face Recognition in Imbalanced Data Regimes", "abstract": "  The growing privacy concerns surrounding face image data demand new\ntechniques that can guarantee user privacy. One such face recognition technique\nthat claims to achieve better user privacy is Federated Face Recognition (FRR),\na subfield of Federated Learning (FL). However, FFR faces challenges due to the\nheterogeneity of the data, given the large number of classes that need to be\nhandled. To overcome this problem, solutions are sought in the field of\npersonalized FL. This work introduces three new data partitions based on the\nCelebA dataset, each with a different form of data heterogeneity. It also\nproposes Hessian-Free Model Agnostic Meta-Learning (HF-MAML) in an FFR setting.\nWe show that HF-MAML scores higher in verification tests than current FFR\nmodels on three different CelebA data partitions. In particular, the\nverification scores improve the most in heterogeneous data partitions. To\nbalance personalization with the development of an effective global model, an\nembedding regularization term is introduced for the loss function. This term\ncan be combined with HF-MAML and is shown to increase global model verification\nperformance. Lastly, this work performs a fairness analysis, showing that\nHF-MAML and its embedding regularization extension can improve fairness by\nreducing the standard deviation over the client evaluation scores.\n", "date": "2024-08-13", "categories": []}
{"id": "2408.16563", "title": "MST-KD: Multiple Specialized Teachers Knowledge Distillation for Fair\n  Face Recognition", "abstract": "  As in school, one teacher to cover all subjects is insufficient to distill\nequally robust information to a student. Hence, each subject is taught by a\nhighly specialised teacher. Following a similar philosophy, we propose a\nmultiple specialized teacher framework to distill knowledge to a student\nnetwork. In our approach, directed at face recognition use cases, we train four\nteachers on one specific ethnicity, leading to four highly specialized and\nbiased teachers. Our strategy learns a project of these four teachers into a\ncommon space and distill that information to a student network. Our results\nhighlighted increased performance and reduced bias for all our experiments. In\naddition, we further show that having biased/specialized teachers is crucial by\nshowing that our approach achieves better results than when knowledge is\ndistilled from four teachers trained on balanced datasets. Our approach\nrepresents a step forward to the understanding of the importance of\nethnicity-specific features.\n", "date": "2024-08-29", "categories": []}
{"id": "2401.15048", "title": "Unrecognizable Yet Identifiable: Image Distortion with Preserved\n  Embeddings", "abstract": "  Biometric authentication systems play a crucial role in modern security\nsystems. However, maintaining the balance of privacy and integrity of stored\nbiometrics derivative data while achieving high recognition accuracy is often\nchallenging. Addressing this issue, we introduce an innovative image\ntransformation technique that effectively renders facial images unrecognizable\nto the eye while maintaining their identifiability by neural network models,\nwhich allows the distorted photo version to be stored for further verification.\nWhile initially intended for biometrics systems, the proposed methodology can\nbe used in various artificial intelligence applications to distort the visual\ndata and keep the derived features close. By experimenting with widely used\ndatasets LFW and MNIST, we show that it is possible to build the distortion\nthat changes the image content by more than 70% while maintaining the same\nrecognition accuracy. We compare our method with previously state-of-the-art\napproaches. We publically release the source code.\n", "date": "2024-01-26", "categories": []}
{"id": "2408.15693", "title": "Synthetic Forehead-creases Biometric Generation for Reliable User\n  Verification", "abstract": "  Recent studies have emphasized the potential of forehead-crease patterns as\nan alternative for face, iris, and periocular recognition, presenting\ncontactless and convenient solutions, particularly in situations where faces\nare covered by surgical masks. However, collecting forehead data presents\nchallenges, including cost and time constraints, as developing and optimizing\nforehead verification methods requires a substantial number of high-quality\nimages. To tackle these challenges, the generation of synthetic biometric data\nhas gained traction due to its ability to protect privacy while enabling\neffective training of deep learning-based biometric verification methods. In\nthis paper, we present a new framework to synthesize forehead-crease image data\nwhile maintaining important features, such as uniqueness and realism. The\nproposed framework consists of two main modules: a Subject-Specific Generation\nModule (SSGM), based on an image-to-image Brownian Bridge Diffusion Model\n(BBDM), which learns a one-to-many mapping between image pairs to generate\nidentity-aware synthetic forehead creases corresponding to real subjects, and a\nSubject-Agnostic Generation Module (SAGM), which samples new synthetic\nidentities with assistance from the SSGM. We evaluate the diversity and realism\nof the generated forehead-crease images primarily using the Fr\\'echet Inception\nDistance (FID) and the Structural Similarity Index Measure (SSIM). In addition,\nwe assess the utility of synthetically generated forehead-crease images using a\nforehead-crease verification system (FHCVS). The results indicate an\nimprovement in the verification accuracy of the FHCVS by utilizing synthetic\ndata.\n", "date": "2024-08-28", "categories": []}
{"id": "2404.09193", "title": "FaceCat: Enhancing Face Recognition Security with a Unified Diffusion\n  Model", "abstract": "  Face anti-spoofing (FAS) and adversarial detection (FAD) have been regarded\nas critical technologies to ensure the safety of face recognition systems.\nHowever, due to limited practicality, complex deployment, and the additional\ncomputational overhead, it is necessary to implement both detection techniques\nwithin a unified framework. This paper aims to achieve this goal by breaking\nthrough two primary obstacles: 1) the suboptimal face feature representation\nand 2) the scarcity of training data. To address the limited performance caused\nby existing feature representations, motivated by the rich structural and\ndetailed features of face diffusion models, we propose FaceCat, the first\napproach leveraging the diffusion model to simultaneously enhance the\nperformance of FAS and FAD. Specifically, FaceCat elaborately designs a\nhierarchical fusion mechanism to capture rich face semantic features of the\ndiffusion model. These features then serve as a robust foundation for a\nlightweight head, designed to execute FAS and FAD simultaneously. Due to the\nlimitations in feature representation that arise from relying solely on\nsingle-modality image data, we further propose a novel text-guided multi-modal\nalignment strategy that utilizes text prompts to enrich feature representation,\nthereby enhancing performance. To combat data scarcity, we build a\ncomprehensive dataset with a wide range of 28 attack types, offering greater\npotential for a unified framework in facial security. Extensive experiments\nvalidate the effectiveness of FaceCat generalizes significantly better and\nobtains excellent robustness against common input transformations.\n", "date": "2024-04-14", "categories": []}
{"id": "2408.13496", "title": "On the Feasibility of Creating Iris Periocular Morphed Images", "abstract": "  In the last few years, face morphing has been shown to be a complex challenge\nfor Face Recognition Systems (FRS). Thus, the evaluation of other biometric\nmodalities such as fingerprint, iris, and others must be explored and evaluated\nto enhance biometric systems. This work proposes an end-to-end framework to\nproduce iris morphs at the image level, creating morphs from Periocular iris\nimages. This framework considers different stages such as pair subject\nselection, segmentation, morph creation, and a new iris recognition system. In\norder to create realistic morphed images, two approaches for subject selection\nare explored: random selection and similar radius size selection. A\nvulnerability analysis and a Single Morphing Attack Detection algorithm were\nalso explored. The results show that this approach obtained very realistic\nimages that can confuse conventional iris recognition systems.\n", "date": "2024-08-24", "categories": []}
{"id": "2408.12934", "title": "WildFusion: Individual Animal Identification with Calibrated Similarity\n  Fusion", "abstract": "  We propose a new method - WildFusion - for individual identification of a\nbroad range of animal species. The method fuses deep scores (e.g.,\nMegaDescriptor or DINOv2) and local matching similarity (e.g., LoFTR and\nLightGlue) to identify individual animals. The global and local information\nfusion is facilitated by similarity score calibration. In a zero-shot setting,\nrelying on local similarity score only, WildFusion achieved mean accuracy,\nmeasured on 17 datasets, of 76.2%. This is better than the state-of-the-art\nmodel, MegaDescriptor-L, whose training set included 15 of the 17 datasets. If\na dataset-specific calibration is applied, mean accuracy increases by 2.3%\npercentage points. WildFusion, with both local and global similarity scores,\noutperforms the state-of-the-art significantly - mean accuracy reached 84.0%,\nan increase of 8.5 percentage points; the mean relative error drops by 35%. We\nmake the code and pre-trained models publicly available5, enabling immediate\nuse in ecology and conservation.\n", "date": "2024-08-23", "categories": []}
{"id": "2408.13251", "title": "Re-evaluation of Face Anti-spoofing Algorithm in Post COVID-19 Era Using\n  Mask Based Occlusion Attack", "abstract": "  Face anti-spoofing algorithms play a pivotal role in the robust deployment of\nface recognition systems against presentation attacks. Conventionally, full\nfacial images are required by such systems to correctly authenticate\nindividuals, but the widespread requirement of masks due to the current\nCOVID-19 pandemic has introduced new challenges for these biometric\nauthentication systems. Hence, in this work, we investigate the performance of\npresentation attack detection (PAD) algorithms under synthetic facial\nocclusions using masks and glasses. We have used five variants of masks to\ncover the lower part of the face with varying coverage areas (low-coverage,\nmedium-coverage, high-coverage, round coverage), and 3D cues. We have also used\ndifferent variants of glasses that cover the upper part of the face. We\nsystematically tested the performance of four PAD algorithms under these\nocclusion attacks using a benchmark dataset. We have specifically looked at\nfour different baseline PAD algorithms that focus on, texture, image quality,\nframe difference/motion, and abstract features through a convolutional neural\nnetwork (CNN). Additionally we have introduced a new hybrid model that uses CNN\nand local binary pattern textures. Our experiment shows that adding the\nocclusions significantly degrades the performance of all of the PAD algorithms.\nOur results show the vulnerability of face anti-spoofing algorithms with\nocclusions, which could be in the usage of such algorithms in the post-pandemic\nera.\n", "date": "2024-08-23", "categories": []}
{"id": "2408.11392", "title": "Fairness measures for biometric quality assessment", "abstract": "  Quality assessment algorithms measure the quality of a captured biometric\nsample. Since the sample quality strongly affects the recognition performance\nof a biometric system, it is essential to only process samples of sufficient\nquality and discard samples of low-quality. Even though quality assessment\nalgorithms are not intended to yield very different quality scores across\ndemographic groups, quality score discrepancies are possible, resulting in\ndifferent discard ratios. To ensure that quality assessment algorithms do not\ntake demographic characteristics into account when assessing sample quality and\nconsequently to ensure that the quality algorithms perform equally for all\nindividuals, it is crucial to develop a fairness measure. In this work we\npropose and compare multiple fairness measures for evaluating quality\ncomponents across demographic groups. Proposed measures, could be used as\npotential candidates for an upcoming standard in this important field.\n", "date": "2024-08-21", "categories": []}
{"id": "2408.05398", "title": "PersonViT: Large-scale Self-supervised Vision Transformer for Person\n  Re-Identification", "abstract": "  Person Re-Identification (ReID) aims to retrieve relevant individuals in\nnon-overlapping camera images and has a wide range of applications in the field\nof public safety. In recent years, with the development of Vision Transformer\n(ViT) and self-supervised learning techniques, the performance of person ReID\nbased on self-supervised pre-training has been greatly improved. Person ReID\nrequires extracting highly discriminative local fine-grained features of the\nhuman body, while traditional ViT is good at extracting context-related global\nfeatures, making it difficult to focus on local human body features. To this\nend, this article introduces the recently emerged Masked Image Modeling (MIM)\nself-supervised learning method into person ReID, and effectively extracts\nhigh-quality global and local features through large-scale unsupervised\npre-training by combining masked image modeling and discriminative contrastive\nlearning, and then conducts supervised fine-tuning training in the person ReID\ntask. This person feature extraction method based on ViT with masked image\nmodeling (PersonViT) has the good characteristics of unsupervised, scalable,\nand strong generalization capabilities, overcoming the problem of difficult\nannotation in supervised person ReID, and achieves state-of-the-art results on\npublicly available benchmark datasets, including MSMT17, Market1501,\nDukeMTMC-reID, and Occluded-Duke. The code and pre-trained models of the\nPersonViT method are released at \\url{https://github.com/hustvl/PersonViT} to\npromote further research in the person ReID field.\n", "date": "2024-08-10", "categories": []}
{"id": "2408.10228", "title": "ECG Unveiled: Analysis of Client Re-identification Risks in Real-World\n  ECG Datasets", "abstract": "  While ECG data is crucial for diagnosing and monitoring heart conditions, it\nalso contains unique biometric information that poses significant privacy\nrisks. Existing ECG re-identification studies rely on exhaustive analysis of\nnumerous deep learning features, confining to ad-hoc explainability towards\nclinicians decision making. In this work, we delve into explainability of ECG\nre-identification risks using transparent machine learning models. We use\nSHapley Additive exPlanations (SHAP) analysis to identify and explain the key\nfeatures contributing to re-identification risks. We conduct an empirical\nanalysis of identity re-identification risks using ECG data from five diverse\nreal-world datasets, encompassing 223 participants. By employing transparent\nmachine learning models, we reveal the diversity among different ECG features\nin contributing towards re-identification of individuals with an accuracy of\n0.76 for gender, 0.67 for age group, and 0.82 for participant ID\nre-identification. Our approach provides valuable insights for clinical experts\nand guides the development of effective privacy-preserving mechanisms. Further,\nour findings emphasize the necessity for robust privacy measures in real-world\nhealth applications and offer detailed, actionable insights for enhancing data\nanonymization techniques.\n", "date": "2024-08-02", "categories": []}
{"id": "2408.10624", "title": "WRIM-Net: Wide-Ranging Information Mining Network for Visible-Infrared\n  Person Re-Identification", "abstract": "  For the visible-infrared person re-identification (VI-ReID) task, one of the\nprimary challenges lies in significant cross-modality discrepancy. Existing\nmethods struggle to conduct modality-invariant information mining. They often\nfocus solely on mining singular dimensions like spatial or channel, and\noverlook the extraction of specific-modality multi-dimension information. To\nfully mine modality-invariant information across a wide range, we introduce the\nWide-Ranging Information Mining Network (WRIM-Net), which mainly comprises a\nMulti-dimension Interactive Information Mining (MIIM) module and an\nAuxiliary-Information-based Contrastive Learning (AICL) approach. Empowered by\nthe proposed Global Region Interaction (GRI), MIIM comprehensively mines\nnon-local spatial and channel information through intra-dimension interaction.\nMoreover, Thanks to the low computational complexity design, separate MIIM can\nbe positioned in shallow layers, enabling the network to better mine\nspecific-modality multi-dimension information. AICL, by introducing the novel\nCross-Modality Key-Instance Contrastive (CMKIC) loss, effectively guides the\nnetwork in extracting modality-invariant information. We conduct extensive\nexperiments not only on the well-known SYSU-MM01 and RegDB datasets but also on\nthe latest large-scale cross-modality LLCM dataset. The results demonstrate\nWRIM-Net's superiority over state-of-the-art methods.\n", "date": "2024-08-20", "categories": []}
{"id": "2408.10823", "title": "Trustworthy Compression? Impact of AI-based Codecs on Biometrics for Law\n  Enforcement", "abstract": "  Image-based biometrics can aid law enforcement in various aspects, for\nexample in iris, fingerprint and soft-biometric recognition. A critical\nprecondition for recognition is the availability of sufficient biometric\ninformation in images. It is visually apparent that strong JPEG compression\nremoves such details. However, latest AI-based image compression seemingly\npreserves many image details even for very strong compression factors. Yet,\nthese perceived details are not necessarily grounded in measurements, which\nraises the question whether these images can still be used for biometric\nrecognition. In this work, we investigate how AI compression impacts iris,\nfingerprint and soft-biometric (fabrics and tattoo) images. We also investigate\nthe recognition performance for iris and fingerprint images after AI\ncompression. It turns out that iris recognition can be strongly affected, while\nfingerprint recognition is quite robust. The loss of detail is qualitatively\nbest seen in fabrics and tattoos images. Overall, our results show that\nAI-compression still permits many biometric tasks, but attention to strong\ncompression factors in sensitive tasks is advisable.\n", "date": "2024-08-20", "categories": []}
{"id": "2408.08918", "title": "Supervised and Unsupervised Alignments for Spoofing Behavioral\n  Biometrics", "abstract": "  Biometric recognition systems are security systems based on intrinsic\nproperties of their users, usually encoded in high dimension representations\ncalled embeddings, which potential theft would represent a greater threat than\na temporary password or a replaceable key. To study the threat of embedding\ntheft, we perform spoofing attacks on two behavioral biometric systems (an\nautomatic speaker verification system and a handwritten digit analysis system)\nusing a set of alignment techniques. Biometric recognition systems based on\nembeddings work in two phases: enrollment - where embeddings are collected and\nstored - then authentication - when new embeddings are compared to the stored\nones -.The threat of stolen enrollment embeddings has been explored by the\ntemplate reconstruction attack literature: reconstructing the original data to\nspoof an authentication system is doable with black-box access to their\nencoder. In this document, we explore the options available to perform template\nreconstruction attacks without any access to the encoder. To perform those\nattacks, we suppose general rules over the distribution of embeddings across\nencoders and use supervised and unsupervised algorithms to align an unlabeled\nset of embeddings with a set from a known encoder. The use of an alignment\nalgorithm from the unsupervised translation literature gives promising results\non spoofing two behavioral biometric systems.\n", "date": "2024-08-14", "categories": []}
{"id": "2408.09426", "title": "A Robust Algorithm for Contactless Fingerprint Enhancement and Matching", "abstract": "  Compared to contact fingerprint images, contactless fingerprint images\nexhibit four distinct characteristics: (1) they contain less noise; (2) they\nhave fewer discontinuities in ridge patterns; (3) the ridge-valley pattern is\nless distinct; and (4) they pose an interoperability problem, as they lack the\nelastic deformation caused by pressing the finger against the capture device.\nThese properties present significant challenges for the enhancement of\ncontactless fingerprint images. In this study, we propose a novel contactless\nfingerprint identification solution that enhances the accuracy of minutiae\ndetection through improved frequency estimation and a new region-quality-based\nminutia extraction algorithm. In addition, we introduce an efficient and highly\naccurate minutiae-based encoding and matching algorithm. We validate the\neffectiveness of our approach through extensive experimental testing. Our\nmethod achieves a minimum Equal Error Rate (EER) of 2.84\\% on the PolyU\ncontactless fingerprint dataset, demonstrating its superior performance\ncompared to existing state-of-the-art techniques. The proposed fingerprint\nidentification method exhibits notable precision and resilience, proving to be\nan effective and feasible solution for contactless fingerprint-based\nidentification systems.\n", "date": "2024-08-18", "categories": []}
{"id": "2408.09676", "title": "Image-based Freeform Handwriting Authentication with Energy-oriented\n  Self-Supervised Learning", "abstract": "  Freeform handwriting authentication verifies a person's identity from their\nwriting style and habits in messy handwriting data. This technique has gained\nwidespread attention in recent years as a valuable tool for various fields,\ne.g., fraud prevention and cultural heritage protection. However, it still\nremains a challenging task in reality due to three reasons: (i) severe damage,\n(ii) complex high-dimensional features, and (iii) lack of supervision. To\naddress these issues, we propose SherlockNet, an energy-oriented two-branch\ncontrastive self-supervised learning framework for robust and fast freeform\nhandwriting authentication. It consists of four stages: (i) pre-processing:\nconverting manuscripts into energy distributions using a novel plug-and-play\nenergy-oriented operator to eliminate the influence of noise; (ii) generalized\npre-training: learning general representation through two-branch momentum-based\nadaptive contrastive learning with the energy distributions, which handles the\nhigh-dimensional features and spatial dependencies of handwriting; (iii)\npersonalized fine-tuning: calibrating the learned knowledge using a small\namount of labeled data from downstream tasks; and (iv) practical application:\nidentifying individual handwriting from scrambled, missing, or forged data\nefficiently and conveniently. Considering the practicality, we construct EN-HA,\na novel dataset that simulates data forgery and severe damage in real\napplications. Finally, we conduct extensive experiments on six benchmark\ndatasets including our EN-HA, and the results prove the robustness and\nefficiency of SherlockNet.\n", "date": "2024-08-19", "categories": []}
{"id": "2408.07883", "title": "To Impute or Not: Recommendations for Multibiometric Fusion", "abstract": "  Combining match scores from different biometric systems via fusion is a\nwell-established approach to improving recognition accuracy. However, missing\nscores can degrade performance as well as limit the possible fusion techniques\nthat can be applied. Imputation is a promising technique in multibiometric\nsystems for replacing missing data. In this paper, we evaluate various score\nimputation approaches on three multimodal biometric score datasets, viz. NIST\nBSSR1, BIOCOP2008, and MIT LL Trimodal, and investigate the factors which might\ninfluence the effectiveness of imputation. Our studies reveal three key\nobservations: (1) Imputation is preferable over not imputing missing scores,\neven when the fusion rule does not require complete score data. (2) Balancing\nthe classes in the training data is crucial to mitigate negative biases in the\nimputation technique towards the under-represented class, even if it involves\ndropping a substantial number of score vectors. (3) Multivariate imputation\napproaches seem to be beneficial when scores between modalities are correlated,\nwhile univariate approaches seem to benefit scenarios where scores between\nmodalities are less correlated.\n", "date": "2024-08-15", "categories": []}
{"id": "2408.07225", "title": "Longitudinal Evaluation of Child Face Recognition and the Impact of\n  Underlying Age", "abstract": "  The need for reliable identification of children in various emerging\napplications has sparked interest in leveraging child face recognition\ntechnology. This study introduces a longitudinal approach to enrollment and\nverification accuracy for child face recognition, focusing on the YFA database\ncollected by Clarkson University CITeR research group over an 8 year period, at\n6 month intervals.\n", "date": "2024-08-01", "categories": []}
{"id": "2408.07642", "title": "Boosting Unconstrained Face Recognition with Targeted Style Adversary", "abstract": "  While deep face recognition models have demonstrated remarkable performance,\nthey often struggle on the inputs from domains beyond their training data.\nRecent attempts aim to expand the training set by relying on computationally\nexpensive and inherently challenging image-space augmentation of image\ngeneration modules. In an orthogonal direction, we present a simple yet\neffective method to expand the training data by interpolating between\ninstance-level feature statistics across labeled and unlabeled sets. Our\nmethod, dubbed Targeted Style Adversary (TSA), is motivated by two\nobservations: (i) the input domain is reflected in feature statistics, and (ii)\nface recognition model performance is influenced by style information. Shifting\ntowards an unlabeled style implicitly synthesizes challenging training\ninstances. We devise a recognizability metric to constraint our framework to\npreserve the inherent identity-related information of labeled instances. The\nefficacy of our method is demonstrated through evaluations on unconstrained\nbenchmarks, outperforming or being on par with its competitors while offering\nnearly a 70\\% improvement in training speed and 40\\% less memory consumption.\n", "date": "2024-08-14", "categories": []}
{"id": "2408.06806", "title": "Unmasking the Uniqueness: A Glimpse into Age-Invariant Face Recognition\n  of Indigenous African Faces", "abstract": "  The task of recognizing the age-separated faces of an individual,\nAge-Invariant Face Recognition (AIFR), has received considerable research\nefforts in Europe, America, and Asia, compared to Africa. Thus, AIFR research\nefforts have often under-represented/misrepresented the African ethnicity with\nnon-indigenous Africans. This work developed an AIFR system for indigenous\nAfrican faces to reduce the misrepresentation of African ethnicity in facial\nimage analysis research. We adopted a pre-trained deep learning model (VGGFace)\nfor AIFR on a dataset of 5,000 indigenous African faces (FAGE\\_v2) collected\nfor this study. FAGE\\_v2 was curated via Internet image searches of 500\nindividuals evenly distributed across 10 African countries. VGGFace was trained\non FAGE\\_v2 to obtain the best accuracy of 81.80\\%. We also performed\nexperiments on an African-American subset of the CACD dataset and obtained the\nbest accuracy of 91.5\\%. The results show a significant difference in the\nrecognition accuracies of indigenous versus non-indigenous Africans.\n", "date": "2024-08-13", "categories": []}
{"id": "2404.05205", "title": "A secure and private ensemble matcher using multi-vault obfuscated\n  templates", "abstract": "  Generative AI has revolutionized modern machine learning by providing\nunprecedented realism, diversity, and efficiency in data generation. This\ntechnology holds immense potential for biometrics, including for securing\nsensitive and personally identifiable information. Given the irrevocability of\nbiometric samples and mounting privacy concerns, biometric template security\nand secure matching are among the most sought-after features of modern\nbiometric systems. This paper proposes a novel obfuscation method using\nGenerative AI to enhance biometric template security. Our approach utilizes\nsynthetic facial images generated by a Generative Adversarial Network (GAN) as\n\"random chaff points\" within a secure vault system. Our method creates n\nsub-templates from the original template, each obfuscated with m GAN chaff\npoints. During verification, s closest vectors to the biometric query are\nretrieved from each vault and combined to generate hash values, which are then\ncompared with the stored hash value. Thus, our method safeguards user\nidentities during the training and deployment phases by employing the\nGAN-generated synthetic images. Our protocol was tested using the AT&T, GT, and\nLFW face datasets, achieving ROC areas under the curve of 0.99, 0.99, and 0.90,\nrespectively. Our results demonstrate that the proposed method can maintain\nhigh accuracy and reasonable computational complexity comparable to those\nunprotected template methods while significantly enhancing security and\nprivacy, underscoring the potential of Generative AI in developing proactive\ndefensive strategies for biometric systems.\n", "date": "2024-04-08", "categories": []}
{"id": "2408.05543", "title": "PixelFade: Privacy-preserving Person Re-identification with Noise-guided\n  Progressive Replacement", "abstract": "  Online person re-identification services face privacy breaches from potential\ndata leakage and recovery attacks, exposing cloud-stored images to malicious\nattackers and triggering public concern. The privacy protection of pedestrian\nimages is crucial. Previous privacy-preserving person re-identification methods\nare unable to resist recovery attacks and compromise accuracy. In this paper,\nwe propose an iterative method (PixelFade) to optimize pedestrian images into\nnoise-like images to resist recovery attacks. We first give an in-depth study\nof protected images from previous privacy methods, which reveal that the chaos\nof protected images can disrupt the learning of recovery models. Accordingly,\nSpecifically, we propose Noise-guided Objective Function with the feature\nconstraints of a specific authorization model, optimizing pedestrian images to\nnormal-distributed noise images while preserving their original identity\ninformation as per the authorization model. To solve the above non-convex\noptimization problem, we propose a heuristic optimization algorithm that\nalternately performs the Constraint Operation and the Partial Replacement\nOperation. This strategy not only safeguards that original pixels are replaced\nwith noises to protect privacy, but also guides the images towards an improved\noptimization direction to effectively preserve discriminative features.\nExtensive experiments demonstrate that our PixelFade outperforms previous\nmethods in resisting recovery attacks and Re-ID performance. The code is\navailable at https://github.com/iSEE-Laboratory/PixelFade.\n", "date": "2024-08-10", "categories": []}
{"id": "2408.05918", "title": "PAFormer: Part Aware Transformer for Person Re-identification", "abstract": "  Within the domain of person re-identification (ReID), partial ReID methods\nare considered mainstream, aiming to measure feature distances through\ncomparisons of body parts between samples. However, in practice, previous\nmethods often lack sufficient awareness of anatomical aspect of body parts,\nresulting in the failure to capture features of the same body parts across\ndifferent samples. To address this issue, we introduce \\textbf{Part Aware\nTransformer (PAFormer)}, a pose estimation based ReID model which can perform\nprecise part-to-part comparison. In order to inject part awareness to pose\ntokens, we introduce learnable parameters called `pose token' which estimate\nthe correlation between each body part and partial regions of the image.\nNotably, at inference phase, PAFormer operates without additional modules\nrelated to body part localization, which is commonly used in previous ReID\nmethodologies leveraging pose estimation models. Additionally, leveraging the\nenhanced awareness of body parts, PAFormer suggests the use of a learning-based\nvisibility predictor to estimate the degree of occlusion for each body part.\nAlso, we introduce a teacher forcing technique using ground truth visibility\nscores which enables PAFormer to be trained only with visible parts. A set of\nextensive experiments show that our method outperforms existing approaches on\nwell-known ReID benchmark datasets.\n", "date": "2024-08-12", "categories": []}
{"id": "2408.04868", "title": "ChatGPT Meets Iris Biometrics", "abstract": "  This study utilizes the advanced capabilities of the GPT-4 multimodal Large\nLanguage Model (LLM) to explore its potential in iris recognition - a field\nless common and more specialized than face recognition. By focusing on this\nniche yet crucial area, we investigate how well AI tools like ChatGPT can\nunderstand and analyze iris images. Through a series of meticulously designed\nexperiments employing a zero-shot learning approach, the capabilities of\nChatGPT-4 was assessed across various challenging conditions including diverse\ndatasets, presentation attacks, occlusions such as glasses, and other\nreal-world variations. The findings convey ChatGPT-4's remarkable adaptability\nand precision, revealing its proficiency in identifying distinctive iris\nfeatures, while also detecting subtle effects like makeup on iris recognition.\nA comparative analysis with Gemini Advanced - Google's AI model - highlighted\nChatGPT-4's better performance and user experience in complex iris analysis\ntasks. This research not only validates the use of LLMs for specialized\nbiometric applications but also emphasizes the importance of nuanced query\nframing and interaction design in extracting significant insights from\nbiometric data. Our findings suggest a promising path for future research and\nthe development of more adaptable, efficient, robust and interactive biometric\nsecurity solutions.\n", "date": "2024-08-09", "categories": []}
{"id": "2408.02750", "title": "Privacy-Safe Iris Presentation Attack Detection", "abstract": "  This paper proposes a framework for a privacy-safe iris presentation attack\ndetection (PAD) method, designed solely with synthetically-generated,\nidentity-leakage-free iris images. Once trained, the method is evaluated in a\nclassical way using state-of-the-art iris PAD benchmarks. We designed two\ngenerative models for the synthesis of ISO/IEC 19794-6-compliant iris images.\nThe first model synthesizes bona fide-looking samples. To avoid ``identity\nleakage,'' the generated samples that accidentally matched those used in the\nmodel's training were excluded. The second model synthesizes images of irises\nwith textured contact lenses and is conditioned by a given contact lens brand\nto have better control over textured contact lens appearance when forming the\ntraining set. Our experiments demonstrate that models trained solely on\nsynthetic data achieve a lower but still reasonable performance when compared\nto solutions trained with iris images collected from human subjects. This is\nthe first-of-its-kind attempt to use solely synthetic data to train a\nfully-functional iris PAD solution, and despite the performance gap between\nregular and the proposed methods, this study demonstrates that with the\nincreasing fidelity of generative models, creating such privacy-safe iris PAD\nmethods may be possible. The source codes and generative models trained for\nthis work are offered along with the paper.\n", "date": "2024-08-05", "categories": []}
{"id": "2408.01588", "title": "Deep Learning Approach for Ear Recognition and Longitudinal Evaluation\n  in Children", "abstract": "  Ear recognition as a biometric modality is becoming increasingly popular,\nwith promising broader application areas. While current applications involve\nadults, one of the challenges in ear recognition for children is the rapid\nstructural changes in the ear as they age. This work introduces a foundational\nlongitudinal dataset collected from children aged 4 to 14 years over a 2.5-year\nperiod and evaluates ear recognition performance in this demographic. We\npresent a deep learning based approach for ear recognition, using an ensemble\nof VGG16 and MobileNet, focusing on both adult and child datasets, with an\nemphasis on longitudinal evaluation for children.\n", "date": "2024-08-02", "categories": []}
{"id": "2405.18320", "title": "Self-Supervised Learning Based Handwriting Verification", "abstract": "  We present SSL-HV: Self-Supervised Learning approaches applied to the task of\nHandwriting Verification. This task involves determining whether a given pair\nof handwritten images originate from the same or different writer distribution.\nWe have compared the performance of multiple generative, contrastive SSL\napproaches against handcrafted feature extractors and supervised learning on\nCEDAR AND dataset. We show that ResNet based Variational Auto-Encoder (VAE)\noutperforms other generative approaches achieving 76.3% accuracy, while\nResNet-18 fine-tuned using Variance-Invariance-Covariance Regularization\n(VICReg) outperforms other contrastive approaches achieving 78% accuracy. Using\na pre-trained VAE and VICReg for the downstream task of writer verification we\nobserved a relative improvement in accuracy of 6.7% and 9% over ResNet-18\nsupervised baseline with 10% writer labels.\n", "date": "2024-05-28", "categories": []}
{"id": "2408.00210", "title": "A Prior Embedding-Driven Architecture for Long Distance Blind Iris\n  Recognition", "abstract": "  Blind iris images, which result from unknown degradation during the process\nof iris recognition at long distances, often lead to decreased iris recognition\nrates. Currently, little existing literature offers a solution to this problem.\nIn response, we propose a prior embedding-driven architecture for long distance\nblind iris recognition. We first proposed a blind iris image restoration\nnetwork called Iris-PPRGAN. To effectively restore the texture of the blind\niris, Iris-PPRGAN includes a Generative Adversarial Network (GAN) used as a\nPrior Decoder, and a DNN used as the encoder. To extract iris features more\nefficiently, we then proposed a robust iris classifier by modifying the\nbottleneck module of InsightFace, which called Insight-Iris. A low-quality\nblind iris image is first restored by Iris-PPRGAN, then the restored iris image\nundergoes recognition via Insight-Iris. Experimental results on the public\nCASIA-Iris-distance dataset demonstrate that our proposed method significantly\nsuperior results to state-of-the-art blind iris restoration methods both\nquantitatively and qualitatively, Specifically, the recognition rate for\nlong-distance blind iris images reaches 90% after processing with our methods,\nrepresenting an improvement of approximately ten percentage points compared to\nimages without restoration.\n", "date": "2024-08-01", "categories": []}
{"id": "2407.21566", "title": "TRGR: Transmissive RIS-aided Gait Recognition Through Walls", "abstract": "  Gait recognition with radio frequency (RF) signals enables many potential\napplications requiring accurate identification. However, current systems\nrequire individuals to be within a line-of-sight (LOS) environment and struggle\nwith low signal-to-noise ratio (SNR) when signals traverse concrete and thick\nwalls. To address these challenges, we present TRGR, a novel transmissive\nreconfigurable intelligent surface (RIS)-aided gait recognition system. TRGR\ncan recognize human identities through walls using only the magnitude\nmeasurements of channel state information (CSI) from a pair of transceivers.\nSpecifically, by leveraging transmissive RIS alongside a configuration\nalternating optimization algorithm, TRGR enhances wall penetration and signal\nquality, enabling accurate gait recognition. Furthermore, a residual\nconvolution network (RCNN) is proposed as the backbone network to learn robust\nhuman information. Experimental results confirm the efficacy of transmissive\nRIS, highlighting the significant potential of transmissive RIS in enhancing\nRF-based gait recognition systems. Extensive experiment results show that TRGR\nachieves an average accuracy of 97.88\\% in identifying persons when signals\ntraverse concrete walls, demonstrating the effectiveness and robustness of\nTRGR.\n", "date": "2024-07-31", "categories": []}
{"id": "2307.13717", "title": "Exploit the Leak: Understanding Risks in Biometric Matchers", "abstract": "  In a biometric authentication or identification system, the matcher compares\na stored and a fresh template to determine whether there is a match. This\nassessment is based on both a similarity score and a predefined threshold. For\nbetter compliance with privacy legislation, the matcher can be built upon a\nprivacy-preserving distance. Beyond the binary output (`yes' or `no'), most\nschemes may perform more precise computations, e.g., the value of the distance.\nSuch precise information is prone to leakage even when not returned by the\nsystem. This can occur due to a malware infection or the use of a weakly\nprivacy-preserving distance, exemplified by side channel attacks or partially\nobfuscated designs. This paper provides an analysis of information leakage\nduring distance evaluation. We provide a catalog of information leakage\nscenarios with their impacts on data privacy. Each scenario gives rise to\nunique attacks with impacts quantified in terms of computational costs, thereby\nproviding a better understanding of the security level.\n", "date": "2023-07-25", "categories": []}
{"id": "2407.15810", "title": "Breaking the Global North Stereotype: A Global South-centric Benchmark\n  Dataset for Auditing and Mitigating Biases in Facial Recognition Systems", "abstract": "  Facial Recognition Systems (FRSs) are being developed and deployed globally\nat unprecedented rates. Most platforms are designed in a limited set of\ncountries but deployed in worldwide, without adequate checkpoints. This is\nespecially problematic for Global South countries which lack strong legislation\nto safeguard persons facing disparate performance of these systems. A\ncombination of unavailability of datasets, lack of understanding of FRS\nfunctionality and low-resource bias mitigation measures accentuate the problem.\nIn this work, we propose a new face dataset composed of 6,579 unique male and\nfemale sportspersons from eight countries around the world. More than 50% of\nthe dataset comprises individuals from the Global South countries and is\ndemographically diverse. To aid adversarial audits and robust model training,\neach image has four adversarial variants, totaling over 40,000 images. We also\nbenchmark five popular FRSs, both commercial and open-source, for the task of\ngender prediction (and country prediction for one of the open-source models as\nan example of red-teaming). Experiments on industrial FRSs reveal accuracies\nranging from 98.2%--38.1%, with a large disparity between males and females in\nthe Global South (max difference of 38.5%). Biases are also observed in all\nFRSs between females of the Global North and South (max difference of ~50%).\nGrad-CAM analysis identifies the nose, forehead and mouth as the regions of\ninterest on one of the open-source FRSs. Utilizing this insight, we design\nsimple, low-resource bias mitigation solutions using few-shot and novel\ncontrastive learning techniques significantly improving the accuracy with\ndisparity between males and females reducing from 50% to 1.5% in one of the\nsettings. In the red-teaming experiment with the open-source Deepface model,\ncontrastive learning proves more effective than simple fine-tuning.\n", "date": "2024-07-22", "categories": []}
{"id": "2211.03679", "title": "Body Part-Based Representation Learning for Occluded Person\n  Re-Identification", "abstract": "  Occluded person re-identification (ReID) is a person retrieval task which\naims at matching occluded person images with holistic ones. For addressing\noccluded ReID, part-based methods have been shown beneficial as they offer\nfine-grained information and are well suited to represent partially visible\nhuman bodies. However, training a part-based model is a challenging task for\ntwo reasons. Firstly, individual body part appearance is not as discriminative\nas global appearance (two distinct IDs might have the same local appearance),\nthis means standard ReID training objectives using identity labels are not\nadapted to local feature learning. Secondly, ReID datasets are not provided\nwith human topographical annotations. In this work, we propose BPBreID, a body\npart-based ReID model for solving the above issues. We first design two modules\nfor predicting body part attention maps and producing body part-based features\nof the ReID target. We then propose GiLt, a novel training scheme for learning\npart-based representations that is robust to occlusions and non-discriminative\nlocal appearance. Extensive experiments on popular holistic and occluded\ndatasets show the effectiveness of our proposed method, which outperforms\nstate-of-the-art methods by 0.7% mAP and 5.6% rank-1 accuracy on the\nchallenging Occluded-Duke dataset. Our code is available at\nhttps://github.com/VlSomers/bpbreid.\n", "date": "2022-11-07", "categories": []}
{"id": "2407.16289", "title": "Federated Learning for Face Recognition via Intra-subject\n  Self-supervised Learning", "abstract": "  Federated Learning (FL) for face recognition aggregates locally optimized\nmodels from individual clients to construct a generalized face recognition\nmodel. However, previous studies present two major challenges: insufficient\nincorporation of self-supervised learning and the necessity for clients to\naccommodate multiple subjects. To tackle these limitations, we propose FedFS\n(Federated Learning for personalized Face recognition via intra-subject\nSelf-supervised learning framework), a novel federated learning architecture\ntailored to train personalized face recognition models without imposing\nsubjects. Our proposed FedFS comprises two crucial components that leverage\naggregated features of the local and global models to cooperate with\nrepresentations of an off-the-shelf model. These components are (1) adaptive\nsoft label construction, utilizing dot product operations to reformat labels\nwithin intra-instances, and (2) intra-subject self-supervised learning,\nemploying cosine similarity operations to strengthen robust intra-subject\nrepresentations. Additionally, we introduce a regularization loss to prevent\noverfitting and ensure the stability of the optimized model. To assess the\neffectiveness of FedFS, we conduct comprehensive experiments on the DigiFace-1M\nand VGGFace datasets, demonstrating superior performance compared to previous\nmethods.\n", "date": "2024-07-23", "categories": []}
{"id": "2402.16399", "title": "Temporal Persistence and Intercorrelation of Embeddings Learned by an\n  End-to-End Deep Learning Eye Movement-driven Biometrics Pipeline", "abstract": "  What qualities make a feature useful for biometric performance? In prior\nresearch, pre-dating the advent of deep learning (DL) approaches to biometric\nanalysis, a strong relationship between temporal persistence, as indexed by the\nintraclass correlation coefficient (ICC), and biometric performance (Equal\nError Rate, EER) was noted. More generally, the claim was made that good\nbiometric performance resulted from a relatively large set of weakly\nintercorrelated features with high ICC. The present study aimed to determine\nwhether the same relationships are found in a state-of-the-art DL-based eye\nmovement biometric system (``Eye-Know-You-Too''), as applied to two publicly\navailable eye movement datasets. To this end, we manipulate various aspects of\neye-tracking signal quality, which produces variation in biometric performance,\nand relate that performance to the temporal persistence and intercorrelation of\nthe resulting embeddings. Data quality indices were related to EER with either\nlinear or logarithmic fits, and the resulting model R^2 was noted. As a general\nmatter, we found that temporal persistence was an important predictor of\nDL-based biometric performance, and also that DL-learned embeddings were\ngenerally weakly intercorrelated.\n", "date": "2024-02-26", "categories": []}
{"id": "2405.12713", "title": "Dynamic Identity-Guided Attention Network for Visible-Infrared Person\n  Re-identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to match people with\nthe same identity between visible and infrared modalities. VI-ReID is a\nchallenging task due to the large differences in individual appearance under\ndifferent modalities. Existing methods generally try to bridge the cross-modal\ndifferences at image or feature level, which lacks exploring the discriminative\nembeddings. Effectively minimizing these cross-modal discrepancies relies on\nobtaining representations that are guided by identity and consistent across\nmodalities, while also filtering out representations that are irrelevant to\nidentity. To address these challenges, we introduce a dynamic identity-guided\nattention network (DIAN) to mine identity-guided and modality-consistent\nembeddings, facilitating effective bridging the gap between different\nmodalities. Specifically, in DIAN, to pursue a semantically richer\nrepresentation, we first use orthogonal projection to fuse the features from\ntwo connected coarse and fine layers. Furthermore, we first use dynamic\nconvolution kernels to mine identity-guided and modality-consistent\nrepresentations. More notably, a cross embedding balancing loss is introduced\nto effectively bridge cross-modal discrepancies by above embeddings.\nExperimental results on SYSU-MM01 and RegDB datasets show that DIAN achieves\nstate-of-the-art performance. Specifically, for indoor search on SYSU-MM01, our\nmethod achieves 86.28% rank-1 accuracy and 87.41% mAP, respectively. Our code\nwill be available soon.\n", "date": "2024-05-21", "categories": []}
{"id": "2407.14087", "title": "Score Normalization for Demographic Fairness in Face Recognition", "abstract": "  Fair biometric algorithms have similar verification performance across\ndifferent demographic groups given a single decision threshold. Unfortunately,\nfor state-of-the-art face recognition networks, score distributions differ\nbetween demographics. Contrary to work that tries to align those distributions\nby extra training or fine-tuning, we solely focus on score post-processing\nmethods. As proved, well-known sample-centered score normalization techniques,\nZ-norm and T-norm, do not improve fairness for high-security operating points.\nThus, we extend the standard Z/T-norm to integrate demographic information in\nnormalization. Additionally, we investigate several possibilities to\nincorporate cohort similarities for both genuine and impostor pairs per\ndemographic to improve fairness across different operating points. We run\nexperiments on two datasets with different demographics (gender and ethnicity)\nand show that our techniques generally improve the overall fairness of five\nstate-of-the-art pre-trained face recognition networks, without downgrading\nverification performance. We also indicate that an equal contribution of False\nMatch Rate (FMR) and False Non-Match Rate (FNMR) in fairness evaluation is\nrequired for the highest gains. Code and protocols are available.\n", "date": "2024-07-19", "categories": []}
{"id": "2407.14812", "title": "GaitMA: Pose-guided Multi-modal Feature Fusion for Gait Recognition", "abstract": "  Gait recognition is a biometric technology that recognizes the identity of\nhumans through their walking patterns. Existing appearance-based methods\nutilize CNN or Transformer to extract spatial and temporal features from\nsilhouettes, while model-based methods employ GCN to focus on the special\ntopological structure of skeleton points. However, the quality of silhouettes\nis limited by complex occlusions, and skeletons lack dense semantic features of\nthe human body. To tackle these problems, we propose a novel gait recognition\nframework, dubbed Gait Multi-model Aggregation Network (GaitMA), which\neffectively combines two modalities to obtain a more robust and comprehensive\ngait representation for recognition. First, skeletons are represented by\njoint/limb-based heatmaps, and features from silhouettes and skeletons are\nrespectively extracted using two CNN-based feature extractors. Second, a\nco-attention alignment module is proposed to align the features by element-wise\nattention. Finally, we propose a mutual learning module, which achieves feature\nfusion through cross-attention, Wasserstein loss is further introduced to\nensure the effective fusion of two modalities. Extensive experimental results\ndemonstrate the superiority of our model on Gait3D, OU-MVLP, and CASIA-B.\n", "date": "2024-07-20", "categories": []}
{"id": "2407.14972", "title": "ARoFace: Alignment Robustness to Improve Low-Quality Face Recognition", "abstract": "  Aiming to enhance Face Recognition (FR) on Low-Quality (LQ) inputs, recent\nstudies suggest incorporating synthetic LQ samples into training. Although\npromising, the quality factors that are considered in these works are general\nrather than FR-specific, \\eg, atmospheric turbulence, resolution, \\etc.\n  Motivated by the observation of the vulnerability of current FR models to\neven small Face Alignment Errors (FAE) in LQ images, we present a simple yet\neffective method that considers FAE as another quality factor that is tailored\nto FR. We seek to improve LQ FR by enhancing FR models' robustness to FAE. To\nthis aim, we formalize the problem as a combination of differentiable spatial\ntransformations and adversarial data augmentation in FR. We perturb the\nalignment of the training samples using a controllable spatial transformation\nand enrich the training with samples expressing FAE.\n  We demonstrate the benefits of the proposed method by conducting evaluations\non IJB-B, IJB-C, IJB-S (+4.3\\% Rank1), and TinyFace (+2.63\\%).\n\\href{https://github.com/msed-Ebrahimi/ARoFace}{https://github.com/msed-Ebrahimi/ARoFace}\n", "date": "2024-07-20", "categories": []}
{"id": "2312.00987", "title": "Deep Generative Attacks and Countermeasures for Data-Driven Offline\n  Signature Verification", "abstract": "  This study investigates the vulnerabilities of data-driven offline signature\nverification (DASV) systems to generative attacks and proposes robust\ncountermeasures. Specifically, we explore the efficacy of Variational\nAutoencoders (VAEs) and Conditional Generative Adversarial Networks (CGANs) in\ncreating deceptive signatures that challenge DASV systems. Using the Structural\nSimilarity Index (SSIM) to evaluate the quality of forged signatures, we assess\ntheir impact on DASV systems built with Xception, ResNet152V2, and DenseNet201\narchitectures. Initial results showed False Accept Rates (FARs) ranging from 0%\nto 5.47% across all models and datasets. However, exposure to synthetic\nsignatures significantly increased FARs, with rates ranging from 19.12% to\n61.64%. The proposed countermeasure, i.e., retraining the models with real +\nsynthetic datasets, was very effective, reducing FARs between 0% and 0.99%.\nThese findings emphasize the necessity of investigating vulnerabilities in\nsecurity systems like DASV and reinforce the role of generative methods in\nenhancing the security of data-driven systems.\n", "date": "2023-12-02", "categories": []}
{"id": "2407.12589", "title": "Privacy-Preserving Adaptive Re-Identification without Image Transfer", "abstract": "  Re-Identification systems (Re-ID) are crucial for public safety but face the\nchallenge of having to adapt to environments that differ from their training\ndistribution. Furthermore, rigorous privacy protocols in public places are\nbeing enforced as apprehensions regarding individual freedom rise, adding\nlayers of complexity to the deployment of accurate Re-ID systems in new\nenvironments. For example, in the European Union, the principles of ``Data\nMinimization'' and ``Purpose Limitation'' restrict the retention and processing\nof images to what is strictly necessary. These regulations pose a challenge to\nthe conventional Re-ID training schemes that rely on centralizing data on\nservers. In this work, we present a novel setting for privacy-preserving\nDistributed Unsupervised Domain Adaptation for person Re-ID (DUDA-Rid) to\naddress the problem of domain shift without requiring any image transfer\noutside the camera devices. To address this setting, we introduce Fed-Protoid,\na novel solution that adapts person Re-ID models directly within the edge\ndevices. Our proposed solution employs prototypes derived from the source\ndomain to align feature statistics within edge devices. Those source prototypes\nare distributed across the edge devices to minimize a distributed Maximum Mean\nDiscrepancy (MMD) loss tailored for the DUDA-Rid setting. Our experiments\nprovide compelling evidence that Fed-Protoid outperforms all evaluated methods\nin terms of both accuracy and communication efficiency, all while maintaining\ndata privacy.\n", "date": "2024-07-17", "categories": []}
{"id": "2407.11141", "title": "UFQA: Utility guided Fingerphoto Quality Assessment", "abstract": "  Quality assessment of fingerprints captured using digital cameras and\nsmartphones, also called fingerphotos, is a challenging problem in biometric\nrecognition systems. As contactless biometric modalities are gaining more\nattention, their reliability should also be improved. Many factors, such as\nillumination, image contrast, camera angle, etc., in fingerphoto acquisition\nintroduce various types of distortion that may render the samples useless.\nCurrent quality estimation methods developed for fingerprints collected using\ncontact-based sensors are inadequate for fingerphotos. We propose Utility\nguided Fingerphoto Quality Assessment (UFQA), a self-supervised dual encoder\nframework to learn meaningful feature representations to assess fingerphoto\nquality. A quality prediction model is trained to assess fingerphoto quality\nwith additional supervision of quality maps. The quality metric is a predictor\nof the utility of fingerphotos in matching scenarios. Therefore, we use a\nholistic approach by including fingerphoto utility and local quality when\nlabeling the training data. Experimental results verify that our approach\nperforms better than the widely used fingerprint quality metric NFIQ2.2 and\nstate-of-the-art image quality assessment algorithms on multiple publicly\navailable fingerphoto datasets.\n", "date": "2024-07-15", "categories": []}
{"id": "2407.10694", "title": "Features Reconstruction Disentanglement Cloth-Changing Person\n  Re-Identification", "abstract": "  Cloth-changing person re-identification (CC-ReID) aims to retrieve specific\npedestrians in a cloth-changing scenario. Its main challenge is to disentangle\nthe clothing-related and clothing-unrelated features. Most existing approaches\nforce the model to learn clothing-unrelated features by changing the color of\nthe clothes. However, due to the lack of ground truth, these methods inevitably\nintroduce noise, which destroys the discriminative features and leads to an\nuncontrollable disentanglement process. In this paper, we propose a new person\nre-identification network called features reconstruction disentanglement ReID\n(FRD-ReID), which can controllably decouple the clothing-unrelated and\nclothing-related features. Specifically, we first introduce the human parsing\nmask as the ground truth of the reconstruction process. At the same time, we\npropose the far away attention (FAA) mechanism and the person contour attention\n(PCA) mechanism for clothing-unrelated features and pedestrian contour features\nto improve the feature reconstruction efficiency. In the testing phase, we\ndirectly discard the clothing-related features for inference,which leads to a\ncontrollable disentanglement process. We conducted extensive experiments on the\nPRCC, LTCC, and Vc-Clothes datasets and demonstrated that our method\noutperforms existing state-of-the-art methods.\n", "date": "2024-07-15", "categories": []}
{"id": "2404.11798", "title": "Establishing a Baseline for Gaze-driven Authentication Performance in\n  VR: A Breadth-First Investigation on a Very Large Dataset", "abstract": "  This paper performs the crucial work of establishing a baseline for\ngaze-driven authentication performance to begin answering fundamental research\nquestions using a very large dataset of gaze recordings from 9202 people with a\nlevel of eye tracking (ET) signal quality equivalent to modern consumer-facing\nvirtual reality (VR) platforms. The size of the employed dataset is at least an\norder-of-magnitude larger than any other dataset from previous related work.\nBinocular estimates of the optical and visual axes of the eyes and a minimum\nduration for enrollment and verification are required for our model to achieve\na false rejection rate (FRR) of below 3% at a false acceptance rate (FAR) of 1\nin 50,000. In terms of identification accuracy which decreases with gallery\nsize, we estimate that our model would fall below chance-level accuracy for\ngallery sizes of 148,000 or more. Our major findings indicate that gaze\nauthentication can be as accurate as required by the FIDO standard when driven\nby a state-of-the-art machine learning architecture and a sufficiently large\ntraining dataset.\n", "date": "2024-04-17", "categories": []}
{"id": "2407.08341", "title": "Adaptive Deep Iris Feature Extractor at Arbitrary Resolutions", "abstract": "  This paper proposes a deep feature extractor for iris recognition at\narbitrary resolutions. Resolution degradation reduces the recognition\nperformance of deep learning models trained by high-resolution images. Using\nvarious-resolution images for training can improve the model's robustness while\nsacrificing recognition performance for high-resolution images. To achieve\nhigher recognition performance at various resolutions, we propose a method of\nresolution-adaptive feature extraction with automatically switching networks.\nOur framework includes resolution expert modules specialized for different\nresolution degradations, including down-sampling and out-of-focus blurring. The\nframework automatically switches them depending on the degradation condition of\nan input image. Lower-resolution experts are trained by knowledge-distillation\nfrom the high-resolution expert in such a manner that both experts can extract\ncommon identity features. We applied our framework to three conventional neural\nnetwork models. The experimental results show that our method enhances the\nrecognition performance at low-resolution in the conventional methods and also\nmaintains their performance at high-resolution.\n", "date": "2024-07-11", "categories": []}
{"id": "2407.09045", "title": "Time-Frequency Analysis of Variable-Length WiFi CSI Signals for Person\n  Re-Identification", "abstract": "  Person re-identification (ReID), as a crucial technology in the field of\nsecurity, plays an important role in security detection and people counting.\nCurrent security and monitoring systems largely rely on visual information,\nwhich may infringe on personal privacy and be susceptible to interference from\npedestrian appearances and clothing in certain scenarios. Meanwhile, the\nwidespread use of routers offers new possibilities for ReID. This letter\nintroduces a method using WiFi Channel State Information (CSI), leveraging the\nmultipath propagation characteristics of WiFi signals as a basis for\ndistinguishing different pedestrian features. We propose a two-stream network\nstructure capable of processing variable-length data, which analyzes the\namplitude in the time domain and the phase in the frequency domain of WiFi\nsignals, fuses time-frequency information through continuous lateral\nconnections, and employs advanced objective functions for representation and\nmetric learning. Tested on a dataset collected in the real world, our method\nachieves 93.68% mAP and 98.13% Rank-1.\n", "date": "2024-07-12", "categories": []}
{"id": "2407.09104", "title": "UserBoost: Generating User-specific Synthetic Data for Faster Enrolment\n  into Behavioural Biometric Systems", "abstract": "  Behavioural biometric authentication systems entail an enrolment period that\nis burdensome for the user. In this work, we explore generating synthetic\ngestures from a few real user gestures with generative deep learning, with the\napplication of training a simple (i.e. non-deep-learned) authentication model.\nSpecifically, we show that utilising synthetic data alongside real data can\nreduce the number of real datapoints a user must provide to enrol into a\nbiometric system. To validate our methods, we use the publicly available\ndataset of WatchAuth, a system proposed in 2022 for authenticating smartwatch\npayments using the physical gesture of reaching towards a payment terminal. We\ndevelop a regularised autoencoder model for generating synthetic user-specific\nwrist motion data representing these physical gestures, and demonstrate the\ndiversity and fidelity of our synthetic gestures. We show that using synthetic\ngestures in training can improve classification ability for a real-world\nsystem. Through this technique we can reduce the number of gestures required to\nenrol a user into a WatchAuth-like system by more than 40% without negatively\nimpacting its error rates.\n", "date": "2024-07-12", "categories": []}
{"id": "2407.08717", "title": "WhisperNetV2: SlowFast Siamese Network For Lip-Based Biometrics", "abstract": "  Lip-based biometric authentication (LBBA) has attracted many researchers\nduring the last decade. The lip is specifically interesting for biometric\nresearchers because it is a twin biometric with the potential to function both\nas a physiological and a behavioral trait. Although much valuable research was\nconducted on LBBA, none of them considered the different emotions of the client\nduring the video acquisition step of LBBA, which can potentially affect the\nclient's facial expressions and speech tempo. We proposed a novel network\nstructure called WhisperNetV2, which extends our previously proposed network\ncalled WhisperNet. Our proposed network leverages a deep Siamese structure with\ntriplet loss having three identical SlowFast networks as embedding networks.\nThe SlowFast network is an excellent candidate for our task since the fast\npathway extracts motion-related features (behavioral lip movements) with a high\nframe rate and low channel capacity. The slow pathway extracts visual features\n(physiological lip appearance) with a low frame rate and high channel capacity.\nUsing an open-set protocol, we trained our network using the CREMA-D dataset\nand acquired an Equal Error Rate (EER) of 0.005 on the test set. Considering\nthat the acquired EER is less than most similar LBBA methods, our method can be\nconsidered as a state-of-the-art LBBA method.\n", "date": "2024-07-11", "categories": []}
{"id": "2407.05045", "title": "Robust Skin Color Driven Privacy Preserving Face Recognition via\n  Function Secret Sharing", "abstract": "  In this work, we leverage the pure skin color patch from the face image as\nthe additional information to train an auxiliary skin color feature extractor\nand face recognition model in parallel to improve performance of\nstate-of-the-art (SOTA) privacy-preserving face recognition (PPFR) systems. Our\nsolution is robust against black-box attacking and well-established generative\nadversarial network (GAN) based image restoration. We analyze the potential\nrisk in previous work, where the proposed cosine similarity computation might\ndirectly leak the protected precomputed embedding stored on the server side. We\npropose a Function Secret Sharing (FSS) based face embedding comparison\nprotocol without any intermediate result leakage. In addition, we show in\nexperiments that the proposed protocol is more efficient compared to the Secret\nSharing (SS) based protocol.\n", "date": "2024-07-06", "categories": []}
{"id": "2405.01199", "title": "Latent Fingerprint Matching via Dense Minutia Descriptor", "abstract": "  Latent fingerprint matching is a daunting task, primarily due to the poor\nquality of latent fingerprints. In this study, we propose a deep-learning based\ndense minutia descriptor (DMD) for latent fingerprint matching. A DMD is\nobtained by extracting the fingerprint patch aligned by its central minutia,\ncapturing detailed minutia information and texture information. Our dense\ndescriptor takes the form of a three-dimensional representation, with two\ndimensions associated with the original image plane and the other dimension\nrepresenting the abstract features. Additionally, the extraction process\noutputs the fingerprint segmentation map, ensuring that the descriptor is only\nvalid in the foreground region. The matching between two descriptors occurs in\ntheir overlapping regions, with a score normalization strategy to reduce the\nimpact brought by the differences outside the valid area. Our descriptor\nachieves state-of-the-art performance on several latent fingerprint datasets.\nOverall, our DMD is more representative and interpretable compared to previous\nmethods.\n", "date": "2024-05-02", "categories": []}
{"id": "2407.02038", "title": "Camera-LiDAR Cross-modality Gait Recognition", "abstract": "  Gait recognition is a crucial biometric identification technique.\nCamera-based gait recognition has been widely applied in both research and\nindustrial fields. LiDAR-based gait recognition has also begun to evolve most\nrecently, due to the provision of 3D structural information. However, in\ncertain applications, cameras fail to recognize persons, such as in low-light\nenvironments and long-distance recognition scenarios, where LiDARs work well.\nOn the other hand, the deployment cost and complexity of LiDAR systems limit\nits wider application. Therefore, it is essential to consider cross-modality\ngait recognition between cameras and LiDARs for a broader range of\napplications. In this work, we propose the first cross-modality gait\nrecognition framework between Camera and LiDAR, namely CL-Gait. It employs a\ntwo-stream network for feature embedding of both modalities. This poses a\nchallenging recognition task due to the inherent matching between 3D and 2D\ndata, exhibiting significant modality discrepancy. To align the feature spaces\nof the two modalities, i.e., camera silhouettes and LiDAR points, we propose a\ncontrastive pre-training strategy to mitigate modality discrepancy. To make up\nfor the absence of paired camera-LiDAR data for pre-training, we also introduce\na strategy for generating data on a large scale. This strategy utilizes\nmonocular depth estimated from single RGB images and virtual cameras to\ngenerate pseudo point clouds for contrastive pre-training. Extensive\nexperiments show that the cross-modality gait recognition is very challenging\nbut still contains potential and feasibility with our proposed model and\npre-training strategy. To the best of our knowledge, this is the first work to\naddress cross-modality gait recognition.\n", "date": "2024-07-02", "categories": []}
{"id": "2309.05180", "title": "What's color got to do with it? Face recognition in grayscale", "abstract": "  State-of-the-art deep CNN face matchers are typically created using extensive\ntraining sets of color face images. Our study reveals that such matchers attain\nvirtually identical accuracy when trained on either grayscale or color versions\nof the training set, even when the evaluation is done using color test images.\nFurthermore, we demonstrate that shallower models, lacking the capacity to\nmodel complex representations, rely more heavily on low-level features such as\nthose associated with color. As a result, they display diminished accuracy when\ntrained with grayscale images. We then consider possible causes for deeper CNN\nface matchers \"not seeing color\". Popular web-scraped face datasets actually\nhave 30 to 60% of their identities with one or more grayscale images. We\nanalyze whether this grayscale element in the training set impacts the accuracy\nachieved, and conclude that it does not. We demonstrate that using only\ngrayscale images for both training and testing achieves accuracy comparable to\nthat achieved using only color images for deeper models. This holds true for\nboth real and synthetic training datasets. HSV color space, which separates\nchroma and luma information, does not improve the network's learning about\ncolor any more than in the RGB color space. We then show that the skin region\nof an individual's images in a web-scraped training set exhibits significant\nvariation in their mapping to color space. This suggests that color carries\nlimited identity-specific information. We also show that when the first\nconvolution layer is restricted to a single filter, models learn a grayscale\nconversion filter and pass a grayscale version of the input color image to the\nnext layer. Finally, we demonstrate that leveraging the lower per-image storage\nfor grayscale to increase the number of images in the training set can improve\naccuracy of the face recognition model.\n", "date": "2023-09-11", "categories": []}
{"id": "2401.05646", "title": "Masked Attribute Description Embedding for Cloth-Changing Person\n  Re-identification", "abstract": "  Cloth-changing person re-identification (CC-ReID) aims to match persons who\nchange clothes over long periods. The key challenge in CC-ReID is to extract\nclothing-independent features, such as face, hairstyle, body shape, and gait.\nCurrent research mainly focuses on modeling body shape using multi-modal\nbiological features (such as silhouettes and sketches). However, it does not\nfully leverage the personal description information hidden in the original RGB\nimage. Considering that there are certain attribute descriptions which remain\nunchanged after the changing of cloth, we propose a Masked Attribute\nDescription Embedding (MADE) method that unifies personal visual appearance and\nattribute description for CC-ReID. Specifically, handling variable\nclothing-sensitive information, such as color and type, is challenging for\neffective modeling. To address this, we mask the clothing and color information\nin the personal attribute description extracted through an attribute detection\nmodel. The masked attribute description is then connected and embedded into\nTransformer blocks at various levels, fusing it with the low-level to\nhigh-level features of the image. This approach compels the model to discard\nclothing information. Experiments are conducted on several CC-ReID benchmarks,\nincluding PRCC, LTCC, Celeb-reID-light, and LaST. Results demonstrate that MADE\neffectively utilizes attribute description, enhancing cloth-changing person\nre-identification performance, and compares favorably with state-of-the-art\nmethods. The code is available at https://github.com/moon-wh/MADE.\n", "date": "2024-01-11", "categories": []}
{"id": "2407.02150", "title": "VRBiom: A New Periocular Dataset for Biometric Applications of HMD", "abstract": "  With advancements in hardware, high-quality HMD devices are being developed\nby numerous companies, driving increased consumer interest in AR, VR, and MR\napplications. In this work, we present a new dataset, called VRBiom, of\nperiocular videos acquired using a Virtual Reality headset. The VRBiom,\ntargeted at biometric applications, consists of 900 short videos acquired from\n25 individuals recorded in the NIR spectrum. These 10s long videos have been\ncaptured using the internal tracking cameras of Meta Quest Pro at 72 FPS. To\nencompass real-world variations, the dataset includes recordings under three\ngaze conditions: steady, moving, and partially closed eyes. We have also\nensured an equal split of recordings without and with glasses to facilitate the\nanalysis of eye-wear. These videos, characterized by non-frontal views of the\neye and relatively low spatial resolutions (400 x 400), can be instrumental in\nadvancing state-of-the-art research across various biometric applications. The\nVRBiom dataset can be utilized to evaluate, train, or adapt models for\nbiometric use-cases such as iris and/or periocular recognition and associated\nsub-tasks such as detection and semantic segmentation.\n  In addition to data from real individuals, we have included around 1100 PA\nconstructed from 92 PA instruments. These PAIs fall into six categories\nconstructed through combinations of print attacks (real and synthetic\nidentities), fake 3D eyeballs, plastic eyes, and various types of masks and\nmannequins. These PA videos, combined with genuine (bona-fide) data, can be\nutilized to address concerns related to spoofing, which is a significant threat\nif these devices are to be used for authentication.\n  The VRBiom dataset is publicly available for research purposes related to\nbiometric applications only.\n", "date": "2024-07-02", "categories": []}
{"id": "2407.01332", "title": "AdaDistill: Adaptive Knowledge Distillation for Deep Face Recognition", "abstract": "  Knowledge distillation (KD) aims at improving the performance of a compact\nstudent model by distilling the knowledge from a high-performing teacher model.\nIn this paper, we present an adaptive KD approach, namely AdaDistill, for deep\nface recognition. The proposed AdaDistill embeds the KD concept into the\nsoftmax loss by training the student using a margin penalty softmax loss with\ndistilled class centers from the teacher. Being aware of the relatively low\ncapacity of the compact student model, we propose to distill less complex\nknowledge at an early stage of training and more complex one at a later stage\nof training. This relative adjustment of the distilled knowledge is controlled\nby the progression of the learning capability of the student over the training\niterations without the need to tune any hyper-parameters. Extensive experiments\nand ablation studies show that AdaDistill can enhance the discriminative\nlearning capability of the student and demonstrate superiority over various\nstate-of-the-art competitors on several challenging benchmarks, such as IJB-B,\nIJB-C, and ICCV2021-MFR\n", "date": "2024-07-01", "categories": []}
{"id": "2406.19672", "title": "Beyond First-Order: A Multi-Scale Approach to Finger Knuckle Print\n  Biometrics", "abstract": "  Recently, finger knuckle prints (FKPs) have gained attention due to their\nrich textural patterns, positioning them as a promising biometric for identity\nrecognition. Prior FKP recognition methods predominantly leverage first-order\nfeature descriptors, which capture intricate texture details but fail to\naccount for structural information. Emerging research, however, indicates that\nsecond-order textures, which describe the curves and arcs of the textures,\nencompass this overlooked structural information. This paper introduces a novel\nFKP recognition approach, the Dual-Order Texture Competition Network (DOTCNet),\ndesigned to capture texture information in FKP images comprehensively. DOTCNet\nincorporates three dual-order texture competitive modules (DTCMs), each\ntargeting textures at different scales. Each DTCM employs a learnable texture\ndescriptor, specifically a learnable Gabor filter (LGF), to extract texture\nfeatures. By leveraging LGFs, the network extracts first and second order\ntextures to describe fine textures and structural features thoroughly.\nFurthermore, an attention mechanism enhances relevant features in the\nfirst-order features, thereby highlighting significant texture details. For\nsecond-order features, a competitive mechanism emphasizes structural\ninformation while reducing noise from higher-order features. Extensive\nexperimental results reveal that DOTCNet significantly outperforms several\nstandard algorithms on the publicly available PolyU-FKP dataset.\n", "date": "2024-06-28", "categories": []}
{"id": "2403.06658", "title": "Towards Zero-Shot Interpretable Human Recognition: A 2D-3D Registration\n  Framework", "abstract": "  Large vision models based in deep learning architectures have been\nconsistently advancing the state-of-the-art in biometric recognition. However,\nthree weaknesses are commonly reported for such kind of approaches: 1) their\nextreme demands in terms of learning data; 2) the difficulties in generalising\nbetween different domains; and 3) the lack of interpretability/explainability,\nwith biometrics being of particular interest, as it is important to provide\nevidence able to be used for forensics/legal purposes (e.g., in courts). To the\nbest of our knowledge, this paper describes the first recognition\nframework/strategy that aims at addressing the three weaknesses simultaneously.\nAt first, it relies exclusively in synthetic samples for learning purposes.\nInstead of requiring a large amount and variety of samples for each subject,\nthe idea is to exclusively enroll a 3D point cloud per identity. Then, using\ngenerative strategies, we synthesize a very large (potentially infinite) number\nof samples, containing all the desired covariates (poses, clothing, distances,\nperspectives, lighting, occlusions,...). Upon the synthesizing method used, it\nis possible to adapt precisely to different kind of domains, which accounts for\ngeneralization purposes. Such data are then used to learn a model that performs\nlocal registration between image pairs, establishing positive correspondences\nbetween body parts that are the key, not only to recognition (according to\ncardinality and distribution), but also to provide an interpretable description\nof the response (e.g.: \"both samples are from the same person, as they have\nsimilar facial shape, hair color and legs thickness\").\n", "date": "2024-03-11", "categories": []}
{"id": "2406.14797", "title": "Camera-Invariant Meta-Learning Network for Single-Camera-Training Person\n  Re-identification", "abstract": "  Single-camera-training person re-identification (SCT re-ID) aims to train a\nre-ID model using SCT datasets where each person appears in only one camera.\nThe main challenge of SCT re-ID is to learn camera-invariant feature\nrepresentations without cross-camera same-person (CCSP) data as supervision.\nPrevious methods address it by assuming that the most similar person should be\nfound in another camera. However, this assumption is not guaranteed to be\ncorrect. In this paper, we propose a Camera-Invariant Meta-Learning Network\n(CIMN) for SCT re-ID. CIMN assumes that the camera-invariant feature\nrepresentations should be robust to camera changes. To this end, we split the\ntraining data into meta-train set and meta-test set based on camera IDs and\nperform a cross-camera simulation via meta-learning strategy, aiming to enforce\nthe representations learned from the meta-train set to be robust to the\nmeta-test set. With the cross-camera simulation, CIMN can learn\ncamera-invariant and identity-discriminative representations even there are no\nCCSP data. However, this simulation also causes the separation of the\nmeta-train set and the meta-test set, which ignores some beneficial relations\nbetween them. Thus, we introduce three losses: meta triplet loss, meta\nclassification loss, and meta camera alignment loss, to leverage the ignored\nrelations. The experiment results demonstrate that our method achieves\ncomparable performance with and without CCSP data, and outperforms the\nstate-of-the-art methods on SCT re-ID benchmarks. In addition, it is also\neffective in improving the domain generalization ability of the model.\n", "date": "2024-06-21", "categories": []}
{"id": "2406.15253", "title": "Fingerprint Membership and Identity Inference Against Generative\n  Adversarial Networks", "abstract": "  Generative models are gaining significant attention as potential catalysts\nfor a novel industrial revolution. Since automated sample generation can be\nuseful to solve privacy and data scarcity issues that usually affect learned\nbiometric models, such technologies became widely spread in this field. In this\npaper, we assess the vulnerabilities of generative machine learning models\nconcerning identity protection by designing and testing an identity inference\nattack on fingerprint datasets created by means of a generative adversarial\nnetwork. Experimental results show that the proposed solution proves to be\neffective under different configurations and easily extendable to other\nbiometric measurements.\n", "date": "2024-06-21", "categories": []}
{"id": "2406.13694", "title": "An Embedded Intelligent System for Attendance Monitoring", "abstract": "  In this paper, we propose an intelligent embedded system for monitoring class\nattendance and sending the attendance list to a remote computer. The proposed\nsystem consists of two parts : an embedded device (Raspberry with PI camera)\nfor facial recognition and a web application for attendance management. The\nproposed solution take into account the different challenges: the limited\nresources of the Raspberry Pi, the need to adapt the facial recognition model\nand achieving acceptable performance using images provided by the Raspberry Pi\ncamera.\n", "date": "2024-06-19", "categories": []}
{"id": "2406.14370", "title": "Enhanced Bank Check Security: Introducing a Novel Dataset and\n  Transformer-Based Approach for Detection and Verification", "abstract": "  Automated signature verification on bank checks is critical for fraud\nprevention and ensuring transaction authenticity. This task is challenging due\nto the coexistence of signatures with other textual and graphical elements on\nreal-world documents. Verification systems must first detect the signature and\nthen validate its authenticity, a dual challenge often overlooked by current\ndatasets and methodologies focusing only on verification. To address this gap,\nwe introduce a novel dataset specifically designed for signature verification\non bank checks. This dataset includes a variety of signature styles embedded\nwithin typical check elements, providing a realistic testing ground for\nadvanced detection methods. Moreover, we propose a novel approach for\nwriter-independent signature verification using an object detection network.\nOur detection-based verification method treats genuine and forged signatures as\ndistinct classes within an object detection framework, effectively handling\nboth detection and verification. We employ a DINO-based network augmented with\na dilation module to detect and verify signatures on check images\nsimultaneously. Our approach achieves an AP of 99.2 for genuine and 99.4 for\nforged signatures, a significant improvement over the DINO baseline, which\nscored 93.1 and 89.3 for genuine and forged signatures, respectively. This\nimprovement highlights our dilation module's effectiveness in reducing both\nfalse positives and negatives. Our results demonstrate substantial advancements\nin detection-based signature verification technology, offering enhanced\nsecurity and efficiency in financial document processing.\n", "date": "2024-06-20", "categories": []}
{"id": "2406.12316", "title": "Enhancing Visible-Infrared Person Re-identification with Modality- and\n  Instance-aware Visual Prompt Learning", "abstract": "  The Visible-Infrared Person Re-identification (VI ReID) aims to match visible\nand infrared images of the same pedestrians across non-overlapped camera views.\nThese two input modalities contain both invariant information, such as shape,\nand modality-specific details, such as color. An ideal model should utilize\nvaluable information from both modalities during training for enhanced\nrepresentational capability. However, the gap caused by modality-specific\ninformation poses substantial challenges for the VI ReID model to handle\ndistinct modality inputs simultaneously. To address this, we introduce the\nModality-aware and Instance-aware Visual Prompts (MIP) network in our work,\ndesigned to effectively utilize both invariant and specific information for\nidentification. Specifically, our MIP model is built on the transformer\narchitecture. In this model, we have designed a series of modality-specific\nprompts, which could enable our model to adapt to and make use of the specific\ninformation inherent in different modality inputs, thereby reducing the\ninterference caused by the modality gap and achieving better identification.\nBesides, we also employ each pedestrian feature to construct a group of\ninstance-specific prompts. These customized prompts are responsible for guiding\nour model to adapt to each pedestrian instance dynamically, thereby capturing\nidentity-level discriminative clues for identification. Through extensive\nexperiments on SYSU-MM01 and RegDB datasets, the effectiveness of both our\ndesigned modules is evaluated. Additionally, our proposed MIP performs better\nthan most state-of-the-art methods.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.12355", "title": "LiCAF: LiDAR-Camera Asymmetric Fusion for Gait Recognition", "abstract": "  Gait recognition is a biometric technology that identifies individuals by\nusing walking patterns. Due to the significant achievements of multimodal\nfusion in gait recognition, we consider employing LiDAR-camera fusion to obtain\nrobust gait representations. However, existing methods often overlook intrinsic\ncharacteristics of modalities, and lack fine-grained fusion and temporal\nmodeling. In this paper, we introduce a novel modality-sensitive network LiCAF\nfor LiDAR-camera fusion, which employs an asymmetric modeling strategy.\nSpecifically, we propose Asymmetric Cross-modal Channel Attention (ACCA) and\nInterlaced Cross-modal Temporal Modeling (ICTM) for cross-modal valuable\nchannel information selection and powerful temporal modeling. Our method\nachieves state-of-the-art performance (93.9% in Rank-1 and 98.8% in Rank-5) on\nthe SUSTech1K dataset, demonstrating its effectiveness.\n", "date": "2024-06-18", "categories": []}
{"id": "2406.09198", "title": "CLIP-Driven Cloth-Agnostic Feature Learning for Cloth-Changing Person\n  Re-Identification", "abstract": "  Contrastive Language-Image Pre-Training (CLIP) has shown impressive\nperformance in short-term Person Re-Identification (ReID) due to its ability to\nextract high-level semantic features of pedestrians, yet its direct application\nto Cloth-Changing Person Re-Identification (CC-ReID) faces challenges due to\nCLIP's image encoder overly focusing on clothes clues. To address this, we\npropose a novel framework called CLIP-Driven Cloth-Agnostic Feature Learning\n(CCAF) for CC-ReID. Accordingly, two modules were custom-designed: the\nInvariant Feature Prompting (IFP) and the Clothes Feature Minimization (CFM).\nThese modules guide the model to extract cloth-agnostic features positively and\nattenuate clothes-related features negatively. Specifically, IFP is designed to\nextract fine-grained semantic features unrelated to clothes from the raw image,\nguided by the cloth-agnostic text prompts. This module first covers the clothes\nin the raw image at the pixel level to obtain the shielding image and then\nutilizes CLIP's knowledge to generate cloth-agnostic text prompts.\nSubsequently, it aligns the raw image-text and the raw image-shielding image in\nthe feature space, emphasizing discriminative clues related to identity but\nunrelated to clothes. Furthermore, CFM is designed to examine and weaken the\nimage encoder's ability to extract clothes features. It first generates text\nprompts corresponding to clothes pixels. Then, guided by these clothes text\nprompts, it iteratively examines and disentangles clothes features from\npedestrian features, ultimately retaining inherent discriminative features.\nExtensive experiments have demonstrated the effectiveness of the proposed CCAF,\nachieving new state-of-the-art performance on several popular CC-ReID\nbenchmarks without any additional inference time.\n", "date": "2024-06-13", "categories": []}
{"id": "2301.03118", "title": "Facial Misrecognition Systems: Simple Weight Manipulations Force DNNs to\n  Err Only on Specific Persons", "abstract": "  In this paper, we describe how to plant novel types of backdoors in any\nfacial recognition model based on the popular architecture of deep Siamese\nneural networks. These backdoors force the system to err only on natural images\nof specific persons who are preselected by the attacker, without controlling\ntheir appearance or inserting any triggers. For example, we show how such a\nbackdoored system can classify any two images of a particular person as\ndifferent people, or any two images of a particular pair of persons as the same\nperson, with almost no effect on the correctness of its decisions for other\npersons. Surprisingly, we show that both types of backdoors can be implemented\nby applying linear transformations to the model's last weight matrix, with no\nadditional training or optimization, using only images of the backdoor\nidentities. A unique property of our attack is that multiple backdoors can be\nindependently installed in the same model by multiple attackers, who may not be\naware of each other's existence, with almost no interference. We have\nexperimentally verified the attacks on a SOTA facial recognition system. When\nwe tried to individually anonymize ten celebrities, the network failed to\nrecognize two of their images as being the same person in $97.02\\%$ to\n$98.31\\%$ of the time. When we tried to confuse between the extremely\ndifferent-looking Morgan Freeman and Scarlett Johansson, for example, their\nimages were declared to be the same person in $98.47 \\%$ of the time. For each\ntype of backdoor, we sequentially installed multiple backdoors with minimal\neffect on the performance of each other (for example, anonymizing all ten\ncelebrities on the same model reduced the success rate for each celebrity by no\nmore than $1.01\\%$). In all of our experiments, the benign accuracy of the\nnetwork on other persons barely degraded (in most cases, it degraded by less\nthan $0.05\\%$).\n", "date": "2023-01-08", "categories": []}
{"id": "2406.06597", "title": "1-D CNN-Based Online Signature Verification with Federated Learning", "abstract": "  Online signature verification plays a pivotal role in security\ninfrastructures. However, conventional online signature verification models\npose significant risks to data privacy, especially during training processes.\nTo mitigate these concerns, we propose a novel federated learning framework\nthat leverages 1-D Convolutional Neural Networks (CNN) for online signature\nverification. Furthermore, our experiments demonstrate the effectiveness of our\nframework regarding 1-D CNN and federated learning. Particularly, the\nexperiment results highlight that our framework 1) minimizes local\ncomputational resources; 2) enhances transfer effects with substantial\ninitialization data; 3) presents remarkable scalability. The centralized 1-D\nCNN model achieves an Equal Error Rate (EER) of 3.33% and an accuracy of\n96.25%. Meanwhile, configurations with 2, 5, and 10 agents yield EERs of 5.42%,\n5.83%, and 5.63%, along with accuracies of 95.21%, 94.17%, and 94.06%,\nrespectively.\n", "date": "2024-06-06", "categories": []}
{"id": "2401.13201", "title": "MLLMReID: Multimodal Large Language Model-based Person Re-identification", "abstract": "  Multimodal large language models (MLLM) have achieved satisfactory results in\nmany tasks. However, their performance in the task of ReID (ReID) has not been\nexplored to date. This paper will investigate how to adapt them for the task of\nReID. An intuitive idea is to fine-tune MLLM with ReID image-text datasets, and\nthen use their visual encoder as a backbone for ReID. However, there still\nexist two apparent issues: (1) Designing instructions for ReID, MLLMs may\noverfit specific instructions, and designing a variety of instructions will\nlead to higher costs. (2) When fine-tuning the visual encoder of a MLLM, it is\nnot trained synchronously with the ReID task. As a result, the effectiveness of\nthe visual encoder fine-tuning cannot be directly reflected in the performance\nof the ReID task. To address these problems, this paper proposes MLLMReID:\nMultimodal Large Language Model-based ReID. Firstly, we proposed Common\nInstruction, a simple approach that leverages the essence ability of LLMs to\ncontinue writing, avoiding complex and diverse instruction design. Secondly, we\npropose a multi-task learning-based synchronization module to ensure that the\nvisual encoder of the MLLM is trained synchronously with the ReID task. The\nexperimental results demonstrate the superiority of our method.\n", "date": "2024-01-24", "categories": []}
{"id": "2406.06045", "title": "Synthesizing Efficient Data with Diffusion Models for Person\n  Re-Identification Pre-Training", "abstract": "  Existing person re-identification (Re-ID) methods principally deploy the\nImageNet-1K dataset for model initialization, which inevitably results in\nsub-optimal situations due to the large domain gap. One of the key challenges\nis that building large-scale person Re-ID datasets is time-consuming. Some\nprevious efforts address this problem by collecting person images from the\ninternet e.g., LUPerson, but it struggles to learn from unlabeled,\nuncontrollable, and noisy data. In this paper, we present a novel paradigm\nDiffusion-ReID to efficiently augment and generate diverse images based on\nknown identities without requiring any cost of data collection and annotation.\nTechnically, this paradigm unfolds in two stages: generation and filtering.\nDuring the generation stage, we propose Language Prompts Enhancement (LPE) to\nensure the ID consistency between the input image sequence and the generated\nimages. In the diffusion process, we propose a Diversity Injection (DI) module\nto increase attribute diversity. In order to make the generated data have\nhigher quality, we apply a Re-ID confidence threshold filter to further remove\nthe low-quality images. Benefiting from our proposed paradigm, we first create\na new large-scale person Re-ID dataset Diff-Person, which consists of over 777K\nimages from 5,183 identities. Next, we build a stronger person Re-ID backbone\npre-trained on our Diff-Person. Extensive experiments are conducted on four\nperson Re-ID benchmarks in six widely used settings. Compared with other\npre-training and self-supervised competitors, our approach shows significant\nsuperiority.\n", "date": "2024-06-10", "categories": []}
{"id": "2406.04129", "title": "LenslessFace: An End-to-End Optimized Lensless System for\n  Privacy-Preserving Face Verification", "abstract": "  Lensless cameras, innovatively replacing traditional lenses for ultra-thin,\nflat optics, encode light directly onto sensors, producing images that are not\nimmediately recognizable. This compact, lightweight, and cost-effective imaging\nsolution offers inherent privacy advantages, making it attractive for\nprivacy-sensitive applications like face verification. Typical lensless face\nverification adopts a two-stage process of reconstruction followed by\nverification, incurring privacy risks from reconstructed faces and high\ncomputational costs. This paper presents an end-to-end optimization approach\nfor privacy-preserving face verification directly on encoded lensless captures,\nensuring that the entire software pipeline remains encoded with no visible\nfaces as intermediate results. To achieve this, we propose several techniques\nto address unique challenges from the lensless setup which precludes\ntraditional face detection and alignment. Specifically, we propose a face\ncenter alignment scheme, an augmentation curriculum to build robustness against\nvariations, and a knowledge distillation method to smooth optimization and\nenhance performance. Evaluations under both simulation and real environment\ndemonstrate our method outperforms two-stage lensless verification while\nenhancing privacy and efficiency. Project website:\n\\url{lenslessface.github.io}.\n", "date": "2024-06-06", "categories": []}
{"id": "2406.02142", "title": "Analyzing the Effect of Combined Degradations on Face Recognition", "abstract": "  A face recognition model is typically trained on large datasets of images\nthat may be collected from controlled environments. This results in performance\ndiscrepancies when applied to real-world scenarios due to the domain gap\nbetween clean and in-the-wild images. Therefore, some researchers have\ninvestigated the robustness of these models by analyzing synthetic\ndegradations. Yet, existing studies have mostly focused on single degradation\nfactors, which may not fully capture the complexity of real-world degradations.\nThis work addresses this problem by analyzing the impact of both single and\ncombined degradations using a real-world degradation pipeline extended with\nunder/over-exposure conditions. We use the LFW dataset for our experiments and\nassess the model's performance based on verification accuracy. Results reveal\nthat single and combined degradations show dissimilar model behavior. The\ncombined effect of degradation significantly lowers performance even if its\nsingle effect is negligible. This work emphasizes the importance of accounting\nfor real-world complexity to assess the robustness of face recognition models\nin real-world settings. The code is publicly available at\nhttps://github.com/ThEnded32/AnalyzingCombinedDegradations.\n", "date": "2024-06-04", "categories": []}
{"id": "2406.00135", "title": "Advancing Ear Biometrics: Enhancing Accuracy and Robustness through Deep\n  Learning", "abstract": "  Biometric identification is a reliable method to verify individuals based on\ntheir unique physical or behavioral traits, offering a secure alternative to\ntraditional methods like passwords or PINs. This study focuses on ear biometric\nidentification, exploiting its distinctive features for enhanced accuracy,\nreliability, and usability. While past studies typically investigate face\nrecognition and fingerprint analysis, our research demonstrates the\neffectiveness of ear biometrics in overcoming limitations such as variations in\nfacial expressions and lighting conditions. We utilized two datasets: AMI (700\nimages from 100 individuals) and EarNV1.0 (28,412 images from 164 individuals).\nTo improve the accuracy and robustness of our ear biometric identification\nsystem, we applied various techniques including data preprocessing and\naugmentation. Our models achieved a testing accuracy of 99.35% on the AMI\nDataset and 98.1% on the EarNV1.0 dataset, showcasing the effectiveness of our\napproach in precisely identifying individuals based on ear biometric\ncharacteristics.\n", "date": "2024-05-31", "categories": []}
{"id": "2406.00409", "title": "Arabic Handwritten Text for Person Biometric Identification: A Deep\n  Learning Approach", "abstract": "  This study thoroughly investigates how well deep learning models can\nrecognize Arabic handwritten text for person biometric identification. It\ncompares three advanced architectures -- ResNet50, MobileNetV2, and\nEfficientNetB7 -- using three widely recognized datasets: AHAWP, Khatt, and\nLAMIS-MSHD. Results show that EfficientNetB7 outperforms the others, achieving\ntest accuracies of 98.57\\%, 99.15\\%, and 99.79\\% on AHAWP, Khatt, and\nLAMIS-MSHD datasets, respectively. EfficientNetB7's exceptional performance is\ncredited to its innovative techniques, including compound scaling, depth-wise\nseparable convolutions, and squeeze-and-excitation blocks. These features allow\nthe model to extract more abstract and distinctive features from handwritten\ntext images. The study's findings hold significant implications for enhancing\nidentity verification and authentication systems, highlighting the potential of\ndeep learning in Arabic handwritten text recognition for person biometric\nidentification.\n", "date": "2024-06-01", "categories": []}
{"id": "2406.00512", "title": "On the use of first and second derivative approximations for biometric\n  online signature recognition", "abstract": "  This paper investigates the impact of different approximation methods in\nfeature extraction for pattern recognition applications, specifically focused\non delta and delta-delta parameters. Using MCYT330 online signature data-base,\nour experiments show that 11-point approximation outperforms 1-point\napproximation, resulting in a 1.4% improvement in identification rate, 36.8%\nreduction in random forgeries and 2.4% reduction in skilled forgeries\n", "date": "2024-06-01", "categories": []}
{"id": "2405.20465", "title": "ENTIRe-ID: An Extensive and Diverse Dataset for Person Re-Identification", "abstract": "  The growing importance of person reidentification in computer vision has\nhighlighted the need for more extensive and diverse datasets. In response, we\nintroduce the ENTIRe-ID dataset, an extensive collection comprising over 4.45\nmillion images from 37 different cameras in varied environments. This dataset\nis uniquely designed to tackle the challenges of domain variability and model\ngeneralization, areas where existing datasets for person re-identification have\nfallen short. The ENTIRe-ID dataset stands out for its coverage of a wide array\nof real-world scenarios, encompassing various lighting conditions, angles of\nview, and diverse human activities. This design ensures a realistic and robust\ntraining platform for ReID models. The ENTIRe-ID dataset is publicly available\nat https://serdaryildiz.github.io/ENTIRe-ID\n", "date": "2024-05-30", "categories": []}
{"id": "2405.19005", "title": "Auto-selected Knowledge Adapters for Lifelong Person Re-identification", "abstract": "  Lifelong Person Re-Identification (LReID) extends traditional ReID by\nrequiring systems to continually learn from non-overlapping datasets across\ndifferent times and locations, adapting to new identities while preserving\nknowledge of previous ones. Existing approaches, either rehearsal-free or\nrehearsal-based, still suffer from the problem of catastrophic forgetting since\nthey try to cram diverse knowledge into one fixed model. To overcome this\nlimitation, we introduce a novel framework AdalReID, that adopts knowledge\nadapters and a parameter-free auto-selection mechanism for lifelong learning.\nConcretely, we incrementally build distinct adapters to learn domain-specific\nknowledge at each step, which can effectively learn and preserve knowledge\nacross different datasets. Meanwhile, the proposed auto-selection strategy\nadaptively calculates the knowledge similarity between the input set and the\nadapters. On the one hand, the appropriate adapters are selected for the inputs\nto process ReID, and on the other hand, the knowledge interaction and fusion\nbetween adapters are enhanced to improve the generalization ability of the\nmodel. Extensive experiments are conducted to demonstrate the superiority of\nour AdalReID, which significantly outperforms SOTAs by about 10$\\sim$20\\% mAP\non both seen and unseen domains.\n", "date": "2024-05-29", "categories": []}
{"id": "2405.20062", "title": "Can the accuracy bias by facial hairstyle be reduced through balancing\n  the training data?", "abstract": "  Appearance of a face can be greatly altered by growing a beard and mustache.\nThe facial hairstyles in a pair of images can cause marked changes to the\nimpostor distribution and the genuine distribution. Also, different\ndistributions of facial hairstyle across demographics could cause a false\nimpression of relative accuracy across demographics. We first show that, even\nthough larger training sets boost the recognition accuracy on all facial\nhairstyles, accuracy variations caused by facial hairstyles persist regardless\nof the size of the training set. Then, we analyze the impact of having\ndifferent fractions of the training data represent facial hairstyles. We\ncreated balanced training sets using a set of identities available in\nWebface42M that both have clean-shaven and facial hair images. We find that,\neven when a face recognition model is trained with a balanced clean-shaven /\nfacial hair training set, accuracy variation on the test data does not\ndiminish. Next, data augmentation is employed to further investigate the effect\nof facial hair distribution in training data by manipulating facial hair pixels\nwith the help of facial landmark points and a facial hair segmentation model.\nOur results show facial hair causes an accuracy gap between clean-shaven and\nfacial hair images, and this impact can be significantly different between\nAfrican-Americans and Caucasians.\n", "date": "2024-05-30", "categories": []}
{"id": "2305.08386", "title": "PLIP: Language-Image Pre-training for Person Representation Learning", "abstract": "  Language-image pre-training is an effective technique for learning powerful\nrepresentations in general domains. However, when directly turning to person\nrepresentation learning, these general pre-training methods suffer from\nunsatisfactory performance. The reason is that they neglect critical\nperson-related characteristics, i.e., fine-grained attributes and identities.\nTo address this issue, we propose a novel language-image pre-training framework\nfor person representation learning, termed PLIP. Specifically, we elaborately\ndesign three pretext tasks: 1) Text-guided Image Colorization, aims to\nestablish the correspondence between the person-related image regions and the\nfine-grained color-part textual phrases. 2) Image-guided Attributes Prediction,\naims to mine fine-grained attribute information of the person body in the\nimage; and 3) Identity-based Vision-Language Contrast, aims to correlate the\ncross-modal representations at the identity level rather than the instance\nlevel. Moreover, to implement our pre-train framework, we construct a\nlarge-scale person dataset with image-text pairs named SYNTH-PEDES by\nautomatically generating textual annotations. We pre-train PLIP on SYNTH-PEDES\nand evaluate our models by spanning downstream person-centric tasks. PLIP not\nonly significantly improves existing methods on all these tasks, but also shows\ngreat ability in the zero-shot and domain generalization settings. The code,\ndataset and weights will be released\nat~\\url{https://github.com/Zplusdragon/PLIP}\n", "date": "2023-05-15", "categories": []}
{"id": "2308.06948", "title": "MixBCT: Towards Self-Adapting Backward-Compatible Training", "abstract": "  Backward-compatible training circumvents the need for expensive updates to\nthe old gallery database when deploying an advanced new model in the retrieval\nsystem. Previous methods achieved backward compatibility by aligning prototypes\nof the new model with the old one, yet they often overlooked the distribution\nof old features, limiting their effectiveness when the low quality of the old\nmodel results in a weakly feature discriminability. Instance-based methods like\nL2 regression take into account the distribution of old features but impose\nstrong constraints on the performance of the new model itself. In this paper,\nwe propose MixBCT, a simple yet highly effective backward-compatible training\nmethod that serves as a unified framework for old models of varying qualities.\nWe construct a single loss function applied to mixed old and new features to\nfacilitate backward-compatible training, which adaptively adjusts the\nconstraint domain for new features based on the distribution of old features.\nWe conducted extensive experiments on the large-scale face recognition datasets\nMS1Mv3 and IJB-C to verify the effectiveness of our method. The experimental\nresults clearly demonstrate its superiority over previous methods. Code is\navailable at https://github.com/yuleung/MixBCT .\n", "date": "2023-08-14", "categories": []}
{"id": "2312.13240", "title": "Efficient Verification-Based Face Identification", "abstract": "  We study the problem of performing face verification with an efficient neural\nmodel $f$. The efficiency of $f$ stems from simplifying the face verification\nproblem from an embedding nearest neighbor search into a binary problem; each\nuser has its own neural network $f$. To allow information sharing between\ndifferent individuals in the training set, we do not train $f$ directly but\ninstead generate the model weights using a hypernetwork $h$. This leads to the\ngeneration of a compact personalized model for face identification that can be\ndeployed on edge devices. Key to the method's success is a novel way of\ngenerating hard negatives and carefully scheduling the training objectives. Our\nmodel leads to a substantially small $f$ requiring only 23k parameters and 5M\nfloating point operations (FLOPS). We use six face verification datasets to\ndemonstrate that our method is on par or better than state-of-the-art models,\nwith a significantly reduced number of parameters and computational burden.\nFurthermore, we perform an extensive ablation study to demonstrate the\nimportance of each element in our method.\n", "date": "2023-12-20", "categories": []}
{"id": "2405.16683", "title": "Toward Digitalization: A Secure Approach to Find a Missing Person Using\n  Facial Recognition Technology", "abstract": "  Facial Recognition is a technique, based on machine learning technology that\ncan recognize a human being analyzing his facial profile, and is applied in\nsolving various types of realworld problems nowadays. In this paper, a common\nreal-world problem, finding a missing person has been solved in a secure and\neffective way with the help of facial recognition technology. Although there\nexist a few works on solving the problem, the proposed work is unique with\nrespect to its security, design, and feasibility. Impeding intruders in\nparticipating in the processes and giving importance to both finders and family\nmembers of a missing person are two of the major features of this work. The\nproofs of the works of our system in finding a missing person have been\ndescribed in the result section of the paper. The advantages that our system\nprovides over the other existing systems can be realized from the comparisons,\ndescribed in the result summary section of the paper. The work is capable of\nproviding a worthy solution to find a missing person on the digital platform.\n", "date": "2024-05-26", "categories": []}
{"id": "2405.16748", "title": "Hypergraph Laplacian Eigenmaps and Face Recognition Problems", "abstract": "  Face recognition is a very important topic in data science and biometric\nsecurity research areas. It has multiple applications in military, finance, and\nretail, to name a few. In this paper, the novel hypergraph Laplacian Eigenmaps\nwill be proposed and combine with the k nearest-neighbor method and/or with the\nkernel ridge regression method to solve the face recognition problem.\nExperimental results illustrate that the accuracy of the combination of the\nnovel hypergraph Laplacian Eigenmaps and one specific classification system is\nsimilar to the accuracy of the combination of the old symmetric normalized\nhypergraph Laplacian Eigenmaps method and one specific classification system.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.16761", "title": "Masked Face Recognition with Generative-to-Discriminative\n  Representations", "abstract": "  Masked face recognition is important for social good but challenged by\ndiverse occlusions that cause insufficient or inaccurate representations. In\nthis work, we propose a unified deep network to learn\ngenerative-to-discriminative representations for facilitating masked face\nrecognition. To this end, we split the network into three modules and learn\nthem on synthetic masked faces in a greedy module-wise pretraining manner.\nFirst, we leverage a generative encoder pretrained for face inpainting and\nfinetune it to represent masked faces into category-aware descriptors.\nAttribute to the generative encoder's ability in recovering context\ninformation, the resulting descriptors can provide occlusion-robust\nrepresentations for masked faces, mitigating the effect of diverse masks. Then,\nwe incorporate a multi-layer convolutional network as a discriminative reformer\nand learn it to convert the category-aware descriptors into identity-aware\nvectors, where the learning is effectively supervised by distilling relation\nknowledge from off-the-shelf face recognition model. In this way, the\ndiscriminative reformer together with the generative encoder serves as the\npretrained backbone, providing general and discriminative representations\ntowards masked faces. Finally, we cascade one fully-connected layer following\nby one softmax layer into a feature classifier and finetune it to identify the\nreformed identity-aware vectors. Extensive experiments on synthetic and\nrealistic datasets demonstrate the effectiveness of our approach in recognizing\nmasked faces.\n", "date": "2024-05-27", "categories": []}
{"id": "2405.13467", "title": "AdaFedFR: Federated Face Recognition with Adaptive Inter-Class\n  Representation Learning", "abstract": "  With the growing attention on data privacy and communication security in face\nrecognition applications, federated learning has been introduced to learn a\nface recognition model with decentralized datasets in a privacy-preserving\nmanner. However, existing works still face challenges such as unsatisfying\nperformance and additional communication costs, limiting their applicability in\nreal-world scenarios. In this paper, we propose a simple yet effective\nfederated face recognition framework called AdaFedFR, by devising an adaptive\ninter-class representation learning algorithm to enhance the generalization of\nthe generic face model and the efficiency of federated training under strict\nprivacy-preservation. In particular, our work delicately utilizes feature\nrepresentations of public identities as learnable negative knowledge to\noptimize the local objective within the feature space, which further encourages\nthe local model to learn powerful representations and optimize personalized\nmodels for clients. Experimental results demonstrate that our method\noutperforms previous approaches on several prevalent face recognition\nbenchmarks within less than 3 communication rounds, which shows\ncommunication-friendly and great efficiency.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.13555", "title": "A Perspective Analysis of Handwritten Signature Technology", "abstract": "  Handwritten signatures are biometric traits at the center of debate in the\nscientific community. Over the last 40 years, the interest in signature studies\nhas grown steadily, having as its main reference the application of automatic\nsignature verification, as previously published reviews in 1989, 2000, and 2008\nbear witness. Ever since, and over the last 10 years, the application of\nhandwritten signature technology has strongly evolved, and much research has\nfocused on the possibility of applying systems based on handwritten signature\nanalysis and processing to a multitude of new fields. After several years of\nhaphazard growth of this research area, it is time to assess its current\ndevelopments for their applicability in order to draw a structured way forward.\nThis perspective reports a systematic review of the last 10 years of the\nliterature on handwritten signatures with respect to the new scenario, focusing\non the most promising domains of research and trying to elicit possible future\nresearch directions in this subject.\n", "date": "2024-05-22", "categories": []}
{"id": "2405.14409", "title": "Investigating the Common Authorship of Signatures by Off-Line Automatic\n  Signature Verification Without the Use of Reference Signatures", "abstract": "  In automatic signature verification, questioned specimens are usually\ncompared with reference signatures. In writer-dependent schemes, a number of\nreference signatures are required to build up the individual signer model while\na writer-independent system requires a set of reference signatures from several\nsigners to develop the model of the system. This paper addresses the problem of\nautomatic signature verification when no reference signatures are available.\nThe scenario we explore consists of a set of signatures, which could be signed\nby the same author or by multiple signers. As such, we discuss three methods\nwhich estimate automatically the common authorship of a set of off-line\nsignatures. The first method develops a score similarity matrix, worked out\nwith the assistance of duplicated signatures; the second uses a\nfeature-distance matrix for each pair of signatures; and the last method\nintroduces pre-classification based on the complexity of each signature.\nPublicly available signatures were used in the experiments, which gave\nencouraging results. As a baseline for the performance obtained by our\napproaches, we carried out a visual Turing Test where forensic and non-forensic\nhuman volunteers, carrying out the same task, performed less well than the\nautomatic schemes.\n", "date": "2024-05-23", "categories": []}
{"id": "2405.12447", "title": "EPL: Empirical Prototype Learning for Deep Face Recognition", "abstract": "  Prototype learning is widely used in face recognition, which takes the row\nvectors of coefficient matrix in the last linear layer of the feature\nextraction model as the prototypes for each class. When the prototypes are\nupdated using the facial sample feature gradients in the model training, they\nare prone to being pulled away from the class center by the hard samples,\nresulting in decreased overall model performance. In this paper, we explicitly\ndefine prototypes as the expectations of sample features in each class and\ndesign the empirical prototypes using the existing samples in the dataset. We\nthen devise a strategy to adaptively update these empirical prototypes during\nthe model training based on the similarity between the sample features and the\nempirical prototypes. Furthermore, we propose an empirical prototype learning\n(EPL) method, which utilizes an adaptive margin parameter with respect to\nsample features. EPL assigns larger margins to the normal samples and smaller\nmargins to the hard samples, allowing the learned empirical prototypes to\nbetter reflect the class center dominated by the normal samples and finally\npull the hard samples towards the empirical prototypes through the learning.\nThe extensive experiments on MFR, IJB-C, LFW, CFP-FP, AgeDB, and MegaFace\ndemonstrate the effectiveness of EPL. Our code is available at\n$\\href{https://github.com/WakingHours-GitHub/EPL}{https://github.com/WakingHours-GitHub/EPL}$.\n", "date": "2024-05-21", "categories": []}
{"id": "2405.12556", "title": "Online Signature Recognition: A Biologically Inspired Feature Vector\n  Splitting Approach", "abstract": "  This research introduces an innovative approach to explore the cognitive and\nbiologically inspired underpinnings of feature vector splitting for analyzing\nthe significance of different attributes in e-security biometric signature\nrecognition applications. Departing from traditional methods of concatenating\nfeatures into an extended set, we employ multiple splitting strategies,\naligning with cognitive principles, to preserve control over the relative\nimportance of each feature subset. Our methodology is applied to three diverse\ndatabases (MCYT100, MCYT300,and SVC) using two classifiers (vector quantization\nand dynamic time warping with one and five training samples). Experimentation\ndemonstrates that the fusion of pressure data with spatial coordinates (x and\ny) consistently enhances performance. However, the inclusion of pen-tip angles\nin the same feature set yields mixed results, with performance improvements\nobserved in select cases. This work delves into the cognitive aspects of\nfeature fusion,shedding light on the cognitive relevance of feature vector\nsplitting in e-security biometric applications.\n", "date": "2024-05-21", "categories": []}
{"id": "2405.12633", "title": "Automating Attendance Management in Human Resources: A Design Science\n  Approach Using Computer Vision and Facial Recognition", "abstract": "  Haar Cascade is a cost-effective and user-friendly machine learning-based\nalgorithm for detecting objects in images and videos. Unlike Deep Learning\nalgorithms, which typically require significant resources and expensive\ncomputing costs, it uses simple image processing techniques like edge detection\nand Haar features that are easy to comprehend and implement. By combining Haar\nCascade with OpenCV2 on an embedded computer like the NVIDIA Jetson Nano, this\nsystem can accurately detect and match faces in a database for attendance\ntracking. This system aims to achieve several specific objectives that set it\napart from existing solutions. It leverages Haar Cascade, enriched with\ncarefully selected Haar features, such as Haar-like wavelets, and employs\nadvanced edge detection techniques. These techniques enable precise face\ndetection and matching in both images and videos, contributing to high accuracy\nand robust performance. By doing so, it minimizes manual intervention and\nreduces errors, thereby strengthening accountability. Additionally, the\nintegration of OpenCV2 and the NVIDIA Jetson Nano optimizes processing\nefficiency, making it suitable for resource-constrained environments. This\nsystem caters to a diverse range of educational institutions, including\nschools, colleges, vocational training centers, and various workplace settings\nsuch as small businesses, offices, and factories. ... The system's\naffordability and efficiency democratize attendance management technology,\nmaking it accessible to a broader audience. Consequently, it has the potential\nto transform attendance tracking and management practices, ultimately leading\nto heightened productivity and accountability. In conclusion, this system\nrepresents a groundbreaking approach to attendance tracking and management...\n", "date": "2024-05-21", "categories": []}
{"id": "2405.12695", "title": "Explainable offline automatic signature verifier to support forensic\n  handwriting examiners", "abstract": "  Signature verification is a critical task in many applications, including\nforensic science, legal judgments, and financial markets. However, current\nsignature verification systems are often difficult to explain, which can limit\ntheir acceptance in these applications. In this paper, we propose a novel\nexplainable offline automatic signature verifier (ASV) to support forensic\nhandwriting examiners. Our ASV is based on a universal background model (UBM)\nconstructed from offline signature images. It allows us to assign a questioned\nsignature to the UBM and to a reference set of known signatures using simple\ndistance measures. This makes it possible to explain the verifier's decision in\na way that is understandable to non experts. We evaluated our ASV on publicly\navailable databases and found that it achieves competitive performance with\nstate of the art ASVs, even when challenging 1 versus 1 comparison are\nconsidered. Our results demonstrate that it is possible to develop an\nexplainable ASV that is also competitive in terms of performance. We believe\nthat our ASV has the potential to improve the acceptance of signature\nverification in critical applications such as forensic science and legal\njudgments.\n", "date": "2024-05-21", "categories": []}
{"id": "2405.11240", "title": "Testing the Performance of Face Recognition for People with Down\n  Syndrome", "abstract": "  The fairness of biometric systems, in particular facial recognition, is often\nanalysed for larger demographic groups, e.g. female vs. male or black vs.\nwhite. In contrast to this, minority groups are commonly ignored. This paper\ninvestigates the performance of facial recognition algorithms on individuals\nwith Down syndrome, a common chromosomal abnormality that affects approximately\none in 1,000 births per year. To do so, a database of 98 individuals with Down\nsyndrome, each represented by at least five facial images, is\nsemi-automatically collected from YouTube. Subsequently, two facial image\nquality assessment algorithms and five recognition algorithms are evaluated on\nthe newly collected database and on the public facial image databases CelebA\nand FRGCv2. The results show that the quality scores of facial images for\nindividuals with Down syndrome are comparable to those of individuals without\nDown syndrome captured under similar conditions. Furthermore, it is observed\nthat face recognition performance decreases significantly for individuals with\nDown syndrome, which is largely attributed to the increased likelihood of false\nmatches.\n", "date": "2024-05-18", "categories": []}
{"id": "2405.11978", "title": "SM-DTW: Stability Modulated Dynamic Time Warping for signature\n  verification", "abstract": "  Building upon findings in computational model of handwriting learning and\nexecution, we introduce the concept of stability to explain the difference\nbetween the actual movements performed during multiple execution of the\nsubject's signature, and conjecture that the most stable parts of the signature\nshould play a paramount role in evaluating the similarity between a questioned\nsignature and the reference ones during signature verification. We then\nintroduce the Stability Modulated Dynamic Time Warping algorithm for\nincorporating the stability regions, i.e. the most similar parts between two\nsignatures, into the distance measure between a pair of signatures computed by\nthe Dynamic Time Warping for signature verification. Experiments were conducted\non two datasets largely adopted for performance evaluation. Experimental\nresults show that the proposed algorithm improves the performance of the\nbaseline system and compares favourably with other top performing signature\nverification systems.\n", "date": "2024-05-20", "categories": []}
{"id": "2405.10868", "title": "Air Signing and Privacy-Preserving Signature Verification for Digital\n  Documents", "abstract": "  This paper presents a novel approach to the digital signing of electronic\ndocuments through the use of a camera-based interaction system, single-finger\ntracking for sign recognition, and multi commands executing hand gestures. The\nproposed solution, referred to as \"Air Signature,\" involves writing the\nsignature in front of the camera, rather than relying on traditional methods\nsuch as mouse drawing or physically signing on paper and showing it to a web\ncamera. The goal is to develop a state-of-the-art method for detecting and\ntracking gestures and objects in real-time. The proposed methods include\napplying existing gesture recognition and object tracking systems, improving\naccuracy through smoothing and line drawing, and maintaining continuity during\nfast finger movements. An evaluation of the fingertip detection, sketching, and\noverall signing process is performed to assess the effectiveness of the\nproposed solution. The secondary objective of this research is to develop a\nmodel that can effectively recognize the unique signature of a user. This type\nof signature can be verified by neural cores that analyze the movement, speed,\nand stroke pixels of the signing in real time. The neural cores use machine\nlearning algorithms to match air signatures to the individual's stored\nsignatures, providing a secure and efficient method of verification. Our\nproposed System does not require sensors or any hardware other than the camera.\n", "date": "2024-05-17", "categories": []}
{"id": "2405.09880", "title": "Deep Learning-Based Quasi-Conformal Surface Registration for Partial 3D\n  Faces Applied to Facial Recognition", "abstract": "  3D face registration is an important process in which a 3D face model is\naligned and mapped to a template face. However, the task of 3D face\nregistration becomes particularly challenging when dealing with partial face\ndata, where only limited facial information is available. To address this\nchallenge, this paper presents a novel deep learning-based approach that\ncombines quasi-conformal geometry with deep neural networks for partial face\nregistration. The proposed framework begins with a Landmark Detection Network\nthat utilizes curvature information to detect the presence of facial features\nand estimate their corresponding coordinates. These facial landmark features\nserve as essential guidance for the registration process. To establish a dense\ncorrespondence between the partial face and the template surface, a\nregistration network based on quasiconformal theories is employed. The\nregistration network establishes a bijective quasiconformal surface mapping\naligning corresponding partial faces based on detected landmarks and curvature\nvalues. It consists of the Coefficients Prediction Network, which outputs the\noptimal Beltrami coefficient representing the surface mapping. The Beltrami\ncoefficient quantifies the local geometric distortion of the mapping. By\ncontrolling the magnitude of the Beltrami coefficient through a suitable\nactivation function, the bijectivity and geometric distortion of the mapping\ncan be controlled. The Beltrami coefficient is then fed into the Beltrami\nsolver network to reconstruct the corresponding mapping. The surface\nregistration enables the acquisition of corresponding regions and the\nestablishment of point-wise correspondence between different partial faces,\nfacilitating precise shape comparison through the evaluation of point-wise\ngeometric differences at these corresponding regions. Experimental results\ndemonstrate the effectiveness of the proposed method.\n", "date": "2024-05-16", "categories": []}
{"id": "2405.09403", "title": "Identity Overlap Between Face Recognition Train/Test Data: Causing\n  Optimistic Bias in Accuracy Measurement", "abstract": "  A fundamental tenet of pattern recognition is that overlap between training\nand testing sets causes an optimistic accuracy estimate. Deep CNNs for face\nrecognition are trained for N-way classification of the identities in the\ntraining set. Accuracy is commonly estimated as average 10-fold classification\naccuracy on image pairs from test sets such as LFW, CALFW, CPLFW, CFP-FP and\nAgeDB-30. Because train and test sets have been independently assembled, images\nand identities in any given test set may also be present in any given training\nset. In particular, our experiments reveal a surprising degree of identity and\nimage overlap between the LFW family of test sets and the MS1MV2 training set.\nOur experiments also reveal identity label noise in MS1MV2. We compare accuracy\nachieved with same-size MS1MV2 subsets that are identity-disjoint and not\nidentity-disjoint with LFW, to reveal the size of the optimistic bias. Using\nmore challenging test sets from the LFW family, we find that the size of the\noptimistic bias is larger for more challenging test sets. Our results highlight\nthe lack of and the need for identity-disjoint train and test methodology in\nface recognition research.\n", "date": "2024-05-15", "categories": []}
{"id": "2405.09487", "title": "Color Space Learning for Cross-Color Person Re-Identification", "abstract": "  The primary color profile of the same identity is assumed to remain\nconsistent in typical Person Re-identification (Person ReID) tasks. However,\nthis assumption may be invalid in real-world situations and images hold variant\ncolor profiles, because of cross-modality cameras or identity with different\nclothing. To address this issue, we propose Color Space Learning (CSL) for\nthose Cross-Color Person ReID problems. Specifically, CSL guides the model to\nbe less color-sensitive with two modules: Image-level Color-Augmentation and\nPixel-level Color-Transformation. The first module increases the color\ndiversity of the inputs and guides the model to focus more on the non-color\ninformation. The second module projects every pixel of input images onto a new\ncolor space. In addition, we introduce a new Person ReID benchmark across RGB\nand Infrared modalities, NTU-Corridor, which is the first with privacy\nagreements from all participants. To evaluate the effectiveness and robustness\nof our proposed CSL, we evaluate it on several Cross-Color Person ReID\nbenchmarks. Our method surpasses the state-of-the-art methods consistently. The\ncode and benchmark are available at: https://github.com/niejiahao1998/CSL\n", "date": "2024-05-15", "categories": []}
{"id": "2405.07571", "title": "TattTRN: Template Reconstruction Network for Tattoo Retrieval", "abstract": "  Tattoos have been used effectively as soft biometrics to assist law\nenforcement in the identification of offenders and victims, as they contain\ndiscriminative information, and are a useful indicator to locate members of a\ncriminal gang or organisation. Due to various privacy issues in the acquisition\nof images containing tattoos, only a limited number of databases exists. This\nlack of databases has delayed the development of new methods to effectively\nretrieve a potential suspect's tattoo images from a candidate gallery. To\nmitigate this issue, in our work, we use an unsupervised generative approach to\ncreate a balanced database consisting of 28,550 semi-synthetic images with\ntattooed subjects from 571 tattoo categories. Further, we introduce a novel\nTattoo Template Reconstruction Network (TattTRN), which learns to map the input\ntattoo sample to its respective tattoo template to enhance the distinguishing\nattributes of the final feature embedding. Experimental results with real data,\ni.e., WebTattoo and BIVTatt databases, demonstrate the soundness of the\npresented approach: an accuracy of up to 99% is achieved for checking at most\nthe first 20 entries of the candidate list.\n", "date": "2024-05-13", "categories": []}
{"id": "2401.04956", "title": "EmMixformer: Mix transformer for eye movement recognition", "abstract": "  Eye movement (EM) is a new highly secure biometric behavioral modality that\nhas received increasing attention in recent years. Although deep neural\nnetworks, such as convolutional neural network (CNN), have recently achieved\npromising performance, current solutions fail to capture local and global\ntemporal dependencies within eye movement data. To overcome this problem, we\npropose in this paper a mixed transformer termed EmMixformer to extract time\nand frequency domain information for eye movement recognition. To this end, we\npropose a mixed block consisting of three modules, transformer, attention Long\nshort-term memory (attention LSTM), and Fourier transformer. We are the first\nto attempt leveraging transformer to learn long temporal dependencies within\neye movement. Second, we incorporate the attention mechanism into LSTM to\npropose attention LSTM with the aim to learn short temporal dependencies.\nThird, we perform self attention in the frequency domain to learn global\nfeatures. As the three modules provide complementary feature representations in\nterms of local and global dependencies, the proposed EmMixformer is capable of\nimproving recognition accuracy. The experimental results on our eye movement\ndataset and two public eye movement datasets show that the proposed EmMixformer\noutperforms the state of the art by achieving the lowest verification error.\n", "date": "2024-01-10", "categories": []}
{"id": "2405.05900", "title": "A Comprehensive Survey of Masked Faces: Recognition, Detection, and\n  Unmasking", "abstract": "  Masked face recognition (MFR) has emerged as a critical domain in biometric\nidentification, especially by the global COVID-19 pandemic, which introduced\nwidespread face masks. This survey paper presents a comprehensive analysis of\nthe challenges and advancements in recognising and detecting individuals with\nmasked faces, which has seen innovative shifts due to the necessity of adapting\nto new societal norms. Advanced through deep learning techniques, MFR, along\nwith Face Mask Recognition (FMR) and Face Unmasking (FU), represent significant\nareas of focus. These methods address unique challenges posed by obscured\nfacial features, from fully to partially covered faces. Our comprehensive\nreview delves into the various deep learning-based methodologies developed for\nMFR, FMR, and FU, highlighting their distinctive challenges and the solutions\nproposed to overcome them. Additionally, we explore benchmark datasets and\nevaluation metrics specifically tailored for assessing performance in MFR\nresearch. The survey also discusses the substantial obstacles still facing\nresearchers in this field and proposes future directions for the ongoing\ndevelopment of more robust and effective masked face recognition systems. This\npaper serves as an invaluable resource for researchers and practitioners,\noffering insights into the evolving landscape of face recognition technologies\nin the face of global health crises and beyond.\n", "date": "2024-05-09", "categories": []}
{"id": "2405.01273", "title": "Towards Inclusive Face Recognition Through Synthetic Ethnicity\n  Alteration", "abstract": "  Numerous studies have shown that existing Face Recognition Systems (FRS),\nincluding commercial ones, often exhibit biases toward certain ethnicities due\nto under-represented data. In this work, we explore ethnicity alteration and\nskin tone modification using synthetic face image generation methods to\nincrease the diversity of datasets. We conduct a detailed analysis by first\nconstructing a balanced face image dataset representing three ethnicities:\nAsian, Black, and Indian. We then make use of existing Generative Adversarial\nNetwork-based (GAN) image-to-image translation and manifold learning models to\nalter the ethnicity from one to another. A systematic analysis is further\nconducted to assess the suitability of such datasets for FRS by studying the\nrealistic skin-tone representation using Individual Typology Angle (ITA).\nFurther, we also analyze the quality characteristics using existing Face image\nquality assessment (FIQA) approaches. We then provide a holistic FRS\nperformance analysis using four different systems. Our findings pave the way\nfor future research works in (i) developing both specific ethnicity and general\n(any to any) ethnicity alteration models, (ii) expanding such approaches to\ncreate databases with diverse skin tones, (iii) creating datasets representing\nvarious ethnicities which further can help in mitigating bias while addressing\nprivacy concerns.\n", "date": "2024-05-02", "categories": []}
{"id": "2404.03443", "title": "Part-Attention Based Model Make Occluded Person Re-Identification\n  Stronger", "abstract": "  The goal of occluded person re-identification (ReID) is to retrieve specific\npedestrians in occluded situations. However, occluded person ReID still suffers\nfrom background clutter and low-quality local feature representations, which\nlimits model performance. In our research, we introduce a new framework called\nPAB-ReID, which is a novel ReID model incorporating part-attention mechanisms\nto tackle the aforementioned issues effectively. Firstly, we introduce the\nhuman parsing label to guide the generation of more accurate human part\nattention maps. In addition, we propose a fine-grained feature focuser for\ngenerating fine-grained human local feature representations while suppressing\nbackground interference. Moreover, We also design a part triplet loss to\nsupervise the learning of human local features, which optimizes\nintra/inter-class distance. We conducted extensive experiments on specialized\nocclusion and regular ReID datasets, showcasing that our approach outperforms\nthe existing state-of-the-art methods.\n", "date": "2024-04-04", "categories": []}
{"id": "2405.00650", "title": "Grains of Saliency: Optimizing Saliency-based Training of Biometric\n  Attack Detection Models", "abstract": "  Incorporating human-perceptual intelligence into model training has shown to\nincrease the generalization capability of models in several difficult biometric\ntasks, such as presentation attack detection (PAD) and detection of synthetic\nsamples. After the initial collection phase, human visual saliency (e.g.,\neye-tracking data, or handwritten annotations) can be integrated into model\ntraining through attention mechanisms, augmented training samples, or through\nhuman perception-related components of loss functions. Despite their successes,\na vital, but seemingly neglected, aspect of any saliency-based training is the\nlevel of salience granularity (e.g., bounding boxes, single saliency maps, or\nsaliency aggregated from multiple subjects) necessary to find a balance between\nreaping the full benefits of human saliency and the cost of its collection. In\nthis paper, we explore several different levels of salience granularity and\ndemonstrate that increased generalization capabilities of PAD and synthetic\nface detection can be achieved by using simple yet effective saliency\npost-processing techniques across several different CNNs.\n", "date": "2024-05-01", "categories": []}
{"id": "2302.08212", "title": "Visible-Infrared Person Re-Identification via Patch-Mixed Cross-Modality\n  Learning", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to retrieve images\nof the same pedestrian from different modalities, where the challenges lie in\nthe significant modality discrepancy. To alleviate the modality gap, recent\nmethods generate intermediate images by GANs, grayscaling, or mixup strategies.\nHowever, these methods could introduce extra data distribution, and the\nsemantic correspondence between the two modalities is not well learned. In this\npaper, we propose a Patch-Mixed Cross-Modality framework (PMCM), where two\nimages of the same person from two modalities are split into patches and\nstitched into a new one for model learning. A part-alignment loss is introduced\nto regularize representation learning, and a patch-mixed modality learning loss\nis proposed to align between the modalities. In this way, the model learns to\nrecognize a person through patches of different styles, thereby the modality\nsemantic correspondence can be inferred. In addition, with the flexible image\ngeneration strategy, the patch-mixed images freely adjust the ratio of\ndifferent modality patches, which could further alleviate the modality\nimbalance problem. On two VI-ReID datasets, we report new state-of-the-art\nperformance with the proposed method.\n", "date": "2023-02-16", "categories": []}
{"id": "2306.01198", "title": "Confidence Intervals for Error Rates in 1:1 Matching Tasks: Critical\n  Statistical Analysis and Recommendations", "abstract": "  Matching algorithms are commonly used to predict matches between items in a\ncollection. For example, in 1:1 face verification, a matching algorithm\npredicts whether two face images depict the same person. Accurately assessing\nthe uncertainty of the error rates of such algorithms can be challenging when\ndata are dependent and error rates are low, two aspects that have been often\noverlooked in the literature. In this work, we review methods for constructing\nconfidence intervals for error rates in 1:1 matching tasks. We derive and\nexamine the statistical properties of these methods, demonstrating how coverage\nand interval width vary with sample size, error rates, and degree of data\ndependence on both analysis and experiments with synthetic and real-world\ndatasets. Based on our findings, we provide recommendations for best practices\nfor constructing confidence intervals for error rates in 1:1 matching tasks.\n", "date": "2023-06-01", "categories": []}
{"id": "2306.11991", "title": "Generalizable Metric Network for Cross-domain Person Re-identification", "abstract": "  Person Re-identification (Re-ID) is a crucial technique for public security\nand has made significant progress in supervised settings. However, the\ncross-domain (i.e., domain generalization) scene presents a challenge in Re-ID\ntasks due to unseen test domains and domain-shift between the training and test\nsets. To tackle this challenge, most existing methods aim to learn\ndomain-invariant or robust features for all domains. In this paper, we observe\nthat the data-distribution gap between the training and test sets is smaller in\nthe sample-pair space than in the sample-instance space. Based on this\nobservation, we propose a Generalizable Metric Network (GMN) to further explore\nsample similarity in the sample-pair space. Specifically, we add a Metric\nNetwork (M-Net) after the main network and train it on positive and negative\nsample-pair features, which is then employed during the test stage.\nAdditionally, we introduce the Dropout-based Perturbation (DP) module to\nenhance the generalization capability of the metric network by enriching the\nsample-pair diversity. Moreover, we develop a Pair-Identity Center (PIC) loss\nto enhance the model's discrimination by ensuring that sample-pair features\nwith the same pair-identity are consistent. We validate the effectiveness of\nour proposed method through a lot of experiments on multiple benchmark datasets\nand confirm the value of each module in our GMN.\n", "date": "2023-06-21", "categories": []}
{"id": "2404.17610", "title": "Regression of Dense Distortion Field from a Single Fingerprint Image", "abstract": "  Skin distortion is a long standing challenge in fingerprint matching, which\ncauses false non-matches. Previous studies have shown that the recognition rate\ncan be improved by estimating the distortion field from a distorted fingerprint\nand then rectifying it into a normal fingerprint. However, existing\nrectification methods are based on principal component representation of\ndistortion fields, which is not accurate and are very sensitive to finger pose.\nIn this paper, we propose a rectification method where a self-reference based\nnetwork is utilized to directly estimate the dense distortion field of\ndistorted fingerprint instead of its low dimensional representation. This\nmethod can output accurate distortion fields of distorted fingerprints with\nvarious finger poses and distortion patterns. We conducted experiments on\nFVC2004 DB1\\_A, expanded Tsinghua Distorted Fingerprint database (with\nadditional distorted fingerprints in diverse finger poses and distortion\npatterns) and a latent fingerprint database. Experimental results demonstrate\nthat our proposed method achieves the state-of-the-art rectification\nperformance in terms of distortion field estimation and rectified fingerprint\nmatching.\n", "date": "2024-04-26", "categories": []}
{"id": "2404.17760", "title": "Adversarial Examples: Generation Proposal in the Context of Facial\n  Recognition Systems", "abstract": "  In this paper we investigate the vulnerability that facial recognition\nsystems present to adversarial examples by introducing a new methodology from\nthe attacker perspective. The technique is based on the use of the autoencoder\nlatent space, organized with principal component analysis. We intend to analyze\nthe potential to craft adversarial examples suitable for both dodging and\nimpersonation attacks, against state-of-the-art systems. Our initial\nhypothesis, which was not strongly favoured by the results, stated that it\nwould be possible to separate between the \"identity\" and \"facial expression\"\nfeatures to produce high-quality examples. Despite the findings not supporting\nit, the results sparked insights into adversarial examples generation and\nopened new research avenues in the area.\n", "date": "2024-04-27", "categories": []}
{"id": "2404.18890", "title": "Hide and Seek: How Does Watermarking Impact Face Recognition?", "abstract": "  The recent progress in generative models has revolutionized the synthesis of\nhighly realistic images, including face images. This technological development\nhas undoubtedly helped face recognition, such as training data augmentation for\nhigher recognition accuracy and data privacy. However, it has also introduced\nnovel challenges concerning the responsible use and proper attribution of\ncomputer generated images. We investigate the impact of digital watermarking, a\ntechnique for embedding ownership signatures into images, on the effectiveness\nof face recognition models. We propose a comprehensive pipeline that integrates\nface image generation, watermarking, and face recognition to systematically\nexamine this question. The proposed watermarking scheme, based on an\nencoder-decoder architecture, successfully embeds and recovers signatures from\nboth real and synthetic face images while preserving their visual fidelity.\nThrough extensive experiments, we unveil that while watermarking enables robust\nimage attribution, it results in a slight decline in face recognition accuracy,\nparticularly evident for face images with challenging poses and expressions.\nAdditionally, we find that directly training face recognition models on\nwatermarked images offers only a limited alleviation of this performance\ndecline. Our findings underscore the intricate trade off between watermarking\nand face recognition accuracy. This work represents a pivotal step towards the\nresponsible utilization of generative models in face recognition and serves to\ninitiate discussions regarding the broader implications of watermarking in\nbiometrics.\n", "date": "2024-04-29", "categories": []}
{"id": "2403.04654", "title": "Audio-Visual Person Verification based on Recursive Fusion of Joint\n  Cross-Attention", "abstract": "  Person or identity verification has been recently gaining a lot of attention\nusing audio-visual fusion as faces and voices share close associations with\neach other. Conventional approaches based on audio-visual fusion rely on\nscore-level or early feature-level fusion techniques. Though existing\napproaches showed improvement over unimodal systems, the potential of\naudio-visual fusion for person verification is not fully exploited. In this\npaper, we have investigated the prospect of effectively capturing both the\nintra- and inter-modal relationships across audio and visual modalities, which\ncan play a crucial role in significantly improving the fusion performance over\nunimodal systems. In particular, we introduce a recursive fusion of a joint\ncross-attentional model, where a joint audio-visual feature representation is\nemployed in the cross-attention framework in a recursive fashion to\nprogressively refine the feature representations that can efficiently capture\nthe intra-and inter-modal relationships. To further enhance the audio-visual\nfeature representations, we have also explored BLSTMs to improve the temporal\nmodeling of audio-visual feature representations. Extensive experiments are\nconducted on the Voxceleb1 dataset to evaluate the proposed model. Results\nindicate that the proposed model shows promising improvement in fusion\nperformance by adeptly capturing the intra-and inter-modal relationships across\naudio and visual modalities.\n", "date": "2024-03-07", "categories": []}
{"id": "2404.03537", "title": "If It's Not Enough, Make It So: Reducing Authentic Data Demand in Face\n  Recognition through Synthetic Faces", "abstract": "  Recent advances in deep face recognition have spurred a growing demand for\nlarge, diverse, and manually annotated face datasets. Acquiring authentic,\nhigh-quality data for face recognition has proven to be a challenge, primarily\ndue to privacy concerns. Large face datasets are primarily sourced from\nweb-based images, lacking explicit user consent. In this paper, we examine\nwhether and how synthetic face data can be used to train effective face\nrecognition models with reduced reliance on authentic images, thereby\nmitigating data collection concerns. First, we explored the performance gap\namong recent state-of-the-art face recognition models, trained with synthetic\ndata only and authentic (scarce) data only. Then, we deepened our analysis by\ntraining a state-of-the-art backbone with various combinations of synthetic and\nauthentic data, gaining insights into optimizing the limited use of the latter\nfor verification accuracy. Finally, we assessed the effectiveness of data\naugmentation approaches on synthetic and authentic data, with the same goal in\nmind. Our results highlighted the effectiveness of FR trained on combined\ndatasets, particularly when combined with appropriate augmentation techniques.\n", "date": "2024-04-04", "categories": []}
{"id": "2404.16840", "title": "Biometrics Employing Neural Network", "abstract": "  Biometrics involves using unique human traits, both physical and behavioral,\nfor the digital identification of individuals to provide access to systems,\ndevices, or information. Within the field of computer science, it acts as a\nmethod for identifying and verifying individuals and controlling access. While\nthe conventional method for personal authentication involves passwords, the\nvulnerability arises when passwords are compromised, allowing unauthorized\naccess to sensitive actions. Biometric authentication presents a viable answer\nto this problem and is the most secure and user-friendly authentication method.\nToday, fingerprints, iris and retina patterns, facial recognition, hand shapes,\npalm prints, and voice recognition are frequently used forms of biometrics.\nDespite the diverse nature of these biometric identifiers, the core objective\nremains consistent ensuring security, recognizing authorized users, and\nrejecting impostors. Hence, it is crucial to determine accurately whether the\ncharacteristics belong to the rightful person. For systems to be effective and\nwidely accepted, the error rate in recognition and verification must approach\nzero. It is acknowledged that current biometric techniques, while advanced, are\nnot infallible and require continuous improvement. A more refined classifier is\ndeemed necessary to classify patterns accurately. Artificial Neural Networks,\nwhich simulate the human brain's operations, present themselves as a promising\napproach. The survey presented herein explores various biometric techniques\nbased on neural networks, emphasizing the ongoing quest for enhanced accuracy\nand reliability. It concludes that The utilization of neural networks along\nwith biometric features not only enhances accuracy but also contributes to\noverall better security.\n", "date": "2024-02-01", "categories": []}
{"id": "2404.17148", "title": "Direct Regression of Distortion Field from a Single Fingerprint Image", "abstract": "  Skin distortion is a long standing challenge in fingerprint matching, which\ncauses false non-matches. Previous studies have shown that the recognition rate\ncan be improved by estimating the distortion field from a distorted fingerprint\nand then rectifying it into a normal fingerprint. However, existing\nrectification methods are based on principal component representation of\ndistortion fields, which is not accurate and are very sensitive to finger pose.\nIn this paper, we propose a rectification method where a self-reference based\nnetwork is utilized to directly estimate the dense distortion field of\ndistorted fingerprint instead of its low dimensional representation. This\nmethod can output accurate distortion fields of distorted fingerprints with\nvarious finger poses. Considering the limited number and variety of distorted\nfingerprints in the existing public dataset, we collected more distorted\nfingerprints with diverse finger poses and distortion patterns as a new\ndatabase. Experimental results demonstrate that our proposed method achieves\nthe state-of-the-art rectification performance in terms of distortion field\nestimation and rectified fingerprint matching.\n", "date": "2024-04-26", "categories": []}
{"id": "2404.17149", "title": "Pose-Specific 3D Fingerprint Unfolding", "abstract": "  In order to make 3D fingerprints compatible with traditional 2D flat\nfingerprints, a common practice is to unfold the 3D fingerprint into a 2D\nrolled fingerprint, which is then matched with the flat fingerprints by\ntraditional 2D fingerprint recognition algorithms. The problem with this method\nis that there may be large elastic deformation between the unfolded rolled\nfingerprint and flat fingerprint, which affects the recognition rate. In this\npaper, we propose a pose-specific 3D fingerprint unfolding algorithm to unfold\nthe 3D fingerprint using the same pose as the flat fingerprint. Our experiments\nshow that the proposed unfolding algorithm improves the compatibility between\n3D fingerprint and flat fingerprint and thus leads to higher genuine matching\nscores.\n", "date": "2024-04-26", "categories": []}
{"id": "2404.16255", "title": "Enhancing Privacy in Face Analytics Using Fully Homomorphic Encryption", "abstract": "  Modern face recognition systems utilize deep neural networks to extract\nsalient features from a face. These features denote embeddings in latent space\nand are often stored as templates in a face recognition system. These\nembeddings are susceptible to data leakage and, in some cases, can even be used\nto reconstruct the original face image. To prevent compromising identities,\ntemplate protection schemes are commonly employed. However, these schemes may\nstill not prevent the leakage of soft biometric information such as age, gender\nand race. To alleviate this issue, we propose a novel technique that combines\nFully Homomorphic Encryption (FHE) with an existing template protection scheme\nknown as PolyProtect. We show that the embeddings can be compressed and\nencrypted using FHE and transformed into a secure PolyProtect template using\npolynomial transformation, for additional protection. We demonstrate the\nefficacy of the proposed approach through extensive experiments on multiple\ndatasets. Our proposed approach ensures irreversibility and unlinkability,\neffectively preventing the leakage of soft biometric attributes from face\nembeddings without compromising recognition accuracy.\n", "date": "2024-04-24", "categories": []}
{"id": "2404.16739", "title": "CBRW: A Novel Approach for Cancelable Biometric Template Generation\n  based on", "abstract": "  Cancelable Biometric is a challenging research field in which security of an\noriginal biometric image is ensured by transforming the original biometric into\nanother irreversible domain. Several approaches have been suggested in\nliterature for generating cancelable biometric templates. In this paper, two\nnovel and simple cancelable biometric template generation methods based on\nRandom Walk (CBRW) have been proposed. By employing random walk and other steps\ngiven in the proposed two algorithms viz. CBRW-BitXOR and CBRW-BitCMP, the\noriginal biometric is transformed into a cancellable template. The performance\nof the proposed methods is compared with other state-of-the-art methods.\nExperiments have been performed on eight publicly available gray and color\ndatasets i.e. CP (ear) (gray and color), UTIRIS (iris) (gray and color), ORL\n(face) (gray), IIT Delhi (iris) (gray and color), and AR (face) (color).\nPerformance of the generated templates is measured in terms of Correlation\nCoefficient (Cr), Root Mean Square Error (RMSE), Peak Signal to Noise Ratio\n(PSNR), Structural Similarity (SSIM), Mean Absolute Error (MAE), Number of\nPixel Change Rate (NPCR), and Unified Average Changing Intensity (UACI). By\nexperimental results, it has been proved that proposed methods are superior\nthan other state-of-the-art methods in qualitative as well as quantitative\nanalysis. Furthermore, CBRW performs better on both gray as well as color\nimages.\n", "date": "2024-04-25", "categories": []}
{"id": "2404.15312", "title": "Realtime Person Identification via Gait Analysis", "abstract": "  Each person has a unique gait, i.e., walking style, that can be used as a\nbiometric for personal identification. Recent works have demonstrated effective\ngait recognition using deep neural networks, however most of these works\npredominantly focus on classification accuracy rather than model efficiency. In\norder to perform gait recognition using wearable devices on the edge, it is\nimperative to develop highly efficient low-power models that can be deployed on\nto small form-factor devices such as microcontrollers. In this paper, we\npropose a small CNN model with 4 layers that is very amenable for edge AI\ndeployment and realtime gait recognition. This model was trained on a public\ngait dataset with 20 classes augmented with data collected by the authors,\naggregating to 24 classes in total. Our model achieves 96.7% accuracy and\nconsumes only 5KB RAM with an inferencing time of 70 ms and 125mW power, while\nrunning continuous inference on Arduino Nano 33 BLE Sense. We successfully\ndemonstrated realtime identification of the authors with the model running on\nArduino, thus underscoring the efficacy and providing a proof of feasiblity for\ndeployment in practical systems in near future.\n", "date": "2024-04-02", "categories": []}
{"id": "2404.15385", "title": "Sum of Group Error Differences: A Critical Examination of Bias\n  Evaluation in Biometric Verification and a Dual-Metric Measure", "abstract": "  Biometric Verification (BV) systems often exhibit accuracy disparities across\ndifferent demographic groups, leading to biases in BV applications. Assessing\nand quantifying these biases is essential for ensuring the fairness of BV\nsystems. However, existing bias evaluation metrics in BV have limitations, such\nas focusing exclusively on match or non-match error rates, overlooking bias on\ndemographic groups with performance levels falling between the best and worst\nperformance levels, and neglecting the magnitude of the bias present.\n  This paper presents an in-depth analysis of the limitations of current bias\nevaluation metrics in BV and, through experimental analysis, demonstrates their\ncontextual suitability, merits, and limitations. Additionally, it introduces a\nnovel general-purpose bias evaluation measure for BV, the ``Sum of Group Error\nDifferences (SEDG)''. Our experimental results on controlled synthetic datasets\ndemonstrate the effectiveness of demographic bias quantification when using\nexisting metrics and our own proposed measure. We discuss the applicability of\nthe bias evaluation metrics in a set of simulated demographic bias scenarios\nand provide scenario-based metric recommendations. Our code is publicly\navailable under \\url{https://github.com/alaaobeid/SEDG}.\n", "date": "2024-04-23", "categories": []}
{"id": "2404.15394", "title": "On Generating Cancelable Biometric Template using Reverse of Boolean XOR", "abstract": "  Cancelable Biometric is repetitive distortion embedded in original Biometric\nimage for keeping it secure from unauthorized access. In this paper, we have\ngenerated Cancelable Biometric templates with Reverse Boolean XOR technique.\nThree different methods have been proposed for generation of Cancelable\nBiometric templates based on Visual Secret Sharing scheme. In each method, one\nSecret image and n-1 Cover images are used as: (M1) One original Biometric\nimage (Secret) with n- 1 randomly chosen Gray Cover images (M2) One original\nSecret image with n-1 Cover images, which are Randomly Permuted version of the\noriginal Secret image (M3) One Secret image with n-1 Cover images, both Secret\nimage and Cover images are Randomly Permuted version of original Biometric\nimage. Experiment works have performed on publicly available ORL Face database\nand IIT Delhi Iris database. The performance of the proposed methods is\ncompared in terms of Co-relation Coefficient (Cr), Mean Square Error (MSE),\nMean Absolute Error (MAE), Structural Similarity (SSIM), Peak Signal to Noise\nRatio (PSNR), Number of Pixel Change Rate (NPCR), and Unified Average Changing\nIntensity (UACI). It is found that among the three proposed method, M3\ngenerates good quality Cancelable templates and gives best performance in terms\nof quality. M3 is also better in quantitative terms on ORL dataset while M2 and\nM3 are comparable on IIT Delhi Iris dataset.\n", "date": "2024-04-23", "categories": []}
{"id": "2404.15608", "title": "Understanding and Improving CNNs with Complex Structure Tensor: A\n  Biometrics Study", "abstract": "  Our study provides evidence that CNNs struggle to effectively extract\norientation features. We show that the use of Complex Structure Tensor, which\ncontains compact orientation features with certainties, as input to CNNs\nconsistently improves identification accuracy compared to using grayscale\ninputs alone. Experiments also demonstrated that our inputs, which were\nprovided by mini complex conv-nets, combined with reduced CNN sizes,\noutperformed full-fledged, prevailing CNN architectures. This suggests that the\nupfront use of orientation features in CNNs, a strategy seen in mammalian\nvision, not only mitigates their limitations but also enhances their\nexplainability and relevance to thin-clients. Experiments were done on publicly\navailable data sets comprising periocular images for biometric identification\nand verification (Close and Open World) using 6 State of the Art CNN\narchitectures. We reduced SOA Equal Error Rate (EER) on the PolyU dataset by\n5-26% depending on data and scenario.\n", "date": "2024-04-24", "categories": []}
{"id": "2403.10558", "title": "Adaptive Hybrid Masking Strategy for Privacy-Preserving Face Recognition\n  Against Model Inversion Attack", "abstract": "  The utilization of personal sensitive data in training face recognition (FR)\nmodels poses significant privacy concerns, as adversaries can employ model\ninversion attacks (MIA) to infer the original training data. Existing defense\nmethods, such as data augmentation and differential privacy, have been employed\nto mitigate this issue. However, these methods often fail to strike an optimal\nbalance between privacy and accuracy. To address this limitation, this paper\nintroduces an adaptive hybrid masking algorithm against MIA. Specifically, face\nimages are masked in the frequency domain using an adaptive MixUp strategy.\nUnlike the traditional MixUp algorithm, which is predominantly used for data\naugmentation, our modified approach incorporates frequency domain mixing.\nPrevious studies have shown that increasing the number of images mixed in MixUp\ncan enhance privacy preservation but at the expense of reduced face recognition\naccuracy. To overcome this trade-off, we develop an enhanced adaptive MixUp\nstrategy based on reinforcement learning, which enables us to mix a larger\nnumber of images while maintaining satisfactory recognition accuracy. To\noptimize privacy protection, we propose maximizing the reward function (i.e.,\nthe loss function of the FR system) during the training of the strategy\nnetwork. While the loss function of the FR network is minimized in the phase of\ntraining the FR network. The strategy network and the face recognition network\ncan be viewed as antagonistic entities in the training process, ultimately\nreaching a more balanced trade-off. Experimental results demonstrate that our\nproposed hybrid masking scheme outperforms existing defense algorithms in terms\nof privacy preservation and recognition accuracy against MIA.\n", "date": "2024-03-14", "categories": []}
{"id": "2404.15234", "title": "Massively Annotated Datasets for Assessment of Synthetic and Real Data\n  in Face Recognition", "abstract": "  Face recognition applications have grown in parallel with the size of\ndatasets, complexity of deep learning models and computational power. However,\nwhile deep learning models evolve to become more capable and computational\npower keeps increasing, the datasets available are being retracted and removed\nfrom public access. Privacy and ethical concerns are relevant topics within\nthese domains. Through generative artificial intelligence, researchers have put\nefforts into the development of completely synthetic datasets that can be used\nto train face recognition systems. Nonetheless, the recent advances have not\nbeen sufficient to achieve performance comparable to the state-of-the-art\nmodels trained on real data. To study the drift between the performance of\nmodels trained on real and synthetic datasets, we leverage a massive attribute\nclassifier (MAC) to create annotations for four datasets: two real and two\nsynthetic. From these annotations, we conduct studies on the distribution of\neach attribute within all four datasets. Additionally, we further inspect the\ndifferences between real and synthetic datasets on the attribute set. When\ncomparing through the Kullback-Leibler divergence we have found differences\nbetween real and synthetic samples. Interestingly enough, we have verified that\nwhile real samples suffice to explain the synthetic distribution, the opposite\ncould not be further from being true.\n", "date": "2024-04-23", "categories": []}
{"id": "2403.04661", "title": "Dynamic Cross Attention for Audio-Visual Person Verification", "abstract": "  Although person or identity verification has been predominantly explored\nusing individual modalities such as face and voice, audio-visual fusion has\nrecently shown immense potential to outperform unimodal approaches. Audio and\nvisual modalities are often expected to pose strong complementary\nrelationships, which plays a crucial role in effective audio-visual fusion.\nHowever, they may not always strongly complement each other, they may also\nexhibit weak complementary relationships, resulting in poor audio-visual\nfeature representations. In this paper, we propose a Dynamic Cross-Attention\n(DCA) model that can dynamically select the cross-attended or unattended\nfeatures on the fly based on the strong or weak complementary relationships,\nrespectively, across audio and visual modalities. In particular, a conditional\ngating layer is designed to evaluate the contribution of the cross-attention\nmechanism and choose cross-attended features only when they exhibit strong\ncomplementary relationships, otherwise unattended features. Extensive\nexperiments are conducted on the Voxceleb1 dataset to demonstrate the\nrobustness of the proposed model. Results indicate that the proposed model\nconsistently improves the performance on multiple variants of cross-attention\nwhile outperforming the state-of-the-art methods.\n", "date": "2024-03-07", "categories": []}
{"id": "2404.13237", "title": "PAFedFV: Personalized and Asynchronous Federated Learning for Finger\n  Vein Recognition", "abstract": "  With the increasing emphasis on user privacy protection, biometric\nrecognition based on federated learning have become the latest research\nhotspot. However, traditional federated learning methods cannot be directly\napplied to finger vein recognition, due to heterogeneity of data and open-set\nverification. Therefore, only a few application cases have been proposed. And\nthese methods still have two drawbacks. (1) Uniform model results in poor\nperformance in some clients, as the finger vein data is highly heterogeneous\nand non-Independently Identically Distributed (non-IID). (2) On individual\nclient, a large amount of time is underutilized, such as the time to wait for\nreturning model from server. To address those problems, this paper proposes a\nPersonalized and Asynchronous Federated Learning for Finger Vein Recognition\n(PAFedFV) framework. PAFedFV designs personalized model aggregation method to\nsolve the heterogeneity among non-IID data. Meanwhile, it employs an\nasynchronized training module for clients to utilize their waiting time.\nFinally, extensive experiments on six finger vein datasets are conducted. Base\non these experiment results, the impact of non-IID finger vein data on\nperformance of federated learning are analyzed, and the superiority of PAFedFV\nin accuracy and robustness are demonstrated.\n", "date": "2024-04-20", "categories": []}
{"id": "2404.13791", "title": "Universal Fingerprint Generation: Controllable Diffusion Model with\n  Multimodal Conditions", "abstract": "  The utilization of synthetic data for fingerprint recognition has garnered\nincreased attention due to its potential to alleviate privacy concerns\nsurrounding sensitive biometric data. However, current methods for generating\nfingerprints have limitations in creating impressions of the same finger with\nuseful intra-class variations. To tackle this challenge, we present GenPrint, a\nframework to produce fingerprint images of various types while maintaining\nidentity and offering humanly understandable control over different appearance\nfactors such as fingerprint class, acquisition type, sensor device, and quality\nlevel. Unlike previous fingerprint generation approaches, GenPrint is not\nconfined to replicating style characteristics from the training dataset alone:\nit enables the generation of novel styles from unseen devices without requiring\nadditional fine-tuning. To accomplish these objectives, we developed GenPrint\nusing latent diffusion models with multimodal conditions (text and image) for\nconsistent generation of style and identity. Our experiments leverage a variety\nof publicly available datasets for training and evaluation. Results demonstrate\nthe benefits of GenPrint in terms of identity preservation, explainable\ncontrol, and universality of generated images. Importantly, the\nGenPrint-generated images yield comparable or even superior accuracy to models\ntrained solely on real data and further enhances performance when augmenting\nthe diversity of existing real fingerprint datasets.\n", "date": "2024-04-21", "categories": []}
{"id": "2404.13827", "title": "Swap It Like Its Hot: Segmentation-based spoof attacks on eye-tracking\n  images", "abstract": "  Video-based eye trackers capture the iris biometric and enable authentication\nto secure user identity. However, biometric authentication is susceptible to\nspoofing another user's identity through physical or digital manipulation. The\ncurrent standard to identify physical spoofing attacks on eye-tracking sensors\nuses liveness detection. Liveness detection classifies gaze data as real or\nfake, which is sufficient to detect physical presentation attacks. However,\nsuch defenses cannot detect a spoofing attack when real eye image inputs are\ndigitally manipulated to swap the iris pattern of another person. We propose\nIrisSwap as a novel attack on gaze-based liveness detection. IrisSwap allows\nattackers to segment and digitally swap in a victim's iris pattern to fool iris\nauthentication. Both offline and online attacks produce gaze data that deceives\nthe current state-of-the-art defense models at rates up to 58% and motivates\nthe need to develop more advanced authentication methods for eye trackers.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.14343", "title": "Heterogeneous Face Recognition Using Domain Invariant Units", "abstract": "  Heterogeneous Face Recognition (HFR) aims to expand the applicability of Face\nRecognition (FR) systems to challenging scenarios, enabling the matching of\nface images across different domains, such as matching thermal images to\nvisible spectra. However, the development of HFR systems is challenging because\nof the significant domain gap between modalities and the lack of availability\nof large-scale paired multi-channel data. In this work, we leverage a\npretrained face recognition model as a teacher network to learn domaininvariant\nnetwork layers called Domain-Invariant Units (DIU) to reduce the domain gap.\nThe proposed DIU can be trained effectively even with a limited amount of\npaired training data, in a contrastive distillation framework. This proposed\napproach has the potential to enhance pretrained models, making them more\nadaptable to a wider range of variations in data. We extensively evaluate our\napproach on multiple challenging benchmarks, demonstrating superior performance\ncompared to state-of-the-art methods.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.14406", "title": "Hyp-OC: Hyperbolic One Class Classification for Face Anti-Spoofing", "abstract": "  Face recognition technology has become an integral part of modern security\nsystems and user authentication processes. However, these systems are\nvulnerable to spoofing attacks and can easily be circumvented. Most prior\nresearch in face anti-spoofing (FAS) approaches it as a two-class\nclassification task where models are trained on real samples and known spoof\nattacks and tested for detection performance on unknown spoof attacks. However,\nin practice, FAS should be treated as a one-class classification task where,\nwhile training, one cannot assume any knowledge regarding the spoof samples a\npriori. In this paper, we reformulate the face anti-spoofing task from a\none-class perspective and propose a novel hyperbolic one-class classification\nframework. To train our network, we use a pseudo-negative class sampled from\nthe Gaussian distribution with a weighted running mean and propose two novel\nloss functions: (1) Hyp-PC: Hyperbolic Pairwise Confusion loss, and (2) Hyp-CE:\nHyperbolic Cross Entropy loss, which operate in the hyperbolic space.\nAdditionally, we employ Euclidean feature clipping and gradient clipping to\nstabilize the training in the hyperbolic space. To the best of our knowledge,\nthis is the first work extending hyperbolic embeddings for face anti-spoofing\nin a one-class manner. With extensive experiments on five benchmark datasets:\nRose-Youtu, MSU-MFSD, CASIA-MFSD, Idiap Replay-Attack, and OULU-NPU, we\ndemonstrate that our method significantly outperforms the state-of-the-art,\nachieving better spoof detection performance.\n", "date": "2024-04-22", "categories": []}
{"id": "2404.12611", "title": "Rethinking Clothes Changing Person ReID: Conflicts, Synthesis, and\n  Optimization", "abstract": "  Clothes-changing person re-identification (CC-ReID) aims to retrieve images\nof the same person wearing different outfits. Mainstream researches focus on\ndesigning advanced model structures and strategies to capture identity\ninformation independent of clothing. However, the same-clothes discrimination\nas the standard ReID learning objective in CC-ReID is persistently ignored in\nprevious researches. In this study, we dive into the relationship between\nstandard and clothes-changing~(CC) learning objectives, and bring the inner\nconflicts between these two objectives to the fore. We try to magnify the\nproportion of CC training pairs by supplementing high-fidelity clothes-varying\nsynthesis, produced by our proposed Clothes-Changing Diffusion model. By\nincorporating the synthetic images into CC-ReID model training, we observe a\nsignificant improvement under CC protocol. However, such improvement sacrifices\nthe performance under the standard protocol, caused by the inner conflict\nbetween standard and CC. For conflict mitigation, we decouple these objectives\nand re-formulate CC-ReID learning as a multi-objective optimization (MOO)\nproblem. By effectively regularizing the gradient curvature across multiple\nobjectives and introducing preference restrictions, our MOO solution surpasses\nthe single-task training paradigm. Our framework is model-agnostic, and\ndemonstrates superior performance under both CC and standard ReID protocols.\n", "date": "2024-04-19", "categories": []}
{"id": "2404.12680", "title": "VoxAtnNet: A 3D Point Clouds Convolutional Neural Network for\n  Generalizable Face Presentation Attack Detection", "abstract": "  Facial biometrics are an essential components of smartphones to ensure\nreliable and trustworthy authentication. However, face biometric systems are\nvulnerable to Presentation Attacks (PAs), and the availability of more\nsophisticated presentation attack instruments such as 3D silicone face masks\nwill allow attackers to deceive face recognition systems easily. In this work,\nwe propose a novel Presentation Attack Detection (PAD) algorithm based on 3D\npoint clouds captured using the frontal camera of a smartphone to detect\npresentation attacks. The proposed PAD algorithm, VoxAtnNet, processes 3D point\nclouds to obtain voxelization to preserve the spatial structure. Then, the\nvoxelized 3D samples were trained using the novel convolutional attention\nnetwork to detect PAs on the smartphone. Extensive experiments were carried out\non the newly constructed 3D face point cloud dataset comprising bona fide and\ntwo different 3D PAIs (3D silicone face mask and wrap photo mask), resulting in\n3480 samples. The performance of the proposed method was compared with existing\nmethods to benchmark the detection performance using three different evaluation\nprotocols. The experimental results demonstrate the improved performance of the\nproposed method in detecting both known and unknown face presentation attacks.\n", "date": "2024-04-19", "categories": []}
{"id": "2404.06211", "title": "Unified Physical-Digital Attack Detection Challenge", "abstract": "  Face Anti-Spoofing (FAS) is crucial to safeguard Face Recognition (FR)\nSystems. In real-world scenarios, FRs are confronted with both physical and\ndigital attacks. However, existing algorithms often address only one type of\nattack at a time, which poses significant limitations in real-world scenarios\nwhere FR systems face hybrid physical-digital threats. To facilitate the\nresearch of Unified Attack Detection (UAD) algorithms, a large-scale\nUniAttackData dataset has been collected. UniAttackData is the largest public\ndataset for Unified Attack Detection, with a total of 28,706 videos, where each\nunique identity encompasses all advanced attack types. Based on this dataset,\nwe organized a Unified Physical-Digital Face Attack Detection Challenge to\nboost the research in Unified Attack Detections. It attracted 136 teams for the\ndevelopment phase, with 13 qualifying for the final round. The results\nre-verified by the organizing team were used for the final ranking. This paper\ncomprehensively reviews the challenge, detailing the dataset introduction,\nprotocol definition, evaluation criteria, and a summary of published results.\nFinally, we focus on the detailed analysis of the highest-performing algorithms\nand offer potential directions for unified physical-digital attack detection\ninspired by this competition. Challenge Website:\nhttps://sites.google.com/view/face-anti-spoofing-challenge/welcome/challengecvpr2024.\n", "date": "2024-04-09", "categories": []}
{"id": "2404.08277", "title": "FaceFilterSense: A Filter-Resistant Face Recognition and Facial\n  Attribute Analysis Framework", "abstract": "  With the advent of social media, fun selfie filters have come into tremendous\nmainstream use affecting the functioning of facial biometric systems as well as\nimage recognition systems. These filters vary from beautification filters and\nAugmented Reality (AR)-based filters to filters that modify facial landmarks.\nHence, there is a need to assess the impact of such filters on the performance\nof existing face recognition systems. The limitation associated with existing\nsolutions is that these solutions focus more on the beautification filters.\nHowever, the current AR-based filters and filters which distort facial key\npoints are in vogue recently and make the faces highly unrecognizable even to\nthe naked eye. Also, the filters considered are mostly obsolete with limited\nvariations. To mitigate these limitations, we aim to perform a holistic impact\nanalysis of the latest filters and propose an user recognition model with the\nfiltered images. We have utilized a benchmark dataset for baseline images, and\napplied the latest filters over them to generate a beautified/filtered dataset.\nNext, we have introduced a model FaceFilterNet for beautified user recognition.\nIn this framework, we also utilize our model to comment on various attributes\nof the person including age, gender, and ethnicity. In addition, we have also\npresented a filter-wise impact analysis on face recognition, age estimation,\ngender, and ethnicity prediction. The proposed method affirms the efficacy of\nour dataset with an accuracy of 87.25% and an optimal accuracy for facial\nattribute analysis.\n", "date": "2024-04-12", "categories": []}
{"id": "2404.08237", "title": "IFViT: Interpretable Fixed-Length Representation for Fingerprint\n  Matching via Vision Transformer", "abstract": "  Determining dense feature points on fingerprints used in constructing deep\nfixed-length representations for accurate matching, particularly at the pixel\nlevel, is of significant interest. To explore the interpretability of\nfingerprint matching, we propose a multi-stage interpretable fingerprint\nmatching network, namely Interpretable Fixed-length Representation for\nFingerprint Matching via Vision Transformer (IFViT), which consists of two\nprimary modules. The first module, an interpretable dense registration module,\nestablishes a Vision Transformer (ViT)-based Siamese Network to capture\nlong-range dependencies and the global context in fingerprint pairs. It\nprovides interpretable dense pixel-wise correspondences of feature points for\nfingerprint alignment and enhances the interpretability in the subsequent\nmatching stage. The second module takes into account both local and global\nrepresentations of the aligned fingerprint pair to achieve an interpretable\nfixed-length representation extraction and matching. It employs the ViTs\ntrained in the first module with the additional fully connected layer and\nretrains them to simultaneously produce the discriminative fixed-length\nrepresentation and interpretable dense pixel-wise correspondences of feature\npoints. Extensive experimental results on diverse publicly available\nfingerprint databases demonstrate that the proposed framework not only exhibits\nsuperior performance on dense registration and matching but also significantly\npromotes the interpretability in deep fixed-length representations-based\nfingerprint matching.\n", "date": "2024-04-12", "categories": []}
{"id": "2404.08450", "title": "Joint Physical-Digital Facial Attack Detection Via Simulating Spoofing\n  Clues", "abstract": "  Face recognition systems are frequently subjected to a variety of physical\nand digital attacks of different types. Previous methods have achieved\nsatisfactory performance in scenarios that address physical attacks and digital\nattacks, respectively. However, few methods are considered to integrate a model\nthat simultaneously addresses both physical and digital attacks, implying the\nnecessity to develop and maintain multiple models. To jointly detect physical\nand digital attacks within a single model, we propose an innovative approach\nthat can adapt to any network architecture. Our approach mainly contains two\ntypes of data augmentation, which we call Simulated Physical Spoofing Clues\naugmentation (SPSC) and Simulated Digital Spoofing Clues augmentation (SDSC).\nSPSC and SDSC augment live samples into simulated attack samples by simulating\nspoofing clues of physical and digital attacks, respectively, which\nsignificantly improve the capability of the model to detect \"unseen\" attack\ntypes. Extensive experiments show that SPSC and SDSC can achieve\nstate-of-the-art generalization in Protocols 2.1 and 2.2 of the UniAttackData\ndataset, respectively. Our method won first place in \"Unified Physical-Digital\nFace Attack Detection\" of the 5th Face Anti-spoofing Challenge@CVPR2024. Our\nfinal submission obtains 3.75% APCER, 0.93% BPCER, and 2.34% ACER,\nrespectively. Our code is available at\nhttps://github.com/Xianhua-He/cvpr2024-face-anti-spoofing-challenge.\n", "date": "2024-04-12", "categories": []}
{"id": "2404.07930", "title": "Parameter Hierarchical Optimization for Visible-Infrared Person\n  Re-Identification", "abstract": "  Visible-infrared person re-identification (VI-reID) aims at matching\ncross-modality pedestrian images captured by disjoint visible or infrared\ncameras. Existing methods alleviate the cross-modality discrepancies via\ndesigning different kinds of network architectures. Different from available\nmethods, in this paper, we propose a novel parameter optimizing paradigm,\nparameter hierarchical optimization (PHO) method, for the task of VI-ReID. It\nallows part of parameters to be directly optimized without any training, which\nnarrows the search space of parameters and makes the whole network more easier\nto be trained. Specifically, we first divide the parameters into different\ntypes, and then introduce a self-adaptive alignment strategy (SAS) to\nautomatically align the visible and infrared images through transformation.\nConsidering that features in different dimension have varying importance, we\ndevelop an auto-weighted alignment learning (AAL) module that can automatically\nweight features according to their importance. Importantly, in the alignment\nprocess of SAS and AAL, all the parameters are immediately optimized with\noptimization principles rather than training the whole network, which yields a\nbetter parameter training manner. Furthermore, we establish the cross-modality\nconsistent learning (CCL) loss to extract discriminative person representations\nwith translation consistency. We provide both theoretical justification and\nempirical evidence that our proposed PHO method outperform existing VI-reID\napproaches.\n", "date": "2024-04-11", "categories": []}
{"id": "2301.04218", "title": "Leveraging Diffusion For Strong and High Quality Face Morphing Attacks", "abstract": "  Face morphing attacks seek to deceive a Face Recognition (FR) system by\npresenting a morphed image consisting of the biometric qualities from two\ndifferent identities with the aim of triggering a false acceptance with one of\nthe two identities, thereby presenting a significant threat to biometric\nsystems. The success of a morphing attack is dependent on the ability of the\nmorphed image to represent the biometric characteristics of both identities\nthat were used to create the image. We present a novel morphing attack that\nuses a Diffusion-based architecture to improve the visual fidelity of the image\nand the ability of the morphing attack to represent characteristics from both\nidentities. We demonstrate the effectiveness of the proposed attack by\nevaluating its visual fidelity via the Frechet Inception Distance (FID). Also,\nextensive experiments are conducted to measure the vulnerability of FR systems\nto the proposed attack. The ability of a morphing attack detector to detect the\nproposed attack is measured and compared against two state-of-the-art GAN-based\nmorphing attacks along with two Landmark-based attacks. Additionally, a novel\nmetric to measure the relative strength between different morphing attacks is\nintroduced and evaluated.\n", "date": "2023-01-10", "categories": []}
{"id": "2403.08092", "title": "Mitigating the Impact of Attribute Editing on Face Recognition", "abstract": "  Through a large-scale study over diverse face images, we show that facial\nattribute editing using modern generative AI models can severely degrade\nautomated face recognition systems. This degradation persists even with\nidentity-preserving generative models. To mitigate this issue, we propose two\nnovel techniques for local and global attribute editing. We empirically ablate\ntwenty-six facial semantic, demographic and expression-based attributes that\nhave been edited using state-of-the-art generative models, and evaluate them\nusing ArcFace and AdaFace matchers on CelebA, CelebAMaskHQ and LFW datasets.\nFinally, we use LLaVA, an emerging visual question-answering framework for\nattribute prediction to validate our editing techniques. Our methods outperform\nthe current state-of-the-art at facial editing (BLIP, InstantID) while\nimproving identity retention by a significant extent.\n", "date": "2024-03-12", "categories": []}
{"id": "2404.06683", "title": "Unsupervised Visible-Infrared ReID via Pseudo-label Correction and\n  Modality-level Alignment", "abstract": "  Unsupervised visible-infrared person re-identification (UVI-ReID) has\nrecently gained great attention due to its potential for enhancing human\ndetection in diverse environments without labeling. Previous methods utilize\nintra-modality clustering and cross-modality feature matching to achieve\nUVI-ReID. However, there exist two challenges: 1) noisy pseudo labels might be\ngenerated in the clustering process, and 2) the cross-modality feature\nalignment via matching the marginal distribution of visible and infrared\nmodalities may misalign the different identities from two modalities. In this\npaper, we first conduct a theoretic analysis where an interpretable\ngeneralization upper bound is introduced. Based on the analysis, we then\npropose a novel unsupervised cross-modality person re-identification framework\n(PRAISE). Specifically, to address the first challenge, we propose a\npseudo-label correction strategy that utilizes a Beta Mixture Model to predict\nthe probability of mis-clustering based network's memory effect and rectifies\nthe correspondence by adding a perceptual term to contrastive learning. Next,\nwe introduce a modality-level alignment strategy that generates paired\nvisible-infrared latent features and reduces the modality gap by aligning the\nlabeling function of visible and infrared features to learn identity\ndiscriminative and modality-invariant features. Experimental results on two\nbenchmark datasets demonstrate that our method achieves state-of-the-art\nperformance than the unsupervised visible-ReID methods.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.06963", "title": "V-MAD: Video-based Morphing Attack Detection in Operational Scenarios", "abstract": "  In response to the rising threat of the face morphing attack, this paper\nintroduces and explores the potential of Video-based Morphing Attack Detection\n(V-MAD) systems in real-world operational scenarios. While current morphing\nattack detection methods primarily focus on a single or a pair of images, V-MAD\nis based on video sequences, exploiting the video streams often acquired by\nface verification tools available, for instance, at airport gates. Through this\nstudy, we show for the first time the advantages that the availability of\nmultiple probe frames can bring to the morphing attack detection task,\nespecially in scenarios where the quality of probe images is varied and might\nbe affected, for instance, by pose or illumination variations. Experimental\nresults on a real operational database demonstrate that video sequences\nrepresent valuable information for increasing the robustness and performance of\nmorphing attack detection systems.\n", "date": "2024-04-10", "categories": []}
{"id": "2404.04580", "title": "SDFR: Synthetic Data for Face Recognition Competition", "abstract": "  Large-scale face recognition datasets are collected by crawling the Internet\nand without individuals' consent, raising legal, ethical, and privacy concerns.\nWith the recent advances in generative models, recently several works proposed\ngenerating synthetic face recognition datasets to mitigate concerns in\nweb-crawled face recognition datasets. This paper presents the summary of the\nSynthetic Data for Face Recognition (SDFR) Competition held in conjunction with\nthe 18th IEEE International Conference on Automatic Face and Gesture\nRecognition (FG 2024) and established to investigate the use of synthetic data\nfor training face recognition models. The SDFR competition was split into two\ntasks, allowing participants to train face recognition systems using new\nsynthetic datasets and/or existing ones. In the first task, the face\nrecognition backbone was fixed and the dataset size was limited, while the\nsecond task provided almost complete freedom on the model backbone, the\ndataset, and the training pipeline. The submitted models were trained on\nexisting and also new synthetic datasets and used clever methods to improve\ntraining with synthetic data. The submissions were evaluated and ranked on a\ndiverse set of seven benchmarking datasets. The paper gives an overview of the\nsubmitted face recognition models and reports achieved performance compared to\nbaseline models trained on real and synthetic datasets. Furthermore, the\nevaluation of submissions is extended to bias assessment across different\ndemography groups. Lastly, an outlook on the current state of the research in\ntraining face recognition models using synthetic data is presented, and\nexisting problems as well as potential future directions are also discussed.\n", "date": "2024-04-06", "categories": []}
{"id": "2305.13600", "title": "SiCL: Silhouette-Driven Contrastive Learning for Unsupervised Person\n  Re-Identification with Clothes Change", "abstract": "  In this paper, we address a highly challenging yet critical task:\nunsupervised long-term person re-identification with clothes change. Existing\nunsupervised person re-id methods are mainly designed for short-term scenarios\nand usually rely on RGB cues so that fail to perceive feature patterns that are\nindependent of the clothes. To crack this bottleneck, we propose a\nsilhouette-driven contrastive learning (SiCL) method, which is designed to\nlearn cross-clothes invariance by integrating both the RGB cues and the\nsilhouette information within a contrastive learning framework. To our\nknowledge, this is the first tailor-made framework for unsupervised long-term\nclothes change \\reid{}, with superior performance on six benchmark datasets. We\nconduct extensive experiments to evaluate our proposed SiCL compared to the\nstate-of-the-art unsupervised person reid methods across all the representative\ndatasets. Experimental results demonstrate that our proposed SiCL significantly\noutperforms other unsupervised re-id methods.\n", "date": "2023-05-23", "categories": []}
{"id": "2401.02634", "title": "AG-ReID.v2: Bridging Aerial and Ground Views for Person\n  Re-identification", "abstract": "  Aerial-ground person re-identification (Re-ID) presents unique challenges in\ncomputer vision, stemming from the distinct differences in viewpoints, poses,\nand resolutions between high-altitude aerial and ground-based cameras. Existing\nresearch predominantly focuses on ground-to-ground matching, with aerial\nmatching less explored due to a dearth of comprehensive datasets. To address\nthis, we introduce AG-ReID.v2, a dataset specifically designed for person Re-ID\nin mixed aerial and ground scenarios. This dataset comprises 100,502 images of\n1,615 unique individuals, each annotated with matching IDs and 15 soft\nattribute labels. Data were collected from diverse perspectives using a UAV,\nstationary CCTV, and smart glasses-integrated camera, providing a rich variety\nof intra-identity variations. Additionally, we have developed an explainable\nattention network tailored for this dataset. This network features a\nthree-stream architecture that efficiently processes pairwise image distances,\nemphasizes key top-down features, and adapts to variations in appearance due to\naltitude differences. Comparative evaluations demonstrate the superiority of\nour approach over existing baselines. We plan to release the dataset and\nalgorithm source code publicly, aiming to advance research in this specialized\nfield of computer vision. For access, please visit\nhttps://github.com/huynguyen792/AG-ReID.v2.\n", "date": "2024-01-05", "categories": []}
{"id": "2312.01734", "title": "Effective Adapter for Face Recognition in the Wild", "abstract": "  In this paper, we tackle the challenge of face recognition in the wild, where\nimages often suffer from low quality and real-world distortions. Traditional\nheuristic approaches-either training models directly on these degraded images\nor their enhanced counterparts using face restoration techniques-have proven\nineffective, primarily due to the degradation of facial features and the\ndiscrepancy in image domains. To overcome these issues, we propose an effective\nadapter for augmenting existing face recognition models trained on high-quality\nfacial datasets. The key of our adapter is to process both the unrefined and\nenhanced images using two similar structures, one fixed and the other\ntrainable. Such design can confer two benefits. First, the dual-input system\nminimizes the domain gap while providing varied perspectives for the face\nrecognition model, where the enhanced image can be regarded as a complex\nnon-linear transformation of the original one by the restoration model. Second,\nboth two similar structures can be initialized by the pre-trained models\nwithout dropping the past knowledge. The extensive experiments in zero-shot\nsettings show the effectiveness of our method by surpassing baselines of about\n3%, 4%, and 7% in three datasets. Our code will be publicly available.\n", "date": "2023-12-04", "categories": []}
{"id": "2404.02345", "title": "GaitSTR: Gait Recognition with Sequential Two-stream Refinement", "abstract": "  Gait recognition aims to identify a person based on their walking sequences,\nserving as a useful biometric modality as it can be observed from long\ndistances without requiring cooperation from the subject. In representing a\nperson's walking sequence, silhouettes and skeletons are the two primary\nmodalities used. Silhouette sequences lack detailed part information when\noverlapping occurs between different body segments and are affected by carried\nobjects and clothing. Skeletons, comprising joints and bones connecting the\njoints, provide more accurate part information for different segments; however,\nthey are sensitive to occlusions and low-quality images, causing\ninconsistencies in frame-wise results within a sequence. In this paper, we\nexplore the use of a two-stream representation of skeletons for gait\nrecognition, alongside silhouettes. By fusing the combined data of silhouettes\nand skeletons, we refine the two-stream skeletons, joints, and bones through\nself-correction in graph convolution, along with cross-modal correction with\ntemporal consistency from silhouettes. We demonstrate that with refined\nskeletons, the performance of the gait recognition model can achieve further\nimprovement on public gait recognition datasets compared with state-of-the-art\nmethods without extra annotations.\n", "date": "2024-04-02", "categories": []}
{"id": "2403.16003", "title": "Diverse Representation Embedding for Lifelong Person Re-Identification", "abstract": "  Lifelong Person Re-Identification (LReID) aims to continuously learn from\nsuccessive data streams, matching individuals across multiple cameras. The key\nchallenge for LReID is how to effectively preserve old knowledge while\nincrementally learning new information, which is caused by task-level domain\ngaps and limited old task datasets. Existing methods based on CNN backbone are\ninsufficient to explore the representation of each instance from different\nperspectives, limiting model performance on limited old task datasets and new\ntask datasets. Unlike these methods, we propose a Diverse Representations\nEmbedding (DRE) framework that first explores a pure transformer for LReID. The\nproposed DRE preserves old knowledge while adapting to new information based on\ninstance-level and task-level layout. Concretely, an Adaptive Constraint Module\n(ACM) is proposed to implement integration and push away operations between\nmultiple overlapping representations generated by transformer-based backbone,\nobtaining rich and discriminative representations for each instance to improve\nadaptive ability of LReID. Based on the processed diverse representations, we\npropose Knowledge Update (KU) and Knowledge Preservation (KP) strategies at the\ntask-level layout by introducing the adjustment model and the learner model. KU\nstrategy enhances the adaptive learning ability of learner models for new\ninformation under the adjustment model prior, and KP strategy preserves old\nknowledge operated by representation-level alignment and logit-level\nsupervision in limited old task datasets while guaranteeing the adaptive\nlearning information capacity of the LReID model. Compared to state-of-the-art\nmethods, our method achieves significantly improved performance in holistic,\nlarge-scale, and occluded datasets.\n", "date": "2024-03-24", "categories": []}
{"id": "2402.03769", "title": "AttackNet: Enhancing Biometric Security via Tailored Convolutional\n  Neural Network Architectures for Liveness Detection", "abstract": "  Biometric security is the cornerstone of modern identity verification and\nauthentication systems, where the integrity and reliability of biometric\nsamples is of paramount importance. This paper introduces AttackNet, a bespoke\nConvolutional Neural Network architecture, meticulously designed to combat\nspoofing threats in biometric systems. Rooted in deep learning methodologies,\nthis model offers a layered defense mechanism, seamlessly transitioning from\nlow-level feature extraction to high-level pattern discernment. Three\ndistinctive architectural phases form the crux of the model, each underpinned\nby judiciously chosen activation functions, normalization techniques, and\ndropout layers to ensure robustness and resilience against adversarial attacks.\nBenchmarking our model across diverse datasets affirms its prowess, showcasing\nsuperior performance metrics in comparison to contemporary models. Furthermore,\na detailed comparative analysis accentuates the model's efficacy, drawing\nparallels with prevailing state-of-the-art methodologies. Through iterative\nrefinement and an informed architectural strategy, AttackNet underscores the\npotential of deep learning in safeguarding the future of biometric security.\n", "date": "2024-02-06", "categories": []}
{"id": "2404.00107", "title": "Robust Ensemble Person Re-Identification via Orthogonal Fusion with\n  Occlusion Handling", "abstract": "  Occlusion remains one of the major challenges in person reidentification\n(ReID) as a result of the diversity of poses and the variation of appearances.\nDeveloping novel architectures to improve the robustness of occlusion-aware\nperson Re-ID requires new insights, especially on low-resolution edge cameras.\nWe propose a deep ensemble model that harnesses both CNN and Transformer\narchitectures to generate robust feature representations. To achieve robust\nRe-ID without the need to manually label occluded regions, we propose to take\nan ensemble learning-based approach derived from the analogy between\narbitrarily shaped occluded regions and robust feature representation. Using\nthe orthogonality principle, our developed deep CNN model makes use of masked\nautoencoder (MAE) and global-local feature fusion for robust person\nidentification. Furthermore, we present a part occlusion-aware transformer\ncapable of learning feature space that is robust to occluded regions.\nExperimental results are reported on several Re-ID datasets to show the\neffectiveness of our developed ensemble model named orthogonal fusion with\nocclusion handling (OFOH). Compared to competing methods, the proposed OFOH\napproach has achieved competent rank-1 and mAP performance.\n", "date": "2024-03-29", "categories": []}
{"id": "2404.00172", "title": "Universal Bovine Identification via Depth Data and Deep Metric Learning", "abstract": "  This paper proposes and evaluates, for the first time, a top-down (dorsal\nview), depth-only deep learning system for accurately identifying individual\ncattle and provides associated code, datasets, and training weights for\nimmediate reproducibility. An increase in herd size skews the cow-to-human\nratio at the farm and makes the manual monitoring of individuals more\nchallenging. Therefore, real-time cattle identification is essential for the\nfarms and a crucial step towards precision livestock farming. Underpinned by\nour previous work, this paper introduces a deep-metric learning method for\ncattle identification using depth data from an off-the-shelf 3D camera. The\nmethod relies on CNN and MLP backbones that learn well-generalised embedding\nspaces from the body shape to differentiate individuals -- requiring neither\nspecies-specific coat patterns nor close-up muzzle prints for operation. The\nnetwork embeddings are clustered using a simple algorithm such as $k$-NN for\nhighly accurate identification, thus eliminating the need to retrain the\nnetwork for enrolling new individuals. We evaluate two backbone architectures,\nResNet, as previously used to identify Holstein Friesians using RGB images, and\nPointNet, which is specialised to operate on 3D point clouds. We also present\nCowDepth2023, a new dataset containing 21,490 synchronised colour-depth image\npairs of 99 cows, to evaluate the backbones. Both ResNet and PointNet\narchitectures, which consume depth maps and point clouds, respectively, led to\nhigh accuracy that is on par with the coat pattern-based backbone.\n", "date": "2024-03-29", "categories": []}
{"id": "2403.08270", "title": "Identity-aware Dual-constraint Network for Cloth-Changing Person\n  Re-identification", "abstract": "  Cloth-Changing Person Re-Identification (CC-ReID) aims to accurately identify\nthe target person in more realistic surveillance scenarios, where pedestrians\nusually change their clothing. Despite great progress, limited cloth-changing\ntraining samples in existing CC-ReID datasets still prevent the model from\nadequately learning cloth-irrelevant features. In addition, due to the absence\nof explicit supervision to keep the model constantly focused on\ncloth-irrelevant areas, existing methods are still hampered by the disruption\nof clothing variations. To solve the above issues, we propose an Identity-aware\nDual-constraint Network (IDNet) for the CC-ReID task. Specifically, to help the\nmodel extract cloth-irrelevant clues, we propose a Clothes Diversity\nAugmentation (CDA), which generates more realistic cloth-changing samples by\nenriching the clothing color while preserving the texture. In addition, a\nMulti-scale Constraint Block (MCB) is designed, which extracts fine-grained\nidentity-related features and effectively transfers cloth-irrelevant knowledge.\nMoreover, a Counterfactual-guided Attention Module (CAM) is presented, which\nlearns cloth-irrelevant features from channel and space dimensions and utilizes\nthe counterfactual intervention for supervising the attention map to highlight\nidentity-related regions. Finally, a Semantic Alignment Constraint (SAC) is\ndesigned to facilitate high-level semantic feature interaction. Comprehensive\nexperiments on four CC-ReID datasets indicate that our method outperforms prior\nstate-of-the-art approaches.\n", "date": "2024-03-13", "categories": []}
{"id": "2403.11708", "title": "Implicit Discriminative Knowledge Learning for Visible-Infrared Person\n  Re-Identification", "abstract": "  Visible-Infrared Person Re-identification (VI-ReID) is a challenging\ncross-modal pedestrian retrieval task, due to significant intra-class\nvariations and cross-modal discrepancies among different cameras. Existing\nworks mainly focus on embedding images of different modalities into a unified\nspace to mine modality-shared features. They only seek distinctive information\nwithin these shared features, while ignoring the identity-aware useful\ninformation that is implicit in the modality-specific features. To address this\nissue, we propose a novel Implicit Discriminative Knowledge Learning (IDKL)\nnetwork to uncover and leverage the implicit discriminative information\ncontained within the modality-specific. First, we extract modality-specific and\nmodality-shared features using a novel dual-stream network. Then, the\nmodality-specific features undergo purification to reduce their modality style\ndiscrepancies while preserving identity-aware discriminative knowledge.\nSubsequently, this kind of implicit knowledge is distilled into the\nmodality-shared feature to enhance its distinctiveness. Finally, an alignment\nloss is proposed to minimize modality discrepancy on enhanced modality-shared\nfeatures. Extensive experiments on multiple public datasets demonstrate the\nsuperiority of IDKL network over the state-of-the-art methods. Code is\navailable at https://github.com/1KK077/IDKL.\n", "date": "2024-03-18", "categories": []}
{"id": "2303.01656", "title": "Feature Completion Transformer for Occluded Person Re-identification", "abstract": "  Occluded person re-identification (Re-ID) is a challenging problem due to the\ndestruction of occluders. Most existing methods focus on visible human body\nparts through some prior information. However, when complementary occlusions\noccur, features in occluded regions can interfere with matching, which affects\nperformance severely. In this paper, different from most previous works that\ndiscard the occluded region, we propose a Feature Completion Transformer\n(FCFormer) to implicitly complement the semantic information of occluded parts\nin the feature space. Specifically, Occlusion Instance Augmentation (OIA) is\nproposed to simulates real and diverse occlusion situations on the holistic\nimage. These augmented images not only enrich the amount of occlusion samples\nin the training set, but also form pairs with the holistic images.\nSubsequently, a dual-stream architecture with a shared encoder is proposed to\nlearn paired discriminative features from pairs of inputs. Without additional\nsemantic information, an occluded-holistic feature sample-label pair can be\nautomatically created. Then, Feature Completion Decoder (FCD) is designed to\ncomplement the features of occluded regions by using learnable tokens to\naggregate possible information from self-generated occluded features. Finally,\nwe propose the Cross Hard Triplet (CHT) loss to further bridge the gap between\ncomplementing features and extracting features under the same ID. In addition,\nFeature Completion Consistency (FC$^2$) loss is introduced to help the\ngenerated completion feature distribution to be closer to the real holistic\nfeature distribution. Extensive experiments over five challenging datasets\ndemonstrate that the proposed FCFormer achieves superior performance and\noutperforms the state-of-the-art methods by significant margins on occluded\ndatasets.\n", "date": "2023-03-03", "categories": []}
{"id": "2403.16172", "title": "Fusion of Minutia Cylinder Codes and Minutia Patch Embeddings for Latent\n  Fingerprint Recognition", "abstract": "  Latent fingerprints are one of the most widely used forensic evidence by law\nenforcement agencies. However, latent recognition performance is far from the\nexemplary performance of sensor fingerprint recognition due to deformations and\nartifacts within these images. In this study, we propose a fusion based local\nmatching approach towards latent fingerprint recognition. Recent latent\nrecognition studies typically relied on local descriptor generation methods, in\nwhich either handcrafted minutiae features or deep neural network features are\nextracted around a minutia of interest, in the latent recognition process.\nProposed approach would integrate these handcrafted features with a recently\nproposed deep neural network embedding features in a multi-stage fusion\napproach to significantly improve latent recognition results. Effectiveness of\nthe proposed approach has been shown on several public and private data sets.\nAs demonstrated in our experimental results, proposed method improves rank-1\nidentification accuracy by considerably for real-world datasets when compared\nto either the single usage of these features or existing state-of-the-art\nmethods in the literature.\n", "date": "2024-03-24", "categories": []}
{"id": "2403.16202", "title": "FH-SSTNet: Forehead Creases based User Verification using Spatio-Spatial\n  Temporal Network", "abstract": "  Biometric authentication, which utilizes contactless features, such as\nforehead patterns, has become increasingly important for identity verification\nand access management. The proposed method is based on learning a 3D\nspatio-spatial temporal convolution to create detailed pictures of forehead\npatterns. We introduce a new CNN model called the Forehead Spatio-Spatial\nTemporal Network (FH-SSTNet), which utilizes a 3D CNN architecture with triplet\nloss to capture distinguishing features. We enhance the model's discrimination\ncapability using Arcloss in the network's head. Experimentation on the Forehead\nCreases version 1 (FH-V1) dataset, containing 247 unique subjects, demonstrates\nthe superior performance of FH-SSTNet compared to existing methods and\npre-trained CNNs like ResNet50, especially for forehead-based user\nverification. The results demonstrate the superior performance of FH-SSTNet for\nforehead-based user verification, confirming its effectiveness in identity\nauthentication.\n", "date": "2024-03-24", "categories": []}
{"id": "2403.14852", "title": "KeyPoint Relative Position Encoding for Face Recognition", "abstract": "  In this paper, we address the challenge of making ViT models more robust to\nunseen affine transformations. Such robustness becomes useful in various\nrecognition tasks such as face recognition when image alignment failures occur.\nWe propose a novel method called KP-RPE, which leverages key points\n(e.g.~facial landmarks) to make ViT more resilient to scale, translation, and\npose variations. We begin with the observation that Relative Position Encoding\n(RPE) is a good way to bring affine transform generalization to ViTs. RPE,\nhowever, can only inject the model with prior knowledge that nearby pixels are\nmore important than far pixels. Keypoint RPE (KP-RPE) is an extension of this\nprinciple, where the significance of pixels is not solely dictated by their\nproximity but also by their relative positions to specific keypoints within the\nimage. By anchoring the significance of pixels around keypoints, the model can\nmore effectively retain spatial relationships, even when those relationships\nare disrupted by affine transformations. We show the merit of KP-RPE in face\nand gait recognition. The experimental results demonstrate the effectiveness in\nimproving face recognition performance from low-quality images, particularly\nwhere alignment is prone to failure. Code and pre-trained models are available.\n", "date": "2024-03-21", "categories": []}
{"id": "2403.15119", "title": "An Open-World, Diverse, Cross-Spatial-Temporal Benchmark for Dynamic\n  Wild Person Re-Identification", "abstract": "  Person re-identification (ReID) has made great strides thanks to the\ndata-driven deep learning techniques. However, the existing benchmark datasets\nlack diversity, and models trained on these data cannot generalize well to\ndynamic wild scenarios. To meet the goal of improving the explicit\ngeneralization of ReID models, we develop a new Open-World, Diverse,\nCross-Spatial-Temporal dataset named OWD with several distinct features. 1)\nDiverse collection scenes: multiple independent open-world and highly dynamic\ncollecting scenes, including streets, intersections, shopping malls, etc. 2)\nDiverse lighting variations: long time spans from daytime to nighttime with\nabundant illumination changes. 3) Diverse person status: multiple camera\nnetworks in all seasons with normal/adverse weather conditions and diverse\npedestrian appearances (e.g., clothes, personal belongings, poses, etc.). 4)\nProtected privacy: invisible faces for privacy critical applications. To\nimprove the implicit generalization of ReID, we further propose a Latent Domain\nExpansion (LDE) method to develop the potential of source data, which decouples\ndiscriminative identity-relevant and trustworthy domain-relevant features and\nimplicitly enforces domain-randomized identity feature space expansion with\nricher domain diversity to facilitate domain invariant representations. Our\ncomprehensive evaluations with most benchmark datasets in the community are\ncrucial for progress, although this work is far from the grand goal toward\nopen-world and dynamic wild applications.\n", "date": "2024-03-22", "categories": []}
{"id": "2403.13820", "title": "Identity information based on human magnetocardiography signals", "abstract": "  We have developed an individual identification system based on\nmagnetocardiography (MCG) signals captured using optically pumped magnetometers\n(OPMs). Our system utilizes pattern recognition to analyze the signals obtained\nat different positions on the body, by scanning the matrices composed of MCG\nsignals with a 2*2 window. In order to make use of the spatial information of\nMCG signals, we transform the signals from adjacent small areas into four\nchannels of a dataset. We further transform the data into time-frequency\nmatrices using wavelet transforms and employ a convolutional neural network\n(CNN) for classification. As a result, our system achieves an accuracy rate of\n97.04% in identifying individuals. This finding indicates that the MCG signal\nholds potential for use in individual identification systems, offering a\nvaluable tool for personalized healthcare management.\n", "date": "2024-03-02", "categories": []}
{"id": "2403.14513", "title": "View-decoupled Transformer for Person Re-identification under\n  Aerial-ground Camera Network", "abstract": "  Existing person re-identification methods have achieved remarkable advances\nin appearance-based identity association across homogeneous cameras, such as\nground-ground matching. However, as a more practical scenario, aerial-ground\nperson re-identification (AGPReID) among heterogeneous cameras has received\nminimal attention. To alleviate the disruption of discriminative identity\nrepresentation by dramatic view discrepancy as the most significant challenge\nin AGPReID, the view-decoupled transformer (VDT) is proposed as a simple yet\neffective framework. Two major components are designed in VDT to decouple\nview-related and view-unrelated features, namely hierarchical subtractive\nseparation and orthogonal loss, where the former separates these two features\ninside the VDT, and the latter constrains these two to be independent. In\naddition, we contribute a large-scale AGPReID dataset called CARGO, consisting\nof five/eight aerial/ground cameras, 5,000 identities, and 108,563 images.\nExperiments on two datasets show that VDT is a feasible and effective solution\nfor AGPReID, surpassing the previous method on mAP/Rank1 by up to 5.0%/2.7% on\nCARGO and 3.7%/5.2% on AG-ReID, keeping the same magnitude of computational\ncomplexity. Our project is available at https://github.com/LinlyAC/VDT-AGPReID\n", "date": "2024-03-21", "categories": []}
{"id": "2302.11427", "title": "Enhanced Face Authentication With Separate Loss Functions", "abstract": "  The overall objective of the main project is to propose and develop a system\nof facial authentication in unlocking phones or applications in phones using\nfacial recognition. The system will include four separate architectures: face\ndetection, face recognition, face spoofing, and classification of closed eyes.\nIn which, we consider the problem of face recognition to be the most important,\ndetermining the true identity of the person standing in front of the screen\nwith absolute accuracy is what facial recognition systems need to achieve.\nAlong with the development of the face recognition problem, the problem of the\nanti-fake face is also gradually becoming popular and equally important. Our\ngoal is to propose and develop two loss functions: LMCot and Double Loss. Then\napply them to the face authentication process.\n", "date": "2023-02-22", "categories": []}
{"id": "2403.06529", "title": "Confidence-Aware RGB-D Face Recognition via Virtual Depth Synthesis", "abstract": "  2D face recognition encounters challenges in unconstrained environments due\nto varying illumination, occlusion, and pose. Recent studies focus on RGB-D\nface recognition to improve robustness by incorporating depth information.\nHowever, collecting sufficient paired RGB-D training data is expensive and\ntime-consuming, hindering wide deployment. In this work, we first construct a\ndiverse depth dataset generated by 3D Morphable Models for depth model\npre-training. Then, we propose a domain-independent pre-training framework that\nutilizes readily available pre-trained RGB and depth models to separately\nperform face recognition without needing additional paired data for retraining.\nTo seamlessly integrate the two distinct networks and harness the complementary\nbenefits of RGB and depth information for improved accuracy, we propose an\ninnovative Adaptive Confidence Weighting (ACW). This mechanism is designed to\nlearn confidence estimates for each modality to achieve modality fusion at the\nscore level. Our method is simple and lightweight, only requiring ACW training\nbeyond the backbone models. Experiments on multiple public RGB-D face\nrecognition benchmarks demonstrate state-of-the-art performance surpassing\nprevious methods based on depth estimation and feature fusion, validating the\nefficacy of our approach.\n", "date": "2024-03-11", "categories": []}
{"id": "2403.10022", "title": "Lifelong Person Re-Identification with Backward-Compatibility", "abstract": "  Lifelong person re-identification (LReID) assumes a practical scenario where\nthe model is sequentially trained on continuously incoming datasets while\nalleviating the catastrophic forgetting in the old datasets. However, not only\nthe training datasets but also the gallery images are incrementally\naccumulated, that requires a huge amount of computational complexity and\nstorage space to extract the features at the inference phase. In this paper, we\naddress the above mentioned problem by incorporating the backward-compatibility\nto LReID for the first time. We train the model using the continuously incoming\ndatasets while maintaining the model's compatibility toward the previously\ntrained old models without re-computing the features of the old gallery images.\nTo this end, we devise the cross-model compatibility loss based on the\ncontrastive learning with respect to the replay features across all the old\ndatasets. Moreover, we also develop the knowledge consolidation method based on\nthe part classification to learn the shared representation across different\ndatasets for the backward-compatibility. We suggest a more practical\nmethodology for performance evaluation as well where all the gallery and query\nimages are considered together. Experimental results demonstrate that the\nproposed method achieves a significantly higher performance of the\nbackward-compatibility compared with the existing methods. It is a promising\ntool for more practical scenarios of LReID.\n", "date": "2024-03-15", "categories": []}
{"id": "2403.11101", "title": "Hierarchical Generative Network for Face Morphing Attacks", "abstract": "  Face morphing attacks circumvent face recognition systems (FRSs) by creating\na morphed image that contains multiple identities. However, existing face\nmorphing attack methods either sacrifice image quality or compromise the\nidentity preservation capability. Consequently, these attacks fail to bypass\nFRSs verification well while still managing to deceive human observers. These\nmethods typically rely on global information from contributing images, ignoring\nthe detailed information from effective facial regions. To address the above\nissues, we propose a novel morphing attack method to improve the quality of\nmorphed images and better preserve the contributing identities. Our proposed\nmethod leverages the hierarchical generative network to capture both local\ndetailed and global consistency information. Additionally, a mask-guided image\nblending module is dedicated to removing artifacts from areas outside the face\nto improve the image's visual quality. The proposed attack method is compared\nto state-of-the-art methods on three public datasets in terms of FRSs'\nvulnerability, attack detectability, and image quality. The results show our\nmethod's potential threat of deceiving FRSs while being capable of passing\nmultiple morphing attack detection (MAD) scenarios.\n", "date": "2024-03-17", "categories": []}
{"id": "2403.11121", "title": "A Versatile Framework for Multi-scene Person Re-identification", "abstract": "  Person Re-identification (ReID) has been extensively developed for a decade\nin order to learn the association of images of the same person across\nnon-overlapping camera views. To overcome significant variations between images\nacross camera views, mountains of variants of ReID models were developed for\nsolving a number of challenges, such as resolution change, clothing change,\nocclusion, modality change, and so on. Despite the impressive performance of\nmany ReID variants, these variants typically function distinctly and cannot be\napplied to other challenges. To our best knowledge, there is no versatile ReID\nmodel that can handle various ReID challenges at the same time. This work\ncontributes to the first attempt at learning a versatile ReID model to solve\nsuch a problem. Our main idea is to form a two-stage prompt-based twin modeling\nframework called VersReID. Our VersReID firstly leverages the scene label to\ntrain a ReID Bank that contains abundant knowledge for handling various scenes,\nwhere several groups of scene-specific prompts are used to encode different\nscene-specific knowledge. In the second stage, we distill a V-Branch model with\nversatile prompts from the ReID Bank for adaptively solving the ReID of\ndifferent scenes, eliminating the demand for scene labels during the inference\nstage. To facilitate training VersReID, we further introduce the multi-scene\nproperties into self-supervised learning of ReID via a multi-scene prioris data\naugmentation (MPDA) strategy. Through extensive experiments, we demonstrate the\nsuccess of learning an effective and versatile ReID model for handling ReID\ntasks under multi-scene conditions without manual assignment of scene labels in\nthe inference stage, including general, low-resolution, clothing change,\nocclusion, and cross-modality scenes. Codes and models are available at\nhttps://github.com/iSEE-Laboratory/VersReID.\n", "date": "2024-03-17", "categories": []}
{"id": "2309.08006", "title": "Facial Kinship Verification from remote photoplethysmography", "abstract": "  Facial Kinship Verification (FKV) aims at automatically determining whether\ntwo subjects have a kinship relation based on human faces. It has potential\napplications in finding missing children and social media analysis. Traditional\nFKV faces challenges as it is vulnerable to spoof attacks and raises privacy\nissues. In this paper, we explore for the first time the FKV with vital\nbio-signals, focusing on remote Photoplethysmography (rPPG). rPPG signals are\nextracted from facial videos, resulting in a one-dimensional signal that\nmeasures the changes in visible light reflection emitted to and detected from\nthe skin caused by the heartbeat. Specifically, in this paper, we employed a\nstraightforward one-dimensional Convolutional Neural Network (1DCNN) with a\n1DCNN-Attention module and kinship contrastive loss to learn the kin similarity\nfrom rPPGs. The network takes multiple rPPG signals extracted from various\nfacial Regions of Interest (ROIs) as inputs. Additionally, the 1DCNN attention\nmodule is designed to learn and capture the discriminative kin features from\nfeature embeddings. Finally, we demonstrate the feasibility of rPPG to detect\nkinship with the experiment evaluation on the UvANEMO Smile Database from\ndifferent kin relations.\n", "date": "2023-09-14", "categories": []}
{"id": "2403.09415", "title": "User Identification via Free Roaming Eye Tracking Data", "abstract": "  We present a new dataset of \"free roaming\" (FR) and \"targeted roaming\" (TR):\na pool of 41 participants is asked to walk around a university campus (FR) or\nis asked to find a particular room within a library (TR). Eye movements are\nrecorded using a commodity wearable eye tracker (Pupil Labs Neon at 200Hz). On\nthis dataset we investigate the accuracy of user identification using a\npreviously known machine learning pipeline where a Radial Basis Function\nNetwork (RBFN) is used as classifier. Our highest accuracies are 87.3% for FR\nand 89.4% for TR. This should be compared to 95.3% which is the (corresponding)\nhighest accuracy we are aware of (achieved in a laboratory setting using the\n\"RAN\" stimulus of the BioEye 2015 competition dataset). To the best of our\nknowledge, our results are the first that study user identification in a non\nlaboratory setting; such settings are often more feasible than laboratory\nsettings and may include further advantages. The minimum duration of each\nrecording is 263s for FR and 154s for TR. Our best accuracies are obtained when\nrestricting to 120s and 140s for FR and TR respectively, always cut from the\nend of the trajectories (both for the training and testing sessions). If we cut\nthe same length from the beginning, then accuracies are 12.2% lower for FR and\naround 6.4% lower for TR. On the full trajectories accuracies are lower by 5%\nand 52% for FR and TR. We also investigate the impact of including higher order\nvelocity derivatives (such as acceleration, jerk, or jounce).\n", "date": "2024-03-14", "categories": []}
{"id": "2403.08161", "title": "LAFS: Landmark-based Facial Self-supervised Learning for Face\n  Recognition", "abstract": "  In this work we focus on learning facial representations that can be adapted\nto train effective face recognition models, particularly in the absence of\nlabels. Firstly, compared with existing labelled face datasets, a vastly larger\nmagnitude of unlabeled faces exists in the real world. We explore the learning\nstrategy of these unlabeled facial images through self-supervised pretraining\nto transfer generalized face recognition performance. Moreover, motivated by\none recent finding, that is, the face saliency area is critical for face\nrecognition, in contrast to utilizing random cropped blocks of images for\nconstructing augmentations in pretraining, we utilize patches localized by\nextracted facial landmarks. This enables our method - namely LAndmark-based\nFacial Self-supervised learning LAFS), to learn key representation that is more\ncritical for face recognition. We also incorporate two landmark-specific\naugmentations which introduce more diversity of landmark information to further\nregularize the learning. With learned landmark-based facial representations, we\nfurther adapt the representation for face recognition with regularization\nmitigating variations in landmark positions. Our method achieves significant\nimprovement over the state-of-the-art on multiple face recognition benchmarks,\nespecially on more challenging few-shot scenarios.\n", "date": "2024-03-13", "categories": []}
{"id": "2403.05344", "title": "Federated Learning Method for Preserving Privacy in Face Recognition\n  System", "abstract": "  The state-of-the-art face recognition systems are typically trained on a\nsingle computer, utilizing extensive image datasets collected from various\nnumber of users. However, these datasets often contain sensitive personal\ninformation that users may hesitate to disclose. To address potential privacy\nconcerns, we explore the application of federated learning, both with and\nwithout secure aggregators, in the context of both supervised and unsupervised\nface recognition systems. Federated learning facilitates the training of a\nshared model without necessitating the sharing of individual private data,\nachieving this by training models on decentralized edge devices housing the\ndata. In our proposed system, each edge device independently trains its own\nmodel, which is subsequently transmitted either to a secure aggregator or\ndirectly to the central server. To introduce diverse data without the need for\ndata transmission, we employ generative adversarial networks to generate\nimposter data at the edge. Following this, the secure aggregator or central\nserver combines these individual models to construct a global model, which is\nthen relayed back to the edge devices. Experimental findings based on the\nCelebA datasets reveal that employing federated learning in both supervised and\nunsupervised face recognition systems offers dual benefits. Firstly, it\nsafeguards privacy since the original data remains on the edge devices.\nSecondly, the experimental results demonstrate that the aggregated model yields\nnearly identical performance compared to the individual models, particularly\nwhen the federated model does not utilize a secure aggregator. Hence, our\nresults shed light on the practical challenges associated with\nprivacy-preserving face image training, particularly in terms of the balance\nbetween privacy and accuracy.\n", "date": "2024-03-08", "categories": []}
{"id": "2403.04036", "title": "Unsupervised Contrastive Learning for Robust RF Device Fingerprinting\n  Under Time-Domain Shift", "abstract": "  Radio Frequency (RF) device fingerprinting has been recognized as a potential\ntechnology for enabling automated wireless device identification and\nclassification. However, it faces a key challenge due to the domain shift that\ncould arise from variations in the channel conditions and environmental\nsettings, potentially degrading the accuracy of RF-based device classification\nwhen testing and training data is collected in different domains. This paper\nintroduces a novel solution that leverages contrastive learning to mitigate\nthis domain shift problem. Contrastive learning, a state-of-the-art\nself-supervised learning approach from deep learning, learns a distance metric\nsuch that positive pairs are closer (i.e. more similar) in the learned metric\nspace than negative pairs. When applied to RF fingerprinting, our model treats\nRF signals from the same transmission as positive pairs and those from\ndifferent transmissions as negative pairs. Through experiments on wireless and\nwired RF datasets collected over several days, we demonstrate that our\ncontrastive learning approach captures domain-invariant features, diminishing\nthe effects of domain-specific variations. Our results show large and\nconsistent improvements in accuracy (10.8\\% to 27.8\\%) over baseline models,\nthus underscoring the effectiveness of contrastive learning in improving device\nclassification under domain shift.\n", "date": "2024-03-06", "categories": []}
{"id": "2403.04183", "title": "YYDS: Visible-Infrared Person Re-Identification with Coarse Descriptions", "abstract": "  Visible-infrared person re-identification (VI-ReID) is challenging due to\nconsiderable cross-modality discrepancies. Existing works mainly focus on\nlearning modality-invariant features while suppressing modality-specific ones.\nHowever, retrieving visible images only depends on infrared samples is an\nextreme problem because of the absence of color information. To this end, we\npresent the Refer-VI-ReID settings, which aims to match target visible images\nfrom both infrared images and coarse language descriptions (e.g., \"a man with\nred top and black pants\") to complement the missing color information. To\naddress this task, we design a Y-Y-shape decomposition structure, dubbed YYDS,\nto decompose and aggregate texture and color features of targets. Specifically,\nthe text-IoU regularization strategy is firstly presented to facilitate the\ndecomposition training, and a joint relation module is then proposed to infer\nthe aggregation. Furthermore, the cross-modal version of k-reciprocal\nre-ranking algorithm is investigated, named CMKR, in which three neighbor\nsearch strategies and one local query expansion method are explored to\nalleviate the modality bias problem of the near neighbors. We conduct\nexperiments on SYSU-MM01, RegDB and LLCM datasets with our manually annotated\ndescriptions. Both YYDS and CMKR achieve remarkable improvements over SOTA\nmethods on all three datasets. Codes are available at\nhttps://github.com/dyhBUPT/YYDS.\n", "date": "2024-03-07", "categories": []}
{"id": "2403.04549", "title": "Explainable Face Verification via Feature-Guided Gradient\n  Backpropagation", "abstract": "  Recent years have witnessed significant advancement in face recognition (FR)\ntechniques, with their applications widely spread in people's lives and\nsecurity-sensitive areas. There is a growing need for reliable interpretations\nof decisions of such systems. Existing studies relying on various mechanisms\nhave investigated the usage of saliency maps as an explanation approach, but\nsuffer from different limitations. This paper first explores the spatial\nrelationship between face image and its deep representation via gradient\nbackpropagation. Then a new explanation approach FGGB has been conceived, which\nprovides precise and insightful similarity and dissimilarity saliency maps to\nexplain the \"Accept\" and \"Reject\" decision of an FR system. Extensive visual\npresentation and quantitative measurement have shown that FGGB achieves\nsuperior performance in both similarity and dissimilarity maps when compared to\ncurrent state-of-the-art explainable face verification approaches.\n", "date": "2024-03-07", "categories": []}
{"id": "2303.05076", "title": "Pedestrian Attribute Editing for Gait Recognition and Anonymization", "abstract": "  As a kind of biometrics, the gait information of pedestrians has attracted\nwidespread attention from both industry and academia since it can be acquired\nfrom long distances without the cooperation of targets. In recent literature,\nthis line of research has brought exciting chances along with alarming\nchallenges: On the positive side, gait recognition used for security\napplications such as suspect retrieval and safety checks is becoming more and\nmore promising. On the negative side, the misuse of gait information may lead\nto privacy concerns, as lawbreakers can track subjects of interest using gait\ncharacteristics even under face-masked and clothes-changed scenarios. To handle\nthis double-edged sword, we propose a gait attribute editing framework termed\nGaitEditor. It can perform various degrees of attribute edits on real gait\nsequences while maintaining the visual authenticity, respectively used for gait\ndata augmentation and de-identification, thereby adaptively enhancing or\ndegrading gait recognition performance according to users' intentions.\nExperimentally, we conduct a comprehensive evaluation under both gait\nrecognition and anonymization protocols on three widely used gait benchmarks.\nNumerous results illustrate that the adaptable utilization of GaitEditor\nefficiently improves gait recognition performance and generates vivid\nvisualizations with de-identification to protect human privacy. To the best of\nour knowledge, GaitEditor is the first framework capable of editing multiple\ngait attributes while simultaneously benefiting gait recognition and gait\nanonymization. The source code of GaitEditor will be available at\nhttps://github.com/ShiqiYu/OpenGait.\n", "date": "2023-03-09", "categories": []}
{"id": "2403.03828", "title": "From Clicks to Security: Investigating Continuous Authentication via\n  Mouse Dynamics", "abstract": "  In the realm of computer security, the importance of efficient and reliable\nuser authentication methods has become increasingly critical. This paper\nexamines the potential of mouse movement dynamics as a consistent metric for\ncontinuous authentication. By analyzing user mouse movement patterns in two\ncontrasting gaming scenarios, \"Team Fortress\" and Poly Bridge we investigate\nthe distinctive behavioral patterns inherent in high-intensity and\nlow-intensity UI interactions. The study extends beyond conventional\nmethodologies by employing a range of machine learning models. These models are\ncarefully selected to assess their effectiveness in capturing and interpreting\nthe subtleties of user behavior as reflected in their mouse movements. This\nmultifaceted approach allows for a more nuanced and comprehensive understanding\nof user interaction patterns. Our findings reveal that mouse movement dynamics\ncan serve as a reliable indicator for continuous user authentication. The\ndiverse machine learning models employed in this study demonstrate competent\nperformance in user verification, marking an improvement over previous methods\nused in this field. This research contributes to the ongoing efforts to enhance\ncomputer security and highlights the potential of leveraging user behavior,\nspecifically mouse dynamics, in developing robust authentication systems.\n", "date": "2024-03-06", "categories": []}
{"id": "2403.03832", "title": "Your device may know you better than you know yourself -- continuous\n  authentication on novel dataset using machine learning", "abstract": "  This research aims to further understanding in the field of continuous\nauthentication using behavioral biometrics. We are contributing a novel dataset\nthat encompasses the gesture data of 15 users playing Minecraft with a Samsung\nTablet, each for a duration of 15 minutes. Utilizing this dataset, we employed\nmachine learning (ML) binary classifiers, being Random Forest (RF), K-Nearest\nNeighbors (KNN), and Support Vector Classifier (SVC), to determine the\nauthenticity of specific user actions. Our most robust model was SVC, which\nachieved an average accuracy of approximately 90%, demonstrating that touch\ndynamics can effectively distinguish users. However, further studies are needed\nto make it viable option for authentication systems\n", "date": "2024-03-06", "categories": []}
{"id": "2403.02892", "title": "Enhancing Long-Term Person Re-Identification Using Global, Local Body\n  Part, and Head Streams", "abstract": "  This work addresses the task of long-term person re-identification.\nTypically, person re-identification assumes that people do not change their\nclothes, which limits its applications to short-term scenarios. To overcome\nthis limitation, we investigate long-term person re-identification, which\nconsiders both clothes-changing and clothes-consistent scenarios. In this\npaper, we propose a novel framework that effectively learns and utilizes both\nglobal and local information. The proposed framework consists of three streams:\nglobal, local body part, and head streams. The global and head streams encode\nidentity-relevant information from an entire image and a cropped image of the\nhead region, respectively. Both streams encode the most distinct, less\ndistinct, and average features using the combinations of adversarial erasing,\nmax pooling, and average pooling. The local body part stream extracts\nidentity-related information for each body part, allowing it to be compared\nwith the same body part from another image. Since body part annotations are not\navailable in re-identification datasets, pseudo-labels are generated using\nclustering. These labels are then utilized to train a body part segmentation\nhead in the local body part stream. The proposed framework is trained by\nbackpropagating the weighted summation of the identity classification loss, the\npair-based loss, and the pseudo body part segmentation loss. To demonstrate the\neffectiveness of the proposed method, we conducted experiments on three\npublicly available datasets (Celeb-reID, PRCC, and VC-Clothes). The\nexperimental results demonstrate that the proposed method outperforms the\nprevious state-of-the-art method.\n", "date": "2024-03-05", "categories": []}
{"id": "2312.14404", "title": "Cross-Covariate Gait Recognition: A Benchmark", "abstract": "  Gait datasets are essential for gait research. However, this paper observes\nthat present benchmarks, whether conventional constrained or emerging\nreal-world datasets, fall short regarding covariate diversity. To bridge this\ngap, we undertake an arduous 20-month effort to collect a cross-covariate gait\nrecognition (CCGR) dataset. The CCGR dataset has 970 subjects and about 1.6\nmillion sequences; almost every subject has 33 views and 53 different\ncovariates. Compared to existing datasets, CCGR has both population and\nindividual-level diversity. In addition, the views and covariates are well\nlabeled, enabling the analysis of the effects of different factors. CCGR\nprovides multiple types of gait data, including RGB, parsing, silhouette, and\npose, offering researchers a comprehensive resource for exploration. In order\nto delve deeper into addressing cross-covariate gait recognition, we propose\nparsing-based gait recognition (ParsingGait) by utilizing the newly proposed\nparsing data. We have conducted extensive experiments. Our main results show:\n1) Cross-covariate emerges as a pivotal challenge for practical applications of\ngait recognition. 2) ParsingGait demonstrates remarkable potential for further\nadvancement. 3) Alarmingly, existing SOTA methods achieve less than 43%\naccuracy on the CCGR, highlighting the urgency of exploring cross-covariate\ngait recognition. Link: https://github.com/ShinanZou/CCGR.\n", "date": "2023-12-22", "categories": []}
{"id": "2403.00889", "title": "Time-bound Contextual Bio-ID Generation for Minimalist Wearables", "abstract": "  As wearable devices become increasingly miniaturized and powerful, a new\nopportunity arises for instant and dynamic device-to-device collaboration and\nhuman-to-device interaction. However, this progress presents a unique\nchallenge: these minimalist wearables lack inherent mechanisms for real-time\nauthentication, posing significant risks to data privacy and overall security.\nTo address this, we introduce Proteus that realizes an innovative concept of\ntime-bound contextual bio-IDs, which are generated from on-device sensor data\nand embedded into a common latent space. These bio-IDs act as a time-bound\nunique user identifier that can be used to identify the wearer in a certain\ncontext. Proteus enables dynamic and contextual device collaboration as well as\nrobust human-to-device interaction. Our evaluations demonstrate the\neffectiveness of our method, particularly in the context of minimalist\nwearables.\n", "date": "2024-03-01", "categories": []}
{"id": "2402.18163", "title": "Ef-QuantFace: Streamlined Face Recognition with Small Data and Low-Bit\n  Precision", "abstract": "  In recent years, model quantization for face recognition has gained\nprominence. Traditionally, compressing models involved vast datasets like the\n5.8 million-image MS1M dataset as well as extensive training times, raising the\nquestion of whether such data enormity is essential. This paper addresses this\nby introducing an efficiency-driven approach, fine-tuning the model with just\nup to 14,000 images, 440 times smaller than MS1M. We demonstrate that effective\nquantization is achievable with a smaller dataset, presenting a new paradigm.\nMoreover, we incorporate an evaluation-based metric loss and achieve an\noutstanding 96.15% accuracy on the IJB-C dataset, establishing a new\nstate-of-the-art compressed model training for face recognition. The subsequent\nanalysis delves into potential applications, emphasizing the transformative\npower of this approach. This paper advances model quantization by highlighting\nthe efficiency and optimal results with small data and training time.\n", "date": "2024-02-28", "categories": []}
{"id": "2401.13641", "title": "How Good is ChatGPT at Face Biometrics? A First Look into Recognition,\n  Soft Biometrics, and Explainability", "abstract": "  Large Language Models (LLMs) such as GPT developed by OpenAI, have already\nshown astonishing results, introducing quick changes in our society. This has\nbeen intensified by the release of ChatGPT which allows anyone to interact in a\nsimple conversational way with LLMs, without any experience in the field\nneeded. As a result, ChatGPT has been rapidly applied to many different tasks\nsuch as code- and song-writer, education, virtual assistants, etc., showing\nimpressive results for tasks for which it was not trained (zero-shot learning).\n  The present study aims to explore the ability of ChatGPT, based on the recent\nGPT-4 multimodal LLM, for the task of face biometrics. In particular, we\nanalyze the ability of ChatGPT to perform tasks such as face verification,\nsoft-biometrics estimation, and explainability of the results. ChatGPT could be\nvery valuable to further increase the explainability and transparency of\nautomatic decisions in human scenarios. Experiments are carried out in order to\nevaluate the performance and robustness of ChatGPT, using popular public\nbenchmarks and comparing the results with state-of-the-art methods in the\nfield. The results achieved in this study show the potential of LLMs such as\nChatGPT for face biometrics, especially to enhance explainability. For\nreproducibility reasons, we release all the code in GitHub.\n", "date": "2024-01-24", "categories": []}
{"id": "2402.15206", "title": "Source-Guided Similarity Preservation for Online Person\n  Re-Identification", "abstract": "  Online Unsupervised Domain Adaptation (OUDA) for person Re-Identification\n(Re-ID) is the task of continuously adapting a model trained on a\nwell-annotated source domain dataset to a target domain observed as a data\nstream. In OUDA, person Re-ID models face two main challenges: catastrophic\nforgetting and domain shift. In this work, we propose a new Source-guided\nSimilarity Preservation (S2P) framework to alleviate these two problems. Our\nframework is based on the extraction of a support set composed of source images\nthat maximizes the similarity with the target data. This support set is used to\nidentify feature similarities that must be preserved during the learning\nprocess. S2P can incorporate multiple existing UDA methods to mitigate\ncatastrophic forgetting. Our experiments show that S2P outperforms previous\nstate-of-the-art methods on multiple real-to-real and synthetic-to-real\nchallenging OUDA benchmarks.\n", "date": "2024-02-23", "categories": []}
{"id": "2210.13664", "title": "Mitigating Gender Bias in Face Recognition Using the von Mises-Fisher\n  Mixture Model", "abstract": "  In spite of the high performance and reliability of deep learning algorithms\nin a wide range of everyday applications, many investigations tend to show that\na lot of models exhibit biases, discriminating against specific subgroups of\nthe population (e.g. gender, ethnicity). This urges the practitioner to develop\nfair systems with a uniform/comparable performance across sensitive groups. In\nthis work, we investigate the gender bias of deep Face Recognition networks. In\norder to measure this bias, we introduce two new metrics, $\\mathrm{BFAR}$ and\n$\\mathrm{BFRR}$, that better reflect the inherent deployment needs of Face\nRecognition systems. Motivated by geometric considerations, we mitigate gender\nbias through a new post-processing methodology which transforms the deep\nembeddings of a pre-trained model to give more representation power to\ndiscriminated subgroups. It consists in training a shallow neural network by\nminimizing a Fair von Mises-Fisher loss whose hyperparameters account for the\nintra-class variance of each gender. Interestingly, we empirically observe that\nthese hyperparameters are correlated with our fairness metrics. In fact,\nextensive numerical experiments on a variety of datasets show that a careful\nselection significantly reduces gender bias. The code used for the experiments\ncan be found at https://github.com/JRConti/EthicalModule_vMF.\n", "date": "2022-10-24", "categories": []}
{"id": "2303.10976", "title": "Attention Disturbance and Dual-Path Constraint Network for Occluded\n  Person Re-identification", "abstract": "  Occluded person re-identification (Re-ID) aims to address the potential\nocclusion problem when matching occluded or holistic pedestrians from different\ncamera views. Many methods use the background as artificial occlusion and rely\non attention networks to exclude noisy interference. However, the significant\ndiscrepancy between simple background occlusion and realistic occlusion can\nnegatively impact the generalization of the network. To address this issue, we\npropose a novel transformer-based Attention Disturbance and Dual-Path\nConstraint Network (ADP) to enhance the generalization of attention networks.\nFirstly, to imitate real-world obstacles, we introduce an Attention Disturbance\nMask (ADM) module that generates an offensive noise, which can distract\nattention like a realistic occluder, as a more complex form of occlusion.\nSecondly, to fully exploit these complex occluded images, we develop a\nDual-Path Constraint Module (DPC) that can obtain preferable supervision\ninformation from holistic images through dual-path interaction. With our\nproposed method, the network can effectively circumvent a wide variety of\nocclusions using the basic ViT baseline. Comprehensive experimental evaluations\nconducted on person re-ID benchmarks demonstrate the superiority of ADP over\nstate-of-the-art methods.\n", "date": "2023-03-20", "categories": []}
{"id": "2402.14454", "title": "CCPA: Long-term Person Re-Identification via Contrastive Clothing and\n  Pose Augmentation", "abstract": "  Long-term Person Re-Identification (LRe-ID) aims at matching an individual\nacross cameras after a long period of time, presenting variations in clothing,\npose, and viewpoint. In this work, we propose CCPA: Contrastive Clothing and\nPose Augmentation framework for LRe-ID. Beyond appearance, CCPA captures body\nshape information which is cloth-invariant using a Relation Graph Attention\nNetwork. Training a robust LRe-ID model requires a wide range of clothing\nvariations and expensive cloth labeling, which is lacked in current LRe-ID\ndatasets. To address this, we perform clothing and pose transfer across\nidentities to generate images of more clothing variations and of different\npersons wearing similar clothing. The augmented batch of images serve as inputs\nto our proposed Fine-grained Contrastive Losses, which not only supervise the\nRe-ID model to learn discriminative person embeddings under long-term scenarios\nbut also ensure in-distribution data generation. Results on LRe-ID datasets\ndemonstrate the effectiveness of our CCPA framework.\n", "date": "2024-02-22", "categories": []}
{"id": "2402.14665", "title": "Quadruplet Loss For Improving the Robustness to Face Morphing Attacks", "abstract": "  Recent advancements in deep learning have revolutionized technology and\nsecurity measures, necessitating robust identification methods. Biometric\napproaches, leveraging personalized characteristics, offer a promising\nsolution. However, Face Recognition Systems are vulnerable to sophisticated\nattacks, notably face morphing techniques, enabling the creation of fraudulent\ndocuments. In this study, we introduce a novel quadruplet loss function for\nincreasing the robustness of face recognition systems against morphing attacks.\nOur approach involves specific sampling of face image quadruplets, combined\nwith face morphs, for network training. Experimental results demonstrate the\nefficiency of our strategy in improving the robustness of face recognition\nnetworks against morphing attacks.\n", "date": "2024-02-22", "categories": []}
{"id": "2211.07245", "title": "Assessing Uncertainty in Similarity Scoring: Performance & Fairness in\n  Face Recognition", "abstract": "  The ROC curve is the major tool for assessing not only the performance but\nalso the fairness properties of a similarity scoring function. In order to draw\nreliable conclusions based on empirical ROC analysis, accurately evaluating the\nuncertainty level related to statistical versions of the ROC curves of interest\nis absolutely necessary, especially for applications with considerable societal\nimpact such as Face Recognition. In this article, we prove asymptotic\nguarantees for empirical ROC curves of similarity functions as well as for\nby-product metrics useful to assess fairness. We also explain that, because the\nfalse acceptance/rejection rates are of the form of U-statistics in the case of\nsimilarity scoring, the naive bootstrap approach may jeopardize the assessment\nprocedure. A dedicated recentering technique must be used instead. Beyond the\ntheoretical analysis carried out, various experiments using real face image\ndatasets provide strong empirical evidence of the practical relevance of the\nmethods promoted here, when applied to several ROC-based measures such as\npopular fairness metrics.\n", "date": "2022-11-14", "categories": []}
{"id": "2304.01580", "title": "Untargeted Near-collision Attacks on Biometrics: Real-world Bounds and\n  Theoretical Limits", "abstract": "  A biometric recognition system can operate in two distinct modes:\nidentification or verification. In the first mode, the system recognizes an\nindividual by searching the enrolled templates of all the users for a match. In\nthe second mode, the system validates a user's identity claim by comparing the\nfresh provided template with the enrolled template. The biometric\ntransformation schemes usually produce binary templates that are better handled\nby cryptographic schemes, and the comparison is based on a distance that leaks\ninformation about the similarities between two biometric templates. Both the\nexperimentally determined false match rate and false non-match rate through\nrecognition threshold adjustment define the recognition accuracy, and hence the\nsecurity of the system. To our knowledge, few works provide a formal treatment\nof security in case of minimal information leakage, i.e., the binary outcome of\na comparison with a threshold. In this paper, we focus on untargeted attacks\nthat can be carried out both online and offline, and in both identification and\nverification modes. On the first hand, we focus our analysis on the accuracy\nmetrics of biometric systems. We provide the complexity of untargeted attacks\nusing the False Match Rate (FMR) and the False Positive Identification Rate\n(FPIR) to address the security of these systems. Studying near-collisions with\nthese metrics allows us to estimate the maximum number of users in a database,\ngiven a chosen FMR, to preserve the security and the accuracy. These results\nare evaluated on systems from the literature. On the other hand, we rely on\nprobabilistic modelling to assess the theoretical security limits of biometric\nsystems. The study of this metric space, and system parameters (template size,\nthreshold and database size), gives us the complexity of untargeted attacks and\nthe probability of a near-collision.\n", "date": "2023-04-04", "categories": []}
{"id": "2402.13771", "title": "Mask-up: Investigating Biases in Face Re-identification for Masked Faces", "abstract": "  AI based Face Recognition Systems (FRSs) are now widely distributed and\ndeployed as MLaaS solutions all over the world, moreso since the COVID-19\npandemic for tasks ranging from validating individuals' faces while buying SIM\ncards to surveillance of citizens. Extensive biases have been reported against\nmarginalized groups in these systems and have led to highly discriminatory\noutcomes. The post-pandemic world has normalized wearing face masks but FRSs\nhave not kept up with the changing times. As a result, these systems are\nsusceptible to mask based face occlusion. In this study, we audit four\ncommercial and nine open-source FRSs for the task of face re-identification\nbetween different varieties of masked and unmasked images across five benchmark\ndatasets (total 14,722 images). These simulate a realistic\nvalidation/surveillance task as deployed in all major countries around the\nworld. Three of the commercial and five of the open-source FRSs are highly\ninaccurate; they further perpetuate biases against non-White individuals, with\nthe lowest accuracy being 0%. A survey for the same task with 85 human\nparticipants also results in a low accuracy of 40%. Thus a human-in-the-loop\nmoderation in the pipeline does not alleviate the concerns, as has been\nfrequently hypothesized in literature. Our large-scale study shows that\ndevelopers, lawmakers and users of such services need to rethink the design\nprinciples behind FRSs, especially for the task of face re-identification,\ntaking cognizance of observed biases.\n", "date": "2024-02-21", "categories": []}
{"id": "2402.11206", "title": "Hand Biometrics in Digital Forensics", "abstract": "  Digital forensic is now an unavoidable part for securing the digital world\nfrom identity theft. Higher order of crimes, dealing with a massive database is\nreally very challenging problem for any intelligent system. Biometric is a\nbetter solution to win over the problems encountered by digital forensics. Many\nbiometric characteristics are playing their significant roles in forensics over\nthe decades. The potential benefits and scope of hand based modes in forensics\nhave been investigated with an illustration of hand geometry verifi-cation\nmethod. It can be applied when effective biometric evidences are properly\nunavailable; gloves are damaged, and dirt or any kind of liquid can minimize\nthe accessibility and reliability of the fingerprint or palmprint. Due to the\ncrisis of pure uniqueness of hand features for a very large database, it may be\nrelevant for verification only. Some unimodal and multimodal hand based\nbiometrics (e.g. hand geometry, palmprint and hand vein) with several feature\nextractions, database and verification methods have been discussed with 2D, 3D\nand infrared images.\n", "date": "2024-02-17", "categories": []}
{"id": "2402.10435", "title": "Dynamic Patch-aware Enrichment Transformer for Occluded Person\n  Re-Identification", "abstract": "  Person re-identification (re-ID) continues to pose a significant challenge,\nparticularly in scenarios involving occlusions. Prior approaches aimed at\ntackling occlusions have predominantly focused on aligning physical body\nfeatures through the utilization of external semantic cues. However, these\nmethods tend to be intricate and susceptible to noise. To address the\naforementioned challenges, we present an innovative end-to-end solution known\nas the Dynamic Patch-aware Enrichment Transformer (DPEFormer). This model\neffectively distinguishes human body information from occlusions automatically\nand dynamically, eliminating the need for external detectors or precise image\nalignment. Specifically, we introduce a dynamic patch token selection module\n(DPSM). DPSM utilizes a label-guided proxy token as an intermediary to identify\ninformative occlusion-free tokens. These tokens are then selected for deriving\nsubsequent local part features. To facilitate the seamless integration of\nglobal classification features with the finely detailed local features selected\nby DPSM, we introduce a novel feature blending module (FBM). FBM enhances\nfeature representation through the complementary nature of information and the\nexploitation of part diversity. Furthermore, to ensure that DPSM and the entire\nDPEFormer can effectively learn with only identity labels, we also propose a\nRealistic Occlusion Augmentation (ROA) strategy. This strategy leverages the\nrecent advances in the Segment Anything Model (SAM). As a result, it generates\nocclusion images that closely resemble real-world occlusions, greatly enhancing\nthe subsequent contrastive learning process. Experiments on occluded and\nholistic re-ID benchmarks signify a substantial advancement of DPEFormer over\nexisting state-of-the-art approaches. The code will be made publicly available.\n", "date": "2024-02-16", "categories": []}
{"id": "2402.08320", "title": "The Paradox of Motion: Evidence for Spurious Correlations in\n  Skeleton-based Gait Recognition Models", "abstract": "  Gait, an unobtrusive biometric, is valued for its capability to identify\nindividuals at a distance, across external outfits and environmental\nconditions. This study challenges the prevailing assumption that vision-based\ngait recognition, in particular skeleton-based gait recognition, relies\nprimarily on motion patterns, revealing a significant role of the implicit\nanthropometric information encoded in the walking sequence. We show through a\ncomparative analysis that removing height information leads to notable\nperformance degradation across three models and two benchmarks (CASIA-B and\nGREW). Furthermore, we propose a spatial transformer model processing\nindividual poses, disregarding any temporal information, which achieves\nunreasonably good accuracy, emphasizing the bias towards appearance information\nand indicating spurious correlations in existing benchmarks. These findings\nunderscore the need for a nuanced understanding of the interplay between motion\nand appearance in vision-based gait recognition, prompting a reevaluation of\nthe methodological assumptions in this field. Our experiments indicate that\n\"in-the-wild\" datasets are less prone to spurious correlations, prompting the\nneed for more diverse and large scale datasets for advancing the field.\n", "date": "2024-02-13", "categories": []}
{"id": "2402.07263", "title": "Trade-off Between Spatial and Angular Resolution in Facial Recognition", "abstract": "  Ensuring robustness in face recognition systems across various challenging\nconditions is crucial for their versatility. State-of-the-art methods often\nincorporate additional information, such as depth, thermal, or angular data, to\nenhance performance. However, light field-based face recognition approaches\nthat leverage angular information face computational limitations. This paper\ninvestigates the fundamental trade-off between spatio-angular resolution in\nlight field representation to achieve improved face recognition performance. By\nutilizing macro-pixels with varying angular resolutions while maintaining the\noverall image size, we aim to quantify the impact of angular information at the\nexpense of spatial resolution, while considering computational constraints. Our\nexperimental results demonstrate a notable performance improvement in face\nrecognition systems by increasing the angular resolution, up to a certain\nextent, at the cost of spatial resolution.\n", "date": "2024-02-11", "categories": []}
{"id": "2402.07685", "title": "Contrastive Multiple Instance Learning for Weakly Supervised Person ReID", "abstract": "  The acquisition of large-scale, precisely labeled datasets for person\nre-identification (ReID) poses a significant challenge. Weakly supervised ReID\nhas begun to address this issue, although its performance lags behind fully\nsupervised methods. In response, we introduce Contrastive Multiple Instance\nLearning (CMIL), a novel framework tailored for more effective weakly\nsupervised ReID. CMIL distinguishes itself by requiring only a single model and\nno pseudo labels while leveraging contrastive losses -- a technique that has\nsignificantly enhanced traditional ReID performance yet is absent in all prior\nMIL-based approaches. Through extensive experiments and analysis across three\ndatasets, CMIL not only matches state-of-the-art performance on the large-scale\nSYSU-30k dataset with fewer assumptions but also consistently outperforms all\nbaselines on the WL-market1501 and Weakly Labeled MUddy racer re-iDentification\ndataset (WL-MUDD) datasets. We introduce and release the WL-MUDD dataset, an\nextension of the MUDD dataset featuring naturally occurring weak labels from\nthe real-world application at PerformancePhoto.co. All our code and data are\naccessible at\nhttps://drive.google.com/file/d/1rjMbWB6m-apHF3Wg_cfqc8QqKgQ21AsT/view?usp=drive_link.\n", "date": "2024-02-12", "categories": []}
{"id": "2312.14001", "title": "Deep Learning Based Face Recognition Method using Siamese Network", "abstract": "  Achieving state-of-the-art results in face verification systems typically\nhinges on the availability of labeled face training data, a resource that often\nproves challenging to acquire in substantial quantities. In this research\nendeavor, we proposed employing Siamese networks for face recognition,\neliminating the need for labeled face images. We achieve this by strategically\nleveraging negative samples alongside nearest neighbor counterparts, thereby\nestablishing positive and negative pairs through an unsupervised methodology.\nThe architectural framework adopts a VGG encoder, trained as a double branch\nsiamese network. Our primary aim is to circumvent the necessity for labeled\nface image data, thus proposing the generation of training pairs in an entirely\nunsupervised manner. Positive training data are selected within a dataset based\non their highest cosine similarity scores with a designated anchor, while\nnegative training data are culled in a parallel fashion, though drawn from an\nalternate dataset. During training, the proposed siamese network conducts\nbinary classification via cross-entropy loss. Subsequently, during the testing\nphase, we directly extract face verification scores from the network's output\nlayer. Experimental results reveal that the proposed unsupervised system\ndelivers a performance on par with a similar but fully supervised baseline.\n", "date": "2023-12-21", "categories": []}
{"id": "2312.14301", "title": "Autoencoder Based Face Verification System", "abstract": "  The primary objective of this work is to present an alternative approach\naimed at reducing the dependency on labeled data. Our proposed method involves\nutilizing autoencoder pre-training within a face image recognition task with\ntwo step processes. Initially, an autoencoder is trained in an unsupervised\nmanner using a substantial amount of unlabeled training dataset. Subsequently,\na deep learning model is trained with initialized parameters from the\npre-trained autoencoder. This deep learning training process is conducted in a\nsupervised manner, employing relatively limited labeled training dataset.\nDuring evaluation phase, face image embeddings is generated as the output of\ndeep neural network layer. Our training is executed on the CelebA dataset,\nwhile evaluation is performed using benchmark face recognition datasets such as\nLabeled Faces in the Wild (LFW) and YouTube Faces (YTF). Experimental results\ndemonstrate that by initializing the deep neural network with pre-trained\nautoencoder parameters achieve comparable results to state-of-the-art methods.\n", "date": "2023-12-21", "categories": []}
{"id": "2308.06944", "title": "One-shot lip-based biometric authentication: extending behavioral\n  features with authentication phrase information", "abstract": "  Lip-based biometric authentication (LBBA) is an authentication method based\non a person's lip movements during speech in the form of video data captured by\na camera sensor. LBBA can utilize both physical and behavioral characteristics\nof lip movements without requiring any additional sensory equipment apart from\nan RGB camera. State-of-the-art (SOTA) approaches use one-shot learning to\ntrain deep siamese neural networks which produce an embedding vector out of\nthese features. Embeddings are further used to compute the similarity between\nan enrolled user and a user being authenticated. A flaw of these approaches is\nthat they model behavioral features as style-of-speech without relation to what\nis being said. This makes the system vulnerable to video replay attacks of the\nclient speaking any phrase. To solve this problem we propose a one-shot\napproach which models behavioral features to discriminate against what is being\nsaid in addition to style-of-speech. We achieve this by customizing the GRID\ndataset to obtain required triplets and training a siamese neural network based\non 3D convolutions and recurrent neural network layers. A custom triplet loss\nfor batch-wise hard-negative mining is proposed. Obtained results using an\nopen-set protocol are 3.2% FAR and 3.8% FRR on the test set of the customized\nGRID dataset. Additional analysis of the results was done to quantify the\ninfluence and discriminatory power of behavioral and physical features for\nLBBA.\n", "date": "2023-08-14", "categories": []}
{"id": "2402.02060", "title": "DiffVein: A Unified Diffusion Network for Finger Vein Segmentation and\n  Authentication", "abstract": "  Finger vein authentication, recognized for its high security and specificity,\nhas become a focal point in biometric research. Traditional methods\npredominantly concentrate on vein feature extraction for discriminative\nmodeling, with a limited exploration of generative approaches. Suffering from\nverification failure, existing methods often fail to obtain authentic vein\npatterns by segmentation. To fill this gap, we introduce DiffVein, a unified\ndiffusion model-based framework which simultaneously addresses vein\nsegmentation and authentication tasks. DiffVein is composed of two dedicated\nbranches: one for segmentation and the other for denoising. For better feature\ninteraction between these two branches, we introduce two specialized modules to\nimprove their collective performance. The first, a mask condition module,\nincorporates the semantic information of vein patterns from the segmentation\nbranch into the denoising process. Additionally, we also propose a Semantic\nDifference Transformer (SD-Former), which employs Fourier-space self-attention\nand cross-attention modules to extract category embedding before feeding it to\nthe segmentation task. In this way, our framework allows for a dynamic\ninterplay between diffusion and segmentation embeddings, thus vein segmentation\nand authentication tasks can inform and enhance each other in the joint\ntraining. To further optimize our model, we introduce a Fourier-space\nStructural Similarity (FourierSIM) loss function, which is tailored to improve\nthe denoising network's learning efficacy. Extensive experiments on the USM and\nTHU-MVFV3V datasets substantiates DiffVein's superior performance, setting new\nbenchmarks in both vein segmentation and authentication tasks.\n", "date": "2024-02-03", "categories": []}
{"id": "2312.10447", "title": "Finger Biometric Recognition With Feature Selection", "abstract": "  Biometrics is indispensable in this modern digital era for secure automated\nhuman authentication in various fields of machine learning and pattern\nrecognition. Hand geometry is a promising physiological biometric trait with\nample deployed application areas for identity verification. Due to the\nintricate anatomic foundation of the thumb and substantial inter-finger posture\nvariation, satisfactory performances cannot be achieved while the thumb is\nincluded in the contact-free environment. To overcome the hindrances associated\nwith the thumb, four finger-based (excluding the thumb) biometric approaches\nhave been devised. In this chapter, a four-finger based biometric method has\nbeen presented. Again, selection of salient features is essential to reduce the\nfeature dimensionality by eliminating the insignificant features. Weights are\nassigned according to the discriminative efficiency of the features to\nemphasize on the essential features. Two different strategies namely, the\nglobal and local feature selection methods are adopted based on the adaptive\nforward-selection and backward-elimination (FoBa) algorithm. The identification\nperformances are evaluated using the weighted k-nearest neighbor (wk-NN) and\nrandom forest (RF) classifiers. The experiments are conducted using the\nselected feature subsets over the 300 subjects of the Bosphorus hand database.\nThe best identification accuracy of 98.67%, and equal error rate (EER) of 4.6%\nhave been achieved using the subset of 25 features which are selected by the\nrank-based local FoBa algorithm.\n", "date": "2023-12-16", "categories": []}
{"id": "2402.00996", "title": "mmID: High-Resolution mmWave Imaging for Human Identification", "abstract": "  Achieving accurate human identification through RF imaging has been a\npersistent challenge, primarily attributed to the limited aperture size and its\nconsequent impact on imaging resolution. The existing imaging solution enables\ntasks such as pose estimation, activity recognition, and human tracking based\non deep neural networks by estimating skeleton joints. In contrast to\nestimating joints, this paper proposes to improve imaging resolution by\nestimating the human figure as a whole using conditional generative adversarial\nnetworks (cGAN). In order to reduce training complexity, we use an estimated\nspatial spectrum using the MUltiple SIgnal Classification (MUSIC) algorithm as\ninput to the cGAN. Our system generates environmentally independent,\nhigh-resolution images that can extract unique physical features useful for\nhuman identification. We use a simple convolution layers-based classification\nnetwork to obtain the final identification result. From the experimental\nresults, we show that resolution of the image produced by our trained generator\nis high enough to enable human identification. Our finding indicates\nhigh-resolution accuracy with 5% mean silhouette difference to the Kinect\ndevice. Extensive experiments in different environments on multiple testers\ndemonstrate that our system can achieve 93% overall test accuracy in unseen\nenvironments for static human target identification.\n", "date": "2024-02-01", "categories": []}
{"id": "2402.01472", "title": "Synthetic Data for the Mitigation of Demographic Biases in Face\n  Recognition", "abstract": "  This study investigates the possibility of mitigating the demographic biases\nthat affect face recognition technologies through the use of synthetic data.\nDemographic biases have the potential to impact individuals from specific\ndemographic groups, and can be identified by observing disparate performance of\nface recognition systems across demographic groups. They primarily arise from\nthe unequal representations of demographic groups in the training data. In\nrecent times, synthetic data have emerged as a solution to some problems that\naffect face recognition systems. In particular, during the generation process\nit is possible to specify the desired demographic and facial attributes of\nimages, in order to control the demographic distribution of the synthesized\ndataset, and fairly represent the different demographic groups. We propose to\nfine-tune with synthetic data existing face recognition systems that present\nsome demographic biases. We use synthetic datasets generated with GANDiffFace,\na novel framework able to synthesize datasets for face recognition with\ncontrollable demographic distribution and realistic intra-class variations. We\nconsider multiple datasets representing different demographic groups for\ntraining and evaluation. Also, we fine-tune different face recognition systems,\nand evaluate their demographic fairness with different metrics. Our results\nsupport the proposed approach and the use of synthetic data to mitigate\ndemographic biases in face recognition.\n", "date": "2024-02-02", "categories": []}
{"id": "2401.17699", "title": "Unified Physical-Digital Face Attack Detection", "abstract": "  Face Recognition (FR) systems can suffer from physical (i.e., print photo)\nand digital (i.e., DeepFake) attacks. However, previous related work rarely\nconsiders both situations at the same time. This implies the deployment of\nmultiple models and thus more computational burden. The main reasons for this\nlack of an integrated model are caused by two factors: (1) The lack of a\ndataset including both physical and digital attacks with ID consistency which\nmeans the same ID covers the real face and all attack types; (2) Given the\nlarge intra-class variance between these two attacks, it is difficult to learn\na compact feature space to detect both attacks simultaneously. To address these\nissues, we collect a Unified physical-digital Attack dataset, called\nUniAttackData. The dataset consists of $1,800$ participations of 2 and 12\nphysical and digital attacks, respectively, resulting in a total of 29,706\nvideos. Then, we propose a Unified Attack Detection framework based on\nVision-Language Models (VLMs), namely UniAttackDetection, which includes three\nmain modules: the Teacher-Student Prompts (TSP) module, focused on acquiring\nunified and specific knowledge respectively; the Unified Knowledge Mining (UKM)\nmodule, designed to capture a comprehensive feature space; and the Sample-Level\nPrompt Interaction (SLPI) module, aimed at grasping sample-level semantics.\nThese three modules seamlessly form a robust unified attack detection\nframework. Extensive experiments on UniAttackData and three other datasets\ndemonstrate the superiority of our approach for unified face attack detection.\n", "date": "2024-01-31", "categories": []}
{"id": "2401.18032", "title": "DROP: Decouple Re-Identification and Human Parsing with Task-specific\n  Features for Occluded Person Re-identification", "abstract": "  The paper introduces the Decouple Re-identificatiOn and human Parsing (DROP)\nmethod for occluded person re-identification (ReID). Unlike mainstream\napproaches using global features for simultaneous multi-task learning of ReID\nand human parsing, or relying on semantic information for attention guidance,\nDROP argues that the inferior performance of the former is due to distinct\ngranularity requirements for ReID and human parsing features. ReID focuses on\ninstance part-level differences between pedestrian parts, while human parsing\ncenters on semantic spatial context, reflecting the internal structure of the\nhuman body. To address this, DROP decouples features for ReID and human\nparsing, proposing detail-preserving upsampling to combine varying resolution\nfeature maps. Parsing-specific features for human parsing are decoupled, and\nhuman position information is exclusively added to the human parsing branch. In\nthe ReID branch, a part-aware compactness loss is introduced to enhance\ninstance-level part differences. Experimental results highlight the efficacy of\nDROP, especially achieving a Rank-1 accuracy of 76.8% on Occluded-Duke,\nsurpassing two mainstream methods. The codebase is accessible at\nhttps://github.com/shuguang-52/DROP.\n", "date": "2024-01-31", "categories": []}
{"id": "2401.16329", "title": "Synthesis of 3D on-air signatures with the Sigma-Lognormal model", "abstract": "  Signature synthesis is a computation technique that generates artificial\nspecimens which can support decision making in automatic signature\nverification. A lot of work has been dedicated to this subject, which centres\non synthesizing dynamic and static two-dimensional handwriting on canvas. This\npaper proposes a framework to generate synthetic 3D on-air signatures\nexploiting the lognormality principle, which mimics the complex neuromotor\ncontrol processes at play as the fingertip moves. Addressing the usual cases\ninvolving the development of artificial individuals and duplicated samples,\nthis paper contributes to the synthesis of: (1) the trajectory and velocity of\nentirely 3D new signatures; (2) kinematic information when only the 3D\ntrajectory of the signature is known, and (3) duplicate samples of 3D real\nsignatures. Validation was conducted by generating synthetic 3D signature\ndatabases mimicking real ones and showing that automatic signature\nverifications of genuine and skilled forgeries report performances similar to\nthose of real and synthetic databases. We also observed that training 3D\nautomatic signature verifiers with duplicates can reduce errors. We further\ndemonstrated that our proposal is also valid for synthesizing 3D air writing\nand gestures. Finally, a perception test confirmed the human likeness of the\ngenerated specimens. The databases generated are publicly available, only for\nresearch purposes, at .\n", "date": "2024-01-29", "categories": []}
{"id": "2401.16559", "title": "IEEE BigData 2023 Keystroke Verification Challenge (KVC)", "abstract": "  This paper describes the results of the IEEE BigData 2023 Keystroke\nVerification Challenge (KVC), that considers the biometric verification\nperformance of Keystroke Dynamics (KD), captured as tweet-long sequences of\nvariable transcript text from over 185,000 subjects. The data are obtained from\ntwo of the largest public databases of KD up to date, the Aalto Desktop and\nMobile Keystroke Databases, guaranteeing a minimum amount of data per subject,\nage and gender annotations, absence of corrupted data, and avoiding excessively\nunbalanced subject distributions with respect to the considered demographic\nattributes. Several neural architectures were proposed by the participants,\nleading to global Equal Error Rates (EERs) as low as 3.33% and 3.61% achieved\nby the best team respectively in the desktop and mobile scenario, outperforming\nthe current state of the art biometric verification performance for KD. Hosted\non CodaLab, the KVC will be made ongoing to represent a useful tool for the\nresearch community to compare different approaches under the same experimental\nconditions and to deepen the knowledge of the field.\n", "date": "2024-01-29", "categories": []}
{"id": "2401.16649", "title": "Using Motion Forecasting for Behavior-Based Virtual Reality (VR)\n  Authentication", "abstract": "  Task-based behavioral biometric authentication of users interacting in\nvirtual reality (VR) environments enables seamless continuous authentication by\nusing only the motion trajectories of the person's body as a unique signature.\nDeep learning-based approaches for behavioral biometrics show high accuracy\nwhen using complete or near complete portions of the user trajectory, but show\nlower performance when using smaller segments from the start of the task. Thus,\nany systems designed with existing techniques are vulnerable while waiting for\nfuture segments of motion trajectories to become available. In this work, we\npresent the first approach that predicts future user behavior using\nTransformer-based forecasting and using the forecasted trajectory to perform\nuser authentication. Our work leverages the notion that given the current\ntrajectory of a user in a task-based environment we can predict the future\ntrajectory of the user as they are unlikely to dramatically shift their\nbehavior since it would preclude the user from successfully completing their\ntask goal. Using the publicly available 41-subject ball throwing dataset of\nMiller et al. we show improvement in user authentication when using forecasted\ndata. When compared to no forecasting, our approach reduces the authentication\nequal error rate (EER) by an average of 23.85% and a maximum reduction of\n36.14%.\n", "date": "2024-01-30", "categories": []}
{"id": "2401.16722", "title": "Optimal-Landmark-Guided Image Blending for Face Morphing Attacks", "abstract": "  In this paper, we propose a novel approach for conducting face morphing\nattacks, which utilizes optimal-landmark-guided image blending. Current face\nmorphing attacks can be categorized into landmark-based and generation-based\napproaches. Landmark-based methods use geometric transformations to warp facial\nregions according to averaged landmarks but often produce morphed images with\npoor visual quality. Generation-based methods, which employ generation models\nto blend multiple face images, can achieve better visual quality but are often\nunsuccessful in generating morphed images that can effectively evade\nstate-of-the-art face recognition systems~(FRSs). Our proposed method overcomes\nthe limitations of previous approaches by optimizing the morphing landmarks and\nusing Graph Convolutional Networks (GCNs) to combine landmark and appearance\nfeatures. We model facial landmarks as nodes in a bipartite graph that is fully\nconnected and utilize GCNs to simulate their spatial and structural\nrelationships. The aim is to capture variations in facial shape and enable\naccurate manipulation of facial appearance features during the warping process,\nresulting in morphed facial images that are highly realistic and visually\nfaithful. Experiments on two public datasets prove that our method inherits the\nadvantages of previous landmark-based and generation-based methods and\ngenerates morphed images with higher quality, posing a more significant threat\nto state-of-the-art FRSs.\n", "date": "2024-01-30", "categories": []}
{"id": "2401.16232", "title": "Cross-Database Liveness Detection: Insights from Comparative Biometric\n  Analysis", "abstract": "  In an era where biometric security serves as a keystone of modern identity\nverification systems, ensuring the authenticity of these biometric samples is\nparamount. Liveness detection, the capability to differentiate between genuine\nand spoofed biometric samples, stands at the forefront of this challenge. This\nresearch presents a comprehensive evaluation of liveness detection models, with\na particular focus on their performance in cross-database scenarios, a test\nparadigm notorious for its complexity and real-world relevance. Our study\ncommenced by meticulously assessing models on individual datasets, revealing\nthe nuances in their performance metrics. Delving into metrics such as the Half\nTotal Error Rate, False Acceptance Rate, and False Rejection Rate, we unearthed\ninvaluable insights into the models' strengths and weaknesses. Crucially, our\nexploration of cross-database testing provided a unique perspective,\nhighlighting the chasm between training on one dataset and deploying on\nanother. Comparative analysis with extant methodologies, ranging from\nconvolutional networks to more intricate strategies, enriched our understanding\nof the current landscape. The variance in performance, even among\nstate-of-the-art models, underscored the inherent challenges in this domain. In\nessence, this paper serves as both a repository of findings and a clarion call\nfor more nuanced, data-diverse, and adaptable approaches in biometric liveness\ndetection. In the dynamic dance between authenticity and deception, our work\noffers a blueprint for navigating the evolving rhythms of biometric security.\n", "date": "2024-01-29", "categories": []}
{"id": "2401.13719", "title": "Inference Attacks Against Face Recognition Model without Classification\n  Layers", "abstract": "  Face recognition (FR) has been applied to nearly every aspect of daily life,\nbut it is always accompanied by the underlying risk of leaking private\ninformation. At present, almost all attack models against FR rely heavily on\nthe presence of a classification layer. However, in practice, the FR model can\nobtain complex features of the input via the model backbone, and then compare\nit with the target for inference, which does not explicitly involve the outputs\nof the classification layer adopting logit or other losses. In this work, we\nadvocate a novel inference attack composed of two stages for practical FR\nmodels without a classification layer. The first stage is the membership\ninference attack. Specifically, We analyze the distances between the\nintermediate features and batch normalization (BN) parameters. The results\nindicate that this distance is a critical metric for membership inference. We\nthus design a simple but effective attack model that can determine whether a\nface image is from the training dataset or not. The second stage is the model\ninversion attack, where sensitive private data is reconstructed using a\npre-trained generative adversarial network (GAN) guided by the attack model in\nthe first stage. To the best of our knowledge, the proposed attack model is the\nvery first in the literature developed for FR models without a classification\nlayer. We illustrate the application of the proposed attack model in the\nestablishment of privacy-preserving FR techniques.\n", "date": "2024-01-24", "categories": []}
{"id": "2308.10623", "title": "GaitPT: Skeletons Are All You Need For Gait Recognition", "abstract": "  The analysis of patterns of walking is an important area of research that has\nnumerous applications in security, healthcare, sports and human-computer\ninteraction. Lately, walking patterns have been regarded as a unique\nfingerprinting method for automatic person identification at a distance. In\nthis work, we propose a novel gait recognition architecture called Gait Pyramid\nTransformer (GaitPT) that leverages pose estimation skeletons to capture unique\nwalking patterns, without relying on appearance information. GaitPT adopts a\nhierarchical transformer architecture that effectively extracts both spatial\nand temporal features of movement in an anatomically consistent manner, guided\nby the structure of the human skeleton. Our results show that GaitPT achieves\nstate-of-the-art performance compared to other skeleton-based gait recognition\nworks, in both controlled and in-the-wild scenarios. GaitPT obtains 82.6%\naverage accuracy on CASIA-B, surpassing other works by a margin of 6%.\nMoreover, it obtains 52.16% Rank-1 accuracy on GREW, outperforming both\nskeleton-based and appearance-based approaches.\n", "date": "2023-08-21", "categories": []}
{"id": "2309.04447", "title": "Impact of Blur and Resolution on Demographic Disparities in 1-to-Many\n  Facial Identification", "abstract": "  Most studies to date that have examined demographic variations in face\nrecognition accuracy have analyzed 1-to-1 matching accuracy, using images that\ncould be described as \"government ID quality\". This paper analyzes the accuracy\nof 1-to-many facial identification across demographic groups, and in the\npresence of blur and reduced resolution in the probe image as might occur in\n\"surveillance camera quality\" images. Cumulative match characteristic curves\n(CMC) are not appropriate for comparing propensity for rank-one recognition\nerrors across demographics, and so we use three metrics for our analysis: (1)\nthe well-known d' metric between mated and non-mated score distributions, and\nintroduced in this work, (2) absolute score difference between thresholds in\nthe high-similarity tail of the non-mated and the low-similarity tail of the\nmated distribution, and (3) distribution of (mated - non-mated rank-one scores)\nacross the set of probe images. We find that demographic variation in 1-to-many\naccuracy does not entirely follow what has been observed in 1-to-1 matching\naccuracy. Also, different from 1-to-1 accuracy, demographic comparison of\n1-to-many accuracy can be affected by different numbers of identities and\nimages across demographics. More importantly, we show that increased blur in\nthe probe image, or reduced resolution of the face in the probe image, can\nsignificantly increase the false positive identification rate. And we show that\nthe demographic variation in these high blur or low resolution conditions is\nmuch larger for male / female than for African-American / Caucasian. The point\nthat 1-to-many accuracy can potentially collapse in the context of processing\n\"surveillance camera quality\" probe images against a \"government ID quality\"\ngallery is an important one.\n", "date": "2023-09-08", "categories": []}
{"id": "2401.13418", "title": "Serial fusion of multi-modal biometric systems", "abstract": "  Serial, or sequential, fusion of multiple biometric matchers has been not\nthoroughly investigated so far. However, this approach exhibits some advantages\nwith respect to the widely adopted parallel approaches. In this paper, we\npropose a novel theoretical framework for the assessment of performance of such\nsystems, based on a previous work of the authors. Benefits in terms of\nperformance are theoretically evaluated, as well as estimation errors in the\nmodel parameters computation. Model is analyzed from the viewpoint of its pros\nand cons, by mean of preliminary experiments performed on NIST Biometric Score\nSet 1.\n", "date": "2024-01-24", "categories": []}
{"id": "2401.13531", "title": "QAGait: Revisit Gait Recognition from a Quality Perspective", "abstract": "  Gait recognition is a promising biometric method that aims to identify\npedestrians from their unique walking patterns. Silhouette modality, renowned\nfor its easy acquisition, simple structure, sparse representation, and\nconvenient modeling, has been widely employed in controlled in-the-lab\nresearch. However, as gait recognition rapidly advances from in-the-lab to\nin-the-wild scenarios, various conditions raise significant challenges for\nsilhouette modality, including 1) unidentifiable low-quality silhouettes\n(abnormal segmentation, severe occlusion, or even non-human shape), and 2)\nidentifiable but challenging silhouettes (background noise, non-standard\nposture, slight occlusion). To address these challenges, we revisit gait\nrecognition pipeline and approach gait recognition from a quality perspective,\nnamely QAGait. Specifically, we propose a series of cost-effective quality\nassessment strategies, including Maxmial Connect Area and Template Match to\neliminate background noises and unidentifiable silhouettes, Alignment strategy\nto handle non-standard postures. We also propose two quality-aware loss\nfunctions to integrate silhouette quality into optimization within the\nembedding space. Extensive experiments demonstrate our QAGait can guarantee\nboth gait reliability and performance enhancement. Furthermore, our quality\nassessment strategies can seamlessly integrate with existing gait datasets,\nshowcasing our superiority. Code is available at\nhttps://github.com/wzb-bupt/QAGait.\n", "date": "2024-01-24", "categories": []}
{"id": "2401.09467", "title": "Offline Handwriting Signature Verification: A Transfer Learning and\n  Feature Selection Approach", "abstract": "  Handwritten signature verification poses a formidable challenge in biometrics\nand document authenticity. The objective is to ascertain the authenticity of a\nprovided handwritten signature, distinguishing between genuine and forged ones.\nThis issue has many applications in sectors such as finance, legal\ndocumentation, and security. Currently, the field of computer vision and\nmachine learning has made significant progress in the domain of handwritten\nsignature verification. The outcomes, however, may be enhanced depending on the\nacquired findings, the structure of the datasets, and the used models. Four\nstages make up our suggested strategy. First, we collected a large dataset of\n12600 images from 420 distinct individuals, and each individual has 30\nsignatures of a certain kind (All authors signatures are genuine). In the\nsubsequent stage, the best features from each image were extracted using a deep\nlearning model named MobileNetV2. During the feature selection step, three\nselectors neighborhood component analysis (NCA), Chi2, and mutual info (MI)\nwere used to pull out 200, 300, 400, and 500 features, giving a total of 12\nfeature vectors. Finally, 12 results have been obtained by applying machine\nlearning techniques such as SVM with kernels (rbf, poly, and linear), KNN, DT,\nLinear Discriminant Analysis, and Naive Bayes. Without employing feature\nselection techniques, our suggested offline signature verification achieved a\nclassification accuracy of 91.3%, whereas using the NCA feature selection\napproach with just 300 features it achieved a classification accuracy of 97.7%.\nHigh classification accuracy was achieved using the designed and suggested\nmodel, which also has the benefit of being a self-organized framework.\nConsequently, using the optimum minimally chosen features, the proposed method\ncould identify the best model performance and result validation prediction\nvectors.\n", "date": "2024-01-05", "categories": []}
{"id": "2401.10139", "title": "Model Compression Techniques in Biometrics Applications: A Survey", "abstract": "  The development of deep learning algorithms has extensively empowered\nhumanity's task automatization capacity. However, the huge improvement in the\nperformance of these models is highly correlated with their increasing level of\ncomplexity, limiting their usefulness in human-oriented applications, which are\nusually deployed in resource-constrained devices. This led to the development\nof compression techniques that drastically reduce the computational and memory\ncosts of deep learning models without significant performance degradation. This\npaper aims to systematize the current literature on this topic by presenting a\ncomprehensive survey of model compression techniques in biometrics\napplications, namely quantization, knowledge distillation and pruning. We\nconduct a critical analysis of the comparative value of these techniques,\nfocusing on their advantages and disadvantages and presenting suggestions for\nfuture work directions that can potentially improve the current methods.\nAdditionally, we discuss and analyze the link between model bias and model\ncompression, highlighting the need to direct compression research toward model\nfairness in future works.\n", "date": "2024-01-18", "categories": []}
{"id": "2401.04448", "title": "A Novel Dataset for Non-Destructive Inspection of Handwritten Documents", "abstract": "  Forensic handwriting examination is a branch of Forensic Science that aims to\nexamine handwritten documents in order to properly define or hypothesize the\nmanuscript's author. These analysis involves comparing two or more (digitized)\ndocuments through a comprehensive comparison of intrinsic local and global\nfeatures. If a correlation exists and specific best practices are satisfied,\nthen it will be possible to affirm that the documents under analysis were\nwritten by the same individual. The need to create sophisticated tools capable\nof extracting and comparing significant features has led to the development of\ncutting-edge software with almost entirely automated processes, improving the\nforensic examination of handwriting and achieving increasingly objective\nevaluations. This is made possible by algorithmic solutions based on purely\nmathematical concepts. Machine Learning and Deep Learning models trained with\nspecific datasets could turn out to be the key elements to best solve the task\nat hand. In this paper, we proposed a new and challenging dataset consisting of\ntwo subsets: the first consists of 21 documents written either by the classic\n``pen and paper\" approach (and later digitized) and directly acquired on common\ndevices such as tablets; the second consists of 362 handwritten manuscripts by\n124 different people, acquired following a specific pipeline. Our study\npioneered a comparison between traditionally handwritten documents and those\nproduced with digital tools (e.g., tablets). Preliminary results on the\nproposed datasets show that 90% classification accuracy can be achieved on the\nfirst subset (documents written on both paper and pen and later digitized and\non tablets) and 96% on the second portion of the data. The datasets are\navailable at\nhttps://iplab.dmi.unict.it/mfs/forensic-handwriting-analysis/novel-dataset-2023/.\n", "date": "2024-01-09", "categories": []}
{"id": "2401.07469", "title": "A Deep Hierarchical Feature Sparse Framework for Occluded Person\n  Re-Identification", "abstract": "  Most existing methods tackle the problem of occluded person re-identification\n(ReID) by utilizing auxiliary models, resulting in a complicated and\ninefficient ReID framework that is unacceptable for real-time applications. In\nthis work, a speed-up person ReID framework named SUReID is proposed to\nmitigate occlusion interference while speeding up inference. The SUReID\nconsists of three key components: hierarchical token sparsification (HTS)\nstrategy, non-parametric feature alignment knowledge distillation (NPKD), and\nnoise occlusion data augmentation (NODA). The HTS strategy works by pruning the\nredundant tokens in the vision transformer to achieve highly effective\nself-attention computation and eliminate interference from occlusions or\nbackground noise. However, the pruned tokens may contain human part features\nthat contaminate the feature representation and degrade the performance. To\nsolve this problem, the NPKD is employed to supervise the HTS strategy,\nretaining more discriminative tokens and discarding meaningless ones.\nFurthermore, the NODA is designed to introduce more noisy samples, which\nfurther trains the ability of the HTS to disentangle different tokens.\nExperimental results show that the SUReID achieves superior performance with\nsurprisingly fast inference.\n", "date": "2024-01-15", "categories": []}
{"id": "2401.08079", "title": "Adversarial Masking Contrastive Learning for vein recognition", "abstract": "  Vein recognition has received increasing attention due to its high security\nand privacy. Recently, deep neural networks such as Convolutional neural\nnetworks (CNN) and Transformers have been introduced for vein recognition and\nachieved state-of-the-art performance. Despite the recent advances, however,\nexisting solutions for finger-vein feature extraction are still not optimal due\nto scarce training image samples. To overcome this problem, in this paper, we\npropose an adversarial masking contrastive learning (AMCL) approach, that\ngenerates challenging samples to train a more robust contrastive learning model\nfor the downstream palm-vein recognition task, by alternatively optimizing the\nencoder in the contrastive learning model and a set of latent variables. First,\na huge number of masks are generated to train a robust generative adversarial\nnetwork (GAN). The trained generator transforms a latent variable from the\nlatent variable space into a mask space. Then, we combine the trained generator\nwith a contrastive learning model to obtain our AMCL, where the generator\nproduces challenging masking images to increase the contrastive loss and the\ncontrastive learning model is trained based on the harder images to learn a\nmore robust feature representation. After training, the trained encoder in the\ncontrastive learning model is combined with a classification layer to build a\nclassifier, which is further fine-tuned on labeled training data for vein\nrecognition. The experimental results on three databases demonstrate that our\napproach outperforms existing contrastive learning approaches in terms of\nimproving identification accuracy of vein classifiers and achieves\nstate-of-the-art recognition results.\n", "date": "2024-01-16", "categories": []}
{"id": "2401.08111", "title": "Mobile Contactless Palmprint Recognition: Use of Multiscale, Multimodel\n  Embeddings", "abstract": "  Contactless palmprints are comprised of both global and local discriminative\nfeatures. Most prior work focuses on extracting global features or local\nfeatures alone for palmprint matching, whereas this research introduces a novel\nframework that combines global and local features for enhanced palmprint\nmatching accuracy. Leveraging recent advancements in deep learning, this study\nintegrates a vision transformer (ViT) and a convolutional neural network (CNN)\nto extract complementary local and global features. Next, a mobile-based,\nend-to-end palmprint recognition system is developed, referred to as Palm-ID.\nOn top of the ViT and CNN features, Palm-ID incorporates a palmprint\nenhancement module and efficient dimensionality reduction (for faster\nmatching). Palm-ID balances the trade-off between accuracy and latency,\nrequiring just 18ms to extract a template of size 516 bytes, which can be\nefficiently searched against a 10,000 palmprint gallery in 0.33ms on an AMD\nEPYC 7543 32-Core CPU utilizing 128-threads. Cross-database matching protocols\nand evaluations on large-scale operational datasets demonstrate the robustness\nof the proposed method, achieving a TAR of 98.06% at FAR=0.01% on a newly\ncollected, time-separated dataset. To show a practical deployment of the\nend-to-end system, the entire recognition pipeline is embedded within a mobile\ndevice for enhanced user privacy and security.\n", "date": "2024-01-16", "categories": []}
{"id": "2307.01838", "title": "EdgeFace: Efficient Face Recognition Model for Edge Devices", "abstract": "  In this paper, we present EdgeFace, a lightweight and efficient face\nrecognition network inspired by the hybrid architecture of EdgeNeXt. By\neffectively combining the strengths of both CNN and Transformer models, and a\nlow rank linear layer, EdgeFace achieves excellent face recognition performance\noptimized for edge devices. The proposed EdgeFace network not only maintains\nlow computational costs and compact storage, but also achieves high face\nrecognition accuracy, making it suitable for deployment on edge devices.\nExtensive experiments on challenging benchmark face datasets demonstrate the\neffectiveness and efficiency of EdgeFace in comparison to state-of-the-art\nlightweight models and deep face recognition models. Our EdgeFace model with\n1.77M parameters achieves state of the art results on LFW (99.73%), IJB-B\n(92.67%), and IJB-C (94.85%), outperforming other efficient models with larger\ncomputational complexities. The code to replicate the experiments will be made\navailable publicly.\n", "date": "2023-07-04", "categories": []}
{"id": "2401.05806", "title": "CLIP-Driven Semantic Discovery Network for Visible-Infrared Person\n  Re-Identification", "abstract": "  Visible-infrared person re-identification (VIReID) primarily deals with\nmatching identities across person images from different modalities. Due to the\nmodality gap between visible and infrared images, cross-modality identity\nmatching poses significant challenges. Recognizing that high-level semantics of\npedestrian appearance, such as gender, shape, and clothing style, remain\nconsistent across modalities, this paper intends to bridge the modality gap by\ninfusing visual features with high-level semantics. Given the capability of\nCLIP to sense high-level semantic information corresponding to visual\nrepresentations, we explore the application of CLIP within the domain of\nVIReID. Consequently, we propose a CLIP-Driven Semantic Discovery Network\n(CSDN) that consists of Modality-specific Prompt Learner, Semantic Information\nIntegration (SII), and High-level Semantic Embedding (HSE). Specifically,\nconsidering the diversity stemming from modality discrepancies in language\ndescriptions, we devise bimodal learnable text tokens to capture\nmodality-private semantic information for visible and infrared images,\nrespectively. Additionally, acknowledging the complementary nature of semantic\ndetails across different modalities, we integrate text features from the\nbimodal language descriptions to achieve comprehensive semantics. Finally, we\nestablish a connection between the integrated text features and the visual\nfeatures across modalities. This process embed rich high-level semantic\ninformation into visual representations, thereby promoting the modality\ninvariance of visual representations. The effectiveness and superiority of our\nproposed CSDN over existing methods have been substantiated through\nexperimental evaluations on multiple widely used benchmarks. The code will be\nreleased at \\url{https://github.com/nengdong96/CSDN}.\n", "date": "2024-01-11", "categories": []}
{"id": "2401.06430", "title": "Mutual Distillation Learning For Person Re-Identification", "abstract": "  With the rapid advancements in deep learning technologies, person\nre-identification (ReID) has witnessed remarkable performance improvements.\nHowever, the majority of prior works have traditionally focused on solving the\nproblem via extracting features solely from a single perspective, such as\nuniform partitioning, hard attention mechanisms, or semantic masks. While these\napproaches have demonstrated efficacy within specific contexts, they fall short\nin diverse situations. In this paper, we propose a novel approach, Mutual\nDistillation Learning For Person Re-identification (termed as MDPR), which\naddresses the challenging problem from multiple perspectives within a single\nunified model, leveraging the power of mutual distillation to enhance the\nfeature representations collectively. Specifically, our approach encompasses\ntwo branches: a hard content branch to extract local features via a uniform\nhorizontal partitioning strategy and a Soft Content Branch to dynamically\ndistinguish between foreground and background and facilitate the extraction of\nmulti-granularity features via a carefully designed attention mechanism. To\nfacilitate knowledge exchange between these two branches, a mutual distillation\nand fusion process is employed, promoting the capability of the outputs of each\nbranch. Extensive experiments are conducted on widely used person ReID datasets\nto validate the effectiveness and superiority of our approach. Notably, our\nmethod achieves an impressive $88.7\\%/94.4\\%$ in mAP/Rank-1 on the\nDukeMTMC-reID dataset, surpassing the current state-of-the-art results. Our\nsource code is available at https://github.com/KuilongCui/MDPR.\n", "date": "2024-01-12", "categories": []}
{"id": "2210.05438", "title": "Parallel Augmentation and Dual Enhancement for Occluded Person\n  Re-identification", "abstract": "  Occluded person re-identification (Re-ID), the task of searching for the same\nperson's images in occluded environments, has attracted lots of attention in\nthe past decades. Recent approaches concentrate on improving performance on\noccluded data by data/feature augmentation or using extra models to predict\nocclusions. However, they ignore the imbalance problem in this task and can not\nfully utilize the information from the training data. To alleviate these two\nissues, we propose a simple yet effective method with Parallel Augmentation and\nDual Enhancement (PADE), which is robust on both occluded and non-occluded data\nand does not require any auxiliary clues. First, we design a parallel\naugmentation mechanism (PAM) to generate more suitable occluded data to\nmitigate the negative effects of unbalanced data. Second, we propose the global\nand local dual enhancement strategy (DES) to promote the context information\nand details. Experimental results on three widely used occluded datasets and\ntwo non-occluded datasets validate the effectiveness of our method. The code is\navailable at\nhttps://github.com/littleprince1121/PADE_Parallel_Augmentation_and_Dual_Enhancement_for_Occluded_Person_ReID\n", "date": "2022-10-11", "categories": []}
{"id": "2303.03301", "title": "Exploring Deep Models for Practical Gait Recognition", "abstract": "  Gait recognition is a rapidly advancing vision technique for person\nidentification from a distance. Prior studies predominantly employed relatively\nshallow networks to extract subtle gait features, achieving impressive\nsuccesses in constrained settings. Nevertheless, experiments revealed that\nexisting methods mostly produce unsatisfactory results when applied to newly\nreleased real-world gait datasets. This paper presents a unified perspective to\nexplore how to construct deep models for state-of-the-art outdoor gait\nrecognition, including the classical CNN-based and emerging Transformer-based\narchitectures. Specifically, we challenge the stereotype of shallow gait models\nand demonstrate the superiority of explicit temporal modeling and deep\ntransformer structure for discriminative gait representation learning.\nConsequently, the proposed CNN-based DeepGaitV2 series and Transformer-based\nSwinGait series exhibit significant performance improvements on Gait3D and\nGREW. As for the constrained gait datasets, the DeepGaitV2 series also reaches\na new state-of-the-art in most cases, convincingly showing its practicality and\ngenerality. The source code is available at\nhttps://github.com/ShiqiYu/OpenGait.\n", "date": "2023-03-06", "categories": []}
{"id": "2401.01227", "title": "IdentiFace : A VGG Based Multimodal Facial Biometric System", "abstract": "  The development of facial biometric systems has contributed greatly to the\ndevelopment of the computer vision field. Nowadays, there's always a need to\ndevelop a multimodal system that combines multiple biometric traits in an\nefficient, meaningful way. In this paper, we introduce \"IdentiFace\" which is a\nmultimodal facial biometric system that combines the core of facial recognition\nwith some of the most important soft biometric traits such as gender, face\nshape, and emotion. We also focused on developing the system using only VGG-16\ninspired architecture with minor changes across different subsystems. This\nunification allows for simpler integration across modalities. It makes it\neasier to interpret the learned features between the tasks which gives a good\nindication about the decision-making process across the facial modalities and\npotential connection. For the recognition problem, we acquired a 99.2% test\naccuracy for five classes with high intra-class variations using data collected\nfrom the FERET database[1]. We achieved 99.4% on our dataset and 95.15% on the\npublic dataset[2] in the gender recognition problem. We were also able to\nachieve a testing accuracy of 88.03% in the face-shape problem using the\ncelebrity face-shape dataset[3]. Finally, we achieved a decent testing accuracy\nof 66.13% in the emotion task which is considered a very acceptable accuracy\ncompared to related work on the FER2013 dataset[4].\n", "date": "2024-01-02", "categories": []}
{"id": "2401.03085", "title": "Consensus-Threshold Criterion for Offline Signature Verification using\n  Convolutional Neural Network Learned Representations", "abstract": "  A genuine signer's signature is naturally unstable even at short\ntime-intervals whereas, expert forgers always try to perfectly mimic a genuine\nsigner's signature. This presents a challenge which puts a genuine signer at\nrisk of being denied access, while a forge signer is granted access. The\nimplication is a high false acceptance rate (FAR) which is the percentage of\nforge signature classified as belonging to a genuine class. Existing work have\nonly scratched the surface of signature verification because the\nmisclassification error remains high. In this paper, a consensus-threshold\ndistance-based classifier criterion is proposed for offline writer-dependent\nsignature verification. Using features extracted from SigNet and SigNet-F deep\nconvolutional neural network models, the proposed classifier minimizes FAR.\nThis is demonstrated via experiments on four datasets: GPDS-300, MCYT, CEDAR\nand Brazilian PUC-PR datasets. On GPDS-300, the consensus threshold classifier\nimproves the state-of-the-art performance by achieving a 1.27% FAR compared to\n8.73% and 17.31% recorded in literature. This performance is consistent across\nother datasets and guarantees that the risk of imposters gaining access to\nsensitive documents or transactions is minimal.\n", "date": "2024-01-05", "categories": []}
{"id": "2401.03037", "title": "CATFace: Cross-Attribute-Guided Transformer with Self-Attention\n  Distillation for Low-Quality Face Recognition", "abstract": "  Although face recognition (FR) has achieved great success in recent years, it\nis still challenging to accurately recognize faces in low-quality images due to\nthe obscured facial details. Nevertheless, it is often feasible to make\npredictions about specific soft biometric (SB) attributes, such as gender, and\nbaldness even in dealing with low-quality images. In this paper, we propose a\nnovel multi-branch neural network that leverages SB attribute information to\nboost the performance of FR. To this end, we propose a cross-attribute-guided\ntransformer fusion (CATF) module that effectively captures the long-range\ndependencies and relationships between FR and SB feature representations. The\nsynergy created by the reciprocal flow of information in the dual\ncross-attention operations of the proposed CATF module enhances the performance\nof FR. Furthermore, we introduce a novel self-attention distillation framework\nthat effectively highlights crucial facial regions, such as landmarks by\naligning low-quality images with those of their high-quality counterparts in\nthe feature space. The proposed self-attention distillation regularizes our\nnetwork to learn a unified quality-invariant feature representation in\nunconstrained environments. We conduct extensive experiments on various FR\nbenchmarks varying in quality. Experimental results demonstrate the superiority\nof our FR method compared to state-of-the-art FR studies.\n", "date": "2024-01-05", "categories": []}
{"id": "2401.02646", "title": "Recent Advancement in 3D Biometrics using Monocular Camera", "abstract": "  Recent literature has witnessed significant interest towards 3D biometrics\nemploying monocular vision for robust authentication methods. Motivated by\nthis, in this work we seek to provide insight on recent development in the area\nof 3D biometrics employing monocular vision. We present the similarity and\ndissimilarity of 3D monocular biometrics and classical biometrics, listing the\nstrengths and challenges. Further, we provide an overview of recent techniques\nin 3D biometrics with monocular vision, as well as application systems adopted\nby the industry. Finally, we discuss open research problems in this area of\nresearch\n", "date": "2024-01-05", "categories": []}
{"id": "2401.02649", "title": "Enhancing 3D-Air Signature by Pen Tip Tail Trajectory Awareness: Dataset\n  and Featuring by Novel Spatio-temporal CNN", "abstract": "  This work proposes a novel process of using pen tip and tail 3D trajectory\nfor air signature. To acquire the trajectories we developed a new pen tool and\na stereo camera was used. We proposed SliT-CNN, a novel 2D spatial-temporal\nconvolutional neural network (CNN) for better featuring of the air signature.\nIn addition, we also collected an air signature dataset from $45$ signers.\nSkilled forgery signatures per user are also collected. A detailed benchmarking\nof the proposed dataset using existing techniques and proposed CNN on existing\nand proposed dataset exhibit the effectiveness of our methodology.\n", "date": "2024-01-05", "categories": []}
{"id": "2401.02861", "title": "Reversing the Irreversible: A Survey on Inverse Biometrics", "abstract": "  With the widespread use of biometric recognition, several issues related to\nthe privacy and security provided by this technology have been recently raised\nand analysed. As a result, the early common belief among the biometrics\ncommunity of templates irreversibility has been proven wrong. It is now an\naccepted fact that it is possible to reconstruct from an unprotected template a\nsynthetic sample that matches the bona fide one. This reverse engineering\nprocess, commonly referred to as \\textit{inverse biometrics}, constitutes a\nsevere threat for biometric systems from two different angles: on the one hand,\nsensitive personal data (i.e., biometric data) can be derived from compromised\nunprotected templates; on the other hand, other powerful attacks can be\nlaunched building upon these reconstructed samples. Given its important\nimplications, biometric stakeholders have produced over the last fifteen years\nnumerous works analysing the different aspects related to inverse biometrics:\ndevelopment of reconstruction algorithms for different characteristics;\nproposal of methodologies to assess the vulnerabilities of biometric systems to\nthe aforementioned algorithms; development of countermeasures to reduce the\npossible effects of attacks. The present article is an effort to condense all\nthis information in one comprehensive review of: the problem itself, the\nevaluation of the problem, and the mitigation of the problem. The present\narticle is an effort to condense all this information in one comprehensive\nreview of: the problem itself, the evaluation of the problem, and the\nmitigation of the problem.\n", "date": "2024-01-05", "categories": []}
{"id": "2401.01839", "title": "Frequency Domain Modality-invariant Feature Learning for\n  Visible-infrared Person Re-Identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) is challenging due to the\nsignificant cross-modality discrepancies between visible and infrared images.\nWhile existing methods have focused on designing complex network architectures\nor using metric learning constraints to learn modality-invariant features, they\noften overlook which specific component of the image causes the modality\ndiscrepancy problem. In this paper, we first reveal that the difference in the\namplitude component of visible and infrared images is the primary factor that\ncauses the modality discrepancy and further propose a novel Frequency Domain\nmodality-invariant feature learning framework (FDMNet) to reduce modality\ndiscrepancy from the frequency domain perspective. Our framework introduces two\nnovel modules, namely the Instance-Adaptive Amplitude Filter (IAF) module and\nthe Phrase-Preserving Normalization (PPNorm) module, to enhance the\nmodality-invariant amplitude component and suppress the modality-specific\ncomponent at both the image- and feature-levels. Extensive experimental results\non two standard benchmarks, SYSU-MM01 and RegDB, demonstrate the superior\nperformance of our FDMNet against state-of-the-art methods.\n", "date": "2024-01-03", "categories": []}
{"id": "2401.02173", "title": "Prompt Decoupling for Text-to-Image Person Re-identification", "abstract": "  Text-to-image person re-identification (TIReID) aims to retrieve the target\nperson from an image gallery via a textual description query. Recently,\npre-trained vision-language models like CLIP have attracted significant\nattention and have been widely utilized for this task due to their robust\ncapacity for semantic concept learning and rich multi-modal knowledge. However,\nrecent CLIP-based TIReID methods commonly rely on direct fine-tuning of the\nentire network to adapt the CLIP model for the TIReID task. Although these\nmethods show competitive performance on this topic, they are suboptimal as they\nnecessitate simultaneous domain adaptation and task adaptation. To address this\nissue, we attempt to decouple these two processes during the training stage.\nSpecifically, we introduce the prompt tuning strategy to enable domain\nadaptation and propose a two-stage training approach to disentangle domain\nadaptation from task adaptation. In the first stage, we freeze the two encoders\nfrom CLIP and solely focus on optimizing the prompts to alleviate domain gap\nbetween the original training data of CLIP and downstream tasks. In the second\nstage, we maintain the fixed prompts and fine-tune the CLIP model to prioritize\ncapturing fine-grained information, which is more suitable for TIReID task.\nFinally, we evaluate the effectiveness of our method on three widely used\ndatasets. Compared to the directly fine-tuned approach, our method achieves\nsignificant improvements.\n", "date": "2024-01-04", "categories": []}
{"id": "2312.11195", "title": "Cross-Age Contrastive Learning for Age-Invariant Face Recognition", "abstract": "  Cross-age facial images are typically challenging and expensive to collect,\nmaking noise-free age-oriented datasets relatively small compared to\nwidely-used large-scale facial datasets. Additionally, in real scenarios,\nimages of the same subject at different ages are usually hard or even\nimpossible to obtain. Both of these factors lead to a lack of supervised data,\nwhich limits the versatility of supervised methods for age-invariant face\nrecognition, a critical task in applications such as security and biometrics.\nTo address this issue, we propose a novel semi-supervised learning approach\nnamed Cross-Age Contrastive Learning (CACon). Thanks to the identity-preserving\npower of recent face synthesis models, CACon introduces a new contrastive\nlearning method that leverages an additional synthesized sample from the input\nimage. We also propose a new loss function in association with CACon to perform\ncontrastive learning on a triplet of samples. We demonstrate that our method\nnot only achieves state-of-the-art performance in homogeneous-dataset\nexperiments on several age-invariant face recognition benchmarks but also\noutperforms other methods by a large margin in cross-dataset experiments.\n", "date": "2023-12-18", "categories": []}
{"id": "2309.11766", "title": "Dictionary Attack on IMU-based Gait Authentication", "abstract": "  We present a novel adversarial model for authentication systems that use gait\npatterns recorded by the inertial measurement unit (IMU) built into\nsmartphones. The attack idea is inspired by and named after the concept of a\ndictionary attack on knowledge (PIN or password) based authentication systems.\nIn particular, this work investigates whether it is possible to build a\ndictionary of IMUGait patterns and use it to launch an attack or find an\nimitator who can actively reproduce IMUGait patterns that match the target's\nIMUGait pattern. Nine physically and demographically diverse individuals walked\nat various levels of four predefined controllable and adaptable gait factors\n(speed, step length, step width, and thigh-lift), producing 178 unique IMUGait\npatterns. Each pattern attacked a wide variety of user authentication models.\nThe deeper analysis of error rates (before and after the attack) challenges the\nbelief that authentication systems based on IMUGait patterns are the most\ndifficult to spoof; further research is needed on adversarial models and\nassociated countermeasures.\n", "date": "2023-09-21", "categories": []}
{"id": "2312.10482", "title": "A new method color MS-BSIF Features learning for the robust kinship\n  verification", "abstract": "  the paper presents a new method color MS-BSIF learning and MS-LBP for the\nkinship verification is the machine's ability to identify the genetic and blood\nthe relationship and its degree between the facial images of humans. Facial\nverification of kinship refers to the task of training a machine to recognize\nthe blood relationship between a pair of faces parent and non-parent\n(verification) based on features extracted from facial images, and determining\nthe exact type or degree of this genetic relationship. We use the LBP and color\nBSIF learning features for the comparison and the TXQDA method for\ndimensionality reduction and data classification. We let's test the kinship\nfacial verification application is namely the kinface Cornell database. This\nsystem improves the robustness of learning while controlling efficiency. The\nexperimental results obtained and compared to other methods have proven the\nreliability of our framework and surpass the performance of other\nstate-of-the-art techniques.\n", "date": "2023-12-16", "categories": []}
{"id": "2312.16226", "title": "Advancing Person Re-Identification: Tensor-based Feature Fusion and\n  Multilinear Subspace Learning", "abstract": "  Person re-identification (PRe-ID) is a computer vision issue, that has been a\nfertile research area in the last few years. It aims to identify persons across\ndifferent non-overlapping camera views. In this paper, We propose a novel\nPRe-ID system that combines tensor feature representation and multilinear\nsubspace learning. Our method exploits the power of pre-trained Convolutional\nNeural Networks (CNNs) as a strong deep feature extractor, along with two\ncomplementary descriptors, Local Maximal Occurrence (LOMO) and Gaussian Of\nGaussian (GOG). Then, Tensor-based Cross-View Quadratic Discriminant Analysis\n(TXQDA) is used to learn a discriminative subspace that enhances the\nseparability between different individuals. Mahalanobis distance is used to\nmatch and similarity computation between query and gallery samples. Finally, we\nevaluate our approach by conducting experiments on three datasets VIPeR, GRID,\nand PRID450s.\n", "date": "2023-12-24", "categories": []}
{"id": "2312.16797", "title": "Multi-Prompts Learning with Cross-Modal Alignment for Attribute-based\n  Person Re-Identification", "abstract": "  The fine-grained attribute descriptions can significantly supplement the\nvaluable semantic information for person image, which is vital to the success\nof person re-identification (ReID) task. However, current ReID algorithms\ntypically failed to effectively leverage the rich contextual information\navailable, primarily due to their reliance on simplistic and coarse utilization\nof image attributes. Recent advances in artificial intelligence generated\ncontent have made it possible to automatically generate plentiful fine-grained\nattribute descriptions and make full use of them. Thereby, this paper explores\nthe potential of using the generated multiple person attributes as prompts in\nReID tasks with off-the-shelf (large) models for more accurate retrieval\nresults. To this end, we present a new framework called Multi-Prompts ReID\n(MP-ReID), based on prompt learning and language models, to fully dip fine\nattributes to assist ReID task. Specifically, MP-ReID first learns to\nhallucinate diverse, informative, and promptable sentences for describing the\nquery images. This procedure includes (i) explicit prompts of which attributes\na person has and furthermore (ii) implicit learnable prompts for\nadjusting/conditioning the criteria used towards this person identity matching.\nExplicit prompts are obtained by ensembling generation models, such as ChatGPT\nand VQA models. Moreover, an alignment module is designed to fuse multi-prompts\n(i.e., explicit and implicit ones) progressively and mitigate the cross-modal\ngap. Extensive experiments on the existing attribute-involved ReID datasets,\nnamely, Market1501 and DukeMTMC-reID, demonstrate the effectiveness and\nrationality of the proposed MP-ReID solution.\n", "date": "2023-12-28", "categories": []}
{"id": "2311.07475", "title": "Masked Face Dataset Generation and Masked Face Recognition", "abstract": "  In the post-pandemic era, wearing face masks has posed great challenge to the\nordinary face recognition. In the previous study, researchers has applied\npretrained VGG16, and ResNet50 to extract features on the elaborate curated\nexisting masked face recognition (MFR) datasets, RMFRD and SMFRD. To make the\nmodel more adaptable to the real world situation where the sample size is\nsmaller and the camera environment has greater changes, we created a more\nchallenging masked face dataset ourselves, by selecting 50 identities with 1702\nimages from Labelled Faces in the Wild (LFW) Dataset, and simulated face masks\nthrough key point detection. The another part of our study is to solve the\nmasked face recognition problem, and we chose models by referring to the former\nstate of the art results, instead of directly using pretrained models, we fine\ntuned the model on our new dataset and use the last linear layer to do the\nclassification directly. Furthermore, we proposed using data augmentation\nstrategy to further increase the test accuracy, and fine tuned a new networks\nbeyond the former study, one of the most SOTA networks, Inception ResNet v1.\nThe best test accuracy on 50 identity MFR has achieved 95%.\n", "date": "2023-11-13", "categories": []}
{"id": "2311.06000", "title": "Keystroke Verification Challenge (KVC): Biometric and Fairness Benchmark\n  Evaluation", "abstract": "  Analyzing keystroke dynamics (KD) for biometric verification has several\nadvantages: it is among the most discriminative behavioral traits; keyboards\nare among the most common human-computer interfaces, being the primary means\nfor users to enter textual data; its acquisition does not require additional\nhardware, and its processing is relatively lightweight; and it allows for\ntransparently recognizing subjects. However, the heterogeneity of experimental\nprotocols and metrics, and the limited size of the databases adopted in the\nliterature impede direct comparisons between different systems, thus\nrepresenting an obstacle in the advancement of keystroke biometrics. To\nalleviate this aspect, we present a new experimental framework to benchmark\nKD-based biometric verification performance and fairness based on tweet-long\nsequences of variable transcript text from over 185,000 subjects, acquired\nthrough desktop and mobile keyboards, extracted from the Aalto Keystroke\nDatabases. The framework runs on CodaLab in the form of the Keystroke\nVerification Challenge (KVC). Moreover, we also introduce a novel fairness\nmetric, the Skewed Impostor Ratio (SIR), to capture inter- and\nintra-demographic group bias patterns in the verification scores. We\ndemonstrate the usefulness of the proposed framework by employing two\nstate-of-the-art keystroke verification systems, TypeNet and TypeFormer, to\ncompare different sets of input features, achieving a less privacy-invasive\nsystem, by discarding the analysis of text content (ASCII codes of the keys\npressed) in favor of extended features in the time domain. Our experiments show\nthat this approach allows to maintain satisfactory performance.\n", "date": "2023-11-10", "categories": []}
{"id": "2312.14544", "title": "Inclusive normalization of face images to passport format", "abstract": "  Face recognition has been used more and more in real world applications in\nrecent years. However, when the skin color bias is coupled with intra-personal\nvariations like harsh illumination, the face recognition task is more likely to\nfail, even during human inspection. Face normalization methods try to deal with\nsuch challenges by removing intra-personal variations from an input image while\nkeeping the identity the same. However, most face normalization methods can\nonly remove one or two variations and ignore dataset biases such as skin color\nbias. The outputs of many face normalization methods are also not realistic to\nhuman observers. In this work, a style based face normalization model\n(StyleFNM) is proposed to remove most intra-personal variations including large\nchanges in pose, bad or harsh illumination, low resolution, blur, facial\nexpressions, and accessories like sunglasses among others. The dataset bias is\nalso dealt with in this paper by controlling a pretrained GAN to generate a\nbalanced dataset of passport-like images. The experimental results show that\nStyleFNM can generate more realistic outputs and can improve significantly the\naccuracy and fairness of face recognition systems.\n", "date": "2023-12-22", "categories": []}
{"id": "2211.14742", "title": "Dynamic Feature Pruning and Consolidation for Occluded Person\n  Re-Identification", "abstract": "  Occluded person re-identification (ReID) is a challenging problem due to\ncontamination from occluders. Existing approaches address the issue with prior\nknowledge cues, such as human body key points and semantic segmentations, which\neasily fail in the presence of heavy occlusion and other humans as occluders.\nIn this paper, we propose a feature pruning and consolidation (FPC) framework\nto circumvent explicit human structure parsing. The framework mainly consists\nof a sparse encoder, a multi-view feature mathcing module, and a feature\nconsolidation decoder. Specifically, the sparse encoder drops less important\nimage tokens, mostly related to background noise and occluders, solely based on\ncorrelation within the class token attention. Subsequently, the matching stage\nrelies on the preserved tokens produced by the sparse encoder to identify\nk-nearest neighbors in the gallery by measuring the image and patch-level\ncombined similarity. Finally, we use the feature consolidation module to\ncompensate pruned features using identified neighbors for recovering essential\ninformation while disregarding disturbance from noise and occlusion.\nExperimental results demonstrate the effectiveness of our proposed framework on\noccluded, partial, and holistic Re-ID datasets. In particular, our method\noutperforms state-of-the-art results by at least 8.6\\% mAP and 6.0\\% Rank-1\naccuracy on the challenging Occluded-Duke dataset.\n", "date": "2022-11-27", "categories": []}
{"id": "2312.05281", "title": "X2-Softmax: Margin Adaptive Loss Function for Face Recognition", "abstract": "  Learning the discriminative features of different faces is an important task\nin face recognition. By extracting face features in neural networks, it becomes\neasy to measure the similarity of different face images, which makes face\nrecognition possible. To enhance the neural network's face feature\nseparability, incorporating an angular margin during training is common\npractice. State-of-the-art loss functions CosFace and ArcFace apply fixed\nmargins between weights of classes to enhance the inter-class separation of\nface features. Since the distribution of samples in the training set is\nimbalanced, similarities between different identities are unequal. Therefore,\nusing an inappropriately fixed angular margin may lead to the problem that the\nmodel is difficult to converge or the face features are not discriminative\nenough. It is more in line with our intuition that the margins are angular\nadaptive, which could increase with the angles between classes growing. In this\npaper, we propose a new angular margin loss named X2-Softmax. X2-Softmax loss\nhas adaptive angular margins, which provide the margin that increases with the\nangle between different classes growing. The angular adaptive margin ensures\nmodel flexibility and effectively improves the effect of face recognition. We\nhave trained the neural network with X2-Softmax loss on the MS1Mv3 dataset and\ntested it on several evaluation benchmarks to demonstrate the effectiveness and\nsuperiority of our loss function.\n", "date": "2023-12-08", "categories": []}
{"id": "2312.07021", "title": "Transferring Modality-Aware Pedestrian Attentive Learning for\n  Visible-Infrared Person Re-identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to search the same\npedestrian of interest across visible and infrared modalities. Existing models\nmainly focus on compensating for modality-specific information to reduce\nmodality variation. However, these methods often lead to a higher computational\noverhead and may introduce interfering information when generating the\ncorresponding images or features. To address this issue, it is critical to\nleverage pedestrian-attentive features and learn modality-complete and\n-consistent representation. In this paper, a novel Transferring Modality-Aware\nPedestrian Attentive Learning (TMPA) model is proposed, focusing on the\npedestrian regions to efficiently compensate for missing modality-specific\nfeatures. Specifically, we propose a region-based data augmentation module\nPedMix to enhance pedestrian region coherence by mixing the corresponding\nregions from different modalities. A lightweight hybrid compensation module,\ni.e., the Modality Feature Transfer (MFT), is devised to integrate cross\nattention and convolution networks to fully explore the discriminative\nmodality-complete features with minimal computational overhead. Extensive\nexperiments conducted on the benchmark SYSU-MM01 and RegDB datasets\ndemonstrated the effectiveness of our proposed TMPA model.\n", "date": "2023-12-12", "categories": []}
{"id": "2311.13444", "title": "SkeletonGait: Gait Recognition Using Skeleton Maps", "abstract": "  The choice of the representations is essential for deep gait recognition\nmethods. The binary silhouettes and skeletal coordinates are two dominant\nrepresentations in recent literature, achieving remarkable advances in many\nscenarios. However, inherent challenges remain, in which silhouettes are not\nalways guaranteed in unconstrained scenes, and structural cues have not been\nfully utilized from skeletons. In this paper, we introduce a novel skeletal\ngait representation named skeleton map, together with SkeletonGait, a\nskeleton-based method to exploit structural information from human skeleton\nmaps. Specifically, the skeleton map represents the coordinates of human joints\nas a heatmap with Gaussian approximation, exhibiting a silhouette-like image\ndevoid of exact body structure. Beyond achieving state-of-the-art performances\nover five popular gait datasets, more importantly, SkeletonGait uncovers novel\ninsights about how important structural features are in describing gait and\nwhen they play a role. Furthermore, we propose a multi-branch architecture,\nnamed SkeletonGait++, to make use of complementary features from both skeletons\nand silhouettes. Experiments indicate that SkeletonGait++ outperforms existing\nstate-of-the-art methods by a significant margin in various scenarios. For\ninstance, it achieves an impressive rank-1 accuracy of over 85% on the\nchallenging GREW dataset. All the source code is available at\nhttps://github.com/ShiqiYu/OpenGait.\n", "date": "2023-11-22", "categories": []}
{"id": "2312.10470", "title": "Enhancing Person Re-Identification through Tensor Feature Fusion", "abstract": "  In this paper, we present a novel person reidentification (PRe-ID) system\nthat based on tensor feature representation and multilinear subspace learning.\nOur approach utilizes pretrained CNNs for high-level feature extraction, along\nwith Local Maximal Occurrence (LOMO) and Gaussian Of Gaussian (GOG )\ndescriptors. Additionally, Cross-View Quadratic Discriminant Analysis (TXQDA)\nalgorithm is used for multilinear subspace learning, which models the data in a\ntensor framework to enhance discriminative capabilities. Similarity measure\nbased on Mahalanobis distance is used for matching between training and test\npedestrian images. Experimental evaluations on VIPeR and PRID450s datasets\ndemonstrate the effectiveness of our method.\n", "date": "2023-12-16", "categories": []}
{"id": "2312.09627", "title": "TF-CLIP: Learning Text-free CLIP for Video-based Person\n  Re-Identification", "abstract": "  Large-scale language-image pre-trained models (e.g., CLIP) have shown\nsuperior performances on many cross-modal retrieval tasks. However, the problem\nof transferring the knowledge learned from such models to video-based person\nre-identification (ReID) has barely been explored. In addition, there is a lack\nof decent text descriptions in current ReID benchmarks. To address these\nissues, in this work, we propose a novel one-stage text-free CLIP-based\nlearning framework named TF-CLIP for video-based person ReID. More\nspecifically, we extract the identity-specific sequence feature as the\nCLIP-Memory to replace the text feature. Meanwhile, we design a\nSequence-Specific Prompt (SSP) module to update the CLIP-Memory online. To\ncapture temporal information, we further propose a Temporal Memory Diffusion\n(TMD) module, which consists of two key components: Temporal Memory\nConstruction (TMC) and Memory Diffusion (MD). Technically, TMC allows the\nframe-level memories in a sequence to communicate with each other, and to\nextract temporal information based on the relations within the sequence. MD\nfurther diffuses the temporal memories to each token in the original features\nto obtain more robust sequence features. Extensive experiments demonstrate that\nour proposed method shows much better results than other state-of-the-art\nmethods on MARS, LS-VID and iLIDS-VID. The code is available at\nhttps://github.com/AsuradaYuci/TF-CLIP.\n", "date": "2023-12-15", "categories": []}
{"id": "2312.09797", "title": "Part Representation Learning with Teacher-Student Decoder for Occluded\n  Person Re-identification", "abstract": "  Occluded person re-identification (ReID) is a very challenging task due to\nthe occlusion disturbance and incomplete target information. Leveraging\nexternal cues such as human pose or parsing to locate and align part features\nhas been proven to be very effective in occluded person ReID. Meanwhile, recent\nTransformer structures have a strong ability of long-range modeling.\nConsidering the above facts, we propose a Teacher-Student Decoder (TSD)\nframework for occluded person ReID, which utilizes the Transformer decoder with\nthe help of human parsing. More specifically, our proposed TSD consists of a\nParsing-aware Teacher Decoder (PTD) and a Standard Student Decoder (SSD). PTD\nemploys human parsing cues to restrict Transformer's attention and imparts this\ninformation to SSD through feature distillation. Thereby, SSD can learn from\nPTD to aggregate information of body parts automatically. Moreover, a mask\ngenerator is designed to provide discriminative regions for better ReID. In\naddition, existing occluded person ReID benchmarks utilize occluded samples as\nqueries, which will amplify the role of alleviating occlusion interference and\nunderestimate the impact of the feature absence issue. Contrastively, we\npropose a new benchmark with non-occluded queries, serving as a complement to\nthe existing benchmark. Extensive experiments demonstrate that our proposed\nmethod is superior and the new benchmark is essential. The source codes are\navailable at https://github.com/hh23333/TSD.\n", "date": "2023-12-15", "categories": []}
{"id": "2311.03828", "title": "Multi-view Information Integration and Propagation for Occluded Person\n  Re-identification", "abstract": "  Occluded person re-identification (re-ID) presents a challenging task due to\nocclusion perturbations. Although great efforts have been made to prevent the\nmodel from being disturbed by occlusion noise, most current solutions only\ncapture information from a single image, disregarding the rich complementary\ninformation available in multiple images depicting the same pedestrian. In this\npaper, we propose a novel framework called Multi-view Information Integration\nand Propagation (MVI$^{2}$P). Specifically, realizing the potential of\nmulti-view images in effectively characterizing the occluded target pedestrian,\nwe integrate feature maps of which to create a comprehensive representation.\nDuring this process, to avoid introducing occlusion noise, we develop a\nCAMs-aware Localization module that selectively integrates information\ncontributing to the identification. Additionally, considering the divergence in\nthe discriminative nature of different images, we design a probability-aware\nQuantification module to emphatically integrate highly reliable information.\nMoreover, as multiple images with the same identity are not accessible in the\ntesting stage, we devise an Information Propagation (IP) mechanism to distill\nknowledge from the comprehensive representation to that of a single occluded\nimage. Extensive experiments and analyses have unequivocally demonstrated the\neffectiveness and superiority of the proposed MVI$^{2}$P. The code will be\nreleased at \\url{https://github.com/nengdong96/MVIIP}.\n", "date": "2023-11-07", "categories": []}
{"id": "2312.07853", "title": "High-Order Structure Based Middle-Feature Learning for Visible-Infrared\n  Person Re-Identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to retrieve images\nof the same persons captured by visible (VIS) and infrared (IR) cameras.\nExisting VI-ReID methods ignore high-order structure information of features\nwhile being relatively difficult to learn a reasonable common feature space due\nto the large modality discrepancy between VIS and IR images. To address the\nabove problems, we propose a novel high-order structure based middle-feature\nlearning network (HOS-Net) for effective VI-ReID. Specifically, we first\nleverage a short- and long-range feature extraction (SLE) module to effectively\nexploit both short-range and long-range features. Then, we propose a high-order\nstructure learning (HSL) module to successfully model the high-order\nrelationship across different local features of each person image based on a\nwhitened hypergraph network.This greatly alleviates model collapse and enhances\nfeature representations. Finally, we develop a common feature space learning\n(CFL) module to learn a discriminative and reasonable common feature space\nbased on middle features generated by aligning features from different\nmodalities and ranges. In particular, a modality-range identity-center\ncontrastive (MRIC) loss is proposed to reduce the distances between the VIS,\nIR, and middle features, smoothing the training process. Extensive experiments\non the SYSU-MM01, RegDB, and LLCM datasets show that our HOS-Net achieves\nsuperior state-of-the-art performance. Our code is available at\n\\url{https://github.com/Jaulaucoeng/HOS-Net}.\n", "date": "2023-12-13", "categories": []}
{"id": "2308.14075", "title": "FaceCoresetNet: Differentiable Coresets for Face Set Recognition", "abstract": "  In set-based face recognition, we aim to compute the most discriminative\ndescriptor from an unbounded set of images and videos showing a single person.\nA discriminative descriptor balances two policies when aggregating information\nfrom a given set. The first is a quality-based policy: emphasizing high-quality\nand down-weighting low-quality images. The second is a diversity-based policy:\nemphasizing unique images in the set and down-weighting multiple occurrences of\nsimilar images as found in video clips which can overwhelm the set\nrepresentation. This work frames face-set representation as a differentiable\ncoreset selection problem. Our model learns how to select a small coreset of\nthe input set that balances quality and diversity policies using a learned\nmetric parameterized by the face quality, optimized end-to-end. The selection\nprocess is a differentiable farthest-point sampling (FPS) realized by\napproximating the non-differentiable Argmax operation with differentiable\nsampling from the Gumbel-Softmax distribution of distances. The small coreset\nis later used as queries in a self and cross-attention architecture to enrich\nthe descriptor with information from the whole set. Our model is\norder-invariant and linear in the input set size. We set a new SOTA to set face\nverification on the IJB-B and IJB-C datasets. Our code is publicly available.\n", "date": "2023-08-27", "categories": []}
{"id": "2312.08111", "title": "Towards Better Morphed Face Images without Ghosting Artifacts", "abstract": "  Automatic generation of morphed face images often produces ghosting artifacts\ndue to poorly aligned structures in the input images. Manual processing can\nmitigate these artifacts. However, this is not feasible for the generation of\nlarge datasets, which are required for training and evaluating robust morphing\nattack detectors. In this paper, we propose a method for automatic prevention\nof ghosting artifacts based on a pixel-wise alignment during morph generation.\nWe evaluate our proposed method on state-of-the-art detectors and show that our\nmorphs are harder to detect, particularly, when combined with\nstyle-transfer-based improvement of low-level image characteristics.\nFurthermore, we show that our approach does not impair the biometric quality,\nwhich is essential for high quality morphs.\n", "date": "2023-12-13", "categories": []}
{"id": "2304.14123", "title": "MCLFIQ: Mobile Contactless Fingerprint Image Quality", "abstract": "  We propose MCLFIQ: Mobile Contactless Fingerprint Image Quality, the first\nquality assessment algorithm for mobile contactless fingerprint samples. To\nthis end, we re-trained the NIST Fingerprint Image Quality (NFIQ) 2 method,\nwhich was originally designed for contact-based fingerprints, with a synthetic\ncontactless fingerprint database. We evaluate the predictive performance of the\nresulting MCLFIQ model in terms of Error-vs.-Discard Characteristic (EDC)\ncurves on three real-world contactless fingerprint databases using three\nrecognition algorithms. In experiments, the MCLFIQ method is compared against\nthe original NFIQ 2.2 method, a sharpness-based quality assessment algorithm\ndeveloped for contactless fingerprint images \\rev{and the general purpose image\nquality assessment method BRISQUE. Furthermore, benchmarks on four\ncontact-based fingerprint datasets are also conducted.}\n  Obtained results show that the fine-tuning of NFIQ 2 on synthetic contactless\nfingerprints is a viable alternative to training on real databases. Moreover,\nthe evaluation shows that our MCLFIQ method works more accurate and robust\ncompared to all baseline methods on contactless fingerprints. We suggest\nconsidering the proposed MCLFIQ method as a \\rev{starting point for the\ndevelopment of} a new standard algorithm for contactless fingerprint quality\nassessment.\n", "date": "2023-04-27", "categories": []}
{"id": "2312.04029", "title": "Improved Face Representation via Joint Label Classification and\n  Supervised Contrastive Clustering", "abstract": "  Face clustering tasks can learn hierarchical semantic information from\nlarge-scale data, which has the potential to help facilitate face recognition.\nHowever, there are few works on this problem. This paper explores it by\nproposing a joint optimization task of label classification and supervised\ncontrastive clustering to introduce the cluster knowledge to the traditional\nface recognition task in two ways. We first extend ArcFace with a\ncluster-guided angular margin to adjust the within-class feature distribution\naccording to the hard level of face clustering. Secondly, we propose a\nsupervised contrastive clustering approach to pull the features to the cluster\ncenter and propose the cluster-aligning procedure to align the cluster center\nand the learnable class center in the classifier for joint training. Finally,\nextensive qualitative and quantitative experiments on popular facial benchmarks\ndemonstrate the effectiveness of our paradigm and its superiority over the\nexisting approaches to face recognition.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.04125", "title": "Forensic Iris Image Synthesis", "abstract": "  Post-mortem iris recognition is an emerging application of iris-based human\nidentification in a forensic setup, able to correctly identify deceased\nsubjects even three weeks post-mortem. This technique thus is considered as an\nimportant component of future forensic toolkits. The current advancements in\nthis field are seriously slowed down by exceptionally difficult data\ncollection, which can happen in mortuary conditions, at crime scenes, or in\n``body farm'' facilities. This paper makes a novel contribution to facilitate\nprogress in post-mortem iris recognition by offering a conditional\nStyleGAN-based iris synthesis model, trained on the largest-available dataset\nof post-mortem iris samples acquired from more than 350 subjects, generating --\nthrough appropriate exploration of StyleGAN latent space -- multiple\nwithin-class (same identity) and between-class (different new identities)\npost-mortem iris images, compliant with ISO/IEC 29794-6, and with decomposition\ndeformations controlled by the requested PMI (post mortem interval). Besides an\nobvious application to enhance the existing, very sparse, post-mortem iris\ndatasets to advance -- among others -- iris presentation attack endeavors, we\nanticipate it may be useful to generate samples that would expose professional\nforensic human examiners to never-seen-before deformations for various PMIs,\nincreasing their training effectiveness. The source codes and model weights are\nmade available with the paper.\n", "date": "2023-12-07", "categories": []}
{"id": "2312.03227", "title": "Human Body Model based ID using Shape and Pose Parameters", "abstract": "  We present a Human Body model based IDentification system (HMID) system that\nis jointly trained for shape, pose and biometric identification. HMID is based\non the Human Mesh Recovery (HMR) network and we propose additional losses to\nimprove and stabilize shape estimation and biometric identification while\nmaintaining the pose and shape output. We show that when our HMID network is\ntrained using additional shape and pose losses, it shows a significant\nimprovement in biometric identification performance when compared to an\nidentical model that does not use such losses. The HMID model uses raw images\ninstead of silhouettes and is able to perform robust recognition on images\ncollected at range and altitude as many anthropometric properties are\nreasonably invariant to clothing, view and range. We show results on the USF\ndataset as well as the BRIAR dataset which includes probes with both clothing\nand view changes. Our approach (using body model losses) shows a significant\nimprovement in Rank20 accuracy and True Accuracy Rate on the BRIAR evaluation\ndataset.\n", "date": "2023-12-06", "categories": []}
{"id": "2312.03562", "title": "Enhancing Kinship Verification through Multiscale Retinex and Combined\n  Deep-Shallow features", "abstract": "  The challenge of kinship verification from facial images represents a\ncutting-edge and formidable frontier in the realms of pattern recognition and\ncomputer vision. This area of study holds a myriad of potential applications,\nspanning from image annotation and forensic analysis to social media research.\nOur research stands out by integrating a preprocessing method named Multiscale\nRetinex (MSR), which elevates image quality and amplifies contrast, ultimately\nbolstering the end results. Strategically, our methodology capitalizes on the\nharmonious blend of deep and shallow texture descriptors, merging them\nproficiently at the score level through the Logistic Regression (LR) method. To\nelucidate, we employ the Local Phase Quantization (LPQ) descriptor to extract\nshallow texture characteristics. For deep feature extraction, we turn to the\nprowess of the VGG16 model, which is pre-trained on a convolutional neural\nnetwork (CNN). The robustness and efficacy of our method have been put to the\ntest through meticulous experiments on three rigorous kinship datasets, namely:\nCornell Kin Face, UB Kin Face, and TS Kin Face.\n", "date": "2023-12-06", "categories": []}
{"id": "2308.11487", "title": "Free Lunch for Gait Recognition: A Novel Relation Descriptor", "abstract": "  Gait recognition is to seek correct matches for query individuals by their\nunique walking patterns. However, current methods focus solely on extracting\nindividual-specific features, overlooking ``interpersonal\" relationships. In\nthis paper, we propose a novel $\\textbf{Relation Descriptor}$ that captures not\nonly individual features but also relations between test gaits and pre-selected\ngait anchors. Specifically, we reinterpret classifier weights as gait anchors\nand compute similarity scores between test features and these anchors, which\nre-expresses individual gait features into a similarity relation distribution.\nIn essence, the relation descriptor offers a holistic perspective that\nleverages the collective knowledge stored within the classifier's weights,\nemphasizing meaningful patterns and enhancing robustness. Despite its\npotential, relation descriptor poses dimensionality challenges since its\ndimension depends on the training set's identity count. To address this, we\npropose Farthest gait-Anchor Selection to identify the most discriminative gait\nanchors and an Orthogonal Regularization Loss to increase diversity within gait\nanchors. Compared to individual-specific features extracted from the backbone,\nour relation descriptor can boost the performance nearly without any extra\ncosts. We evaluate the effectiveness of our method on the popular GREW, Gait3D,\nOU-MVLP, CASIA-B, and CCPG, showing that our method consistently outperforms\nthe baselines and achieves state-of-the-art performance.\n", "date": "2023-08-22", "categories": []}
{"id": "2312.00072", "title": "CRAFT: Contextual Re-Activation of Filters for face recognition Training", "abstract": "  The first layer of a deep CNN backbone applies filters to an image to extract\nthe basic features available to later layers. During training, some filters may\ngo inactive, mean ing all weights in the filter approach zero. An inactive fil\nter in the final model represents a missed opportunity to extract a useful\nfeature. This phenomenon is especially prevalent in specialized CNNs such as\nfor face recogni tion (as opposed to, e.g., ImageNet). For example, in one the\nmost widely face recognition model (ArcFace), about half of the convolution\nfilters in the first layer are inactive. We propose a novel approach designed\nand tested specif ically for face recognition networks, known as \"CRAFT:\nContextual Re-Activation of Filters for Face Recognition Training\". CRAFT\nidentifies inactive filters during training and reinitializes them based on the\ncontext of strong filters at that stage in training. We show that CRAFT reduces\nfraction of inactive filters from 44% to 32% on average and discovers filter\npatterns not found by standard training. Compared to standard training without\nreactivation, CRAFT demonstrates enhanced model accuracy on standard\nface-recognition benchmark datasets including AgeDB-30, CPLFW, LFW, CALFW, and\nCFP-FP, as well as on more challenging datasets like IJBB and IJBC.\n", "date": "2023-11-29", "categories": []}
{"id": "2312.00802", "title": "Continuous Authentication Using Mouse Clickstream Data Analysis", "abstract": "  Biometrics is used to authenticate an individual based on physiological or\nbehavioral traits. Mouse dynamics is an example of a behavioral biometric that\ncan be used to perform continuous authentication as protection against security\nbreaches. Recent research on mouse dynamics has shown promising results in\nidentifying users; however, it has not yet reached an acceptable level of\naccuracy. In this paper, an empirical evaluation of different classification\ntechniques is conducted on a mouse dynamics dataset, the Balabit Mouse\nChallenge dataset. User identification is carried out using three mouse\nactions: mouse move, point and click, and drag and drop. Verification and\nauthentication methods are conducted using three machine-learning classifiers:\nthe Decision Tree classifier, the K-Nearest Neighbors classifier, and the\nRandom Forest classifier. The results show that the three classifiers can\ndistinguish between a genuine user and an impostor with a relatively high\ndegree of accuracy. In the verification mode, all the classifiers achieve a\nperfect accuracy of 100%. In authentication mode, all three classifiers\nachieved the highest accuracy (ACC) and Area Under Curve (AUC) from scenario B\nusing the point and click action data: (Decision Tree ACC:87.6%, AUC:90.3%),\n(K-Nearest Neighbors ACC:99.3%, AUC:99.9%), and (Random Forest ACC:89.9%,\nAUC:92.5%).\n", "date": "2023-11-23", "categories": []}
{"id": "2312.01367", "title": "DiFace: Cross-Modal Face Recognition through Controlled Diffusion", "abstract": "  Diffusion probabilistic models (DPMs) have exhibited exceptional proficiency\nin generating visual media of outstanding quality and realism. Nonetheless,\ntheir potential in non-generative domains, such as face recognition, has yet to\nbe thoroughly investigated. Meanwhile, despite the extensive development of\nmulti-modal face recognition methods, their emphasis has predominantly centered\non visual modalities. In this context, face recognition through textual\ndescription presents a unique and promising solution that not only transcends\nthe limitations from application scenarios but also expands the potential for\nresearch in the field of cross-modal face recognition. It is regrettable that\nthis avenue remains unexplored and underutilized, a consequence from the\nchallenges mainly associated with three aspects: 1) the intrinsic imprecision\nof verbal descriptions; 2) the significant gaps between texts and images; and\n3) the immense hurdle posed by insufficient databases.To tackle this problem,\nwe present DiFace, a solution that effectively achieves face recognition via\ntext through a controllable diffusion process, by establishing its theoretical\nconnection with probability transport. Our approach not only unleashes the\npotential of DPMs across a broader spectrum of tasks but also achieves, to the\nbest of our knowledge, a significant accuracy in text-to-image face recognition\nfor the first time, as demonstrated by our experiments on verification and\nidentification.\n", "date": "2023-12-03", "categories": []}
{"id": "2312.01605", "title": "TextAug: Test time Text Augmentation for Multimodal Person\n  Re-identification", "abstract": "  Multimodal Person Reidentification is gaining popularity in the research\ncommunity due to its effectiveness compared to counter-part unimodal\nframeworks. However, the bottleneck for multimodal deep learning is the need\nfor a large volume of multimodal training examples. Data augmentation\ntechniques such as cropping, flipping, rotation, etc. are often employed in the\nimage domain to improve the generalization of deep learning models. Augmenting\nin other modalities than images, such as text, is challenging and requires\nsignificant computational resources and external data sources. In this study,\nwe investigate the effectiveness of two computer vision data augmentation\ntechniques: cutout and cutmix, for text augmentation in multi-modal person\nre-identification. Our approach merges these two augmentation strategies into\none strategy called CutMixOut which involves randomly removing words or\nsub-phrases from a sentence (Cutout) and blending parts of two or more\nsentences to create diverse examples (CutMix) with a certain probability\nassigned to each operation. This augmentation was implemented at inference time\nwithout any prior training. Our results demonstrate that the proposed technique\nis simple and effective in improving the performance on multiple multimodal\nperson re-identification benchmarks.\n", "date": "2023-12-04", "categories": []}
{"id": "2312.00019", "title": "The theoretical limits of biometry", "abstract": "  Biometry has proved its capability in terms of recognition accuracy. Now, it\nis widely used for automated border control with the biometric passport, to\nunlock a smartphone or a computer with a fingerprint or a face recognition\nalgorithm. While identity verification is widely democratized, pure\nidentification with no additional clues is still a work in progress. The\nidentification difficulty depends on the population size, as the larger the\ngroup is, the larger the confusion risk. For collision prevention, biometric\ntraits must be sufficiently distinguishable to scale to considerable groups,\nand algorithms should be able to capture their differences accurately.\n  Most biometric works are purely experimental, and it is impossible to\nextrapolate the results to a smaller or a larger group. In this work, we\npropose a theoretical analysis of the distinguishability problem, which governs\nthe error rates of biometric systems. We demonstrate simple relationships\nbetween the population size and the number of independent bits necessary to\nprevent collision in the presence of noise. This work provides the lowest lower\nbound for memory requirements. The results are very encouraging, as the\nbiometry of the whole Earth population can fit in a regular disk, leaving some\nspace for noise and redundancy.\n", "date": "2023-11-06", "categories": []}
{"id": "2312.00040", "title": "Presentation Attack detection using Wavelet Transform and Deep Residual\n  Neural Net", "abstract": "  Biometric authentication is becoming more prevalent for secured\nauthentication systems. However, the biometric substances can be deceived by\nthe imposters in several ways. Among other imposter attacks, print attacks,\nmask attacks, and replay attacks fall under the presentation attack category.\nThe bio-metric images, especially the iris and face, are vulnerable to\ndifferent presentation attacks. This research applies deep learning approaches\nto mitigate presentation attacks in a biometric access control system. Our\ncontribution in this paper is two-fold: First, we applied the wavelet transform\nto extract the features from the biometric images. Second, we modified the deep\nresidual neural net and applied it to the spoof datasets in an attempt to\ndetect the presentation attacks. This research applied the proposed approach to\nbiometric spoof datasets, namely ATVS, CASIA two class, and CASIA cropped image\nsets. The datasets used in this research contain images that are captured in\nboth a controlled and uncontrolled environment along with different resolutions\nand sizes. We obtained the best accuracy of 93% on the ATVS Iris datasets. For\nCASIA two class and CASIA cropped datasets, we achieved test accuracies of 91%\nand 82%, respectively.\n", "date": "2023-11-23", "categories": []}
{"id": "2312.00041", "title": "Presentation Attack Detection using Convolutional Neural Networks and\n  Local Binary Patterns", "abstract": "  The use of biometrics to authenticate users and control access to secure\nareas has become extremely popular in recent years, and biometric access\ncontrol systems are frequently used by both governments and private\ncorporations. However, these systems may represent risks to security when\ndeployed without considering the possibility of biometric presentation attacks\n(also known as spoofing). Presentation attacks are a serious threat because\nthey do not require significant time, expense, or skill to carry out while\nremaining effective against many biometric systems in use today. This research\ncompares three different software-based methods for facial and iris\npresentation attack detection in images. The first method uses Inception-v3, a\npre-trained deep Convolutional Neural Network (CNN) made by Google for the\nImageNet challenge, which is retrained for this problem. The second uses a\nshallow CNN based on a modified Spoofnet architecture, which is trained\nnormally. The third is a texture-based method using Local Binary Patterns\n(LBP). The datasets used are the ATVS-FIr dataset, which contains real and fake\niris images, and the CASIA Face Anti-Spoofing Dataset, which contains real\nimages as well as warped photos, cut photos, and video replay presentation\nattacks. We also present a third set of results, based on cropped versions of\nthe CASIA images.\n", "date": "2023-11-23", "categories": []}
{"id": "2311.17339", "title": "RADAP: A Robust and Adaptive Defense Against Diverse Adversarial Patches\n  on Face Recognition", "abstract": "  Face recognition (FR) systems powered by deep learning have become widely\nused in various applications. However, they are vulnerable to adversarial\nattacks, especially those based on local adversarial patches that can be\nphysically applied to real-world objects. In this paper, we propose RADAP, a\nrobust and adaptive defense mechanism against diverse adversarial patches in\nboth closed-set and open-set FR systems. RADAP employs innovative techniques,\nsuch as FCutout and F-patch, which use Fourier space sampling masks to improve\nthe occlusion robustness of the FR model and the performance of the patch\nsegmenter. Moreover, we introduce an edge-aware binary cross-entropy (EBCE)\nloss function to enhance the accuracy of patch detection. We also present the\nsplit and fill (SAF) strategy, which is designed to counter the vulnerability\nof the patch segmenter to complete white-box adaptive attacks. We conduct\ncomprehensive experiments to validate the effectiveness of RADAP, which shows\nsignificant improvements in defense performance against various adversarial\npatches, while maintaining clean accuracy higher than that of the undefended\nVanilla model.\n", "date": "2023-11-29", "categories": []}
{"id": "2311.17583", "title": "CLIPC8: Face liveness detection algorithm based on image-text pairs and\n  contrastive learning", "abstract": "  Face recognition technology is widely used in the financial field, and\nvarious types of liveness attack behaviors need to be addressed. Existing\nliveness detection algorithms are trained on specific training datasets and\ntested on testing datasets, but their performance and robustness in\ntransferring to unseen datasets are relatively poor. To tackle this issue, we\npropose a face liveness detection method based on image-text pairs and\ncontrastive learning, dividing liveness attack problems in the financial field\ninto eight categories and using text information to describe the images of\nthese eight types of attacks. The text encoder and image encoder are used to\nextract feature vector representations for the classification description text\nand face images, respectively. By maximizing the similarity of positive samples\nand minimizing the similarity of negative samples, the model learns shared\nrepresentations between images and texts. The proposed method is capable of\neffectively detecting specific liveness attack behaviors in certain scenarios,\nsuch as those occurring in dark environments or involving the tampering of ID\ncard photos. Additionally, it is also effective in detecting traditional\nliveness attack methods, such as printing photo attacks and screen remake\nattacks. The zero-shot capabilities of face liveness detection on five public\ndatasets, including NUAA, CASIA-FASD, Replay-Attack, OULU-NPU and MSU-MFSD also\nreaches the level of commercial algorithms. The detection capability of\nproposed algorithm was verified on 5 types of testing datasets, and the results\nshow that the method outperformed commercial algorithms, and the detection\nrates reached 100% on multiple datasets. Demonstrating the effectiveness and\nrobustness of introducing image-text pairs and contrastive learning into\nliveness detection tasks as proposed in this paper.\n", "date": "2023-11-29", "categories": []}
{"id": "2311.17655", "title": "Vulnerability of Automatic Identity Recognition to Audio-Visual\n  Deepfakes", "abstract": "  The task of deepfakes detection is far from being solved by speech or vision\nresearchers. Several publicly available databases of fake synthetic video and\nspeech were built to aid the development of detection methods. However,\nexisting databases typically focus on visual or voice modalities and provide no\nproof that their deepfakes can in fact impersonate any real person. In this\npaper, we present the first realistic audio-visual database of deepfakes\nSWAN-DF, where lips and speech are well synchronized and video have high visual\nand audio qualities. We took the publicly available SWAN dataset of real videos\nwith different identities to create audio-visual deepfakes using several models\nfrom DeepFaceLab and blending techniques for face swapping and HiFiVC, DiffVC,\nYourTTS, and FreeVC models for voice conversion. From the publicly available\nspeech dataset LibriTTS, we also created a separate database of only audio\ndeepfakes LibriTTS-DF using several latest text to speech methods: YourTTS,\nAdaspeech, and TorToiSe. We demonstrate the vulnerability of a state of the art\nspeaker recognition system, such as ECAPA-TDNN-based model from SpeechBrain, to\nthe synthetic voices. Similarly, we tested face recognition system based on the\nMobileFaceNet architecture to several variants of our visual deepfakes. The\nvulnerability assessment show that by tuning the existing pretrained deepfake\nmodels to specific identities, one can successfully spoof the face and speaker\nrecognition systems in more than 90% of the time and achieve a very realistic\nlooking and sounding fake video of a given person.\n", "date": "2023-11-29", "categories": []}
{"id": "2307.07187", "title": "Erasing, Transforming, and Noising Defense Network for Occluded Person\n  Re-Identification", "abstract": "  Occlusion perturbation presents a significant challenge in person\nre-identification (re-ID), and existing methods that rely on external visual\ncues require additional computational resources and only consider the issue of\nmissing information caused by occlusion. In this paper, we propose a simple yet\neffective framework, termed Erasing, Transforming, and Noising Defense Network\n(ETNDNet), which treats occlusion as a noise disturbance and solves occluded\nperson re-ID from the perspective of adversarial defense. In the proposed\nETNDNet, we introduce three strategies: Firstly, we randomly erase the feature\nmap to create an adversarial representation with incomplete information,\nenabling adversarial learning of identity loss to protect the re-ID system from\nthe disturbance of missing information. Secondly, we introduce random\ntransformations to simulate the position misalignment caused by occlusion,\ntraining the extractor and classifier adversarially to learn robust\nrepresentations immune to misaligned information. Thirdly, we perturb the\nfeature map with random values to address noisy information introduced by\nobstacles and non-target pedestrians, and employ adversarial gaming in the\nre-ID system to enhance its resistance to occlusion noise. Without bells and\nwhistles, ETNDNet has three key highlights: (i) it does not require any\nexternal modules with parameters, (ii) it effectively handles various issues\ncaused by occlusion from obstacles and non-target pedestrians, and (iii) it\ndesigns the first GAN-based adversarial defense paradigm for occluded person\nre-ID. Extensive experiments on five public datasets fully demonstrate the\neffectiveness, superiority, and practicality of the proposed ETNDNet. The code\nwill be released at \\url{https://github.com/nengdong96/ETNDNet}.\n", "date": "2023-07-14", "categories": []}
{"id": "2311.15163", "title": "Deep Learning-Based Approaches for Contactless Fingerprints Segmentation\n  and Extraction", "abstract": "  Fingerprints are widely recognized as one of the most unique and reliable\ncharacteristics of human identity. Most modern fingerprint authentication\nsystems rely on contact-based fingerprints, which require the use of\nfingerprint scanners or fingerprint sensors for capturing fingerprints during\nthe authentication process. Various types of fingerprint sensors, such as\noptical, capacitive, and ultrasonic sensors, employ distinct techniques to\ngather and analyze fingerprint data. This dependency on specific hardware or\nsensors creates a barrier or challenge for the broader adoption of fingerprint\nbased biometric systems. This limitation hinders the widespread adoption of\nfingerprint authentication in various applications and scenarios. Border\ncontrol, healthcare systems, educational institutions, financial transactions,\nand airport security face challenges when fingerprint sensors are not\nuniversally available. To mitigate the dependence on additional hardware, the\nuse of contactless fingerprints has emerged as an alternative. Developing\nprecise fingerprint segmentation methods, accurate fingerprint extraction\ntools, and reliable fingerprint matchers are crucial for the successful\nimplementation of a robust contactless fingerprint authentication system. This\npaper focuses on the development of a deep learning-based segmentation tool for\ncontactless fingerprint localization and segmentation. Our system leverages\ndeep learning techniques to achieve high segmentation accuracy and reliable\nextraction of fingerprints from contactless fingerprint images. In our\nevaluation, our segmentation method demonstrated an average mean absolute error\n(MAE) of 30 pixels, an error in angle prediction (EAP) of 5.92 degrees, and a\nlabeling accuracy of 97.46%. These results demonstrate the effectiveness of our\nnovel contactless fingerprint segmentation and extraction tools.\n", "date": "2023-11-26", "categories": []}
{"id": "2311.15326", "title": "Lightweight Face Recognition: An Improved MobileFaceNet Model", "abstract": "  This paper presents an extensive exploration and comparative analysis of\nlightweight face recognition (FR) models, specifically focusing on\nMobileFaceNet and its modified variant, MMobileFaceNet. The need for efficient\nFR models on devices with limited computational resources has led to the\ndevelopment of models with reduced memory footprints and computational demands\nwithout sacrificing accuracy. Our research delves into the impact of dataset\nselection, model architecture, and optimization algorithms on the performance\nof FR models. We highlight our participation in the EFaR-2023 competition,\nwhere our models showcased exceptional performance, particularly in categories\nrestricted by the number of parameters. By employing a subset of the Webface42M\ndataset and integrating sharpness-aware minimization (SAM) optimization, we\nachieved significant improvements in accuracy across various benchmarks,\nincluding those that test for cross-pose, cross-age, and cross-ethnicity\nperformance. The results underscore the efficacy of our approach in crafting\nmodels that are not only computationally efficient but also maintain high\naccuracy in diverse conditions.\n", "date": "2023-11-26", "categories": []}
{"id": "2311.15571", "title": "Video-based Visible-Infrared Person Re-Identification with Auxiliary\n  Samples", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to match persons\ncaptured by visible and infrared cameras, allowing person retrieval and\ntracking in 24-hour surveillance systems. Previous methods focus on learning\nfrom cross-modality person images in different cameras. However, temporal\ninformation and single-camera samples tend to be neglected. To crack this nut,\nin this paper, we first contribute a large-scale VI-ReID dataset named\nBUPTCampus. Different from most existing VI-ReID datasets, it 1) collects\ntracklets instead of images to introduce rich temporal information, 2) contains\npixel-aligned cross-modality sample pairs for better modality-invariant\nlearning, 3) provides one auxiliary set to help enhance the optimization, in\nwhich each identity only appears in a single camera. Based on our constructed\ndataset, we present a two-stream framework as baseline and apply Generative\nAdversarial Network (GAN) to narrow the gap between the two modalities. To\nexploit the advantages introduced by the auxiliary set, we propose a curriculum\nlearning based strategy to jointly learn from both primary and auxiliary sets.\nMoreover, we design a novel temporal k-reciprocal re-ranking method to refine\nthe ranking list with fine-grained temporal correlation cues. Experimental\nresults demonstrate the effectiveness of the proposed methods. We also\nreproduce 9 state-of-the-art image-based and video-based VI-ReID methods on\nBUPTCampus and our methods show substantial superiority to them. The codes and\ndataset are available at: https://github.com/dyhBUPT/BUPTCampus.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.15888", "title": "Towards Adaptive RF Fingerprint-based Authentication of IIoT devices", "abstract": "  As IoT technologies mature, they are increasingly finding their way into more\nsensitive domains, such as Medical and Industrial IoT, in which safety and\ncyber-security are of great importance. While the number of deployed IoT\ndevices continues to increase exponentially, they still present severe\ncyber-security vulnerabilities. Effective authentication is paramount to\nsupport trustworthy IIoT communications, however, current solutions focus on\nupper-layer identity verification or key-based cryptography which are often\ninadequate to the heterogeneous IIoT environment. In this work, we present a\nfirst step towards achieving powerful and flexible IIoT device authentication,\nby leveraging AI adaptive Radio Frequency Fingerprinting technique selection\nand tuning, at the PHY layer for highly accurate device authentication over\nchallenging RF environments.\n", "date": "2023-11-27", "categories": []}
{"id": "2311.14195", "title": "Touch Analysis: An Empirical Evaluation of Machine Learning\n  Classification Algorithms on Touch Data", "abstract": "  Our research aims at classifying individuals based on their unique\ninteractions on touchscreen-based smartphones. In this research, we use\nTouch-Analytics datasets, which include 41 subjects and 30 different behavioral\nfeatures. Furthermore, we derived new features from the raw data to improve the\noverall authentication performance. Previous research has already been done on\nthe Touch-Analytics datasets with the state-of-the-art classifiers, including\nSupport Vector Machine (SVM) and k-nearest neighbor (kNN), and achieved equal\nerror rates (EERs) between 0% to 4%. Here, we propose a novel Deep Neural Net\n(DNN) architecture to classify the individuals correctly. The proposed DNN\narchitecture has three dense layers and uses many-to-many mapping techniques.\nWhen we combine the new features with the existing ones, SVM and kNN achieved\nthe classification accuracy of 94.7% and 94.6%, respectively. This research\nexplored seven other classifiers and out of them, the decision tree and our\nproposed DNN classifiers resulted in the highest accuracy of 100%. The others\nincluded: Logistic Regression (LR), Linear Discriminant Analysis (LDA),\nGaussian Naive Bayes (NB), Neural Network, and VGGNet with the following\naccuracy scores of 94.7%, 95.9%, 31.9%, 88.8%, and 96.1%, respectively.\n", "date": "2023-11-23", "categories": []}
{"id": "2311.12764", "title": "Investigating Weight-Perturbed Deep Neural Networks With Application in\n  Iris Presentation Attack Detection", "abstract": "  Deep neural networks (DNNs) exhibit superior performance in various machine\nlearning tasks, e.g., image classification, speech recognition, biometric\nrecognition, object detection, etc. However, it is essential to analyze their\nsensitivity to parameter perturbations before deploying them in real-world\napplications. In this work, we assess the sensitivity of DNNs against\nperturbations to their weight and bias parameters. The sensitivity analysis\ninvolves three DNN architectures (VGG, ResNet, and DenseNet), three types of\nparameter perturbations (Gaussian noise, weight zeroing, and weight scaling),\nand two settings (entire network and layer-wise). We perform experiments in the\ncontext of iris presentation attack detection and evaluate on two publicly\navailable datasets: LivDet-Iris-2017 and LivDet-Iris-2020. Based on the\nsensitivity analysis, we propose improved models simply by perturbing\nparameters of the network without undergoing training. We further combine these\nperturbed models at the score-level and at the parameter-level to improve the\nperformance over the original model. The ensemble at the parameter-level shows\nan average improvement of 43.58% on the LivDet-Iris-2017 dataset and 9.25% on\nthe LivDet-Iris-2020 dataset. The source code is available at\nhttps://github.com/redwankarimsony/WeightPerturbation-MSU.\n", "date": "2023-11-21", "categories": []}
{"id": "2311.11512", "title": "Seeing through the Mask: Multi-task Generative Mask Decoupling Face\n  Recognition", "abstract": "  The outbreak of COVID-19 pandemic make people wear masks more frequently than\never. Current general face recognition system suffers from serious performance\ndegradation,when encountering occluded scenes. The potential reason is that\nface features are corrupted by occlusions on key facial regions. To tackle this\nproblem, previous works either extract identity-related embeddings on feature\nlevel by additional mask prediction, or restore the occluded facial part by\ngenerative models. However, the former lacks visual results for model\ninterpretation, while the latter suffers from artifacts which may affect\ndownstream recognition. Therefore, this paper proposes a Multi-task gEnerative\nmask dEcoupling face Recognition (MEER) network to jointly handle these two\ntasks, which can learn occlusionirrelevant and identity-related representation\nwhile achieving unmasked face synthesis. We first present a novel mask\ndecoupling module to disentangle mask and identity information, which makes the\nnetwork obtain purer identity features from visible facial components. Then, an\nunmasked face is restored by a joint-training strategy, which will be further\nused to refine the recognition network with an id-preserving loss. Experiments\non masked face recognition under realistic and synthetic occlusions benchmarks\ndemonstrate that the MEER can outperform the state-ofthe-art methods.\n", "date": "2023-11-20", "categories": []}
{"id": "2304.04400", "title": "Identity-Guided Collaborative Learning for Cloth-Changing Person\n  Reidentification", "abstract": "  Cloth-changing person reidentification (ReID) is a newly emerging research\ntopic that is aimed at addressing the issues of large feature variations due to\ncloth-changing and pedestrian view/pose changes. Although significant progress\nhas been achieved by introducing extra information (e.g., human contour\nsketching information, human body keypoints, and 3D human information),\ncloth-changing person ReID is still challenging due to impressionable\npedestrian representations. Moreover, human semantic information and pedestrian\nidentity information are not fully explored. To solve these issues, we propose\na novel identity-guided collaborative learning scheme (IGCL) for cloth-changing\nperson ReID, where the human semantic is fully utilized and the identity is\nunchangeable to guide collaborative learning. First, we design a novel clothing\nattention degradation stream to reasonably reduce the interference caused by\nclothing information where clothing attention and mid-level collaborative\nlearning are employed. Second, we propose a human semantic attention and body\njigsaw stream to highlight the human semantic information and simulate\ndifferent poses of the same identity. In this way, the extraction features not\nonly focus on human semantic information that is unrelated to the background\nbut also are suitable for pedestrian pose variations. Moreover, a pedestrian\nidentity enhancement stream is further proposed to enhance the identity\nimportance and extract more favorable identity robust features. Most\nimportantly, all these streams are jointly explored in an end-to-end unified\nframework, and the identity is utilized to guide the optimization. Extensive\nexperiments on five public clothing person ReID datasets demonstrate that the\nproposed IGCL significantly outperforms SOTA methods and that the extracted\nfeature is more robust, discriminative, and clothing-irrelevant.\n", "date": "2023-04-10", "categories": []}
{"id": "2308.14113", "title": "Semantic-aware Consistency Network for Cloth-changing Person\n  Re-Identification", "abstract": "  Cloth-changing Person Re-Identification (CC-ReID) is a challenging task that\naims to retrieve the target person across multiple surveillance cameras when\nclothing changes might happen. Despite recent progress in CC-ReID, existing\napproaches are still hindered by the interference of clothing variations since\nthey lack effective constraints to keep the model consistently focused on\nclothing-irrelevant regions. To address this issue, we present a Semantic-aware\nConsistency Network (SCNet) to learn identity-related semantic features by\nproposing effective consistency constraints. Specifically, we generate the\nblack-clothing image by erasing pixels in the clothing area, which explicitly\nmitigates the interference from clothing variations. In addition, to fully\nexploit the fine-grained identity information, a head-enhanced attention module\nis introduced, which learns soft attention maps by utilizing the proposed\npart-based matching loss to highlight head information. We further design a\nsemantic consistency loss to facilitate the learning of high-level\nidentity-related semantic features, forcing the model to focus on semantically\nconsistent cloth-irrelevant regions. By using the consistency constraint, our\nmodel does not require any extra auxiliary segmentation module to generate the\nblack-clothing image or locate the head region during the inference stage.\nExtensive experiments on four cloth-changing person Re-ID datasets (LTCC, PRCC,\nVc-Clothes, and DeepChange) demonstrate that our proposed SCNet makes\nsignificant improvements over prior state-of-the-art approaches. Our code is\navailable at: https://github.com/Gpn-star/SCNet.\n", "date": "2023-08-27", "categories": []}
{"id": "2311.10389", "title": "Two-Factor Authentication Approach Based on Behavior Patterns for\n  Defeating Puppet Attacks", "abstract": "  Fingerprint traits are widely recognized for their unique qualities and\nsecurity benefits. Despite their extensive use, fingerprint features can be\nvulnerable to puppet attacks, where attackers manipulate a reluctant but\ngenuine user into completing the authentication process. Defending against such\nattacks is challenging due to the coexistence of a legitimate identity and an\nillegitimate intent. In this paper, we propose PUPGUARD, a solution designed to\nguard against puppet attacks. This method is based on user behavioral patterns,\nspecifically, the user needs to press the capture device twice successively\nwith different fingers during the authentication process. PUPGUARD leverages\nboth the image features of fingerprints and the timing characteristics of the\npressing intervals to establish two-factor authentication. More specifically,\nafter extracting image features and timing characteristics, and performing\nfeature selection on the image features, PUPGUARD fuses these two features into\na one-dimensional feature vector, and feeds it into a one-class classifier to\nobtain the classification result. This two-factor authentication method\nemphasizes dynamic behavioral patterns during the authentication process,\nthereby enhancing security against puppet attacks. To assess PUPGUARD's\neffectiveness, we conducted experiments on datasets collected from 31 subjects,\nincluding image features and timing characteristics. Our experimental results\ndemonstrate that PUPGUARD achieves an impressive accuracy rate of 97.87% and a\nremarkably low false positive rate (FPR) of 1.89%. Furthermore, we conducted\ncomparative experiments to validate the superiority of combining image features\nand timing characteristics within PUPGUARD for enhancing resistance against\npuppet attacks.\n", "date": "2023-11-17", "categories": []}
{"id": "2311.10476", "title": "FRCSyn Challenge at WACV 2024:Face Recognition Challenge in the Era of\n  Synthetic Data", "abstract": "  Despite the widespread adoption of face recognition technology around the\nworld, and its remarkable performance on current benchmarks, there are still\nseveral challenges that must be covered in more detail. This paper offers an\noverview of the Face Recognition Challenge in the Era of Synthetic Data\n(FRCSyn) organized at WACV 2024. This is the first international challenge\naiming to explore the use of synthetic data in face recognition to address\nexisting limitations in the technology. Specifically, the FRCSyn Challenge\ntargets concerns related to data privacy issues, demographic biases,\ngeneralization to unseen scenarios, and performance limitations in challenging\nscenarios, including significant age disparities between enrollment and\ntesting, pose variations, and occlusions. The results achieved in the FRCSyn\nChallenge, together with the proposed benchmark, contribute significantly to\nthe application of synthetic data to improve face recognition technology.\n", "date": "2023-11-17", "categories": []}
{"id": "2301.09702", "title": "Illumination Variation Correction Using Image Synthesis For Unsupervised\n  Domain Adaptive Person Re-Identification", "abstract": "  Unsupervised domain adaptive (UDA) person re-identification (re-ID) aims to\nlearn identity information from labeled images in source domains and apply it\nto unlabeled images in a target domain. One major issue with many unsupervised\nre-identification methods is that they do not perform well relative to large\ndomain variations such as illumination, viewpoint, and occlusions. In this\npaper, we propose a Synthesis Model Bank (SMB) to deal with illumination\nvariation in unsupervised person re-ID. The proposed SMB consists of several\nconvolutional neural networks (CNN) for feature extraction and Mahalanobis\nmatrices for distance metrics. They are trained using synthetic data with\ndifferent illumination conditions such that their synergistic effect makes the\nSMB robust against illumination variation. To better quantify the illumination\nintensity and improve the quality of synthetic images, we introduce a new 3D\nvirtual-human dataset for GAN-based image synthesis. From our experiments, the\nproposed SMB outperforms other synthesis methods on several re-ID benchmarks.\n", "date": "2023-01-23", "categories": []}
{"id": "2311.08148", "title": "Cattle Identification Using Muzzle Images and Deep Learning Techniques", "abstract": "  Traditional animal identification methods such as ear-tagging, ear notching,\nand branding have been effective but pose risks to the animal and have\nscalability issues. Electrical methods offer better tracking and monitoring but\nrequire specialized equipment and are susceptible to attacks. Biometric\nidentification using time-immutable dermatoglyphic features such as muzzle\nprints and iris patterns is a promising solution. This project explores cattle\nidentification using 4923 muzzle images collected from 268 beef cattle. Two\ndeep learning classification models are implemented - wide ResNet50 and\nVGG16\\_BN and image compression is done to lower the image quality and adapt\nthe models to work for the African context. From the experiments run, a maximum\naccuracy of 99.5\\% is achieved while using the wide ResNet50 model with a\ncompression retaining 25\\% of the original image. From the study, it is noted\nthat the time required by the models to train and converge as well as\nrecognition time are dependent on the machine used to run the model.\n", "date": "2023-11-14", "categories": []}
{"id": "2311.05725", "title": "Whole-body Detection, Recognition and Identification at Altitude and\n  Range", "abstract": "  In this paper, we address the challenging task of whole-body biometric\ndetection, recognition, and identification at distances of up to 500m and large\npitch angles of up to 50 degree. We propose an end-to-end system evaluated on\ndiverse datasets, including the challenging Biometric Recognition and\nIdentification at Range (BRIAR) dataset. Our approach involves pre-training the\ndetector on common image datasets and fine-tuning it on BRIAR's complex videos\nand images. After detection, we extract body images and employ a feature\nextractor for recognition. We conduct thorough evaluations under various\nconditions, such as different ranges and angles in indoor, outdoor, and aerial\nscenarios. Our method achieves an average F1 score of 98.29% at IoU = 0.7 and\ndemonstrates strong performance in recognition accuracy and true acceptance\nrate at low false acceptance rates compared to existing models. On a test set\nof 100 subjects with 444 distractors, our model achieves a rank-20 recognition\naccuracy of 75.13% and a TAR@1%FAR of 54.09%.\n", "date": "2023-11-09", "categories": []}
{"id": "2311.06145", "title": "An Evaluation of Forensic Facial Recognition", "abstract": "  Recent advances in machine learning and computer vision have led to reported\nfacial recognition accuracies surpassing human performance. We question if\nthese systems will translate to real-world forensic scenarios in which a\npotentially low-resolution, low-quality, partially-occluded image is compared\nagainst a standard facial database. We describe the construction of a\nlarge-scale synthetic facial dataset along with a controlled facial forensic\nlineup, the combination of which allows for a controlled evaluation of facial\nrecognition under a range of real-world conditions. Using this synthetic\ndataset, and a popular dataset of real faces, we evaluate the accuracy of two\npopular neural-based recognition systems. We find that previously reported face\nrecognition accuracies of more than 95% drop to as low as 65% in this more\nchallenging forensic scenario.\n", "date": "2023-11-10", "categories": []}
{"id": "2311.05071", "title": "On the Behavior of Audio-Visual Fusion Architectures in Identity\n  Verification Tasks", "abstract": "  We train an identity verification architecture and evaluate modifications to\nthe part of the model that combines audio and visual representations, including\nin scenarios where one input is missing in either of two examples to be\ncompared. We report results on the Voxceleb1-E test set that suggest averaging\nthe output embeddings improves error rate in the full-modality setting and when\na single modality is missing, and makes more complete use of the embedding\nspace than systems which use shared layers and discuss possible reasons for\nthis behavior.\n", "date": "2023-11-09", "categories": []}
{"id": "2311.05336", "title": "SynFacePAD 2023: Competition on Face Presentation Attack Detection Based\n  on Privacy-aware Synthetic Training Data", "abstract": "  This paper presents a summary of the Competition on Face Presentation Attack\nDetection Based on Privacy-aware Synthetic Training Data (SynFacePAD 2023) held\nat the 2023 International Joint Conference on Biometrics (IJCB 2023). The\ncompetition attracted a total of 8 participating teams with valid submissions\nfrom academia and industry. The competition aimed to motivate and attract\nsolutions that target detecting face presentation attacks while considering\nsynthetic-based training data motivated by privacy, legal and ethical concerns\nassociated with personal data. To achieve that, the training data used by the\nparticipants was limited to synthetic data provided by the organizers. The\nsubmitted solutions presented innovations and novel approaches that led to\noutperforming the considered baseline in the investigated benchmarks.\n", "date": "2023-11-09", "categories": []}
{"id": "2311.05383", "title": "Improving Hand Recognition in Uncontrolled and Uncooperative\n  Environments using Multiple Spatial Transformers and Loss Functions", "abstract": "  The prevalence of smartphone and consumer camera has led to more evidence in\nthe form of digital images, which are mostly taken in uncontrolled and\nuncooperative environments. In these images, criminals likely hide or cover\ntheir faces while their hands are observable in some cases, creating a\nchallenging use case for forensic investigation. Many existing hand-based\nrecognition methods perform well for hand images collected in controlled\nenvironments with user cooperation. However, their performance deteriorates\nsignificantly in uncontrolled and uncooperative environments. A recent work has\nexposed the potential of hand recognition in these environments. However, only\nthe palmar regions were considered, and the recognition performance is still\nfar from satisfactory. To improve the recognition accuracy, an algorithm\nintegrating a multi-spatial transformer network (MSTN) and multiple loss\nfunctions is proposed to fully utilize information in full hand images. MSTN is\nfirstly employed to localize the palms and fingers and estimate the alignment\nparameters. Then, the aligned images are further fed into pretrained\nconvolutional neural networks, where features are extracted. Finally, a\ntraining scheme with multiple loss functions is used to train the network\nend-to-end. To demonstrate the effectiveness of the proposed algorithm, the\ntrained model is evaluated on NTU-PI-v1 database and six benchmark databases\nfrom different domains. Experimental results show that the proposed algorithm\nperforms significantly better than the existing methods in these uncontrolled\nand uncooperative environments and has good generalization capabilities to\nsamples from different domains.\n", "date": "2023-11-09", "categories": []}
{"id": "2311.05579", "title": "SigScatNet: A Siamese + Scattering based Deep Learning Approach for\n  Signature Forgery Detection and Similarity Assessment", "abstract": "  The surge in counterfeit signatures has inflicted widespread inconveniences\nand formidable challenges for both individuals and organizations. This\ngroundbreaking research paper introduces SigScatNet, an innovative solution to\ncombat this issue by harnessing the potential of a Siamese deep learning\nnetwork, bolstered by Scattering wavelets, to detect signature forgery and\nassess signature similarity. The Siamese Network empowers us to ascertain the\nauthenticity of signatures through a comprehensive similarity index, enabling\nprecise validation and comparison. Remarkably, the integration of Scattering\nwavelets endows our model with exceptional efficiency, rendering it light\nenough to operate seamlessly on cost-effective hardware systems. To validate\nthe efficacy of our approach, extensive experimentation was conducted on two\nopen-sourced datasets: the ICDAR SigComp Dutch dataset and the CEDAR dataset.\nThe experimental results demonstrate the practicality and resounding success of\nour proposed SigScatNet, yielding an unparalleled Equal Error Rate of 3.689%\nwith the ICDAR SigComp Dutch dataset and an astonishing 0.0578% with the CEDAR\ndataset. Through the implementation of SigScatNet, our research spearheads a\nnew state-of-the-art in signature analysis in terms of EER scores and\ncomputational efficiency, offering an advanced and accessible solution for\ndetecting forgery and quantifying signature similarities. By employing\ncutting-edge Siamese deep learning and Scattering wavelets, we provide a robust\nframework that paves the way for secure and efficient signature verification\nsystems.\n", "date": "2023-11-09", "categories": []}
{"id": "2311.04496", "title": "PersonMAE: Person Re-Identification Pre-Training with Masked\n  AutoEncoders", "abstract": "  Pre-training is playing an increasingly important role in learning generic\nfeature representation for Person Re-identification (ReID). We argue that a\nhigh-quality ReID representation should have three properties, namely,\nmulti-level awareness, occlusion robustness, and cross-region invariance. To\nthis end, we propose a simple yet effective pre-training framework, namely\nPersonMAE, which involves two core designs into masked autoencoders to better\nserve the task of Person Re-ID. 1) PersonMAE generates two regions from the\ngiven image with RegionA as the input and \\textit{RegionB} as the prediction\ntarget. RegionA is corrupted with block-wise masking to mimic common occlusion\nin ReID and its remaining visible parts are fed into the encoder. 2) Then\nPersonMAE aims to predict the whole RegionB at both pixel level and semantic\nfeature level. It encourages its pre-trained feature representations with the\nthree properties mentioned above. These properties make PersonMAE compatible\nwith downstream Person ReID tasks, leading to state-of-the-art performance on\nfour downstream ReID tasks, i.e., supervised (holistic and occluded setting),\nand unsupervised (UDA and USL setting). Notably, on the commonly adopted\nsupervised setting, PersonMAE with ViT-B backbone achieves 79.8% and 69.5% mAP\non the MSMT17 and OccDuke datasets, surpassing the previous state-of-the-art by\na large margin of +8.0 mAP, and +5.3 mAP, respectively.\n", "date": "2023-11-08", "categories": []}
{"id": "2311.04633", "title": "General Framework to Evaluate Unlinkability in Biometric Template\n  Protection Systems", "abstract": "  The wide deployment of biometric recognition systems in the last two decades\nhas raised privacy concerns regarding the storage and use of biometric data. As\na consequence, the ISO/IEC 24745 international standard on biometric\ninformation protection has established two main requirements for protecting\nbiometric templates: irreversibility and unlinkability. Numerous efforts have\nbeen directed to the development and analysis of irreversible templates.\nHowever, there is still no systematic quantitative manner to analyse the\nunlinkability of such templates. In this paper we address this shortcoming by\nproposing a new general framework for the evaluation of biometric templates'\nunlinkability. To illustrate the potential of the approach, it is applied to\nassess the unlinkability of four state-of-the-art techniques for biometric\ntemplate protection: biometric salting, Bloom filters, Homomorphic Encryption\nand block re-mapping. For the last technique, the proposed framework is\ncompared with other existing metrics to show its advantages.\n", "date": "2023-11-08", "categories": []}
{"id": "2304.13409", "title": "Efficient Explainable Face Verification based on Similarity Score\n  Argument Backpropagation", "abstract": "  Explainable Face Recognition is gaining growing attention as the use of the\ntechnology is gaining ground in security-critical applications. Understanding\nwhy two faces images are matched or not matched by a given face recognition\nsystem is important to operators, users, anddevelopers to increase trust,\naccountability, develop better systems, and highlight unfair behavior. In this\nwork, we propose xSSAB, an approach to back-propagate similarity score-based\narguments that support or oppose the face matching decision to visualize\nspatial maps that indicate similar and dissimilar areas as interpreted by the\nunderlying FR model. Furthermore, we present Patch-LFW, a new explainable face\nverification benchmark that enables along with a novel evaluation protocol, the\nfirst quantitative evaluation of the validity of similarity and dissimilarity\nmaps in explainable face recognition approaches. We compare our efficient\napproach to state-of-the-art approaches demonstrating a superior trade-off\nbetween efficiency and performance. The code as well as the proposed Patch-LFW\nis publicly available at: https://github.com/marcohuber/xSSAB.\n", "date": "2023-04-26", "categories": []}
{"id": "2311.03970", "title": "Bias and Diversity in Synthetic-based Face Recognition", "abstract": "  Synthetic data is emerging as a substitute for authentic data to solve\nethical and legal challenges in handling authentic face data. The current\nmodels can create real-looking face images of people who do not exist. However,\nit is a known and sensitive problem that face recognition systems are\nsusceptible to bias, i.e. performance differences between different demographic\nand non-demographics attributes, which can lead to unfair decisions. In this\nwork, we investigate how the diversity of synthetic face recognition datasets\ncompares to authentic datasets, and how the distribution of the training data\nof the generative models affects the distribution of the synthetic data. To do\nthis, we looked at the distribution of gender, ethnicity, age, and head\nposition. Furthermore, we investigated the concrete bias of three recent\nsynthetic-based face recognition models on the studied attributes in comparison\nto a baseline model trained on authentic data. Our results show that the\ngenerator generate a similar distribution as the used training data in terms of\nthe different attributes. With regard to bias, it can be seen that the\nsynthetic-based models share a similar bias behavior with the authentic-based\nmodels. However, with the uncovered lower intra-identity attribute consistency\nseems to be beneficial in reducing bias.\n", "date": "2023-11-07", "categories": []}
{"id": "2311.04148", "title": "Contactless Fingerprint Biometric Anti-Spoofing: An Unsupervised Deep\n  Learning Approach", "abstract": "  Contactless fingerprint recognition offers a higher level of user comfort and\naddresses hygiene concerns more effectively. However, it is also more\nvulnerable to presentation attacks such as photo paper, paper-printout, and\nvarious display attacks, which makes it more challenging to implement in\nbiometric systems compared to contact-based modalities. Limited research has\nbeen conducted on presentation attacks in contactless fingerprint systems, and\nthese studies have encountered challenges in terms of generalization and\nscalability since both bonafide samples and presentation attacks are utilized\nduring training model. Although this approach appears promising, it lacks the\nability to handle unseen attacks, which is a crucial factor for developing PAD\nmethods that can generalize effectively. We introduced an innovative\nanti-spoofing approach that combines an unsupervised autoencoder with a\nconvolutional block attention module to address the limitations of existing\nmethods. Our model is exclusively trained on bonafide images without exposure\nto any spoofed samples during the training phase. It is then evaluated against\nvarious types of presentation attack images in the testing phase. The scheme we\nproposed has achieved an average BPCER of 0.96\\% with an APCER of 1.6\\% for\npresentation attacks involving various types of spoofed samples.\n", "date": "2023-11-07", "categories": []}
{"id": "2311.02523", "title": "UniTSFace: Unified Threshold Integrated Sample-to-Sample Loss for Face\n  Recognition", "abstract": "  Sample-to-class-based face recognition models can not fully explore the\ncross-sample relationship among large amounts of facial images, while\nsample-to-sample-based models require sophisticated pairing processes for\ntraining. Furthermore, neither method satisfies the requirements of real-world\nface verification applications, which expect a unified threshold separating\npositive from negative facial pairs. In this paper, we propose a unified\nthreshold integrated sample-to-sample based loss (USS loss), which features an\nexplicit unified threshold for distinguishing positive from negative pairs.\nInspired by our USS loss, we also derive the sample-to-sample based softmax and\nBCE losses, and discuss their relationship. Extensive evaluation on multiple\nbenchmark datasets, including MFR, IJB-C, LFW, CFP-FP, AgeDB, and MegaFace,\ndemonstrates that the proposed USS loss is highly efficient and can work\nseamlessly with sample-to-class-based losses. The embedded loss (USS and\nsample-to-class Softmax loss) overcomes the pitfalls of previous approaches and\nthe trained facial model UniTSFace exhibits exceptional performance,\noutperforming state-of-the-art methods, such as CosFace, ArcFace, VPL,\nAnchorFace, and UNPG. Our code is available.\n", "date": "2023-11-04", "categories": []}
{"id": "2311.02803", "title": "Fast and Interpretable Face Identification for Out-Of-Distribution Data\n  Using Vision Transformers", "abstract": "  Most face identification approaches employ a Siamese neural network to\ncompare two images at the image embedding level. Yet, this technique can be\nsubject to occlusion (e.g. faces with masks or sunglasses) and\nout-of-distribution data. DeepFace-EMD (Phan et al. 2022) reaches\nstate-of-the-art accuracy on out-of-distribution data by first comparing two\nimages at the image level, and then at the patch level. Yet, its later\npatch-wise re-ranking stage admits a large $O(n^3 \\log n)$ time complexity (for\n$n$ patches in an image) due to the optimal transport optimization. In this\npaper, we propose a novel, 2-image Vision Transformers (ViTs) that compares two\nimages at the patch level using cross-attention. After training on 2M pairs of\nimages on CASIA Webface (Yi et al. 2014), our model performs at a comparable\naccuracy as DeepFace-EMD on out-of-distribution data, yet at an inference speed\nmore than twice as fast as DeepFace-EMD (Phan et al. 2022). In addition, via a\nhuman study, our model shows promising explainability through the visualization\nof cross-attention. We believe our work can inspire more explorations in using\nViTs for face identification.\n", "date": "2023-11-06", "categories": []}
{"id": "2311.01651", "title": "Keypoint Description by Symmetry Assessment -- Applications in\n  Biometrics", "abstract": "  We present a model-based feature extractor to describe neighborhoods around\nkeypoints by finite expansion, estimating the spatially varying orientation by\nharmonic functions. The iso-curves of such functions are highly symmetric\nw.r.t. the origin (a keypoint) and the estimated parameters have well defined\ngeometric interpretations. The origin is also a unique singularity of all\nharmonic functions, helping to determine the location of a keypoint precisely,\nwhereas the functions describe the object shape of the neighborhood. This is\nnovel and complementary to traditional texture features which describe\ntexture-shape properties i.e. they are purposively invariant to translation\n(within a texture). We report on experiments of verification and identification\nof keypoints in forensic fingerprints by using publicly available data (NIST\nSD27) and discuss the results in comparison to other studies. These support our\nconclusions that the novel features can equip single cores or single minutia\nwith a significant verification power at 19% EER, and an identification power\nof 24-78% for ranks of 1-20. Additionally, we report verification results of\nperiocular biometrics using near-infrared images, reaching an EER performance\nof 13%, which is comparable to the state of the art. More importantly, fusion\nof two systems, our and texture features (Gabor), result in a measurable\nperformance improvement. We report reduction of the EER to 9%, supporting the\nview that the novel features capture relevant visual information, which\ntraditional texture features do not.\n", "date": "2023-11-03", "categories": []}
{"id": "2209.08762", "title": "Fairness on Synthetic Visual and Thermal Mask Images", "abstract": "  In this paper, we study performance and fairness on visual and thermal images\nand expand the assessment to masked synthetic images. Using the SpeakingFace\nand Thermal-Mask dataset, we propose a process to assess fairness on real\nimages and show how the same process can be applied to synthetic images. The\nresulting process shows a demographic parity difference of 1.59 for random\nguessing and increases to 5.0 when the recognition performance increases to a\nprecision and recall rate of 99.99\\%. We indicate that inherently biased\ndatasets can deeply impact the fairness of any biometric system. A primary\ncause of a biased dataset is the class imbalance due to the data collection\nprocess. To address imbalanced datasets, the classes with fewer samples can be\naugmented with synthetic images to generate a more balanced dataset resulting\nin less bias when training a machine learning system. For biometric-enabled\nsystems, fairness is of critical importance, while the related concept of\nEquity, Diversity, and Inclusion (EDI) is well suited for the generalization of\nfairness in biometrics, in this paper, we focus on the 3 most common\ndemographic groups age, gender, and ethnicity.\n", "date": "2022-09-19", "categories": []}
{"id": "2311.01155", "title": "Learning Intra and Inter-Camera Invariance for Isolated Camera\n  Supervised Person Re-identification", "abstract": "  Supervised person re-identification assumes that a person has images captured\nunder multiple cameras. However when cameras are placed in distance, a person\nrarely appears in more than one camera. This paper thus studies person re-ID\nunder such isolated camera supervised (ISCS) setting. Instead of trying to\ngenerate fake cross-camera features like previous methods, we explore a novel\nperspective by making efficient use of the variation in training data. Under\nISCS setting, a person only has limited images from a single camera, so the\ncamera bias becomes a critical issue confounding ID discrimination.\nCross-camera images are prone to being recognized as different IDs simply by\ncamera style. To eliminate the confounding effect of camera bias, we propose to\nlearn both intra- and inter-camera invariance under a unified framework. First,\nwe construct style-consistent environments via clustering, and perform\nprototypical contrastive learning within each environment. Meanwhile, strongly\naugmented images are contrasted with original prototypes to enforce\nintra-camera augmentation invariance. For inter-camera invariance, we further\ndesign a much improved variant of multi-camera negative loss that optimizes the\ndistance of multi-level negatives. The resulting model learns to be invariant\nto both subtle and severe style variation within and cross-camera. On multiple\nbenchmarks, we conduct extensive experiments and validate the effectiveness and\nsuperiority of the proposed method. Code will be available at\nhttps://github.com/Terminator8758/IICI.\n", "date": "2023-11-02", "categories": []}
{"id": "2311.01237", "title": "Log-Likelihood Score Level Fusion for Improved Cross-Sensor Smartphone\n  Periocular Recognition", "abstract": "  The proliferation of cameras and personal devices results in a wide\nvariability of imaging conditions, producing large intra-class variations and a\nsignificant performance drop when images from heterogeneous environments are\ncompared. However, many applications require to deal with data from different\nsources regularly, thus needing to overcome these interoperability problems.\nHere, we employ fusion of several comparators to improve periocular performance\nwhen images from different smartphones are compared. We use a probabilistic\nfusion framework based on linear logistic regression, in which fused scores\ntend to be log-likelihood ratios, obtaining a reduction in cross-sensor EER of\nup to 40% due to the fusion. Our framework also provides an elegant and simple\nsolution to handle signals from different devices, since same-sensor and\ncross-sensor score distributions are aligned and mapped to a common\nprobabilistic domain. This allows the use of Bayes thresholds for optimal\ndecision-making, eliminating the need of sensor-specific thresholds, which is\nessential in operational conditions because the threshold setting critically\ndetermines the accuracy of the authentication process in many applications.\n", "date": "2023-11-02", "categories": []}
{"id": "2311.01241", "title": "Exploring Deep Learning Image Super-Resolution for Iris Recognition", "abstract": "  In this work we test the ability of deep learning methods to provide an\nend-to-end mapping between low and high resolution images applying it to the\niris recognition problem. Here, we propose the use of two deep learning\nsingle-image super-resolution approaches: Stacked Auto-Encoders (SAE) and\nConvolutional Neural Networks (CNN) with the most possible lightweight\nstructure to achieve fast speed, preserve local information and reduce\nartifacts at the same time. We validate the methods with a database of 1.872\nnear-infrared iris images with quality assessment and recognition experiments\nshowing the superiority of deep learning approaches over the compared\nalgorithms.\n", "date": "2023-11-02", "categories": []}
{"id": "2311.00400", "title": "Open-Set Face Recognition with Maximal Entropy and Objectosphere Loss", "abstract": "  Open-set face recognition characterizes a scenario where unknown individuals,\nunseen during the training and enrollment stages, appear on operation time.\nThis work concentrates on watchlists, an open-set task that is expected to\noperate at a low False Positive Identification Rate and generally includes only\na few enrollment samples per identity. We introduce a compact adapter network\nthat benefits from additional negative face images when combined with distinct\ncost functions, such as Objectosphere Loss (OS) and the proposed Maximal\nEntropy Loss (MEL). MEL modifies the traditional Cross-Entropy loss in favor of\nincreasing the entropy for negative samples and attaches a penalty to known\ntarget classes in pursuance of gallery specialization. The proposed approach\nadopts pre-trained deep neural networks (DNNs) for face recognition as feature\nextractors. Then, the adapter network takes deep feature representations and\nacts as a substitute for the output layer of the pre-trained DNN in exchange\nfor an agile domain adaptation. Promising results have been achieved following\nopen-set protocols for three different datasets: LFW, IJB-C, and UCCS as well\nas state-of-the-art performance when supplementary negative data is properly\nselected to fine-tune the adapter network.\n", "date": "2023-11-01", "categories": []}
{"id": "2311.00603", "title": "Occluded Person Re-Identification with Deep Learning: A Survey and\n  Perspectives", "abstract": "  Person re-identification (Re-ID) technology plays an increasingly crucial\nrole in intelligent surveillance systems. Widespread occlusion significantly\nimpacts the performance of person Re-ID. Occluded person Re-ID refers to a\npedestrian matching method that deals with challenges such as pedestrian\ninformation loss, noise interference, and perspective misalignment. It has\ngarnered extensive attention from researchers. Over the past few years, several\nocclusion-solving person Re-ID methods have been proposed, tackling various\nsub-problems arising from occlusion. However, there is a lack of comprehensive\nstudies that compare, summarize, and evaluate the potential of occluded person\nRe-ID methods in detail. In this review, we start by providing a detailed\noverview of the datasets and evaluation scheme used for occluded person Re-ID.\nNext, we scientifically classify and analyze existing deep learning-based\noccluded person Re-ID methods from various perspectives, summarizing them\nconcisely. Furthermore, we conduct a systematic comparison among these methods,\nidentify the state-of-the-art approaches, and present an outlook on the future\ndevelopment of occluded person Re-ID.\n", "date": "2023-11-01", "categories": []}
{"id": "2310.18438", "title": "Exploring Shape Embedding for Cloth-Changing Person Re-Identification\n  via 2D-3D Correspondences", "abstract": "  Cloth-Changing Person Re-Identification (CC-ReID) is a common and realistic\nproblem since fashion constantly changes over time and people's aesthetic\npreferences are not set in stone. While most existing cloth-changing ReID\nmethods focus on learning cloth-agnostic identity representations from coarse\nsemantic cues (e.g. silhouettes and part segmentation maps), they neglect the\ncontinuous shape distributions at the pixel level. In this paper, we propose\nContinuous Surface Correspondence Learning (CSCL), a new shape embedding\nparadigm for cloth-changing ReID. CSCL establishes continuous correspondences\nbetween a 2D image plane and a canonical 3D body surface via pixel-to-vertex\nclassification, which naturally aligns a person image to the surface of a 3D\nhuman model and simultaneously obtains pixel-wise surface embeddings. We\nfurther extract fine-grained shape features from the learned surface embeddings\nand then integrate them with global RGB features via a carefully designed\ncross-modality fusion module. The shape embedding paradigm based on 2D-3D\ncorrespondences remarkably enhances the model's global understanding of human\nbody shape. To promote the study of ReID under clothing change, we construct 3D\nDense Persons (DP3D), which is the first large-scale cloth-changing ReID\ndataset that provides densely annotated 2D-3D correspondences and a precise 3D\nmesh for each person image, while containing diverse cloth-changing cases over\nall four seasons. Experiments on both cloth-changing and cloth-consistent ReID\nbenchmarks validate the effectiveness of our method.\n", "date": "2023-10-27", "categories": []}
{"id": "2310.18668", "title": "FinBTech: Blockchain-Based Video and Voice Authentication System for\n  Enhanced Security in Financial Transactions Utilizing FaceNet512 and Gaussian\n  Mixture Models", "abstract": "  In the digital age, it is crucial to make sure that financial transactions\nare as secure and reliable as possible. This abstract offers a ground-breaking\nmethod that combines smart contracts, blockchain technology, FaceNet512 for\nimproved face recognition, and Gaussian Mixture Models (GMM) for speech\nauthentication to create a system for video and audio verification that is\nunmatched. Smart contracts and the immutable ledger of the blockchain are\ncombined to offer a safe and open environment for financial transactions.\nFaceNet512 and GMM offer multi-factor biometric authentication simultaneously,\nenhancing security to new heights. By combining cutting-edge technology, this\nsystem offers a strong defense against identity theft and illegal access,\nestablishing a new benchmark for safe financial transactions.\n", "date": "2023-10-28", "categories": []}
{"id": "2310.19024", "title": "FPGAN-Control: A Controllable Fingerprint Generator for Training with\n  Synthetic Data", "abstract": "  Training fingerprint recognition models using synthetic data has recently\ngained increased attention in the biometric community as it alleviates the\ndependency on sensitive personal data. Existing approaches for fingerprint\ngeneration are limited in their ability to generate diverse impressions of the\nsame finger, a key property for providing effective data for training\nrecognition models. To address this gap, we present FPGAN-Control, an identity\npreserving image generation framework which enables control over the\nfingerprint's image appearance (e.g., fingerprint type, acquisition device,\npressure level) of generated fingerprints. We introduce a novel appearance loss\nthat encourages disentanglement between the fingerprint's identity and\nappearance properties. In our experiments, we used the publicly available NIST\nSD302 (N2N) dataset for training the FPGAN-Control model. We demonstrate the\nmerits of FPGAN-Control, both quantitatively and qualitatively, in terms of\nidentity preservation level, degree of appearance control, and low\nsynthetic-to-real domain gap. Finally, training recognition models using only\nsynthetic datasets generated by FPGAN-Control lead to recognition accuracies\nthat are on par or even surpass models trained using real data. To the best of\nour knowledge, this is the first work to demonstrate this.\n", "date": "2023-10-29", "categories": []}
{"id": "2310.19418", "title": "GaitFormer: Learning Gait Representations with Noisy Multi-Task Learning", "abstract": "  Gait analysis is proven to be a reliable way to perform person identification\nwithout relying on subject cooperation. Walking is a biometric that does not\nsignificantly change in short periods of time and can be regarded as unique to\neach person. So far, the study of gait analysis focused mostly on\nidentification and demographics estimation, without considering many of the\npedestrian attributes that appearance-based methods rely on. In this work,\nalongside gait-based person identification, we explore pedestrian attribute\nidentification solely from movement patterns. We propose DenseGait, the largest\ndataset for pretraining gait analysis systems containing 217K anonymized\ntracklets, annotated automatically with 42 appearance attributes. DenseGait is\nconstructed by automatically processing video streams and offers the full array\nof gait covariates present in the real world. We make the dataset available to\nthe research community. Additionally, we propose GaitFormer, a\ntransformer-based model that after pretraining in a multi-task fashion on\nDenseGait, achieves 92.5% accuracy on CASIA-B and 85.33% on FVG, without\nutilizing any manually annotated data. This corresponds to a +14.2% and +9.67%\naccuracy increase compared to similar methods. Moreover, GaitFormer is able to\naccurately identify gender information and a multitude of appearance attributes\nutilizing only movement patterns. The code to reproduce the experiments is made\npublicly.\n", "date": "2023-10-30", "categories": []}
{"id": "2210.00527", "title": "Comparison of Data Representations and Machine Learning Architectures\n  for User Identification on Arbitrary Motion Sequences", "abstract": "  Reliable and robust user identification and authentication are important and\noften necessary requirements for many digital services. It becomes paramount in\nsocial virtual reality (VR) to ensure trust, specifically in digital encounters\nwith lifelike realistic-looking avatars as faithful replications of real\npersons. Recent research has shown that the movements of users in extended\nreality (XR) systems carry user-specific information and can thus be used to\nverify their identities. This article compares three different potential\nencodings of the motion data from head and hands (scene-relative,\nbody-relative, and body-relative velocities), and the performances of five\ndifferent machine learning architectures (random forest, multi-layer\nperceptron, fully recurrent neural network, long-short term memory, gated\nrecurrent unit). We use the publicly available dataset \"Talking with Hands\" and\npublish all code to allow reproducibility and to provide baselines for future\nwork. After hyperparameter optimization, the combination of a long-short term\nmemory architecture and body-relative data outperformed competing combinations:\nthe model correctly identifies any of the 34 subjects with an accuracy of 100%\nwithin 150 seconds. Altogether, our approach provides an effective foundation\nfor behaviometric-based identification and authentication to guide researchers\nand practitioners. Data and code are published under\nhttps://go.uniwue.de/58w1r.\n", "date": "2022-10-02", "categories": []}
{"id": "2304.03510", "title": "Multispectral Imaging for Differential Face Morphing Attack Detection: A\n  Preliminary Study", "abstract": "  Face morphing attack detection is emerging as an increasingly challenging\nproblem owing to advancements in high-quality and realistic morphing attack\ngeneration. Reliable detection of morphing attacks is essential because these\nattacks are targeted for border control applications. This paper presents a\nmultispectral framework for differential morphing-attack detection (D-MAD). The\nD-MAD methods are based on using two facial images that are captured from the\nePassport (also called the reference image) and the trusted device (for\nexample, Automatic Border Control (ABC) gates) to detect whether the face image\npresented in ePassport is morphed. The proposed multispectral D-MAD framework\nintroduce a multispectral image captured as a trusted capture to acquire seven\ndifferent spectral bands to detect morphing attacks. Extensive experiments were\nconducted on the newly created Multispectral Morphed Datasets (MSMD) with 143\nunique data subjects that were captured using both visible and multispectral\ncameras in multiple sessions. The results indicate the superior performance of\nthe proposed multispectral framework compared to visible images.\n", "date": "2023-04-07", "categories": []}
{"id": "2310.16364", "title": "Towards Large-scale Masked Face Recognition", "abstract": "  During the COVID-19 coronavirus epidemic, almost everyone is wearing masks,\nwhich poses a huge challenge for deep learning-based face recognition\nalgorithms. In this paper, we will present our \\textbf{championship} solutions\nin ICCV MFR WebFace260M and InsightFace unconstrained tracks. We will focus on\nfour challenges in large-scale masked face recognition, i.e., super-large scale\ntraining, data noise handling, masked and non-masked face recognition accuracy\nbalancing, and how to design inference-friendly model architecture. We hope\nthat the discussion on these four aspects can guide future research towards\nmore robust masked face recognition systems.\n", "date": "2023-10-25", "categories": []}
{"id": "2310.16808", "title": "Fingervein Verification using Convolutional Multi-Head Attention Network", "abstract": "  Biometric verification systems are deployed in various security-based\naccess-control applications that require user-friendly and reliable person\nverification. Among the different biometric characteristics, fingervein\nbiometrics have been extensively studied owing to their reliable verification\nperformance. Furthermore, fingervein patterns reside inside the skin and are\nnot visible outside; therefore, they possess inherent resistance to\npresentation attacks and degradation due to external factors. In this paper, we\nintroduce a novel fingervein verification technique using a convolutional\nmultihead attention network called VeinAtnNet. The proposed VeinAtnNet is\ndesigned to achieve light weight with a smaller number of learnable parameters\nwhile extracting discriminant information from both normal and enhanced\nfingervein images. The proposed VeinAtnNet was trained on the newly constructed\nfingervein dataset with 300 unique fingervein patterns that were captured in\nmultiple sessions to obtain 92 samples per unique fingervein. Extensive\nexperiments were performed on the newly collected dataset FV-300 and the\npublicly available FV-USM and FV-PolyU fingervein dataset. The performance of\nthe proposed method was compared with five state-of-the-art fingervein\nverification systems, indicating the efficacy of the proposed VeinAtnNet.\n", "date": "2023-10-25", "categories": []}
{"id": "2302.08927", "title": "Unique Identification of 50,000+ Virtual Reality Users from Head & Hand\n  Motion Data", "abstract": "  With the recent explosive growth of interest and investment in virtual\nreality (VR) and the so-called \"metaverse,\" public attention has rightly\nshifted toward the unique security and privacy threats that these platforms may\npose. While it has long been known that people reveal information about\nthemselves via their motion, the extent to which this makes an individual\nglobally identifiable within virtual reality has not yet been widely\nunderstood. In this study, we show that a large number of real VR users\n(N=55,541) can be uniquely and reliably identified across multiple sessions\nusing just their head and hand motion relative to virtual objects. After\ntraining a classification model on 5 minutes of data per person, a user can be\nuniquely identified amongst the entire pool of 50,000+ with 94.33% accuracy\nfrom 100 seconds of motion, and with 73.20% accuracy from just 10 seconds of\nmotion. This work is the first to truly demonstrate the extent to which\nbiomechanics may serve as a unique identifier in VR, on par with widely used\nbiometrics such as facial or fingerprint recognition.\n", "date": "2023-02-17", "categories": []}
{"id": "2305.08546", "title": "Towards Visual Saliency Explanations of Face Verification", "abstract": "  In the past years, deep convolutional neural networks have been pushing the\nfrontier of face recognition (FR) techniques in both verification and\nidentification scenarios. Despite the high accuracy, they are often criticized\nfor lacking explainability. There has been an increasing demand for\nunderstanding the decision-making process of deep face recognition systems.\nRecent studies have investigated the usage of visual saliency maps as an\nexplanation, but they often lack a discussion and analysis in the context of\nface recognition. This paper concentrates on explainable face verification\ntasks and conceives a new explanation framework. Firstly, a definition of the\nsaliency-based explanation method is provided, which focuses on the decisions\nmade by the deep FR model. Secondly, a new model-agnostic explanation method\nnamed CorrRISE is proposed to produce saliency maps, which reveal both the\nsimilar and dissimilar regions of any given pair of face images. Then, an\nevaluation methodology is designed to measure the performance of general visual\nsaliency explanation methods in face verification. Finally, substantial visual\nand quantitative results have shown that the proposed CorrRISE method\ndemonstrates promising results in comparison with other state-of-the-art\nexplainable face verification approaches.\n", "date": "2023-05-15", "categories": []}
{"id": "2310.13573", "title": "Boosting Generalization with Adaptive Style Techniques for Fingerprint\n  Liveness Detection", "abstract": "  We introduce a high-performance fingerprint liveness feature extraction\ntechnique that secured first place in LivDet 2023 Fingerprint Representation\nChallenge. Additionally, we developed a practical fingerprint recognition\nsystem with 94.68% accuracy, earning second place in LivDet 2023 Liveness\nDetection in Action. By investigating various methods, particularly style\ntransfer, we demonstrate improvements in accuracy and generalization when faced\nwith limited training data. As a result, our approach achieved state-of-the-art\nperformance in LivDet 2023 Challenges.\n", "date": "2023-10-20", "categories": []}
{"id": "2310.15913", "title": "Mitigate Domain Shift by Primary-Auxiliary Objectives Association for\n  Generalizing Person ReID", "abstract": "  While deep learning has significantly improved ReID model accuracy under the\nindependent and identical distribution (IID) assumption, it has also become\nclear that such models degrade notably when applied to an unseen novel domain\ndue to unpredictable/unknown domain shift. Contemporary domain generalization\n(DG) ReID models struggle in learning domain-invariant representation solely\nthrough training on an instance classification objective. We consider that a\ndeep learning model is heavily influenced and therefore biased towards\ndomain-specific characteristics, e.g., background clutter, scale and viewpoint\nvariations, limiting the generalizability of the learned model, and hypothesize\nthat the pedestrians are domain invariant owning they share the same structural\ncharacteristics. To enable the ReID model to be less domain-specific from these\npure pedestrians, we introduce a method that guides model learning of the\nprimary ReID instance classification objective by a concurrent auxiliary\nlearning objective on weakly labeled pedestrian saliency detection. To solve\nthe problem of conflicting optimization criteria in the model parameter space\nbetween the two learning objectives, we introduce a Primary-Auxiliary\nObjectives Association (PAOA) mechanism to calibrate the loss gradients of the\nauxiliary task towards the primary learning task gradients. Benefiting from the\nharmonious multitask learning design, our model can be extended with the recent\ntest-time diagram to form the PAOA+, which performs on-the-fly optimization\nagainst the auxiliary objective in order to maximize the model's generative\ncapacity in the test target domain. Experiments demonstrate the superiority of\nthe proposed PAOA model.\n", "date": "2023-10-24", "categories": []}
{"id": "2310.15946", "title": "ShARc: Shape and Appearance Recognition for Person Identification\n  In-the-wild", "abstract": "  Identifying individuals in unconstrained video settings is a valuable yet\nchallenging task in biometric analysis due to variations in appearances,\nenvironments, degradations, and occlusions. In this paper, we present ShARc, a\nmultimodal approach for video-based person identification in uncontrolled\nenvironments that emphasizes 3-D body shape, pose, and appearance. We introduce\ntwo encoders: a Pose and Shape Encoder (PSE) and an Aggregated Appearance\nEncoder (AAE). PSE encodes the body shape via binarized silhouettes, skeleton\nmotions, and 3-D body shape, while AAE provides two levels of temporal\nappearance feature aggregation: attention-based feature aggregation and\naveraging aggregation. For attention-based feature aggregation, we employ\nspatial and temporal attention to focus on key areas for person distinction.\nFor averaging aggregation, we introduce a novel flattening layer after\naveraging to extract more distinguishable information and reduce overfitting of\nattention. We utilize centroid feature averaging for gallery registration. We\ndemonstrate significant improvements over existing state-of-the-art methods on\npublic datasets, including CCVID, MEVID, and BRIAR.\n", "date": "2023-10-24", "categories": []}
{"id": "2310.15044", "title": "A Universal Anti-Spoofing Approach for Contactless Fingerprint Biometric\n  Systems", "abstract": "  With the increasing integration of smartphones into our daily lives,\nfingerphotos are becoming a potential contactless authentication method. While\nit offers convenience, it is also more vulnerable to spoofing using various\npresentation attack instruments (PAI). The contactless fingerprint is an\nemerging biometric authentication but has not yet been heavily investigated for\nanti-spoofing. While existing anti-spoofing approaches demonstrated fair\nresults, they have encountered challenges in terms of universality and\nscalability to detect any unseen/unknown spoofed samples. To address this\nissue, we propose a universal presentation attack detection method for\ncontactless fingerprints, despite having limited knowledge of presentation\nattack samples. We generated synthetic contactless fingerprints using StyleGAN\nfrom live finger photos and integrating them to train a semi-supervised\nResNet-18 model. A novel joint loss function, combining the Arcface and Center\nloss, is introduced with a regularization to balance between the two loss\nfunctions and minimize the variations within the live samples while enhancing\nthe inter-class variations between the deepfake and live samples. We also\nconducted a comprehensive comparison of different regularizations' impact on\nthe joint loss function for presentation attack detection (PAD) and explored\nthe performance of a modified ResNet-18 architecture with different activation\nfunctions (i.e., leaky ReLU and RelU) in conjunction with Arcface and center\nloss. Finally, we evaluate the performance of the model using unseen types of\nspoof attacks and live data. Our proposed method achieves a Bona Fide\nClassification Error Rate (BPCER) of 0.12\\%, an Attack Presentation\nClassification Error Rate (APCER) of 0.63\\%, and an Average Classification\nError Rate (ACER) of 0.37\\%.\n", "date": "2023-10-23", "categories": []}
{"id": "2310.11640", "title": "Free-text Keystroke Authentication using Transformers: A Comparative\n  Study of Architectures and Loss Functions", "abstract": "  Keystroke biometrics is a promising approach for user identification and\nverification, leveraging the unique patterns in individuals' typing behavior.\nIn this paper, we propose a Transformer-based network that employs\nself-attention to extract informative features from keystroke sequences,\nsurpassing the performance of traditional Recurrent Neural Networks. We explore\ntwo distinct architectures, namely bi-encoder and cross-encoder, and compare\ntheir effectiveness in keystroke authentication. Furthermore, we investigate\ndifferent loss functions, including triplet, batch-all triplet, and WDCL loss,\nalong with various distance metrics such as Euclidean, Manhattan, and cosine\ndistances. These experiments allow us to optimize the training process and\nenhance the performance of our model. To evaluate our proposed model, we employ\nthe Aalto desktop keystroke dataset. The results demonstrate that the\nbi-encoder architecture with batch-all triplet loss and cosine distance\nachieves the best performance, yielding an exceptional Equal Error Rate of\n0.0186%. Furthermore, alternative algorithms for calculating similarity scores\nare explored to enhance accuracy. Notably, the utilization of a one-class\nSupport Vector Machine reduces the Equal Error Rate to an impressive 0.0163%.\nThe outcomes of this study indicate that our model surpasses the previous\nstate-of-the-art in free-text keystroke authentication. These findings\ncontribute to advancing the field of keystroke authentication and offer\npractical implications for secure user verification systems.\n", "date": "2023-10-18", "categories": []}
{"id": "2310.09563", "title": "Learning Unified Representations for Multi-Resolution Face Recognition", "abstract": "  In this work, we propose Branch-to-Trunk network (BTNet), a representation\nlearning method for multi-resolution face recognition. It consists of a trunk\nnetwork (TNet), namely a unified encoder, and multiple branch networks (BNets),\nnamely resolution adapters. As per the input, a resolution-specific BNet is\nused and the output are implanted as feature maps in the feature pyramid of\nTNet, at a layer with the same resolution. The discriminability of tiny faces\nis significantly improved, as the interpolation error introduced by rescaling,\nespecially up-sampling, is mitigated on the inputs. With branch distillation\nand backward-compatible training, BTNet transfers discriminative\nhigh-resolution information to multiple branches while guaranteeing\nrepresentation compatibility. Our experiments demonstrate strong performance on\nface recognition benchmarks, both for multi-resolution identity matching and\nfeature aggregation, with much less computation amount and parameter storage.\nWe establish new state-of-the-art on the challenging QMUL-SurvFace 1: N face\nidentification task. Our code is available at\nhttps://github.com/StevenSmith2000/BTNet.\n", "date": "2023-10-14", "categories": []}
{"id": "2310.09756", "title": "New Benchmarks for Asian Facial Recognition Tasks: Face Classification\n  with Large Foundation Models", "abstract": "  The face classification system is an important tool for recognizing personal\nidentity properly. This paper introduces a new Large-Scale Korean Influencer\nDataset named KoIn. Our presented dataset contains many real-world photos of\nKorean celebrities in various environments that might contain stage lighting,\nbackup dancers, and background objects. These various images can be useful for\ntraining classification models classifying K-influencers. Most of the images in\nour proposed dataset have been collected from social network services (SNS)\nsuch as Instagram. Our dataset, KoIn, contains over 100,000 K-influencer photos\nfrom over 100 Korean celebrity classes. Moreover, our dataset provides\nadditional hard case samples such as images including human faces with masks\nand hats. We note that the hard case samples are greatly useful in evaluating\nthe robustness of the classification systems. We have extensively conducted\nseveral experiments utilizing various classification models to validate the\neffectiveness of our proposed dataset. Specifically, we demonstrate that recent\nstate-of-the-art (SOTA) foundation architectures show decent classification\nperformance when trained on our proposed dataset. In this paper, we also\nanalyze the robustness performance against hard case samples of large-scale\nfoundation models when we fine-tune the foundation models on the normal cases\nof the proposed dataset, KoIn. Our presented dataset and codes will be publicly\navailable at https://github.com/dukong1/KoIn_Benchmark_Dataset.\n", "date": "2023-10-15", "categories": []}
{"id": "2301.11986", "title": "Enhancing Face Recognition with Latent Space Data Augmentation and\n  Facial Posture Reconstruction", "abstract": "  The small amount of training data for many state-of-the-art deep\nlearning-based Face Recognition (FR) systems causes a marked deterioration in\ntheir performance. Although a considerable amount of research has addressed\nthis issue by inventing new data augmentation techniques, using either input\nspace transformations or Generative Adversarial Networks (GAN) for feature\nspace augmentations, these techniques have yet to satisfy expectations. In this\npaper, we propose an approach named the Face Representation Augmentation (FRA)\nfor augmenting face datasets. To the best of our knowledge, FRA is the first\nmethod that shifts its focus towards manipulating the face embeddings generated\nby any face representation learning algorithm to create new embeddings\nrepresenting the same identity and facial emotion but with an altered posture.\nExtensive experiments conducted in this study convince of the efficacy of our\nmethodology and its power to provide noiseless, completely new facial\nrepresentations to improve the training procedure of any FR algorithm.\nTherefore, FRA can help the recent state-of-the-art FR methods by providing\nmore data for training FR systems. The proposed method, using experiments\nconducted on the Karolinska Directed Emotional Faces (KDEF) dataset, improves\nthe identity classification accuracies by 9.52 %, 10.04 %, and 16.60 %, in\ncomparison with the base models of MagFace, ArcFace, and CosFace, respectively.\n", "date": "2023-01-27", "categories": []}
{"id": "2310.06107", "title": "Developing and Refining a Multifunctional Facial Recognition System for\n  Older Adults with Cognitive Impairments: A Journey Towards Enhanced Quality\n  of Life", "abstract": "  In an era where the global population is aging significantly, cognitive\nimpairments among the elderly have become a major health concern. The need for\neffective assistive technologies is clear, and facial recognition systems are\nemerging as promising tools to address this issue. This document discusses the\ndevelopment and evaluation of a new Multifunctional Facial Recognition System\n(MFRS), designed specifically to assist older adults with cognitive\nimpairments. The MFRS leverages face_recognition [1], a powerful open-source\nlibrary capable of extracting, identifying, and manipulating facial features.\nOur system integrates the face recognition and retrieval capabilities of\nface_recognition, along with additional functionalities to capture images and\nrecord voice memos. This combination of features notably enhances the system's\nusability and versatility, making it a more user-friendly and universally\napplicable tool for end-users. The source code for this project can be accessed\nat https://github.com/Li-8023/Multi-function-face-recognition.git.\n", "date": "2023-10-09", "categories": []}
{"id": "2310.04541", "title": "Iris Liveness Detection Competition (LivDet-Iris) -- The 2023 Edition", "abstract": "  This paper describes the results of the 2023 edition of the ''LivDet'' series\nof iris presentation attack detection (PAD) competitions. New elements in this\nfifth competition include (1) GAN-generated iris images as a category of\npresentation attack instruments (PAI), and (2) an evaluation of human accuracy\nat detecting PAI as a reference benchmark. Clarkson University and the\nUniversity of Notre Dame contributed image datasets for the competition,\ncomposed of samples representing seven different PAI categories, as well as\nbaseline PAD algorithms. Fraunhofer IGD, Beijing University of Civil\nEngineering and Architecture, and Hochschule Darmstadt contributed results for\na total of eight PAD algorithms to the competition. Accuracy results are\nanalyzed by different PAI types, and compared to human accuracy. Overall, the\nFraunhofer IGD algorithm, using an attention-based pixel-wise binary\nsupervision network, showed the best-weighted accuracy results (average\nclassification error rate of 37.31%), while the Beijing University of Civil\nEngineering and Architecture's algorithm won when equal weights for each PAI\nwere given (average classification rate of 22.15%). These results suggest that\niris PAD is still a challenging problem.\n", "date": "2023-10-06", "categories": []}
{"id": "2310.04059", "title": "DEFT: A new distance-based feature set for keystroke dynamics", "abstract": "  Keystroke dynamics is a behavioural biometric utilised for user\nidentification and authentication. We propose a new set of features based on\nthe distance between keys on the keyboard, a concept that has not been\nconsidered before in keystroke dynamics. We combine flight times, a popular\nmetric, with the distance between keys on the keyboard and call them as\nDistance Enhanced Flight Time features (DEFT). This novel approach provides\ncomprehensive insights into a person's typing behaviour, surpassing typing\nvelocity alone. We build a DEFT model by combining DEFT features with other\npreviously used keystroke dynamic features. The DEFT model is designed to be\ndevice-agnostic, allowing us to evaluate its effectiveness across three\ncommonly used devices: desktop, mobile, and tablet. The DEFT model outperforms\nthe existing state-of-the-art methods when we evaluate its effectiveness across\ntwo datasets. We obtain accuracy rates exceeding 99% and equal error rates\nbelow 10% on all three devices.\n", "date": "2023-10-06", "categories": []}
{"id": "2309.06262", "title": "Modality Unifying Network for Visible-Infrared Person Re-Identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) is a challenging task due\nto large cross-modality discrepancies and intra-class variations. Existing\nmethods mainly focus on learning modality-shared representations by embedding\ndifferent modalities into the same feature space. As a result, the learned\nfeature emphasizes the common patterns across modalities while suppressing\nmodality-specific and identity-aware information that is valuable for Re-ID. To\naddress these issues, we propose a novel Modality Unifying Network (MUN) to\nexplore a robust auxiliary modality for VI-ReID. First, the auxiliary modality\nis generated by combining the proposed cross-modality learner and\nintra-modality learner, which can dynamically model the modality-specific and\nmodality-shared representations to alleviate both cross-modality and\nintra-modality variations. Second, by aligning identity centres across the\nthree modalities, an identity alignment loss function is proposed to discover\nthe discriminative feature representations. Third, a modality alignment loss is\nintroduced to consistently reduce the distribution distance of visible and\ninfrared images by modality prototype modeling. Extensive experiments on\nmultiple public datasets demonstrate that the proposed method surpasses the\ncurrent state-of-the-art methods by a significant margin.\n", "date": "2023-09-12", "categories": []}
{"id": "2310.03091", "title": "Privacy-preserving Multi-biometric Indexing based on Frequent Binary\n  Patterns", "abstract": "  The development of large-scale identification systems that ensure the privacy\nprotection of enrolled subjects represents a major challenge. Biometric\ndeployments that provide interoperability and usability by including efficient\nmulti-biometric solutions are a recent requirement. In the context of privacy\nprotection, several template protection schemes have been proposed in the past.\nHowever, these schemes seem inadequate for indexing (workload reduction) in\nbiometric identification systems. More specifically, they have been used in\nidentification systems that perform exhaustive searches, leading to a\ndegradation of computational efficiency. To overcome these limitations, we\npropose an efficient privacy-preserving multi-biometric identification system\nthat retrieves protected deep cancelable templates and is agnostic with respect\nto biometric characteristics and biometric template protection schemes. To this\nend, a multi-biometric binning scheme is designed to exploit the low\nintra-class variation properties contained in the frequent binary patterns\nextracted from different types of biometric characteristics. Experimental\nresults reported on publicly available databases using state-of-the-art Deep\nNeural Network (DNN)-based embedding extractors show that the protected\nmulti-biometric identification system can reduce the computational workload to\napproximately 57\\% (indexing up to three types of biometric characteristics)\nand 53% (indexing up to two types of biometric characteristics), while\nsimultaneously improving the biometric performance of the baseline biometric\nsystem at the high-security thresholds. The source code of the proposed\nmulti-biometric indexing approach together with the composed multi-biometric\ndataset, will be made available to the research community once the article is\naccepted.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.02997", "title": "Optimizing Key-Selection for Face-based One-Time Biometrics via Morphing", "abstract": "  Nowadays, facial recognition systems are still vulnerable to adversarial\nattacks. These attacks vary from simple perturbations of the input image to\nmodifying the parameters of the recognition model to impersonate an authorised\nsubject. So-called privacy-enhancing facial recognition systems have been\nmostly developed to provide protection of stored biometric reference data, i.e.\ntemplates. In the literature, privacy-enhancing facial recognition approaches\nhave focused solely on conventional security threats at the template level,\nignoring the growing concern related to adversarial attacks. Up to now, few\nworks have provided mechanisms to protect face recognition against adversarial\nattacks while maintaining high security at the template level. In this paper,\nwe propose different key selection strategies to improve the security of a\ncompetitive cancelable scheme operating at the signal level. Experimental\nresults show that certain strategies based on signal-level key selection can\nlead to complete blocking of the adversarial attack based on an iterative\noptimization for the most secure threshold, while for the most practical\nthreshold, the attack success chance can be decreased to approximately 5.0%.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.03005", "title": "Reversing Deep Face Embeddings with Probable Privacy Protection", "abstract": "  Generally, privacy-enhancing face recognition systems are designed to offer\npermanent protection of face embeddings. Recently, so-called soft-biometric\nprivacy-enhancement approaches have been introduced with the aim of canceling\nsoft-biometric attributes. These methods limit the amount of soft-biometric\ninformation (gender or skin-colour) that can be inferred from face embeddings.\nPrevious work has underlined the need for research into rigorous evaluations\nand standardised evaluation protocols when assessing privacy protection\ncapabilities. Motivated by this fact, this paper explores to what extent the\nnon-invertibility requirement can be met by methods that claim to provide\nsoft-biometric privacy protection. Additionally, a detailed vulnerability\nassessment of state-of-the-art face embedding extractors is analysed in terms\nof the transformation complexity used for privacy protection. In this context,\na well-known state-of-the-art face image reconstruction approach has been\nevaluated on protected face embeddings to break soft biometric privacy\nprotection. Experimental results show that biometric privacy-enhanced face\nembeddings can be reconstructed with an accuracy of up to approximately 98%,\ndepending on the complexity of the protection algorithm.\n", "date": "2023-10-04", "categories": []}
{"id": "2310.02140", "title": "PAD-Phys: Exploiting Physiology for Presentation Attack Detection in\n  Face Biometrics", "abstract": "  Presentation Attack Detection (PAD) is a crucial stage in facial recognition\nsystems to avoid leakage of personal information or spoofing of identity to\nentities. Recently, pulse detection based on remote photoplethysmography (rPPG)\nhas been shown to be effective in face presentation attack detection.\n  This work presents three different approaches to the presentation attack\ndetection based on rPPG: (i) The physiological domain, a domain using\nrPPG-based models, (ii) the Deepfakes domain, a domain where models were\nretrained from the physiological domain to specific Deepfakes detection tasks;\nand (iii) a new Presentation Attack domain was trained by applying transfer\nlearning from the two previous domains to improve the capability to\ndifferentiate between bona-fides and attacks.\n  The results show the efficiency of the rPPG-based models for presentation\nattack detection, evidencing a 21.70% decrease in average classification error\nrate (ACER) (from 41.03% to 19.32%) when the presentation attack domain is\ncompared to the physiological and Deepfakes domains. Our experiments highlight\nthe efficiency of transfer learning in rPPG-based models and perform well in\npresentation attack detection in instruments that do not allow copying of this\nphysiological feature.\n", "date": "2023-10-03", "categories": []}
{"id": "2210.11817", "title": "Motion Matters: A Novel Motion Modeling For Cross-View Gait Feature\n  Learning", "abstract": "  As a unique biometric that can be perceived at a distance, gait has broad\napplications in person authentication, social security, and so on. Existing\ngait recognition methods suffer from changes in viewpoint and clothing and\nbarely consider extracting diverse motion features, a fundamental\ncharacteristic in gaits, from gait sequences. This paper proposes a novel\nmotion modeling method to extract the discriminative and robust representation.\nSpecifically, we first extract the motion features from the encoded motion\nsequences in the shallow layer. Then we continuously enhance the motion feature\nin deep layers. This motion modeling approach is independent of mainstream work\nin building network architectures. As a result, one can apply this motion\nmodeling method to any backbone to improve gait recognition performance. In\nthis paper, we combine motion modeling with one commonly used backbone~(GaitGL)\nas GaitGL-M to illustrate motion modeling. Extensive experimental results on\ntwo commonly-used cross-view gait datasets demonstrate the superior performance\nof GaitGL-M over existing state-of-the-art methods.\n", "date": "2022-10-21", "categories": []}
{"id": "2303.13006", "title": "Controllable Inversion of Black-Box Face Recognition Models via\n  Diffusion", "abstract": "  Face recognition models embed a face image into a low-dimensional identity\nvector containing abstract encodings of identity-specific facial features that\nallow individuals to be distinguished from one another. We tackle the\nchallenging task of inverting the latent space of pre-trained face recognition\nmodels without full model access (i.e. black-box setting). A variety of methods\nhave been proposed in literature for this task, but they have serious\nshortcomings such as a lack of realistic outputs and strong requirements for\nthe data set and accessibility of the face recognition model. By analyzing the\nblack-box inversion problem, we show that the conditional diffusion model loss\nnaturally emerges and that we can effectively sample from the inverse\ndistribution even without an identity-specific loss. Our method, named identity\ndenoising diffusion probabilistic model (ID3PM), leverages the stochastic\nnature of the denoising diffusion process to produce high-quality,\nidentity-preserving face images with various backgrounds, lighting, poses, and\nexpressions. We demonstrate state-of-the-art performance in terms of identity\npreservation and diversity both qualitatively and quantitatively, and our\nmethod is the first black-box face recognition model inversion method that\noffers intuitive control over the generation process.\n", "date": "2023-03-23", "categories": []}
{"id": "2310.00629", "title": "Finger-UNet: A U-Net based Multi-Task Architecture for Deep Fingerprint\n  Enhancement", "abstract": "  For decades, fingerprint recognition has been prevalent for security,\nforensics, and other biometric applications. However, the availability of\ngood-quality fingerprints is challenging, making recognition difficult.\nFingerprint images might be degraded with a poor ridge structure and noisy or\nless contrasting backgrounds. Hence, fingerprint enhancement plays a vital role\nin the early stages of the fingerprint recognition/verification pipeline. In\nthis paper, we investigate and improvise the encoder-decoder style architecture\nand suggest intuitive modifications to U-Net to enhance low-quality\nfingerprints effectively. We investigate the use of Discrete Wavelet Transform\n(DWT) for fingerprint enhancement and use a wavelet attention module instead of\nmax pooling which proves advantageous for our task. Moreover, we replace\nregular convolutions with depthwise separable convolutions, which significantly\nreduces the memory footprint of the model without degrading the performance. We\nalso demonstrate that incorporating domain knowledge with fingerprint minutiae\nprediction task can improve fingerprint reconstruction through multi-task\nlearning. Furthermore, we also integrate the orientation estimation task to\npropagate the knowledge of ridge orientations to enhance the performance\nfurther. We present the experimental results and evaluate our model on FVC 2002\nand NIST SD302 databases to show the effectiveness of our approach compared to\nprevious works.\n", "date": "2023-10-01", "categories": []}
{"id": "2310.00659", "title": "Liveness Detection Competition -- Noncontact-based Fingerprint\n  Algorithms and Systems (LivDet-2023 Noncontact Fingerprint)", "abstract": "  Liveness Detection (LivDet) is an international competition series open to\nacademia and industry with the objec-tive to assess and report state-of-the-art\nin Presentation Attack Detection (PAD). LivDet-2023 Noncontact Fingerprint is\nthe first edition of the noncontact fingerprint-based PAD competition for\nalgorithms and systems. The competition serves as an important benchmark in\nnoncontact-based fingerprint PAD, offering (a) independent assessment of the\nstate-of-the-art in noncontact-based fingerprint PAD for algorithms and\nsystems, and (b) common evaluation protocol, which includes finger photos of a\nvariety of Presentation Attack Instruments (PAIs) and live fingers to the\nbiometric research community (c) provides standard algorithm and system\nevaluation protocols, along with the comparative analysis of state-of-the-art\nalgorithms from academia and industry with both old and new android\nsmartphones. The winning algorithm achieved an APCER of 11.35% averaged overall\nPAIs and a BPCER of 0.62%. The winning system achieved an APCER of 13.0.4%,\naveraged over all PAIs tested over all the smartphones, and a BPCER of 1.68%\nover all smartphones tested. Four-finger systems that make individual\nfinger-based PAD decisions were also tested. The dataset used for competition\nwill be available 1 to all researchers as per data share protocol\n", "date": "2023-10-01", "categories": []}
{"id": "2210.13905", "title": "Confidence-Calibrated Face and Kinship Verification", "abstract": "  In this paper, we investigate the problem of prediction confidence in face\nand kinship verification. Most existing face and kinship verification methods\nfocus on accuracy performance while ignoring confidence estimation for their\nprediction results. However, confidence estimation is essential for modeling\nreliability and trustworthiness in such high-risk tasks. To address this, we\nintroduce an effective confidence measure that allows verification models to\nconvert a similarity score into a confidence score for any given face pair. We\nfurther propose a confidence-calibrated approach, termed Angular Scaling\nCalibration (ASC). ASC is easy to implement and can be readily applied to\nexisting verification models without model modifications, yielding\naccuracy-preserving and confidence-calibrated probabilistic verification\nmodels. In addition, we introduce the uncertainty in the calibrated confidence\nto boost the reliability and trustworthiness of the verification models in the\npresence of noisy data. To the best of our knowledge, our work presents the\nfirst comprehensive confidence-calibrated solution for modern face and kinship\nverification tasks. We conduct extensive experiments on four widely used face\nand kinship verification datasets, and the results demonstrate the\neffectiveness of our proposed approach. Code and models are available at\nhttps://github.com/cnulab/ASC.\n", "date": "2022-10-25", "categories": []}
{"id": "2309.17156", "title": "Age Group Discrimination via Free Handwriting Indicators", "abstract": "  The growing global elderly population is expected to increase the prevalence\nof frailty, posing significant challenges to healthcare systems. Frailty, a\nsyndrome associated with ageing, is characterised by progressive health\ndecline, increased vulnerability to stressors and increased risk of mortality.\nIt represents a significant burden on public health and reduces the quality of\nlife of those affected. The lack of a universally accepted method to assess\nfrailty and a standardised definition highlights a critical research gap. Given\nthis lack and the importance of early prevention, this study presents an\ninnovative approach using an instrumented ink pen to ecologically assess\nhandwriting for age group classification. Content-free handwriting data from 80\nhealthy participants in different age groups (20-40, 41-60, 61-70 and 70+) were\nanalysed. Fourteen gesture- and tremor-related indicators were computed from\nthe raw data and used in five classification tasks. These tasks included\ndiscriminating between adjacent and non-adjacent age groups using Catboost and\nLogistic Regression classifiers. Results indicate exceptional classifier\nperformance, with accuracy ranging from 82.5% to 97.5%, precision from 81.8% to\n100%, recall from 75% to 100% and ROC-AUC from 92.2% to 100%. Model\ninterpretability, facilitated by SHAP analysis, revealed age-dependent\nsensitivity of temporal and tremor-related handwriting features. Importantly,\nthis classification method offers potential for early detection of abnormal\nsigns of ageing in uncontrolled settings such as remote home monitoring,\nthereby addressing the critical issue of frailty detection and contributing to\nimproved care for older adults.\n", "date": "2023-09-29", "categories": []}
{"id": "2309.15578", "title": "LivDet2023 -- Fingerprint Liveness Detection Competition: Advancing\n  Generalization", "abstract": "  The International Fingerprint Liveness Detection Competition (LivDet) is a\nbiennial event that invites academic and industry participants to prove their\nadvancements in Fingerprint Presentation Attack Detection (PAD). This edition,\nLivDet2023, proposed two challenges, Liveness Detection in Action and\nFingerprint Representation, to evaluate the efficacy of PAD embedded in\nverification systems and the effectiveness and compactness of feature sets. A\nthird, hidden challenge is the inclusion of two subsets in the training set\nwhose sensor information is unknown, testing participants ability to generalize\ntheir models. Only bona fide fingerprint samples were provided to participants,\nand the competition reports and assesses the performance of their algorithms\nsuffering from this limitation in data availability.\n", "date": "2023-09-27", "categories": []}
{"id": "2309.15734", "title": "Synthetic Latent Fingerprint Generation Using Style Transfer", "abstract": "  Limited data availability is a challenging problem in the latent fingerprint\ndomain. Synthetically generated fingerprints are vital for training data-hungry\nneural network-based algorithms. Conventional methods distort clean\nfingerprints to generate synthetic latent fingerprints. We propose a simple and\neffective approach using style transfer and image blending to synthesize\nrealistic latent fingerprints. Our evaluation criteria and experiments\ndemonstrate that the generated synthetic latent fingerprints preserve the\nidentity information from the input contact-based fingerprints while possessing\nsimilar characteristics as real latent fingerprints. Additionally, we show that\nthe generated fingerprints exhibit several qualities and styles, suggesting\nthat the proposed method can generate multiple samples from a single\nfingerprint.\n", "date": "2023-09-27", "categories": []}
{"id": "2309.15780", "title": "AaP-ReID: Improved Attention-Aware Person Re-identification", "abstract": "  Person re-identification (ReID) is a well-known problem in the field of\ncomputer vision. The primary objective is to identify a specific individual\nwithin a gallery of images. However, this task is challenging due to various\nfactors, such as pose variations, illumination changes, obstructions, and the\npresence ofconfusing backgrounds. Existing ReID methods often fail to capture\ndiscriminative features (e.g., head, shoes, backpacks) and instead capture\nirrelevant features when the target is occluded. Motivated by the success of\npart-based and attention-based ReID methods, we improve AlignedReID++ and\npresent AaP-ReID, a more effective method for person ReID that incorporates\nchannel-wise attention into a ResNet-based architecture. Our method\nincorporates the Channel-Wise Attention Bottleneck (CWAbottleneck) block and\ncan learn discriminating features by dynamically adjusting the importance\nofeach channel in the feature maps. We evaluated Aap-ReID on three benchmark\ndatasets: Market-1501, DukeMTMC-reID, and CUHK03. When compared with\nstate-of-the-art person ReID methods, we achieve competitive results with\nrank-1 accuracies of 95.6% on Market-1501, 90.6% on DukeMTMC-reID, and 82.4% on\nCUHK03.\n", "date": "2023-09-27", "categories": []}
{"id": "2309.14806", "title": "3D printed realistic finger vein phantoms", "abstract": "  Finger vein pattern recognition is an emerging biometric with a good\nresistance to presentation attacks and low error rates. One problem is that it\nis hard to obtain ground truth finger vein patterns from live fingers. In this\npaper we propose an advanced method to create finger vein phantoms using 3D\nprinting where we mimic the optical properties of the various tissues inside\nthe fingers, like bone, veins and soft tissues using different printing\nmaterials and parameters. We demonstrate that we are able to create finger\nphantoms that result in realistic finger vein images and precisely known vein\npatterns. These phantoms can be used to develop and evaluate finger vein\nextraction and recognition methods. In addition, we show that the finger vein\nphantoms can be used to spoof a finger vein recognition system. This paper is\nbased on the Master's thesis of Rasmus van der Grift.\n", "date": "2023-09-26", "categories": []}
{"id": "2309.13046", "title": "Privacy Preserving Machine Learning for Behavioral Authentication\n  Systems", "abstract": "  A behavioral authentication (BA) system uses the behavioral characteristics\nof users to verify their identity claims. A BA verification algorithm can be\nconstructed by training a neural network (NN) classifier on users' profiles.\nThe trained NN model classifies the presented verification data, and if the\nclassification matches the claimed identity, the verification algorithm accepts\nthe claim. This classification-based approach removes the need to maintain a\nprofile database. However, similar to other NN architectures, the NN classifier\nof the BA system is vulnerable to privacy attacks. To protect the privacy of\ntraining and test data used in an NN different techniques are widely used. In\nthis paper, our focus is on a non-crypto-based approach, and we used random\nprojection (RP) to ensure data privacy in an NN model. RP is a\ndistance-preserving transformation based on a random matrix. Before sharing the\nprofiles with the verifier, users will transform their profiles by RP and keep\ntheir matrices secret. To reduce the computation load in RP, we use sparse\nrandom projection, which is very effective for low-compute devices. Along with\ncorrectness and security properties, our system can ensure the changeability\nproperty of the BA system. We also introduce an ML-based privacy attack, and\nour proposed system is robust against this and other privacy and security\nattacks. We implemented our approach on three existing behavioral BA systems\nand achieved a below 2.0% FRR and a below 1.0% FAR rate. Moreover, the machine\nlearning-based privacy attacker can only recover below 3.0% to 12.0% of\nfeatures from a portion of the projected profiles. However, these recovered\nfeatures are not sufficient to know details about the users' behavioral pattern\nor to be used in a subsequent attack. Our approach is general and can be used\nin other NN-based BA systems as well as in traditional biometric systems.\n", "date": "2023-08-31", "categories": []}
{"id": "2309.13137", "title": "Trading-off Mutual Information on Feature Aggregation for Face\n  Recognition", "abstract": "  Despite the advances in the field of Face Recognition (FR), the precision of\nthese methods is not yet sufficient. To improve the FR performance, this paper\nproposes a technique to aggregate the outputs of two state-of-the-art (SOTA)\ndeep FR models, namely ArcFace and AdaFace. In our approach, we leverage the\ntransformer attention mechanism to exploit the relationship between different\nparts of two feature maps. By doing so, we aim to enhance the overall\ndiscriminative power of the FR system. One of the challenges in feature\naggregation is the effective modeling of both local and global dependencies.\nConventional transformers are known for their ability to capture long-range\ndependencies, but they often struggle with modeling local dependencies\naccurately. To address this limitation, we augment the self-attention mechanism\nto capture both local and global dependencies effectively. This allows our\nmodel to take advantage of the overlapping receptive fields present in\ncorresponding locations of the feature maps. However, fusing two feature maps\nfrom different FR models might introduce redundancies to the face embedding.\nSince these models often share identical backbone architectures, the resulting\nfeature maps may contain overlapping information, which can mislead the\ntraining process. To overcome this problem, we leverage the principle of\nInformation Bottleneck to obtain a maximally informative facial representation.\nThis ensures that the aggregated features retain the most relevant and\ndiscriminative information while minimizing redundant or misleading details. To\nevaluate the effectiveness of our proposed method, we conducted experiments on\npopular benchmarks and compared our results with state-of-the-art algorithms.\nThe consistent improvement we observed in these benchmarks demonstrates the\nefficacy of our approach in enhancing FR performance.\n", "date": "2023-09-22", "categories": []}
{"id": "2309.13317", "title": "Class Attendance System in Education with Deep Learning Method", "abstract": "  With the advancing technology, the hardware gain of computers and the\nincrease in the processing capacity of processors have facilitated the\nprocessing of instantaneous and real-time images. Face recognition processes\nare also studies in the field of image processing. Facial recognition processes\nare frequently used in security applications and commercial applications.\nEspecially in the last 20 years, the high performances of artificial\nintelligence (AI) studies have contributed to the spread of these studies in\nmany different fields. Education is one of them. The potential and advantages\nof using AI in education; can be grouped under three headings: student,\nteacher, and institution. One of the institutional studies may be the security\nof educational environments and the contribution of automation to education and\ntraining processes. From this point of view, deep learning methods, one of the\nsub-branches of AI, were used in this study. For object detection from images,\na pioneering study has been designed and successfully implemented to keep\nrecords of students' entrance to the educational institution and to perform\nclass attendance with images taken from the camera using image processing\nalgorithms. The application of the study to real-life problems will be carried\nout in a school determined in the 2022-2023 academic year.\n", "date": "2023-09-23", "categories": []}
{"id": "2309.13704", "title": "Sound-Print: Generalised Face Presentation Attack Detection using Deep\n  Representation of Sound Echoes", "abstract": "  Facial biometrics are widely deployed in smartphone-based applications\nbecause of their usability and increased verification accuracy in unconstrained\nscenarios. The evolving applications of smartphone-based facial recognition\nhave also increased Presentation Attacks (PAs), where an attacker can present a\nPresentation Attack Instrument (PAI) to maliciously gain access to the\napplication. Because the materials used to generate PAI are not deterministic,\nthe detection of unknown presentation attacks is challenging. In this paper, we\npresent an acoustic echo-based face Presentation Attack Detection (PAD) on a\nsmartphone in which the PAs are detected based on the reflection profiles of\nthe transmitted signal. We propose a novel transmission signal based on the\nwide pulse that allows us to model the background noise before transmitting the\nsignal and increase the Signal-to-Noise Ratio (SNR). The received signal\nreflections were processed to remove background noise and accurately represent\nreflection characteristics. The reflection profiles of the bona fide and PAs\nare different owing to the different reflection characteristics of the human\nskin and artefact materials. Extensive experiments are presented using the\nnewly collected Acoustic Sound Echo Dataset (ASED) with 4807 samples captured\nfrom bona fide and four different types of PAIs, including print (two types),\ndisplay, and silicone face-mask attacks. The obtained results indicate the\nrobustness of the proposed method for detecting unknown face presentation\nattacks.\n", "date": "2023-09-24", "categories": []}
{"id": "2309.13763", "title": "Combining Two Adversarial Attacks Against Person Re-Identification\n  Systems", "abstract": "  The field of Person Re-Identification (Re-ID) has received much attention\nrecently, driven by the progress of deep neural networks, especially for image\nclassification. The problem of Re-ID consists in identifying individuals\nthrough images captured by surveillance cameras in different scenarios.\nGovernments and companies are investing a lot of time and money in Re-ID\nsystems for use in public safety and identifying missing persons. However,\nseveral challenges remain for successfully implementing Re-ID, such as\nocclusions and light reflections in people's images. In this work, we focus on\nadversarial attacks on Re-ID systems, which can be a critical threat to the\nperformance of these systems. In particular, we explore the combination of\nadversarial attacks against Re-ID models, trying to strengthen the decrease in\nthe classification results. We conduct our experiments on three datasets:\nDukeMTMC-ReID, Market-1501, and CUHK03. We combine the use of two types of\nadversarial attacks, P-FGSM and Deep Mis-Ranking, applied to two popular Re-ID\nmodels: IDE (ResNet-50) and AlignedReID. The best result demonstrates a\ndecrease of 3.36% in the Rank-10 metric for AlignedReID applied to CUHK03. We\nalso try to use Dropout during the inference as a defense method.\n", "date": "2023-09-24", "categories": []}
{"id": "2308.10658", "title": "Learning Clothing and Pose Invariant 3D Shape Representation for\n  Long-Term Person Re-Identification", "abstract": "  Long-Term Person Re-Identification (LT-ReID) has become increasingly crucial\nin computer vision and biometrics. In this work, we aim to extend LT-ReID\nbeyond pedestrian recognition to include a wider range of real-world human\nactivities while still accounting for cloth-changing scenarios over large time\ngaps. This setting poses additional challenges due to the geometric\nmisalignment and appearance ambiguity caused by the diversity of human pose and\nclothing. To address these challenges, we propose a new approach 3DInvarReID\nfor (i) disentangling identity from non-identity components (pose, clothing\nshape, and texture) of 3D clothed humans, and (ii) reconstructing accurate 3D\nclothed body shapes and learning discriminative features of naked body shapes\nfor person ReID in a joint manner. To better evaluate our study of LT-ReID, we\ncollect a real-world dataset called CCDA, which contains a wide variety of\nhuman activities and clothing changes. Experimentally, we show the superior\nperformance of our approach for person ReID.\n", "date": "2023-08-21", "categories": []}
{"id": "2309.12118", "title": "Vulnerability of 3D Face Recognition Systems to Morphing Attacks", "abstract": "  In recent years face recognition systems have been brought to the mainstream\ndue to development in hardware and software. Consistent efforts are being made\nto make them better and more secure. This has also brought developments in 3D\nface recognition systems at a rapid pace. These 3DFR systems are expected to\novercome certain vulnerabilities of 2DFR systems. One such problem that the\ndomain of 2DFR systems face is face image morphing. A substantial amount of\nresearch is being done for generation of high quality face morphs along with\ndetection of attacks from these morphs. Comparatively the understanding of\nvulnerability of 3DFR systems against 3D face morphs is less. But at the same\ntime an expectation is set from 3DFR systems to be more robust against such\nattacks. This paper attempts to research and gain more information on this\nmatter. The paper describes a couple of methods that can be used to generate 3D\nface morphs. The face morphs that are generated using this method are then\ncompared to the contributing faces to obtain similarity scores. The highest\nMMPMR is obtained around 40% with RMMR of 41.76% when 3DFRS are attacked with\nlook-a-like morphs.\n", "date": "2023-09-21", "categories": []}
{"id": "2309.12237", "title": "t-EER: Parameter-Free Tandem Evaluation of Countermeasures and Biometric\n  Comparators", "abstract": "  Presentation attack (spoofing) detection (PAD) typically operates alongside\nbiometric verification to improve reliablity in the face of spoofing attacks.\nEven though the two sub-systems operate in tandem to solve the single task of\nreliable biometric verification, they address different detection tasks and are\nhence typically evaluated separately. Evidence shows that this approach is\nsuboptimal. We introduce a new metric for the joint evaluation of PAD solutions\noperating in situ with biometric verification. In contrast to the tandem\ndetection cost function proposed recently, the new tandem equal error rate\n(t-EER) is parameter free. The combination of two classifiers nonetheless leads\nto a \\emph{set} of operating points at which false alarm and miss rates are\nequal and also dependent upon the prevalence of attacks. We therefore introduce\nthe \\emph{concurrent} t-EER, a unique operating point which is invariable to\nthe prevalence of attacks. Using both modality (and even application) agnostic\nsimulated scores, as well as real scores for a voice biometrics application, we\ndemonstrate application of the t-EER to a wide range of biometric system\nevaluations under attack. The proposed approach is a strong candidate metric\nfor the tandem evaluation of PAD systems and biometric comparators.\n", "date": "2023-09-21", "categories": []}
{"id": "2309.10641", "title": "KFC: Kinship Verification with Fair Contrastive Loss and Multi-Task\n  Learning", "abstract": "  Kinship verification is an emerging task in computer vision with multiple\npotential applications. However, there's no large enough kinship dataset to\ntrain a representative and robust model, which is a limitation for achieving\nbetter performance. Moreover, face verification is known to exhibit bias, which\nhas not been dealt with by previous kinship verification works and sometimes\neven results in serious issues. So we first combine existing kinship datasets\nand label each identity with the correct race in order to take race information\ninto consideration and provide a larger and complete dataset, called KinRace\ndataset. Secondly, we propose a multi-task learning model structure with\nattention module to enhance accuracy, which surpasses state-of-the-art\nperformance. Lastly, our fairness-aware contrastive loss function with\nadversarial learning greatly mitigates racial bias. We introduce a debias term\ninto traditional contrastive loss and implement gradient reverse in race\nclassification task, which is an innovative idea to mix two fairness methods to\nalleviate bias. Exhaustive experimental evaluation demonstrates the\neffectiveness and superior performance of the proposed KFC in both standard\ndeviation and accuracy at the same time.\n", "date": "2023-09-19", "categories": []}
{"id": "2305.07710", "title": "Zero-shot racially balanced dataset generation using an existing biased\n  StyleGAN2", "abstract": "  Facial recognition systems have made significant strides thanks to data-heavy\ndeep learning models, but these models rely on large privacy-sensitive\ndatasets. Further, many of these datasets lack diversity in terms of ethnicity\nand demographics, which can lead to biased models that can have serious\nsocietal and security implications. To address these issues, we propose a\nmethodology that leverages the biased generative model StyleGAN2 to create\ndemographically diverse images of synthetic individuals. The synthetic dataset\nis created using a novel evolutionary search algorithm that targets specific\ndemographic groups. By training face recognition models with the resulting\nbalanced dataset containing 50,000 identities per race (13.5 million images in\ntotal), we can improve their performance and minimize biases that might have\nbeen present in a model trained on a real dataset.\n", "date": "2023-05-12", "categories": []}
{"id": "2306.06196", "title": "ElectroCardioGuard: Preventing Patient Misidentification in\n  Electrocardiogram Databases through Neural Networks", "abstract": "  Electrocardiograms (ECGs) are commonly used by cardiologists to detect\nheart-related pathological conditions. Reliable collections of ECGs are crucial\nfor precise diagnosis. However, in clinical practice, the assignment of\ncaptured ECG recordings to incorrect patients can occur inadvertently. In\ncollaboration with a clinical and research facility which recognized this\nchallenge and reached out to us, we present a study that addresses this issue.\nIn this work, we propose a small and efficient neural-network based model for\ndetermining whether two ECGs originate from the same patient. Our model\ndemonstrates great generalization capabilities and achieves state-of-the-art\nperformance in gallery-probe patient identification on PTB-XL while utilizing\n760x fewer parameters. Furthermore, we present a technique leveraging our model\nfor detection of recording-assignment mistakes, showcasing its applicability in\na realistic scenario. Finally, we evaluate our model on a newly collected ECG\ndataset specifically curated for this study, and make it public for the\nresearch community.\n", "date": "2023-06-09", "categories": []}
{"id": "2307.12917", "title": "Hierarchical Skeleton Meta-Prototype Contrastive Learning with Hard\n  Skeleton Mining for Unsupervised Person Re-Identification", "abstract": "  With rapid advancements in depth sensors and deep learning, skeleton-based\nperson re-identification (re-ID) models have recently achieved remarkable\nprogress with many advantages. Most existing solutions learn single-level\nskeleton features from body joints with the assumption of equal skeleton\nimportance, while they typically lack the ability to exploit more informative\nskeleton features from various levels such as limb level with more global body\npatterns. The label dependency of these methods also limits their flexibility\nin learning more general skeleton representations. This paper proposes a\ngeneric unsupervised Hierarchical skeleton Meta-Prototype Contrastive learning\n(Hi-MPC) approach with Hard Skeleton Mining (HSM) for person re-ID with\nunlabeled 3D skeletons. Firstly, we construct hierarchical representations of\nskeletons to model coarse-to-fine body and motion features from the levels of\nbody joints, components, and limbs. Then a hierarchical meta-prototype\ncontrastive learning model is proposed to cluster and contrast the most typical\nskeleton features (\"prototypes\") from different-level skeletons. By converting\noriginal prototypes into meta-prototypes with multiple homogeneous\ntransformations, we induce the model to learn the inherent consistency of\nprototypes to capture more effective skeleton features for person re-ID.\nFurthermore, we devise a hard skeleton mining mechanism to adaptively infer the\ninformative importance of each skeleton, so as to focus on harder skeletons to\nlearn more discriminative skeleton representations. Extensive evaluations on\nfive datasets demonstrate that our approach outperforms a wide variety of\nstate-of-the-art skeleton-based methods. We further show the general\napplicability of our method to cross-view person re-ID and RGB-based scenarios\nwith estimated skeletons.\n", "date": "2023-07-24", "categories": []}
{"id": "2308.03322", "title": "Part-Aware Transformer for Generalizable Person Re-identification", "abstract": "  Domain generalization person re-identification (DG-ReID) aims to train a\nmodel on source domains and generalize well on unseen domains. Vision\nTransformer usually yields better generalization ability than common CNN\nnetworks under distribution shifts. However, Transformer-based ReID models\ninevitably over-fit to domain-specific biases due to the supervised learning\nstrategy on the source domain. We observe that while the global images of\ndifferent IDs should have different features, their similar local parts (e.g.,\nblack backpack) are not bounded by this constraint. Motivated by this, we\npropose a pure Transformer model (termed Part-aware Transformer) for DG-ReID by\ndesigning a proxy task, named Cross-ID Similarity Learning (CSL), to mine local\nvisual information shared by different IDs. This proxy task allows the model to\nlearn generic features because it only cares about the visual similarity of the\nparts regardless of the ID labels, thus alleviating the side effect of\ndomain-specific biases. Based on the local similarity obtained in CSL, a\nPart-guided Self-Distillation (PSD) is proposed to further improve the\ngeneralization of global features. Our method achieves state-of-the-art\nperformance under most DG ReID settings. Under the Market$\\to$Duke setting, our\nmethod exceeds state-of-the-art by 10.9% and 12.8% in Rank1 and mAP,\nrespectively. The code is available at\nhttps://github.com/liyuke65535/Part-Aware-Transformer.\n", "date": "2023-08-07", "categories": []}
{"id": "2309.09480", "title": "Stealthy Physical Masked Face Recognition Attack via Adversarial Style\n  Optimization", "abstract": "  Deep neural networks (DNNs) have achieved state-of-the-art performance on\nface recognition (FR) tasks in the last decade. In real scenarios, the\ndeployment of DNNs requires taking various face accessories into consideration,\nlike glasses, hats, and masks. In the COVID-19 pandemic era, wearing face masks\nis one of the most effective ways to defend against the novel coronavirus.\nHowever, DNNs are known to be vulnerable to adversarial examples with a small\nbut elaborated perturbation. Thus, a facial mask with adversarial perturbations\nmay pose a great threat to the widely used deep learning-based FR models. In\nthis paper, we consider a challenging adversarial setting: targeted attack\nagainst FR models. We propose a new stealthy physical masked FR attack via\nadversarial style optimization. Specifically, we train an adversarial style\nmask generator that hides adversarial perturbations inside style masks.\nMoreover, to ameliorate the phenomenon of sub-optimization with one fixed\nstyle, we propose to discover the optimal style given a target through style\noptimization in a continuous relaxation manner. We simultaneously optimize the\ngenerator and the style selection for generating strong and stealthy\nadversarial style masks. We evaluated the effectiveness and transferability of\nour proposed method via extensive white-box and black-box digital experiments.\nFurthermore, we also conducted physical attack experiments against local FR\nmodels and online platforms.\n", "date": "2023-09-18", "categories": []}
{"id": "2309.08372", "title": "Beyond Domain Gap: Exploiting Subjectivity in Sketch-Based Person\n  Retrieval", "abstract": "  Person re-identification (re-ID) requires densely distributed cameras. In\npractice, the person of interest may not be captured by cameras and, therefore,\nneeds to be retrieved using subjective information (e.g., sketches from\nwitnesses). Previous research defines this case using the sketch as sketch\nre-identification (Sketch re-ID) and focuses on eliminating the domain gap.\nActually, subjectivity is another significant challenge. We model and\ninvestigate it by posing a new dataset with multi-witness descriptions. It\nfeatures two aspects. 1) Large-scale. It contains over 4,763 sketches and\n32,668 photos, making it the largest Sketch re-ID dataset. 2) Multi-perspective\nand multi-style. Our dataset offers multiple sketches for each identity.\nWitnesses' subjective cognition provides multiple perspectives on the same\nindividual, while different artists' drawing styles provide variation in sketch\nstyles. We further have two novel designs to alleviate the challenge of\nsubjectivity. 1) Fusing subjectivity. We propose a non-local (NL) fusion module\nthat gathers sketches from different witnesses for the same identity. 2)\nIntroducing objectivity. An AttrAlign module utilizes attributes as an implicit\nmask to align cross-domain features. To push forward the advance of Sketch\nre-ID, we set three benchmarks (large-scale, multi-style, cross-style).\nExtensive experiments demonstrate our leading performance in these benchmarks.\nDataset and Codes are publicly available at:\nhttps://github.com/Lin-Kayla/subjectivity-sketch-reid\n", "date": "2023-09-15", "categories": []}
{"id": "2309.08442", "title": "Toward responsible face datasets: modeling the distribution of a\n  disentangled latent space for sampling face images from demographic groups", "abstract": "  Recently, it has been exposed that some modern facial recognition systems\ncould discriminate specific demographic groups and may lead to unfair attention\nwith respect to various facial attributes such as gender and origin. The main\nreason are the biases inside datasets, unbalanced demographics, used to train\ntheses models. Unfortunately, collecting a large-scale balanced dataset with\nrespect to various demographics is impracticable.\n  In this paper, we investigate as an alternative the generation of a balanced\nand possibly bias-free synthetic dataset that could be used to train, to\nregularize or to evaluate deep learning-based facial recognition models. We\npropose to use a simple method for modeling and sampling a disentangled\nprojection of a StyleGAN latent space to generate any combination of\ndemographic groups (e.g. $hispanic-female$). Our experiments show that we can\nsynthesis any combination of demographic groups effectively and the identities\nare different from the original training dataset. We also released the source\ncode.\n", "date": "2023-09-15", "categories": []}
{"id": "2309.03558", "title": "Region Generation and Assessment Network for Occluded Person\n  Re-Identification", "abstract": "  Person Re-identification (ReID) plays a more and more crucial role in recent\nyears with a wide range of applications. Existing ReID methods are suffering\nfrom the challenges of misalignment and occlusions, which degrade the\nperformance dramatically. Most methods tackle such challenges by utilizing\nexternal tools to locate body parts or exploiting matching strategies.\nNevertheless, the inevitable domain gap between the datasets utilized for\nexternal tools and the ReID datasets and the complicated matching process make\nthese methods unreliable and sensitive to noises. In this paper, we propose a\nRegion Generation and Assessment Network (RGANet) to effectively and\nefficiently detect the human body regions and highlight the important regions.\nIn the proposed RGANet, we first devise a Region Generation Module (RGM) which\nutilizes the pre-trained CLIP to locate the human body regions using semantic\nprototypes extracted from text descriptions. Learnable prompt is designed to\neliminate domain gap between CLIP datasets and ReID datasets. Then, to measure\nthe importance of each generated region, we introduce a Region Assessment\nModule (RAM) that assigns confidence scores to different regions and reduces\nthe negative impact of the occlusion regions by lower scores. The RAM consists\nof a discrimination-aware indicator and an invariance-aware indicator, where\nthe former indicates the capability to distinguish from different identities\nand the latter represents consistency among the images of the same class of\nhuman body regions. Extensive experimental results for six widely-used\nbenchmarks including three tasks (occluded, partial, and holistic) demonstrate\nthe superiority of RGANet against state-of-the-art methods.\n", "date": "2023-09-07", "categories": []}
{"id": "2309.05879", "title": "Generalized Attacks on Face Verification Systems", "abstract": "  Face verification (FV) using deep neural network models has made tremendous\nprogress in recent years, surpassing human accuracy and seeing deployment in\nvarious applications such as border control and smartphone unlocking. However,\nFV systems are vulnerable to Adversarial Attacks, which manipulate input images\nto deceive these systems in ways usually unnoticeable to humans. This paper\nprovides an in-depth study of attacks on FV systems. We introduce the\nDodgePersonation Attack that formulates the creation of face images that\nimpersonate a set of given identities while avoiding being identified as any of\nthe identities in a separate, disjoint set. A taxonomy is proposed to provide a\nunified view of different types of Adversarial Attacks against FV systems,\nincluding Dodging Attacks, Impersonation Attacks, and Master Face Attacks.\nFinally, we propose the ''One Face to Rule Them All'' Attack which implements\nthe DodgePersonation Attack with state-of-the-art performance on a well-known\nscenario (Master Face Attack) and which can also be used for the new scenarios\nintroduced in this paper. While the state-of-the-art Master Face Attack can\nproduce a set of 9 images to cover 43.82% of the identities in their test\ndatabase, with 9 images our attack can cover 57.27% to 58.5% of these\nidentifies while giving the attacker the choice of the identity to use to\ncreate the impersonation. Moreover, the 9 generated attack images appear\nidentical to a casual observer.\n", "date": "2023-09-12", "categories": []}
{"id": "2304.13800", "title": "Latent Fingerprint Recognition: Fusion of Local and Global Embeddings", "abstract": "  One of the most challenging problems in fingerprint recognition continues to\nbe establishing the identity of a suspect associated with partial and smudgy\nfingerprints left at a crime scene (i.e., latent prints or fingermarks).\nDespite the success of fixed-length embeddings for rolled and slap fingerprint\nrecognition, the features learned for latent fingerprint matching have mostly\nbeen limited to local minutiae-based embeddings and have not directly leveraged\nglobal representations for matching. In this paper, we combine global\nembeddings with local embeddings for state-of-the-art latent to rolled matching\naccuracy with high throughput. The combination of both local and global\nrepresentations leads to improved recognition accuracy across NIST SD 27, NIST\nSD 302, MSP, MOLF DB1/DB4, and MOLF DB2/DB4 latent fingerprint datasets for\nboth closed-set (84.11%, 54.36%, 84.35%, 70.43%, 62.86% rank-1 retrieval rate,\nrespectively) and open-set (0.50, 0.74, 0.44, 0.60, 0.68 FNIR at FPIR=0.02,\nrespectively) identification scenarios on a gallery of 100K rolled\nfingerprints. Not only do we fuse the complimentary representations, we also\nuse the local features to guide the global representations to focus on\ndiscriminatory regions in two fingerprint images to be compared. This leads to\na multi-stage matching paradigm in which subsets of the retrieved candidate\nlists for each probe image are passed to subsequent stages for further\nprocessing, resulting in a considerable reduction in latency (requiring just\n0.068 ms per latent to rolled comparison on a AMD EPYC 7543 32-Core Processor,\nroughly 15K comparisons per second). Finally, we show the generalizability of\nthe fused representations for improving authentication accuracy across several\nrolled, plain, and contactless fingerprint datasets.\n", "date": "2023-04-26", "categories": []}
{"id": "2305.01905", "title": "Localization using Multi-Focal Spatial Attention for Masked Face\n  Recognition", "abstract": "  Since the beginning of world-wide COVID-19 pandemic, facial masks have been\nrecommended to limit the spread of the disease. However, these masks hide\ncertain facial attributes. Hence, it has become difficult for existing face\nrecognition systems to perform identity verification on masked faces. In this\ncontext, it is necessary to develop masked Face Recognition (MFR) for\ncontactless biometric recognition systems. Thus, in this paper, we propose\nComplementary Attention Learning and Multi-Focal Spatial Attention that\nprecisely removes masked region by training complementary spatial attention to\nfocus on two distinct regions: masked regions and backgrounds. In our method,\nstandard spatial attention and networks focus on unmasked regions, and extract\nmask-invariant features while minimizing the loss of the conventional Face\nRecognition (FR) performance. For conventional FR, we evaluate the performance\non the IJB-C, Age-DB, CALFW, and CPLFW datasets. We evaluate the MFR\nperformance on the ICCV2021-MFR/Insightface track, and demonstrate the improved\nperformance on the both MFR and FR datasets. Additionally, we empirically\nverify that spatial attention of proposed method is more precisely activated in\nunmasked regions.\n", "date": "2023-05-03", "categories": []}
{"id": "2306.17206", "title": "FarSight: A Physics-Driven Whole-Body Biometric System at Large Distance\n  and Altitude", "abstract": "  Whole-body biometric recognition is an important area of research due to its\nvast applications in law enforcement, border security, and surveillance. This\npaper presents the end-to-end design, development and evaluation of FarSight,\nan innovative software system designed for whole-body (fusion of face, gait and\nbody shape) biometric recognition. FarSight accepts videos from elevated\nplatforms and drones as input and outputs a candidate list of identities from a\ngallery. The system is designed to address several challenges, including (i)\nlow-quality imagery, (ii) large yaw and pitch angles, (iii) robust feature\nextraction to accommodate large intra-person variabilities and large\ninter-person similarities, and (iv) the large domain gap between training and\ntest sets. FarSight combines the physics of imaging and deep learning models to\nenhance image restoration and biometric feature encoding. We test FarSight's\neffectiveness using the newly acquired IARPA Biometric Recognition and\nIdentification at Altitude and Range (BRIAR) dataset. Notably, FarSight\ndemonstrated a substantial performance increase on the BRIAR dataset, with\ngains of +11.82% Rank-20 identification and +11.3% TAR@1% FAR.\n", "date": "2023-06-29", "categories": []}
{"id": "2309.03390", "title": "A novel method for iris recognition using BP neural network and parallel\n  computing by the aid of GPUs (Graphics Processing Units)", "abstract": "  In this paper, we seek a new method in designing an iris recognition system.\nIn this method, first the Haar wavelet features are extracted from iris images.\nThe advantage of using these features is the high-speed extraction, as well as\nbeing unique to each iris. Then the back propagation neural network (BPNN) is\nused as a classifier. In this system, the BPNN parallel algorithms and their\nimplementation on GPUs have been used by the aid of CUDA in order to speed up\nthe learning process. Finally, the system performance and the speeding outcomes\nin a way that this algorithm is done in series are presented.\n", "date": "2023-09-06", "categories": []}
{"id": "2206.09410", "title": "Low-Mid Adversarial Perturbation against Unauthorized Face Recognition\n  System", "abstract": "  In light of the growing concerns regarding the unauthorized use of facial\nrecognition systems and its implications on individual privacy, the exploration\nof adversarial perturbations as a potential countermeasure has gained traction.\nHowever, challenges arise in effectively deploying this approach against\nunauthorized facial recognition systems due to the effects of JPEG compression\non image distribution across the internet, which ultimately diminishes the\nefficacy of adversarial perturbations. Existing JPEG compression-resistant\ntechniques struggle to strike a balance between resistance, transferability,\nand attack potency. To address these limitations, we propose a novel solution\nreferred to as \\emph{low frequency adversarial perturbation} (LFAP). This\nmethod conditions the source model to leverage low-frequency characteristics\nthrough adversarial training. To further enhance the performance, we introduce\nan improved \\emph{low-mid frequency adversarial perturbation} (LMFAP) that\nincorporates mid-frequency components for an additive benefit. Our study\nencompasses a range of settings to replicate genuine application scenarios,\nincluding cross backbones, supervisory heads, training datasets, and testing\ndatasets. Moreover, we evaluated our approaches on a commercial black-box API,\n\\texttt{Face++}. The empirical results validate the cutting-edge performance\nachieved by our proposed solutions.\n", "date": "2022-06-19", "categories": []}
{"id": "2206.13964", "title": "Learning Gait Representation from Massive Unlabelled Walking Videos: A\n  Benchmark", "abstract": "  Gait depicts individuals' unique and distinguishing walking patterns and has\nbecome one of the most promising biometric features for human identification.\nAs a fine-grained recognition task, gait recognition is easily affected by many\nfactors and usually requires a large amount of completely annotated data that\nis costly and insatiable. This paper proposes a large-scale self-supervised\nbenchmark for gait recognition with contrastive learning, aiming to learn the\ngeneral gait representation from massive unlabelled walking videos for\npractical applications via offering informative walking priors and diverse\nreal-world variations. Specifically, we collect a large-scale unlabelled gait\ndataset GaitLU-1M consisting of 1.02M walking sequences and propose a\nconceptually simple yet empirically powerful baseline model GaitSSB.\nExperimentally, we evaluate the pre-trained model on four widely-used gait\nbenchmarks, CASIA-B, OU-MVLP, GREW and Gait3D with or without transfer\nlearning. The unsupervised results are comparable to or even better than the\nearly model-based and GEI-based methods. After transfer learning, our method\noutperforms existing methods by a large margin in most cases. Theoretically, we\ndiscuss the critical issues for gait-specific contrastive framework and present\nsome insights for further study. As far as we know, GaitLU-1M is the first\nlarge-scale unlabelled gait dataset, and GaitSSB is the first method that\nachieves remarkable unsupervised results on the aforementioned benchmarks. The\nsource code of GaitSSB will be integrated into OpenGait which is available at\nhttps://github.com/ShiqiYu/OpenGait.\n", "date": "2022-06-28", "categories": []}
{"id": "2209.04920", "title": "Vec2Face-v2: Unveil Human Faces from their Blackbox Features via\n  Attention-based Network in Face Recognition", "abstract": "  In this work, we investigate the problem of face reconstruction given a\nfacial feature representation extracted from a blackbox face recognition\nengine. Indeed, it is a very challenging problem in practice due to the\nlimitations of abstracted information from the engine. We, therefore, introduce\na new method named Attention-based Bijective Generative Adversarial Networks in\na Distillation framework (DAB-GAN) to synthesize the faces of a subject given\nhis/her extracted face recognition features. Given any unconstrained unseen\nfacial features of a subject, the DAB-GAN can reconstruct his/her facial images\nin high definition. The DAB-GAN method includes a novel attention-based\ngenerative structure with the newly defined Bijective Metrics Learning\napproach. The framework starts by introducing a bijective metric so that the\ndistance measurement and metric learning process can be directly adopted in the\nimage domain for an image reconstruction task. The information from the\nblackbox face recognition engine will be optimally exploited using the global\ndistillation process. Then an attention-based generator is presented for a\nhighly robust generator to synthesize realistic faces with ID preservation. We\nhave evaluated our method on the challenging face recognition databases, i.e.,\nCelebA, LFW, CFP-FP, CP-LFW, AgeDB, CA-LFW, and consistently achieved\nstate-of-the-art results. The advancement of DAB-GAN is also proven in both\nimage realism and ID preservation properties.\n", "date": "2022-09-11", "categories": []}
{"id": "2303.08665", "title": "Cross-resolution Face Recognition via Identity-Preserving Network and\n  Knowledge Distillation", "abstract": "  Cross-resolution face recognition has become a challenging problem for modern\ndeep face recognition systems. It aims at matching a low-resolution probe image\nwith high-resolution gallery images registered in a database. Existing methods\nmainly leverage prior information from high-resolution images by either\nreconstructing facial details with super-resolution techniques or learning a\nunified feature space. To address this challenge, this paper proposes a new\napproach that enforces the network to focus on the discriminative information\nstored in the low-frequency components of a low-resolution image. A\ncross-resolution knowledge distillation paradigm is first employed as the\nlearning framework. Then, an identity-preserving network, WaveResNet, and a\nwavelet similarity loss are designed to capture low-frequency details and boost\nperformance. Finally, an image degradation model is conceived to simulate more\nrealistic low-resolution training data. Consequently, extensive experimental\nresults show that the proposed method consistently outperforms the baseline\nmodel and other state-of-the-art methods across a variety of image resolutions.\n", "date": "2023-03-15", "categories": []}
{"id": "2309.00705", "title": "Indexing Irises by Intrinsic Dimension", "abstract": "  28,000+ high-quality iris images of 1350 distinct eyes from 650+ different\nindividuals from a relatively diverse university town population were\ncollected. A small defined unobstructed portion of the normalized iris image is\nselected as a key portion for quickly identifying an unknown individual when\nsubmitting an iris image to be matched to a database of enrolled irises of the\n1350 distinct eyes. The intrinsic dimension of a set of these key portions of\nthe 1350 enrolled irises is measured to be about four (4). This set is mapped\nto a four-dimensional intrinsic space by principal components analysis (PCA).\nWhen an iris image is presented to the iris database for identification, the\nsearch begins in the neighborhood of the location of its key portion in the 4D\nintrinsic space, typically finding a correct identifying match after comparison\nto only a few percent of the database.\n", "date": "2023-09-01", "categories": []}
{"id": "2309.00794", "title": "FastPoseGait: A Toolbox and Benchmark for Efficient Pose-based Gait\n  Recognition", "abstract": "  We present FastPoseGait, an open-source toolbox for pose-based gait\nrecognition based on PyTorch. Our toolbox supports a set of cutting-edge\npose-based gait recognition algorithms and a variety of related benchmarks.\nUnlike other pose-based projects that focus on a single algorithm, FastPoseGait\nintegrates several state-of-the-art (SOTA) algorithms under a unified\nframework, incorporating both the latest advancements and best practices to\nease the comparison of effectiveness and efficiency. In addition, to promote\nfuture research on pose-based gait recognition, we provide numerous pre-trained\nmodels and detailed benchmark results, which offer valuable insights and serve\nas a reference for further investigations. By leveraging the highly modular\nstructure and diverse methods offered by FastPoseGait, researchers can quickly\ndelve into pose-based gait recognition and promote development in the field. In\nthis paper, we outline various features of this toolbox, aiming that our\ntoolbox and benchmarks can further foster collaboration, facilitate\nreproducibility, and encourage the development of innovative algorithms for\npose-based gait recognition. FastPoseGait is available at\nhttps://github.com//BNU-IVC/FastPoseGait and is actively maintained. We will\ncontinue updating this report as we add new features.\n", "date": "2023-09-02", "categories": []}
{"id": "2309.01166", "title": "Spatial-temporal Vehicle Re-identification", "abstract": "  Vehicle re-identification (ReID) in a large-scale camera network is important\nin public safety, traffic control, and security. However, due to the appearance\nambiguities of vehicle, the previous appearance-based ReID methods often fail\nto track vehicle across multiple cameras. To overcome the challenge, we propose\na spatial-temporal vehicle ReID framework that estimates reliable camera\nnetwork topology based on the adaptive Parzen window method and optimally\ncombines the appearance and spatial-temporal similarities through the fusion\nnetwork. Based on the proposed methods, we performed superior performance on\nthe public dataset (VeRi776) by 99.64% of rank-1 accuracy. The experimental\nresults support that utilizing spatial and temporal information for ReID can\nleverage the accuracy of appearance-based methods and effectively deal with\nappearance ambiguities.\n", "date": "2023-09-03", "categories": []}
{"id": "2309.01353", "title": "Real-time pedestrian recognition on low computational resources", "abstract": "  Pedestrian recognition has successfully been applied to security, autonomous\ncars, Aerial photographs. For most applications, pedestrian recognition on\nsmall mobile devices is important. However, the limitations of the computing\nhardware make this a challenging task. In this work, we investigate real-time\npedestrian recognition on small physical-size computers with low computational\nresources for faster speed. This paper presents three methods that work on the\nsmall physical size CPUs system. First, we improved the Local Binary Pattern\n(LBP) features and Adaboost classifier. Second, we optimized the Histogram of\nOriented Gradients (HOG) and Support Vector Machine. Third, We implemented fast\nConvolutional Neural Networks (CNNs). The results demonstrate that the three\nmethods achieved real-time pedestrian recognition at an accuracy of more than\n95% and a speed of more than 5 fps on a small physical size computational\nplatform with a 1.8 GHz Intel i5 CPU. Our methods can be easily applied to\nsmall mobile devices with high compatibility and generality.\n", "date": "2023-09-04", "categories": []}
{"id": "2309.02404", "title": "Voice Morphing: Two Identities in One Voice", "abstract": "  In a biometric system, each biometric sample or template is typically\nassociated with a single identity. However, recent research has demonstrated\nthe possibility of generating \"morph\" biometric samples that can successfully\nmatch more than a single identity. Morph attacks are now recognized as a\npotential security threat to biometric systems. However, most morph attacks\nhave been studied on biometric modalities operating in the image domain, such\nas face, fingerprint, and iris. In this preliminary work, we introduce Voice\nIdentity Morphing (VIM) - a voice-based morph attack that can synthesize speech\nsamples that impersonate the voice characteristics of a pair of individuals.\nOur experiments evaluate the vulnerabilities of two popular speaker recognition\nsystems, ECAPA-TDNN and x-vector, to VIM, with a success rate (MMPMR) of over\n80% at a false match rate of 1% on the Librispeech dataset.\n", "date": "2023-09-05", "categories": []}
{"id": "2309.00549", "title": "Impact of Image Context for Single Deep Learning Face Morphing Attack\n  Detection", "abstract": "  The increase in security concerns due to technological advancements has led\nto the popularity of biometric approaches that utilize physiological or\nbehavioral characteristics for enhanced recognition. Face recognition systems\n(FRSs) have become prevalent, but they are still vulnerable to image\nmanipulation techniques such as face morphing attacks. This study investigates\nthe impact of the alignment settings of input images on deep learning face\nmorphing detection performance. We analyze the interconnections between the\nface contour and image context and suggest optimal alignment conditions for\nface morphing detection.\n", "date": "2023-09-01", "categories": []}
{"id": "2308.16486", "title": "Illumination Distillation Framework for Nighttime Person\n  Re-Identification and A New Benchmark", "abstract": "  Nighttime person Re-ID (person re-identification in the nighttime) is a very\nimportant and challenging task for visual surveillance but it has not been\nthoroughly investigated. Under the low illumination condition, the performance\nof person Re-ID methods usually sharply deteriorates. To address the low\nillumination challenge in nighttime person Re-ID, this paper proposes an\nIllumination Distillation Framework (IDF), which utilizes illumination\nenhancement and illumination distillation schemes to promote the learning of\nRe-ID models. Specifically, IDF consists of a master branch, an illumination\nenhancement branch, and an illumination distillation module. The master branch\nis used to extract the features from a nighttime image. The illumination\nenhancement branch first estimates an enhanced image from the nighttime image\nusing a nonlinear curve mapping method and then extracts the enhanced features.\nHowever, nighttime and enhanced features usually contain data noise due to\nunstable lighting conditions and enhancement failures. To fully exploit the\ncomplementary benefits of nighttime and enhanced features while suppressing\ndata noise, we propose an illumination distillation module. In particular, the\nillumination distillation module fuses the features from two branches through a\nbottleneck fusion model and then uses the fused features to guide the learning\nof both branches in a distillation manner. In addition, we build a real-world\nnighttime person Re-ID dataset, named Night600, which contains 600 identities\ncaptured from different viewpoints and nighttime illumination conditions under\ncomplex outdoor environments. Experimental results demonstrate that our IDF can\nachieve state-of-the-art performance on two nighttime person Re-ID datasets\n(i.e., Night600 and Knight ). We will release our code and dataset at\nhttps://github.com/Alexadlu/IDF.\n", "date": "2023-08-31", "categories": []}
{"id": "2308.15740", "title": "Beard Segmentation and Recognition Bias", "abstract": "  A person's facial hairstyle, such as presence and size of beard, can\nsignificantly impact face recognition accuracy. There are publicly-available\ndeep networks that achieve reasonable accuracy at binary attribute\nclassification, such as beard / no beard, but few if any that segment the\nfacial hair region. To investigate the effect of facial hair in a rigorous\nmanner, we first created a set of fine-grained facial hair annotations to train\na segmentation model and evaluate its accuracy across African-American and\nCaucasian face images. We then use our facial hair segmentations to categorize\nimage pairs according to the degree of difference or similarity in the facial\nhairstyle. We find that the False Match Rate (FMR) for image pairs with\ndifferent categories of facial hairstyle varies by a factor of over 10 for\nAfrican-American males and over 25 for Caucasian males. To reduce the bias\nacross image pairs with different facial hairstyles, we propose a scheme for\nadaptive thresholding based on facial hairstyle similarity. Evaluation on a\nsubject-disjoint set of images shows that adaptive similarity thresholding\nbased on facial hairstyles of the image pair reduces the ratio between the\nhighest and lowest FMR across facial hairstyle categories for African-American\nfrom 10.7 to 1.8 and for Caucasians from 25.9 to 1.3. Facial hair annotations\nand facial hair segmentation model will be publicly available.\n", "date": "2023-08-30", "categories": []}
{"id": "2308.14852", "title": "SynthDistill: Face Recognition with Knowledge Distillation from\n  Synthetic Data", "abstract": "  State-of-the-art face recognition networks are often computationally\nexpensive and cannot be used for mobile applications. Training lightweight face\nrecognition models also requires large identity-labeled datasets. Meanwhile,\nthere are privacy and ethical concerns with collecting and using large face\nrecognition datasets. While generating synthetic datasets for training face\nrecognition models is an alternative option, it is challenging to generate\nsynthetic data with sufficient intra-class variations. In addition, there is\nstill a considerable gap between the performance of models trained on real and\nsynthetic data. In this paper, we propose a new framework (named SynthDistill)\nto train lightweight face recognition models by distilling the knowledge of a\npretrained teacher face recognition model using synthetic data. We use a\npretrained face generator network to generate synthetic face images and use the\nsynthesized images to learn a lightweight student network. We use synthetic\nface images without identity labels, mitigating the problems in the intra-class\nvariation generation of synthetic datasets. Instead, we propose a novel dynamic\nsampling strategy from the intermediate latent space of the face generator\nnetwork to include new variations of the challenging images while further\nexploring new face images in the training batch. The results on five different\nface recognition datasets demonstrate the superiority of our lightweight model\ncompared to models trained on previous synthetic datasets, achieving a\nverification accuracy of 99.52% on the LFW dataset with a lightweight network.\nThe results also show that our proposed framework significantly reduces the gap\nbetween training with real and synthetic data. The source code for replicating\nthe experiments is publicly released.\n", "date": "2023-08-28", "categories": []}
{"id": "2308.15063", "title": "Learning Cross-modality Information Bottleneck Representation for\n  Heterogeneous Person Re-Identification", "abstract": "  Visible-Infrared person re-identification (VI-ReID) is an important and\nchallenging task in intelligent video surveillance. Existing methods mainly\nfocus on learning a shared feature space to reduce the modality discrepancy\nbetween visible and infrared modalities, which still leave two problems\nunderexplored: information redundancy and modality complementarity. To this\nend, properly eliminating the identity-irrelevant information as well as making\nup for the modality-specific information are critical and remains a challenging\nendeavor. To tackle the above problems, we present a novel mutual information\nand modality consensus network, namely CMInfoNet, to extract modality-invariant\nidentity features with the most representative information and reduce the\nredundancies. The key insight of our method is to find an optimal\nrepresentation to capture more identity-relevant information and compress the\nirrelevant parts by optimizing a mutual information bottleneck trade-off.\nBesides, we propose an automatically search strategy to find the most prominent\nparts that identify the pedestrians. To eliminate the cross- and intra-modality\nvariations, we also devise a modality consensus module to align the visible and\ninfrared modalities for task-specific guidance. Moreover, the global-local\nfeature representations can also be acquired for key parts discrimination.\nExperimental results on four benchmarks, i.e., SYSU-MM01, RegDB,\nOccluded-DukeMTMC, Occluded-REID, Partial-REID and Partial\\_iLIDS dataset, have\ndemonstrated the effectiveness of CMInfoNet.\n", "date": "2023-08-29", "categories": []}
{"id": "2308.15346", "title": "Enhancing Mobile Face Anti-Spoofing: A Robust Framework for Diverse\n  Attack Types under Screen Flash", "abstract": "  Face anti-spoofing (FAS) is crucial for securing face recognition systems.\nHowever, existing FAS methods with handcrafted binary or pixel-wise labels have\nlimitations due to diverse presentation attacks (PAs). In this paper, we\npropose an attack type robust face anti-spoofing framework under light flash,\ncalled ATR-FAS. Due to imaging differences caused by various attack types,\ntraditional FAS methods based on single binary classification network may\nresult in excessive intra-class distance of spoof faces, leading to a challenge\nof decision boundary learning. Therefore, we employed multiple networks to\nreconstruct multi-frame depth maps as auxiliary supervision, and each network\nexperts in one type of attack. A dual gate module (DGM) consisting of a type\ngate and a frame-attention gate is introduced, which perform attack type\nrecognition and multi-frame attention generation, respectively. The outputs of\nDGM are utilized as weight to mix the result of multiple expert networks. The\nmulti-experts mixture enables ATR-FAS to generate spoof-differentiated depth\nmaps, and stably detects spoof faces without being affected by different types\nof PAs. Moreover, we design a differential normalization procedure to convert\noriginal flash frames into differential frames. This simple but effective\nprocessing enhances the details in flash frames, aiding in the generation of\ndepth maps. To verify the effectiveness of our framework, we collected a\nlarge-scale dataset containing 12,660 live and spoof videos with diverse PAs\nunder dynamic flash from the smartphone screen. Extensive experiments\nillustrate that the proposed ATR-FAS significantly outperforms existing\nstate-of-the-art methods. The code and dataset will be available at\nhttps://github.com/Chaochao-Lin/ATR-FAS.\n", "date": "2023-08-29", "categories": []}
{"id": "2211.07383", "title": "Attacking Face Recognition with T-shirts: Database, Vulnerability\n  Assessment and Detection", "abstract": "  Face recognition systems are widely deployed for biometric authentication.\nDespite this, it is well-known that, without any safeguards, face recognition\nsystems are highly vulnerable to presentation attacks. In response to this\nsecurity issue, several promising methods for detecting presentation attacks\nhave been proposed which show high performance on existing benchmarks. However,\nan ongoing challenge is the generalization of presentation attack detection\nmethods to unseen and new attack types. To this end, we propose a new T-shirt\nFace Presentation Attack (TFPA) database of 1,608 T-shirt attacks using 100\nunique presentation attack instruments. In an extensive evaluation, we show\nthat this type of attack can compromise the security of face recognition\nsystems and that some state-of-the-art attack detection mechanisms trained on\npopular benchmarks fail to robustly generalize to the new attacks. Further, we\npropose three new methods for detecting T-shirt attack images, one which relies\non the statistical differences between depth maps of bona fide images and\nT-shirt attacks, an anomaly detection approach trained on features only\nextracted from bona fide RGB images, and a fusion approach which achieves\ncompetitive detection performance.\n", "date": "2022-11-14", "categories": []}
{"id": "2308.13340", "title": "TriGait: Aligning and Fusing Skeleton and Silhouette Gait Data via a\n  Tri-Branch Network", "abstract": "  Gait recognition is a promising biometric technology for identification due\nto its non-invasiveness and long-distance. However, external variations such as\nclothing changes and viewpoint differences pose significant challenges to gait\nrecognition. Silhouette-based methods preserve body shape but neglect internal\nstructure information, while skeleton-based methods preserve structure\ninformation but omit appearance. To fully exploit the complementary nature of\nthe two modalities, a novel triple branch gait recognition framework, TriGait,\nis proposed in this paper. It effectively integrates features from the skeleton\nand silhouette data in a hybrid fusion manner, including a two-stream network\nto extract static and motion features from appearance, a simple yet effective\nmodule named JSA-TC to capture dependencies between all joints, and a third\nbranch for cross-modal learning by aligning and fusing low-level features of\ntwo modalities. Experimental results demonstrate the superiority and\neffectiveness of TriGait for gait recognition. The proposed method achieves a\nmean rank-1 accuracy of 96.0% over all conditions on CASIA-B dataset and 94.3%\naccuracy for CL, significantly outperforming all the state-of-the-art methods.\nThe source code will be available at https://github.com/feng-xueling/TriGait/.\n", "date": "2023-08-25", "categories": []}
{"id": "2304.08134", "title": "Tackling Face Verification Edge Cases: In-Depth Analysis and\n  Human-Machine Fusion Approach", "abstract": "  Nowadays, face recognition systems surpass human performance on several\ndatasets. However, there are still edge cases that the machine can't correctly\nclassify. This paper investigates the effect of a combination of machine and\nhuman operators in the face verification task. First, we look closer at the\nedge cases for several state-of-the-art models to discover common datasets'\nchallenging settings. Then, we conduct a study with 60 participants on these\nselected tasks with humans and provide an extensive analysis. Finally, we\ndemonstrate that combining machine and human decisions can further improve the\nperformance of state-of-the-art face verification systems on various benchmark\ndatasets. Code and data are publicly available on GitHub.\n", "date": "2023-04-17", "categories": []}
{"id": "2304.09818", "title": "What Should Be Balanced in a \"Balanced\" Face Recognition Dataset?", "abstract": "  The issue of demographic disparities in face recognition accuracy has\nattracted increasing attention in recent years. Various face image datasets\nhave been proposed as 'fair' or 'balanced' to assess the accuracy of face\nrecognition algorithms across demographics. These datasets typically balance\nthe number of identities and images across demographics. It is important to\nnote that the number of identities and images in an evaluation dataset are {\\em\nnot} driving factors for 1-to-1 face matching accuracy. Moreover, balancing the\nnumber of identities and images does not ensure balance in other factors known\nto impact accuracy, such as head pose, brightness, and image quality. We\ndemonstrate these issues using several recently proposed datasets. To improve\nthe ability to perform less biased evaluations, we propose a bias-aware toolkit\nthat facilitates creation of cross-demographic evaluation datasets balanced on\nfactors mentioned in this paper.\n", "date": "2023-04-17", "categories": []}
{"id": "2308.07445", "title": "Open-set Face Recognition using Ensembles trained on Clustered Data", "abstract": "  Open-set face recognition describes a scenario where unknown subjects, unseen\nduring the training stage, appear on test time. Not only it requires methods\nthat accurately identify individuals of interest, but also demands approaches\nthat effectively deal with unfamiliar faces. This work details a scalable\nopen-set face identification approach to galleries composed of hundreds and\nthousands of subjects. It is composed of clustering and an ensemble of binary\nlearning algorithms that estimates when query face samples belong to the face\ngallery and then retrieves their correct identity. The approach selects the\nmost suitable gallery subjects and uses the ensemble to improve prediction\nperformance. We carry out experiments on well-known LFW and YTF benchmarks.\nResults show that competitive performance can be achieved even when targeting\nscalability.\n", "date": "2023-08-14", "categories": []}
{"id": "2308.12371", "title": "Open-set Face Recognition with Neural Ensemble, Maximal Entropy Loss and\n  Feature Augmentation", "abstract": "  Open-set face recognition refers to a scenario in which biometric systems\nhave incomplete knowledge of all existing subjects. Therefore, they are\nexpected to prevent face samples of unregistered subjects from being identified\nas previously enrolled identities. This watchlist context adds an arduous\nrequirement that calls for the dismissal of irrelevant faces by focusing mainly\non subjects of interest. As a response, this work introduces a novel method\nthat associates an ensemble of compact neural networks with a margin-based cost\nfunction that explores additional samples. Supplementary negative samples can\nbe obtained from external databases or synthetically built at the\nrepresentation level in training time with a new mix-up feature augmentation\napproach. Deep neural networks pre-trained on large face datasets serve as the\npreliminary feature extraction module. We carry out experiments on well-known\nLFW and IJB-C datasets where results show that the approach is able to boost\nclosed and open-set identification rates.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.11732", "title": "(Un)fair Exposure in Deep Face Rankings at a Distance", "abstract": "  Law enforcement regularly faces the challenge of ranking suspects from their\nfacial images. Deep face models aid this process but frequently introduce\nbiases that disproportionately affect certain demographic segments. While bias\ninvestigation is common in domains like job candidate ranking, the field of\nforensic face rankings remains underexplored. In this paper, we propose a novel\nexperimental framework, encompassing six state-of-the-art face encoders and two\npublic data sets, designed to scrutinize the extent to which demographic groups\nsuffer from biases in exposure in the context of forensic face rankings.\nThrough comprehensive experiments that cover both re-identification and\nidentification tasks, we show that exposure biases within this domain are far\nfrom being countered, demanding attention towards establishing ad-hoc policies\nand corrective measures. The source code is available at\nhttps://github.com/atzoriandrea/ijcb2023-unfair-face-rankings\n", "date": "2023-08-22", "categories": []}
{"id": "2308.11757", "title": "Weakly Supervised Face and Whole Body Recognition in Turbulent\n  Environments", "abstract": "  Face and person recognition have recently achieved remarkable success under\nchallenging scenarios, such as off-pose and cross-spectrum matching. However,\nlong-range recognition systems are often hindered by atmospheric turbulence,\nleading to spatially and temporally varying distortions in the image. Current\nsolutions rely on generative models to reconstruct a turbulent-free image, but\noften preserve photo-realism instead of discriminative features that are\nessential for recognition. This can be attributed to the lack of large-scale\ndatasets of turbulent and pristine paired images, necessary for optimal\nreconstruction. To address this issue, we propose a new weakly supervised\nframework that employs a parameter-efficient self-attention module to generate\ndomain agnostic representations, aligning turbulent and pristine images into a\ncommon subspace. Additionally, we introduce a new tilt map estimator that\npredicts geometric distortions observed in turbulent images. This estimate is\nused to re-rank gallery matches, resulting in up to 13.86\\% improvement in\nrank-1 accuracy. Our method does not require synthesizing turbulent-free images\nor ground-truth paired images, and requires significantly fewer annotated\nsamples, enabling more practical and rapid utility of increasingly large\ndatasets. We analyze our framework using two datasets -- Long-Range Face\nIdentification Dataset (LRFID) and BRIAR Government Collection 1 (BGC1) --\nachieving enhanced discriminability under varying turbulence and standoff\ndistance.\n", "date": "2023-08-22", "categories": []}
{"id": "2308.11840", "title": "Compressed Models Decompress Race Biases: What Quantized Models Forget\n  for Fair Face Recognition", "abstract": "  With the ever-growing complexity of deep learning models for face\nrecognition, it becomes hard to deploy these systems in real life. Researchers\nhave two options: 1) use smaller models; 2) compress their current models.\nSince the usage of smaller models might lead to concerning biases, compression\ngains relevance. However, compressing might be also responsible for an increase\nin the bias of the final model. We investigate the overall performance, the\nperformance on each ethnicity subgroup and the racial bias of a\nState-of-the-Art quantization approach when used with synthetic and real data.\nThis analysis provides a few more details on potential benefits of performing\nquantization with synthetic data, for instance, the reduction of biases on the\nmajority of test scenarios. We tested five distinct architectures and three\ndifferent training datasets. The models were evaluated on a fourth dataset\nwhich was collected to infer and compare the performance of face recognition\nmodels on different ethnicity.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.11900", "title": "HashReID: Dynamic Network with Binary Codes for Efficient Person\n  Re-identification", "abstract": "  Biometric applications, such as person re-identification (ReID), are often\ndeployed on energy constrained devices. While recent ReID methods prioritize\nhigh retrieval performance, they often come with large computational costs and\nhigh search time, rendering them less practical in real-world settings. In this\nwork, we propose an input-adaptive network with multiple exit blocks, that can\nterminate computation early if the retrieval is straightforward or noisy,\nsaving a lot of computation. To assess the complexity of the input, we\nintroduce a temporal-based classifier driven by a new training strategy.\nFurthermore, we adopt a binary hash code generation approach instead of relying\non continuous-valued features, which significantly improves the search process\nby a factor of 20. To ensure similarity preservation, we utilize a new ranking\nregularizer that bridges the gap between continuous and binary features.\nExtensive analysis of our proposed method is conducted on three datasets:\nMarket1501, MSMT17 (Multi-Scene Multi-Time), and the BGC1 (BRIAR Government\nCollection). Using our approach, more than 70% of the samples with compact hash\ncodes exit early on the Market1501 dataset, saving 80% of the networks\ncomputational cost and improving over other hash-based methods by 60%. These\nresults demonstrate a significant improvement over dynamic networks and\nshowcase comparable accuracy performance to conventional ReID methods. Code\nwill be made available.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.11901", "title": "Camera-Driven Representation Learning for Unsupervised Domain Adaptive\n  Person Re-identification", "abstract": "  We present a novel unsupervised domain adaption method for person\nre-identification (reID) that generalizes a model trained on a labeled source\ndomain to an unlabeled target domain. We introduce a camera-driven curriculum\nlearning (CaCL) framework that leverages camera labels of person images to\ntransfer knowledge from source to target domains progressively. To this end, we\ndivide target domain dataset into multiple subsets based on the camera labels,\nand initially train our model with a single subset (i.e., images captured by a\nsingle camera). We then gradually exploit more subsets for training, according\nto a curriculum sequence obtained with a camera-driven scheduling rule. The\nscheduler considers maximum mean discrepancies (MMD) between each subset and\nthe source domain dataset, such that the subset closer to the source domain is\nexploited earlier within the curriculum. For each curriculum sequence, we\ngenerate pseudo labels of person images in a target domain to train a reID\nmodel in a supervised way. We have observed that the pseudo labels are highly\nbiased toward cameras, suggesting that person images obtained from the same\ncamera are likely to have the same pseudo labels, even for different IDs. To\naddress the camera bias problem, we also introduce a camera-diversity (CD) loss\nencouraging person images of the same pseudo label, but captured across various\ncameras, to involve more for discriminative feature learning, providing person\nrepresentations robust to inter-camera variations. Experimental results on\nstandard benchmarks, including real-to-real and synthetic-to-real scenarios,\ndemonstrate the effectiveness of our framework.\n", "date": "2023-08-23", "categories": []}
{"id": "2308.11442", "title": "SDeMorph: Towards Better Facial De-morphing from Single Morph", "abstract": "  Face Recognition Systems (FRS) are vulnerable to morph attacks. A face morph\nis created by combining multiple identities with the intention to fool FRS and\nmaking it match the morph with multiple identities. Current Morph Attack\nDetection (MAD) can detect the morph but are unable to recover the identities\nused to create the morph with satisfactory outcomes. Existing work in\nde-morphing is mostly reference-based, i.e. they require the availability of\none identity to recover the other. Sudipta et al. \\cite{ref9} proposed a\nreference-free de-morphing technique but the visual realism of outputs produced\nwere feeble. In this work, we propose SDeMorph (Stably Diffused De-morpher), a\nnovel de-morphing method that is reference-free and recovers the identities of\nbona fides. Our method produces feature-rich outputs that are of significantly\nhigh quality in terms of definition and facial fidelity. Our method utilizes\nDenoising Diffusion Probabilistic Models (DDPM) by destroying the input morphed\nsignal and then reconstructing it back using a branched-UNet. Experiments on\nASML, FRLL-FaceMorph, FRLL-MorDIFF, and SMDD datasets support the effectiveness\nof the proposed method.\n", "date": "2023-08-22", "categories": []}
{"id": "2308.10461", "title": "Privacy-Preserving Face Recognition Using Random Frequency Components", "abstract": "  The ubiquitous use of face recognition has sparked increasing privacy\nconcerns, as unauthorized access to sensitive face images could compromise the\ninformation of individuals. This paper presents an in-depth study of the\nprivacy protection of face images' visual information and against recovery.\nDrawing on the perceptual disparity between humans and models, we propose to\nconceal visual information by pruning human-perceivable low-frequency\ncomponents. For impeding recovery, we first elucidate the seeming paradox\nbetween reducing model-exploitable information and retaining high recognition\naccuracy. Based on recent theoretical insights and our observation on model\nattention, we propose a solution to the dilemma, by advocating for the training\nand inference of recognition models on randomly selected frequency components.\nWe distill our findings into a novel privacy-preserving face recognition\nmethod, PartialFace. Extensive experiments demonstrate that PartialFace\neffectively balances privacy protection goals and recognition accuracy. Code is\navailable at: https://github.com/Tencent/TFace.\n", "date": "2023-08-21", "categories": []}
{"id": "2308.10716", "title": "Color Prompting for Data-Free Continual Unsupervised Domain Adaptive\n  Person Re-Identification", "abstract": "  Unsupervised domain adaptive person re-identification (Re-ID) methods\nalleviate the burden of data annotation through generating pseudo supervision\nmessages. However, real-world Re-ID systems, with continuously accumulating\ndata streams, simultaneously demand more robust adaptation and anti-forgetting\ncapabilities. Methods based on image rehearsal addresses the forgetting issue\nwith limited extra storage but carry the risk of privacy leakage. In this work,\nwe propose a Color Prompting (CoP) method for data-free continual unsupervised\ndomain adaptive person Re-ID. Specifically, we employ a light-weighted prompter\nnetwork to fit the color distribution of the current task together with Re-ID\ntraining. Then for the incoming new tasks, the learned color distribution\nserves as color style transfer guidance to transfer the images into past\nstyles. CoP achieves accurate color style recovery for past tasks with adequate\ndata diversity, leading to superior anti-forgetting effects compared with image\nrehearsal methods. Moreover, CoP demonstrates strong generalization performance\nfor fast adaptation into new domains, given only a small amount of unlabeled\nimages. Extensive experiments demonstrate that after the continual training\npipeline the proposed CoP achieves 6.7% and 8.1% average rank-1 improvements\nover the replay method on seen and unseen domains, respectively. The source\ncode for this work is publicly available in\nhttps://github.com/vimar-gu/ColorPromptReID.\n", "date": "2023-08-21", "categories": []}
{"id": "2308.10717", "title": "Rethinking Person Re-identification from a Projection-on-Prototypes\n  Perspective", "abstract": "  Person Re-IDentification (Re-ID) as a retrieval task, has achieved tremendous\ndevelopment over the past decade. Existing state-of-the-art methods follow an\nanalogous framework to first extract features from the input images and then\ncategorize them with a classifier. However, since there is no identity overlap\nbetween training and testing sets, the classifier is often discarded during\ninference. Only the extracted features are used for person retrieval via\ndistance metrics. In this paper, we rethink the role of the classifier in\nperson Re-ID, and advocate a new perspective to conceive the classifier as a\nprojection from image features to class prototypes. These prototypes are\nexactly the learned parameters of the classifier. In this light, we describe\nthe identity of input images as similarities to all prototypes, which are then\nutilized as more discriminative features to perform person Re-ID. We thereby\npropose a new baseline ProNet, which innovatively reserves the function of the\nclassifier at the inference stage. To facilitate the learning of class\nprototypes, both triplet loss and identity classification loss are applied to\nfeatures that undergo the projection by the classifier. An improved version of\nProNet++ is presented by further incorporating multi-granularity designs.\nExperiments on four benchmarks demonstrate that our proposed ProNet is simple\nyet effective, and significantly beats previous baselines. ProNet++ also\nachieves competitive or even better results than transformer-based competitors.\n", "date": "2023-08-21", "categories": []}
{"id": "2308.08887", "title": "Identity-Seeking Self-Supervised Representation Learning for\n  Generalizable Person Re-identification", "abstract": "  This paper aims to learn a domain-generalizable (DG) person re-identification\n(ReID) representation from large-scale videos \\textbf{without any annotation}.\nPrior DG ReID methods employ limited labeled data for training due to the high\ncost of annotation, which restricts further advances. To overcome the barriers\nof data and annotation, we propose to utilize large-scale unsupervised data for\ntraining. The key issue lies in how to mine identity information. To this end,\nwe propose an Identity-seeking Self-supervised Representation learning (ISR)\nmethod. ISR constructs positive pairs from inter-frame images by modeling the\ninstance association as a maximum-weight bipartite matching problem. A\nreliability-guided contrastive loss is further presented to suppress the\nadverse impact of noisy positive pairs, ensuring that reliable positive pairs\ndominate the learning process. The training cost of ISR scales approximately\nlinearly with the data size, making it feasible to utilize large-scale data for\ntraining. The learned representation exhibits superior generalization ability.\n\\textbf{Without human annotation and fine-tuning, ISR achieves 87.0\\% Rank-1 on\nMarket-1501 and 56.4\\% Rank-1 on MSMT17}, outperforming the best supervised\ndomain-generalizable method by 5.0\\% and 19.5\\%, respectively. In the\npre-training$\\rightarrow$fine-tuning scenario, ISR achieves state-of-the-art\nperformance, with 88.4\\% Rank-1 on MSMT17. The code is at\n\\url{https://github.com/dcp15/ISR_ICCV2023_Oral}.\n", "date": "2023-08-17", "categories": []}
{"id": "2308.09230", "title": "CCFace: Classification Consistency for Low-Resolution Face Recognition", "abstract": "  In recent years, deep face recognition methods have demonstrated impressive\nresults on in-the-wild datasets. However, these methods have shown a\nsignificant decline in performance when applied to real-world low-resolution\nbenchmarks like TinyFace or SCFace. To address this challenge, we propose a\nnovel classification consistency knowledge distillation approach that transfers\nthe learned classifier from a high-resolution model to a low-resolution\nnetwork. This approach helps in finding discriminative representations for\nlow-resolution instances. To further improve the performance, we designed a\nknowledge distillation loss using the adaptive angular penalty inspired by the\nsuccess of the popular angular margin loss function. The adaptive penalty\nreduces overfitting on low-resolution samples and alleviates the convergence\nissue of the model integrated with data augmentation. Additionally, we utilize\nan asymmetric cross-resolution learning approach based on the state-of-the-art\nsemi-supervised representation learning paradigm to improve discriminability on\nlow-resolution instances and prevent them from forming a cluster. Our proposed\nmethod outperforms state-of-the-art approaches on low-resolution benchmarks,\nwith a three percent improvement on TinyFace while maintaining performance on\nhigh-resolution benchmarks.\n", "date": "2023-08-18", "categories": []}
{"id": "2308.09234", "title": "Deep Boosting Multi-Modal Ensemble Face Recognition with Sample-Level\n  Weighting", "abstract": "  Deep convolutional neural networks have achieved remarkable success in face\nrecognition (FR), partly due to the abundant data availability. However, the\ncurrent training benchmarks exhibit an imbalanced quality distribution; most\nimages are of high quality. This poses issues for generalization on hard\nsamples since they are underrepresented during training. In this work, we\nemploy the multi-model boosting technique to deal with this issue. Inspired by\nthe well-known AdaBoost, we propose a sample-level weighting approach to\nincorporate the importance of different samples into the FR loss. Individual\nmodels of the proposed framework are experts at distinct levels of sample\nhardness. Therefore, the combination of models leads to a robust feature\nextractor without losing the discriminability on the easy samples. Also, for\nincorporating the sample hardness into the training criterion, we analytically\nshow the effect of sample mining on the important aspects of current angular\nmargin loss functions, i.e., margin and scale. The proposed method shows\nsuperior performance in comparison with the state-of-the-art algorithms in\nextensive experiments on the CFP-FP, LFW, CPLFW, CALFW, AgeDB, TinyFace, IJB-B,\nand IJB-C evaluation datasets.\n", "date": "2023-08-18", "categories": []}
{"id": "2308.07967", "title": "Boosting Cross-Quality Face Verification using Blind Face Restoration", "abstract": "  In recent years, various Blind Face Restoration (BFR) techniques were\ndeveloped. These techniques transform low quality faces suffering from multiple\ndegradations to more realistic and natural face images with high perceptual\nquality. However, it is crucial for the task of face verification to not only\nenhance the perceptual quality of the low quality images but also to improve\nthe biometric-utility face quality metrics. Furthermore, preserving the\nvaluable identity information is of great importance. In this paper, we\ninvestigate the impact of applying three state-of-the-art blind face\nrestoration techniques namely, GFP-GAN, GPEN and SGPN on the performance of\nface verification system under very challenging environment characterized by\nvery low quality images. Extensive experimental results on the recently\nproposed cross-quality LFW database using three state-of-the-art deep face\nrecognition models demonstrate the effectiveness of GFP-GAN in boosting\nsignificantly the face verification accuracy.\n", "date": "2023-08-15", "categories": []}
{"id": "2303.08597", "title": "Aerial-Ground Person Re-ID", "abstract": "  Person re-ID matches persons across multiple non-overlapping cameras. Despite\nthe increasing deployment of airborne platforms in surveillance, current\nexisting person re-ID benchmarks' focus is on ground-ground matching and very\nlimited efforts on aerial-aerial matching. We propose a new benchmark dataset -\nAG-ReID, which performs person re-ID matching in a new setting: across aerial\nand ground cameras. Our dataset contains 21,983 images of 388 identities and 15\nsoft attributes for each identity. The data was collected by a UAV flying at\naltitudes between 15 to 45 meters and a ground-based CCTV camera on a\nuniversity campus. Our dataset presents a novel elevated-viewpoint challenge\nfor person re-ID due to the significant difference in person appearance across\nthese cameras. We propose an explainable algorithm to guide the person re-ID\nmodel's training with soft attributes to address this challenge. Experiments\ndemonstrate the efficacy of our method on the aerial-ground person re-ID task.\nThe dataset will be published and the baseline codes will be open-sourced at\nhttps://github.com/huynguyen792/AG-ReID to facilitate research in this area.\n", "date": "2023-03-15", "categories": []}
{"id": "2308.07024", "title": "PGT-Net: Progressive Guided Multi-task Neural Network for Small-area Wet\n  Fingerprint Denoising and Recognition", "abstract": "  Fingerprint recognition on mobile devices is an important method for identity\nverification. However, real fingerprints usually contain sweat and moisture\nwhich leads to poor recognition performance. In addition, for rolling out\nslimmer and thinner phones, technology companies reduce the size of recognition\nsensors by embedding them with the power button. Therefore, the limited size of\nfingerprint data also increases the difficulty of recognition. Denoising the\nsmall-area wet fingerprint images to clean ones becomes crucial to improve\nrecognition performance. In this paper, we propose an end-to-end trainable\nprogressive guided multi-task neural network (PGT-Net). The PGT-Net includes a\nshared stage and specific multi-task stages, enabling the network to train\nbinary and non-binary fingerprints sequentially. The binary information is\nregarded as guidance for output enhancement which is enriched with the ridge\nand valley details. Moreover, a novel residual scaling mechanism is introduced\nto stabilize the training process. Experiment results on the FW9395 and\nFT-lightnoised dataset provided by FocalTech shows that PGT-Net has promising\nperformance on the wet-fingerprint denoising and significantly improves the\nfingerprint recognition rate (FRR). On the FT-lightnoised dataset, the FRR of\nfingerprint recognition can be declined from 17.75% to 4.47%. On the FW9395\ndataset, the FRR of fingerprint recognition can be declined from 9.45% to\n1.09%.\n", "date": "2023-08-14", "categories": []}
{"id": "2308.07243", "title": "AAFACE: Attribute-aware Attentional Network for Face Recognition", "abstract": "  In this paper, we present a new multi-branch neural network that\nsimultaneously performs soft biometric (SB) prediction as an auxiliary modality\nand face recognition (FR) as the main task. Our proposed network named AAFace\nutilizes SB attributes to enhance the discriminative ability of FR\nrepresentation. To achieve this goal, we propose an attribute-aware attentional\nintegration (AAI) module to perform weighted integration of FR with SB feature\nmaps. Our proposed AAI module is not only fully context-aware but also capable\nof learning complex relationships between input features by means of the\nsequential multi-scale channel and spatial sub-modules. Experimental results\nverify the superiority of our proposed network compared with the\nstate-of-the-art (SoTA) SB prediction and FR methods.\n", "date": "2023-08-14", "categories": []}
{"id": "2307.03903", "title": "Adversarial Self-Attack Defense and Spatial-Temporal Relation Mining for\n  Visible-Infrared Video Person Re-Identification", "abstract": "  In visible-infrared video person re-identification (re-ID), extracting\nfeatures not affected by complex scenes (such as modality, camera views,\npedestrian pose, background, etc.) changes, and mining and utilizing motion\ninformation are the keys to solving cross-modal pedestrian identity matching.\nTo this end, the paper proposes a new visible-infrared video person re-ID\nmethod from a novel perspective, i.e., adversarial self-attack defense and\nspatial-temporal relation mining. In this work, the changes of views, posture,\nbackground and modal discrepancy are considered as the main factors that cause\nthe perturbations of person identity features. Such interference information\ncontained in the training samples is used as an adversarial perturbation. It\nperforms adversarial attacks on the re-ID model during the training to make the\nmodel more robust to these unfavorable factors. The attack from the adversarial\nperturbation is introduced by activating the interference information contained\nin the input samples without generating adversarial samples, and it can be thus\ncalled adversarial self-attack. This design allows adversarial attack and\ndefense to be integrated into one framework. This paper further proposes a\nspatial-temporal information-guided feature representation network to use the\ninformation in video sequences. The network cannot only extract the information\ncontained in the video-frame sequences but also use the relation of the local\ninformation in space to guide the network to extract more robust features. The\nproposed method exhibits compelling performance on large-scale cross-modality\nvideo datasets. The source code of the proposed method will be released at\nhttps://github.com/lhf12278/xxx.\n", "date": "2023-07-08", "categories": []}
{"id": "2308.04995", "title": "IDiff-Face: Synthetic-based Face Recognition through Fizzy\n  Identity-Conditioned Diffusion Models", "abstract": "  The availability of large-scale authentic face databases has been crucial to\nthe significant advances made in face recognition research over the past\ndecade. However, legal and ethical concerns led to the recent retraction of\nmany of these databases by their creators, raising questions about the\ncontinuity of future face recognition research without one of its key\nresources. Synthetic datasets have emerged as a promising alternative to\nprivacy-sensitive authentic data for face recognition development. However,\nrecent synthetic datasets that are used to train face recognition models suffer\neither from limitations in intra-class diversity or cross-class (identity)\ndiscrimination, leading to less optimal accuracies, far away from the\naccuracies achieved by models trained on authentic data. This paper targets\nthis issue by proposing IDiff-Face, a novel approach based on conditional\nlatent diffusion models for synthetic identity generation with realistic\nidentity variations for face recognition training. Through extensive\nevaluations, our proposed synthetic-based face recognition approach pushed the\nlimits of state-of-the-art performances, achieving, for example, 98.00%\naccuracy on the Labeled Faces in the Wild (LFW) benchmark, far ahead from the\nrecent synthetic-based face recognition solutions with 95.40% and bridging the\ngap to authentic-based face recognition with 99.82% accuracy.\n", "date": "2023-08-09", "categories": []}
{"id": "2308.05441", "title": "Benchmarking Algorithmic Bias in Face Recognition: An Experimental\n  Approach Using Synthetic Faces and Human Evaluation", "abstract": "  We propose an experimental method for measuring bias in face recognition\nsystems. Existing methods to measure bias depend on benchmark datasets that are\ncollected in the wild and annotated for protected (e.g., race, gender) and\nnon-protected (e.g., pose, lighting) attributes. Such observational datasets\nonly permit correlational conclusions, e.g., \"Algorithm A's accuracy is\ndifferent on female and male faces in dataset X.\". By contrast, experimental\nmethods manipulate attributes individually and thus permit causal conclusions,\ne.g., \"Algorithm A's accuracy is affected by gender and skin color.\"\n  Our method is based on generating synthetic faces using a neural face\ngenerator, where each attribute of interest is modified independently while\nleaving all other attributes constant. Human observers crucially provide the\nground truth on perceptual identity similarity between synthetic image pairs.\nWe validate our method quantitatively by evaluating race and gender biases of\nthree research-grade face recognition models. Our synthetic pipeline reveals\nthat for these algorithms, accuracy is lower for Black and East Asian\npopulation subgroups. Our method can also quantify how perceptual changes in\nattributes affect face identity distances reported by these models. Our large\nsynthetic dataset, consisting of 48,000 synthetic face image pairs (10,200\nunique synthetic faces) and 555,000 human annotations (individual attributes\nand pairwise identity comparisons) is available to researchers in this\nimportant area.\n", "date": "2023-08-10", "categories": []}
{"id": "2304.07916", "title": "GaitRef: Gait Recognition with Refined Sequential Skeletons", "abstract": "  Identifying humans with their walking sequences, known as gait recognition,\nis a useful biometric understanding task as it can be observed from a long\ndistance and does not require cooperation from the subject. Two common\nmodalities used for representing the walking sequence of a person are\nsilhouettes and joint skeletons. Silhouette sequences, which record the\nboundary of the walking person in each frame, may suffer from the variant\nappearances from carried-on objects and clothes of the person. Framewise joint\ndetections are noisy and introduce some jitters that are not consistent with\nsequential detections. In this paper, we combine the silhouettes and skeletons\nand refine the framewise joint predictions for gait recognition. With temporal\ninformation from the silhouette sequences, we show that the refined skeletons\ncan improve gait recognition performance without extra annotations. We compare\nour methods on four public datasets, CASIA-B, OUMVLP, Gait3D and GREW, and show\nstate-of-the-art performance.\n", "date": "2023-04-16", "categories": []}
{"id": "2307.14016", "title": "RPG-Palm: Realistic Pseudo-data Generation for Palmprint Recognition", "abstract": "  Palmprint recently shows great potential in recognition applications as it is\na privacy-friendly and stable biometric. However, the lack of large-scale\npublic palmprint datasets limits further research and development of palmprint\nrecognition. In this paper, we propose a novel realistic pseudo-palmprint\ngeneration (RPG) model to synthesize palmprints with massive identities. We\nfirst introduce a conditional modulation generator to improve the intra-class\ndiversity. Then an identity-aware loss is proposed to ensure identity\nconsistency against unpaired training. We further improve the B\\'ezier palm\ncreases generation strategy to guarantee identity independence. Extensive\nexperimental results demonstrate that synthetic pretraining significantly\nboosts the recognition model performance. For example, our model improves the\nstate-of-the-art B\\'ezierPalm by more than $5\\%$ and $14\\%$ in terms of\nTAR@FAR=1e-6 under the $1:1$ and $1:3$ Open-set protocol. When accessing only\n$10\\%$ of the real training data, our method still outperforms ArcFace with\n$100\\%$ real training data, indicating that we are closer to real-data-free\npalmprint recognition.\n", "date": "2023-07-26", "categories": []}
{"id": "2308.04168", "title": "EFaR 2023: Efficient Face Recognition Competition", "abstract": "  This paper presents the summary of the Efficient Face Recognition Competition\n(EFaR) held at the 2023 International Joint Conference on Biometrics (IJCB\n2023). The competition received 17 submissions from 6 different teams. To drive\nfurther development of efficient face recognition models, the submitted\nsolutions are ranked based on a weighted score of the achieved verification\naccuracies on a diverse set of benchmarks, as well as the deployability given\nby the number of floating-point operations and model size. The evaluation of\nsubmissions is extended to bias, cross-quality, and large-scale recognition\nbenchmarks. Overall, the paper gives an overview of the achieved performance\nvalues of the submitted solutions as well as a diverse set of baselines. The\nsubmitted solutions use small, efficient network architectures to reduce the\ncomputational cost, some solutions apply model quantization. An outlook on\npossible techniques that are underrepresented in current solutions is given as\nwell.\n", "date": "2023-08-08", "categories": []}
{"id": "2308.04224", "title": "Will your Doorbell Camera still recognize you as you grow old", "abstract": "  Robust authentication for low-power consumer devices such as doorbell cameras\nposes a valuable and unique challenge. This work explores the effect of age and\naging on the performance of facial authentication methods. Two public age\ndatasets, AgeDB and Morph-II have been used as baselines in this work. A\nphoto-realistic age transformation method has been employed to augment a set of\nhigh-quality facial images with various age effects. Then the effect of these\nsynthetic aging data on the high-performance deep-learning-based face\nrecognition model is quantified by using various metrics including Receiver\nOperating Characteristic (ROC) curves and match score distributions.\nExperimental results demonstrate that long-term age effects are still a\nsignificant challenge for the state-of-the-art facial authentication method.\n", "date": "2023-08-08", "categories": []}
{"id": "2308.03189", "title": "Understanding Biometric Entropy and Iris Capacity: Avoiding Identity\n  Collisions on National Scales", "abstract": "  The numbers of persons who can be enrolled by their iris patterns with no\nidentity collisions is studied in relation to the biometric entropy extracted,\nand the decision operating threshold. The population size at which identity\ncollision becomes likelier than not, given those variables, defines iris\n\"capacity.\" The general solution to this combinatorial problem is derived, in\nanalogy with the well-known \"birthday problem.\" Its application to unique\nbiometric identification on national population scales is shown, referencing\nempirical data from US NIST (National Institute of Standards and Technology)\ntrials involving 1.2 trillion (1.2 x 10^(12) ) iris comparisons. The entropy of\na given person's two iris patterns suffices for global identity uniqueness.\n", "date": "2023-08-06", "categories": []}
{"id": "2308.03703", "title": "Video-based Person Re-identification with Long Short-Term Representation\n  Learning", "abstract": "  Video-based person Re-Identification (V-ReID) aims to retrieve specific\npersons from raw videos captured by non-overlapped cameras. As a fundamental\ntask, it spreads many multimedia and computer vision applications. However, due\nto the variations of persons and scenes, there are still many obstacles that\nmust be overcome for high performance. In this work, we notice that both the\nlong-term and short-term information of persons are important for robust video\nrepresentations. Thus, we propose a novel deep learning framework named Long\nShort-Term Representation Learning (LSTRL) for effective V-ReID. More\nspecifically, to extract long-term representations, we propose a\nMulti-granularity Appearance Extractor (MAE), in which four granularity\nappearances are effectively captured across multiple frames. Meanwhile, to\nextract short-term representations, we propose a Bi-direction Motion Estimator\n(BME), in which reciprocal motion information is efficiently extracted from\nconsecutive frames. The MAE and BME are plug-and-play and can be easily\ninserted into existing networks for efficient feature learning. As a result,\nthey significantly improve the feature representation ability for V-ReID.\nExtensive experiments on three widely used benchmarks show that our proposed\napproach can deliver better performances than most state-of-the-arts.\n", "date": "2023-08-07", "categories": []}
{"id": "2308.02065", "title": "On the Biometric Capacity of Generative Face Models", "abstract": "  There has been tremendous progress in generating realistic faces with high\nfidelity over the past few years. Despite this progress, a crucial question\nremains unanswered: \"Given a generative face model, how many unique identities\ncan it generate?\" In other words, what is the biometric capacity of the\ngenerative face model? A scientific basis for answering this question will\nbenefit evaluating and comparing different generative face models and establish\nan upper bound on their scalability. This paper proposes a statistical approach\nto estimate the biometric capacity of generated face images in a hyperspherical\nfeature space. We employ our approach on multiple generative models, including\nunconditional generators like StyleGAN, Latent Diffusion Model, and \"Generated\nPhotos,\" as well as DCFace, a class-conditional generator. We also estimate\ncapacity w.r.t. demographic attributes such as gender and age. Our capacity\nestimates indicate that (a) under ArcFace representation at a false acceptance\nrate (FAR) of 0.1%, StyleGAN3 and DCFace have a capacity upper bound of\n$1.43\\times10^6$ and $1.190\\times10^4$, respectively; (b) the capacity reduces\ndrastically as we lower the desired FAR with an estimate of $1.796\\times10^4$\nand $562$ at FAR of 1% and 10%, respectively, for StyleGAN3; (c) there is no\ndiscernible disparity in the capacity w.r.t gender; and (d) for some generative\nmodels, there is an appreciable disparity in the capacity w.r.t age. Code is\navailable at https://github.com/human-analysis/capacity-generative-face-models.\n", "date": "2023-08-03", "categories": []}
{"id": "2308.00451", "title": "Physics-Driven Spectrum-Consistent Federated Learning for Palmprint\n  Verification", "abstract": "  Palmprint as biometrics has gained increasing attention recently due to its\ndiscriminative ability and robustness. However, existing methods mainly improve\npalmprint verification within one spectrum, which is challenging to verify\nacross different spectrums. Additionally, in distributed server-client-based\ndeployment, palmprint verification systems predominantly necessitate clients to\ntransmit private data for model training on the centralized server, thereby\nengendering privacy apprehensions. To alleviate the above issues, in this\npaper, we propose a physics-driven spectrum-consistent federated learning\nmethod for palmprint verification, dubbed as PSFed-Palm. PSFed-Palm draws upon\nthe inherent physical properties of distinct wavelength spectrums, wherein\nimages acquired under similar wavelengths display heightened resemblances. Our\napproach first partitions clients into short- and long-spectrum groups\naccording to the wavelength range of their local spectrum images. Subsequently,\nwe introduce anchor models for short- and long-spectrum, which constrain the\noptimization directions of local models associated with long- and\nshort-spectrum images. Specifically, a spectrum-consistent loss that enforces\nthe model parameters and feature representation to align with their\ncorresponding anchor models is designed. Finally, we impose constraints on the\nlocal models to ensure their consistency with the global model, effectively\npreventing model drift. This measure guarantees spectrum consistency while\nprotecting data privacy, as there is no need to share local data. Extensive\nexperiments are conducted to validate the efficacy of our proposed PSFed-Palm\napproach. The proposed PSFed-Palm demonstrates compelling performance despite\nonly a limited number of training data. The codes will be released at\nhttps://github.com/Zi-YuanYang/PSFed-Palm.\n", "date": "2023-08-01", "categories": []}
{"id": "2210.10090", "title": "How to Boost Face Recognition with StyleGAN?", "abstract": "  State-of-the-art face recognition systems require vast amounts of labeled\ntraining data. Given the priority of privacy in face recognition applications,\nthe data is limited to celebrity web crawls, which have issues such as limited\nnumbers of identities. On the other hand, self-supervised revolution in the\nindustry motivates research on the adaptation of related techniques to facial\nrecognition. One of the most popular practical tricks is to augment the dataset\nby the samples drawn from generative models while preserving the identity. We\nshow that a simple approach based on fine-tuning pSp encoder for StyleGAN\nallows us to improve upon the state-of-the-art facial recognition and performs\nbetter compared to training on synthetic face identities. We also collect\nlarge-scale unlabeled datasets with controllable ethnic constitution --\nAfricanFaceSet-5M (5 million images of different people) and AsianFaceSet-3M (3\nmillion images of different people) -- and we show that pretraining on each of\nthem improves recognition of the respective ethnicities (as well as others),\nwhile combining all unlabeled datasets results in the biggest performance\nincrease. Our self-supervised strategy is the most useful with limited amounts\nof labeled training data, which can be beneficial for more tailored face\nrecognition tasks and when facing privacy concerns. Evaluation is based on a\nstandard RFW dataset and a new large-scale RB-WebFace benchmark. The code and\ndata are made publicly available at\nhttps://github.com/seva100/stylegan-for-facerec.\n", "date": "2022-10-18", "categories": []}
{"id": "2307.16416", "title": "MRA-GNN: Minutiae Relation-Aware Model over Graph Neural Network for\n  Fingerprint Embedding", "abstract": "  Deep learning has achieved remarkable results in fingerprint embedding, which\nplays a critical role in modern Automated Fingerprint Identification Systems.\nHowever, previous works including CNN-based and Transformer-based approaches\nfail to exploit the nonstructural data, such as topology and correlation in\nfingerprints, which is essential to facilitate the identifiability and\nrobustness of embedding. To address this challenge, we propose a novel paradigm\nfor fingerprint embedding, called Minutiae Relation-Aware model over Graph\nNeural Network (MRA-GNN). Our proposed approach incorporates a GNN-based\nframework in fingerprint embedding to encode the topology and correlation of\nfingerprints into descriptive features, achieving fingerprint representation in\nthe form of graph embedding. Specifically, we reinterpret fingerprint data and\ntheir relative connections as vertices and edges respectively, and introduce a\nminutia graph and fingerprint graph to represent the topological relations and\ncorrelation structures of fingerprints. We equip MRA-GNN with a Topological\nrelation Reasoning Module (TRM) and Correlation-Aware Module (CAM) to learn the\nfingerprint embedding from these graphs successfully. To tackle the\nover-smoothing problem in GNN models, we incorporate Feed-Forward Module and\ngraph residual connections into proposed modules. The experimental results\ndemonstrate that our proposed approach outperforms state-of-the-art methods on\nvarious fingerprint datasets, indicating the effectiveness of our approach in\nexploiting nonstructural information of fingerprints.\n", "date": "2023-07-31", "categories": []}
{"id": "2307.14617", "title": "Multiscale Dynamic Graph Representation for Biometric Recognition with\n  Occlusions", "abstract": "  Occlusion is a common problem with biometric recognition in the wild. The\ngeneralization ability of CNNs greatly decreases due to the adverse effects of\nvarious occlusions. To this end, we propose a novel unified framework\nintegrating the merits of both CNNs and graph models to overcome occlusion\nproblems in biometric recognition, called multiscale dynamic graph\nrepresentation (MS-DGR). More specifically, a group of deep features reflected\non certain subregions is recrafted into a feature graph (FG). Each node inside\nthe FG is deemed to characterize a specific local region of the input sample,\nand the edges imply the co-occurrence of non-occluded regions. By analyzing the\nsimilarities of the node representations and measuring the topological\nstructures stored in the adjacent matrix, the proposed framework leverages\ndynamic graph matching to judiciously discard the nodes corresponding to the\noccluded parts. The multiscale strategy is further incorporated to attain more\ndiverse nodes representing regions of various sizes. Furthermore, the proposed\nframework exhibits a more illustrative and reasonable inference by showing the\npaired nodes. Extensive experiments demonstrate the superiority of the proposed\nframework, which boosts the accuracy in both natural and occlusion-simulated\ncases by a large margin compared with that of baseline methods.\n", "date": "2023-07-27", "categories": []}
{"id": "2307.14713", "title": "GaitMorph: Transforming Gait by Optimally Transporting Discrete Codes", "abstract": "  Gait, the manner of walking, has been proven to be a reliable biometric with\nuses in surveillance, marketing and security. A promising new direction for the\nfield is training gait recognition systems without explicit human annotations,\nthrough self-supervised learning approaches. Such methods are heavily reliant\non strong augmentations for the same walking sequence to induce more data\nvariability and to simulate additional walking variations. Current data\naugmentation schemes are heuristic and cannot provide the necessary data\nvariation as they are only able to provide simple temporal and spatial\ndistortions. In this work, we propose GaitMorph, a novel method to modify the\nwalking variation for an input gait sequence. Our method entails the training\nof a high-compression model for gait skeleton sequences that leverages\nunlabelled data to construct a discrete and interpretable latent space, which\npreserves identity-related features. Furthermore, we propose a method based on\noptimal transport theory to learn latent transport maps on the discrete\ncodebook that morph gait sequences between variations. We perform extensive\nexperiments and show that our method is suitable to synthesize additional views\nfor an input sequence.\n", "date": "2023-07-27", "categories": []}
{"id": "2302.14831", "title": "FacEDiM: A Face Embedding Distribution Model for Few-Shot Biometric\n  Authentication of Cattle", "abstract": "  This work proposes to solve the problem of few-shot biometric authentication\nby computing the Mahalanobis distance between testing embeddings and a\nmultivariate Gaussian distribution of training embeddings obtained using\npre-trained CNNs. Experimental results show that models pre-trained on the\nImageNet dataset significantly outperform models pre-trained on human faces.\nWith a VGG16 model, we obtain a FRR of 1.25% for a FAR of 1.18% on a dataset of\n20 cattle identities.\n", "date": "2023-02-28", "categories": []}
{"id": "2307.14111", "title": "Periocular biometrics: databases, algorithms and directions", "abstract": "  Periocular biometrics has been established as an independent modality due to\nconcerns on the performance of iris or face systems in uncontrolled conditions.\nPeriocular refers to the facial region in the eye vicinity, including eyelids,\nlashes and eyebrows. It is available over a wide range of acquisition\ndistances, representing a trade-off between the whole face (which can be\noccluded at close distances) and the iris texture (which do not have enough\nresolution at long distances). Since the periocular region appears in face or\niris images, it can be used also in conjunction with these modalities. Features\nextracted from the periocular region have been also used successfully for\ngender classification and ethnicity classification, and to study the impact of\ngender transformation or plastic surgery in the recognition performance. This\npaper presents a review of the state of the art in periocular biometric\nresearch, providing an insight of the most relevant issues and giving a\nthorough coverage of the existing literature. Future research trends are also\nbriefly discussed.\n", "date": "2023-07-26", "categories": []}
{"id": "2307.14278", "title": "Large-scale Fully-Unsupervised Re-Identification", "abstract": "  Fully-unsupervised Person and Vehicle Re-Identification have received\nincreasing attention due to their broad applicability in surveillance,\nforensics, event understanding, and smart cities, without requiring any manual\nannotation. However, most of the prior art has been evaluated in datasets that\nhave just a couple thousand samples. Such small-data setups often allow the use\nof costly techniques in time and memory footprints, such as Re-Ranking, to\nimprove clustering results. Moreover, some previous work even pre-selects the\nbest clustering hyper-parameters for each dataset, which is unrealistic in a\nlarge-scale fully-unsupervised scenario. In this context, this work tackles a\nmore realistic scenario and proposes two strategies to learn from large-scale\nunlabeled data. The first strategy performs a local neighborhood sampling to\nreduce the dataset size in each iteration without violating neighborhood\nrelationships. A second strategy leverages a novel Re-Ranking technique, which\nhas a lower time upper bound complexity and reduces the memory complexity from\nO(n^2) to O(kn) with k << n. To avoid the pre-selection of specific\nhyper-parameter values for the clustering algorithm, we also present a novel\nscheduling algorithm that adjusts the density parameter during training, to\nleverage the diversity of samples and keep the learning robust to noisy\nlabeling. Finally, due to the complementary knowledge learned by different\nmodels, we also introduce a co-training strategy that relies upon the\npermutation of predicted pseudo-labels, among the backbones, with no need for\nany hyper-parameters or weighting optimization. The proposed methodology\noutperforms the state-of-the-art methods in well-known benchmarks and in the\nchallenging large-scale Veri-Wild dataset, with a faster and memory-efficient\nRe-Ranking strategy, and a large-scale, noisy-robust, and ensemble-based\nlearning approach.\n", "date": "2023-07-26", "categories": []}
{"id": "2307.13428", "title": "An Explainable Model-Agnostic Algorithm for CNN-based Biometrics\n  Verification", "abstract": "  This paper describes an adaptation of the Local Interpretable Model-Agnostic\nExplanations (LIME) AI method to operate under a biometric verification\nsetting. LIME was initially proposed for networks with the same output classes\nused for training, and it employs the softmax probability to determine which\nregions of the image contribute the most to classification. However, in a\nverification setting, the classes to be recognized have not been seen during\ntraining. In addition, instead of using the softmax output, face descriptors\nare usually obtained from a layer before the classification layer. The model is\nadapted to achieve explainability via cosine similarity between feature vectors\nof perturbated versions of the input image. The method is showcased for face\nbiometrics with two CNN models based on MobileNetv2 and ResNet50.\n", "date": "2023-07-25", "categories": []}
{"id": "2307.08092", "title": "Gait Data Augmentation using Physics-Based Biomechanical Simulation", "abstract": "  This paper focuses on addressing the problem of data scarcity for gait\nanalysis. Standard augmentation methods may produce gait sequences that are not\nconsistent with the biomechanical constraints of human walking. To address this\nissue, we propose a novel framework for gait data augmentation by using\nOpenSIM, a physics-based simulator, to synthesize biomechanically plausible\nwalking sequences. The proposed approach is validated by augmenting the WBDS\nand CASIA-B datasets and then training gait-based classifiers for 3D gender\ngait classification and 2D gait person identification respectively.\nExperimental results indicate that our augmentation approach can improve the\nperformance of model-based gait classifiers and deliver state-of-the-art\nresults for gait-based person identification with an accuracy of up to 96.11%\non the CASIA-B dataset.\n", "date": "2023-07-16", "categories": []}
{"id": "2212.13792", "title": "Periocular Biometrics: A Modality for Unconstrained Scenarios", "abstract": "  Periocular refers to the externally visible region of the face that surrounds\nthe eye socket. This feature-rich area can provide accurate identification in\nunconstrained or uncooperative scenarios, where the iris or face modalities may\nnot offer sufficient biometric cues due to factors such as partial occlusion or\nhigh subject-to-camera distance. The COVID-19 pandemic has further highlighted\nits importance, as the ocular region remained the only visible facial area even\nin controlled settings due to the widespread use of masks. This paper discusses\nthe state of the art in periocular biometrics, presenting an overall framework\nencompassing its most significant research aspects, which include: (a) ocular\ndefinition, acquisition, and detection; (b) identity recognition, including\ncombination with other modalities and use of various spectra; and (c) ocular\nsoft-biometric analysis. Finally, we conclude by addressing current challenges\nand proposing future directions.\n", "date": "2022-12-28", "categories": []}
{"id": "2307.10237", "title": "CoNAN: Conditional Neural Aggregation Network For Unconstrained Face\n  Feature Fusion", "abstract": "  Face recognition from image sets acquired under unregulated and uncontrolled\nsettings, such as at large distances, low resolutions, varying viewpoints,\nillumination, pose, and atmospheric conditions, is challenging. Face feature\naggregation, which involves aggregating a set of N feature representations\npresent in a template into a single global representation, plays a pivotal role\nin such recognition systems. Existing works in traditional face feature\naggregation either utilize metadata or high-dimensional intermediate feature\nrepresentations to estimate feature quality for aggregation. However,\ngenerating high-quality metadata or style information is not feasible for\nextremely low-resolution faces captured in long-range and high altitude\nsettings. To overcome these limitations, we propose a feature distribution\nconditioning approach called CoNAN for template aggregation. Specifically, our\nmethod aims to learn a context vector conditioned over the distribution\ninformation of the incoming feature set, which is utilized to weigh the\nfeatures based on their estimated informativeness. The proposed method produces\nstate-of-the-art results on long-range unconstrained face recognition datasets\nsuch as BTS, and DroneSURF, validating the advantages of such an aggregation\nstrategy.\n", "date": "2023-07-16", "categories": []}
{"id": "2307.10697", "title": "SqueezerFaceNet: Reducing a Small Face Recognition CNN Even More Via\n  Filter Pruning", "abstract": "  The widespread use of mobile devices for various digital services has created\na need for reliable and real-time person authentication. In this context,\nfacial recognition technologies have emerged as a dependable method for\nverifying users due to the prevalence of cameras in mobile devices and their\nintegration into everyday applications. The rapid advancement of deep\nConvolutional Neural Networks (CNNs) has led to numerous face verification\narchitectures. However, these models are often large and impractical for mobile\napplications, reaching sizes of hundreds of megabytes with millions of\nparameters. We address this issue by developing SqueezerFaceNet, a light face\nrecognition network which less than 1M parameters. This is achieved by applying\na network pruning method based on Taylor scores, where filters with small\nimportance scores are removed iteratively. Starting from an already small\nnetwork (of 1.24M) based on SqueezeNet, we show that it can be further reduced\n(up to 40%) without an appreciable loss in performance. To the best of our\nknowledge, we are the first to evaluate network pruning methods for the task of\nface recognition.\n", "date": "2023-07-20", "categories": []}
{"id": "2307.10011", "title": "Towards Fair Face Verification: An In-depth Analysis of Demographic\n  Biases", "abstract": "  Deep learning-based person identification and verification systems have\nremarkably improved in terms of accuracy in recent years; however, such\nsystems, including widely popular cloud-based solutions, have been found to\nexhibit significant biases related to race, age, and gender, a problem that\nrequires in-depth exploration and solutions. This paper presents an in-depth\nanalysis, with a particular emphasis on the intersectionality of these\ndemographic factors. Intersectional bias refers to the performance\ndiscrepancies w.r.t. the different combinations of race, age, and gender\ngroups, an area relatively unexplored in current literature. Furthermore, the\nreliance of most state-of-the-art approaches on accuracy as the principal\nevaluation metric often masks significant demographic disparities in\nperformance. To counter this crucial limitation, we incorporate five additional\nmetrics in our quantitative analysis, including disparate impact and\nmistreatment metrics, which are typically ignored by the relevant\nfairness-aware approaches. Results on the Racial Faces in-the-Wild (RFW)\nbenchmark indicate pervasive biases in face recognition systems, extending\nbeyond race, with different demographic factors yielding significantly\ndisparate outcomes. In particular, Africans demonstrate an 11.25% lower True\nPositive Rate (TPR) compared to Caucasians, while only a 3.51% accuracy drop is\nobserved. Even more concerning, the intersections of multiple protected groups,\nsuch as African females over 60 years old, demonstrate a +39.89% disparate\nmistreatment rate compared to the highest Caucasians rate. By shedding light on\nthese biases and their implications, this paper aims to stimulate further\nresearch towards developing fairer, more equitable face recognition and\nverification systems.\n", "date": "2023-07-19", "categories": []}
{"id": "2307.09183", "title": "Pixel-wise Graph Attention Networks for Person Re-identification", "abstract": "  Graph convolutional networks (GCN) is widely used to handle irregular data\nsince it updates node features by using the structure information of graph.\nWith the help of iterated GCN, high-order information can be obtained to\nfurther enhance the representation of nodes. However, how to apply GCN to\nstructured data (such as pictures) has not been deeply studied. In this paper,\nwe explore the application of graph attention networks (GAT) in image feature\nextraction. First of all, we propose a novel graph generation algorithm to\nconvert images into graphs through matrix transformation. It is one magnitude\nfaster than the algorithm based on K Nearest Neighbors (KNN). Then, GAT is used\non the generated graph to update the node features. Thus, a more robust\nrepresentation is obtained. These two steps are combined into a module called\npixel-wise graph attention module (PGA). Since the graph obtained by our graph\ngeneration algorithm can still be transformed into a picture after processing,\nPGA can be well combined with CNN. Based on these two modules, we consulted the\nResNet and design a pixel-wise graph attention network (PGANet). The PGANet is\napplied to the task of person re-identification in the datasets Market1501,\nDukeMTMC-reID and Occluded-DukeMTMC (outperforms state-of-the-art by 0.8\\%,\n1.1\\% and 11\\% respectively, in mAP scores). Experiment results show that it\nachieves the state-of-the-art performance.\n\\href{https://github.com/wenyu1009/PGANet}{The code is available here}.\n", "date": "2023-07-18", "categories": []}
{"id": "2307.06533", "title": "Domain-adaptive Person Re-identification without Cross-camera Paired\n  Samples", "abstract": "  Existing person re-identification (re-ID) research mainly focuses on\npedestrian identity matching across cameras in adjacent areas. However, in\nreality, it is inevitable to face the problem of pedestrian identity matching\nacross long-distance scenes. The cross-camera pedestrian samples collected from\nlong-distance scenes often have no positive samples. It is extremely\nchallenging to use cross-camera negative samples to achieve cross-region\npedestrian identity matching. Therefore, a novel domain-adaptive person re-ID\nmethod that focuses on cross-camera consistent discriminative feature learning\nunder the supervision of unpaired samples is proposed. This method mainly\nincludes category synergy co-promotion module (CSCM) and cross-camera\nconsistent feature learning module (CCFLM). In CSCM, a task-specific feature\nrecombination (FRT) mechanism is proposed. This mechanism first groups features\naccording to their contributions to specific tasks. Then an interactive\npromotion learning (IPL) scheme between feature groups is developed and\nembedded in this mechanism to enhance feature discriminability. Since the\ncontrol parameters of the specific task model are reduced after division by\ntask, the generalization ability of the model is improved. In CCFLM,\ninstance-level feature distribution alignment and cross-camera identity\nconsistent learning methods are constructed. Therefore, the supervised model\ntraining is achieved under the style supervision of the target domain by\nexchanging styles between source-domain samples and target-domain samples, and\nthe challenges caused by the lack of cross-camera paired samples are solved by\nutilizing cross-camera similar samples. In experiments, three challenging\ndatasets are used as target domains, and the effectiveness of the proposed\nmethod is demonstrated through four experimental settings.\n", "date": "2023-07-13", "categories": []}
{"id": "2307.08615", "title": "Benchmarking fixed-length Fingerprint Representations across different\n  Embedding Sizes and Sensor Types", "abstract": "  Traditional minutiae-based fingerprint representations consist of a\nvariable-length set of minutiae. This necessitates a more complex comparison\ncausing the drawback of high computational cost in one-to-many comparison.\nRecently, deep neural networks have been proposed to extract fixed-length\nembeddings from fingerprints. In this paper, we explore to what extent\nfingerprint texture information contained in such embeddings can be reduced in\nterms of dimension while preserving high biometric performance. This is of\nparticular interest since it would allow to reduce the number of operations\nincurred at comparisons. We also study the impact in terms of recognition\nperformance of the fingerprint textural information for two sensor types, i.e.\noptical and capacitive. Furthermore, the impact of rotation and translation of\nfingerprint images on the extraction of fingerprint embeddings is analysed.\nExperimental results conducted on a publicly available database reveal an\noptimal embedding size of 512 feature elements for the texture-based embedding\npart of fixed-length fingerprint representations. In addition, differences in\nperformance between sensor types can be perceived.\n", "date": "2023-07-17", "categories": []}
{"id": "2307.05529", "title": "Keystroke Dynamics for User Identification", "abstract": "  In previous research, keystroke dynamics has shown promise for user\nauthentication, based on both fixed-text and free-text data. In this research,\nwe consider the more challenging multiclass user identification problem, based\non free-text data. We experiment with a complex image-like feature that has\npreviously been used to achieve state-of-the-art authentication results over\nfree-text data. Using this image-like feature and multiclass Convolutional\nNeural Networks, we are able to obtain a classification (i.e., identification)\naccuracy of 0.78 over a set of 148 users. However, we find that a Random Forest\nclassifier trained on a slightly modified version of this same feature yields\nan accuracy of 0.93.\n", "date": "2023-07-07", "categories": []}
{"id": "2307.05534", "title": "Face Image Quality Enhancement Study for Face Recognition", "abstract": "  Unconstrained face recognition is an active research area among computer\nvision and biometric researchers for many years now. Still the problem of face\nrecognition in low quality photos has not been well-studied so far. In this\npaper, we explore the face recognition performance on low quality photos, and\nwe try to improve the accuracy in dealing with low quality face images. We\nassemble a large database with low quality photos, and examine the performance\nof face recognition algorithms for three different quality sets. Using\nstate-of-the-art facial image enhancement approaches, we explore the face\nrecognition performance for the enhanced face images. To perform this without\nexperimental bias, we have developed a new protocol for recognition with low\nquality face photos and validate the performance experimentally. Our designed\nprotocol for face recognition with low quality face images can be useful to\nother researchers. Moreover, experiment results show some of the challenging\naspects of this problem.\n", "date": "2023-07-08", "categories": []}
{"id": "2307.05563", "title": "RidgeBase: A Cross-Sensor Multi-Finger Contactless Fingerprint Dataset", "abstract": "  Contactless fingerprint matching using smartphone cameras can alleviate major\nchallenges of traditional fingerprint systems including hygienic acquisition,\nportability and presentation attacks. However, development of practical and\nrobust contactless fingerprint matching techniques is constrained by the\nlimited availability of large scale real-world datasets. To motivate further\nadvances in contactless fingerprint matching across sensors, we introduce the\nRidgeBase benchmark dataset. RidgeBase consists of more than 15,000 contactless\nand contact-based fingerprint image pairs acquired from 88 individuals under\ndifferent background and lighting conditions using two smartphone cameras and\none flatbed contact sensor. Unlike existing datasets, RidgeBase is designed to\npromote research under different matching scenarios that include Single Finger\nMatching and Multi-Finger Matching for both contactless- to-contactless (CL2CL)\nand contact-to-contactless (C2CL) verification and identification. Furthermore,\ndue to the high intra-sample variance in contactless fingerprints belonging to\nthe same finger, we propose a set-based matching protocol inspired by the\nadvances in facial recognition datasets. This protocol is specifically designed\nfor pragmatic contactless fingerprint matching that can account for variances\nin focus, polarity and finger-angles. We report qualitative and quantitative\nbaseline results for different protocols using a COTS fingerprint matcher\n(Verifinger) and a Deep CNN based approach on the RidgeBase dataset. The\ndataset can be downloaded here:\nhttps://www.buffalo.edu/cubs/research/datasets/ridgebase-benchmark-dataset.html\n", "date": "2023-07-09", "categories": []}
{"id": "2307.05128", "title": "One-Shot Learning for Periocular Recognition: Exploring the Effect of\n  Domain Adaptation and Data Bias on Deep Representations", "abstract": "  One weakness of machine-learning algorithms is the need to train the models\nfor a new task. This presents a specific challenge for biometric recognition\ndue to the dynamic nature of databases and, in some instances, the reliance on\nsubject collaboration for data collection. In this paper, we investigate the\nbehavior of deep representations in widely used CNN models under extreme data\nscarcity for One-Shot periocular recognition, a biometric recognition task. We\nanalyze the outputs of CNN layers as identity-representing feature vectors. We\nexamine the impact of Domain Adaptation on the network layers' output for\nunseen data and evaluate the method's robustness concerning data normalization\nand generalization of the best-performing layer. We improved state-of-the-art\nresults that made use of networks trained with biometric datasets with millions\nof images and fine-tuned for the target periocular dataset by utilizing\nout-of-the-box CNNs trained for the ImageNet Recognition Challenge and standard\ncomputer vision algorithms. For example, for the Cross-Eyed dataset, we could\nreduce the EER by 67% and 79% (from 1.70% and 3.41% to 0.56% and 0.71%) in the\nClose-World and Open-World protocols, respectively, for the periocular case. We\nalso demonstrate that traditional algorithms like SIFT can outperform CNNs in\nsituations with limited data or scenarios where the network has not been\ntrained with the test classes like the Open-World mode. SIFT alone was able to\nreduce the EER by 64% and 71.6% (from 1.7% and 3.41% to 0.6% and 0.97%) for\nCross-Eyed in the Close-World and Open-World protocols, respectively, and a\nreduction of 4.6% (from 3.94% to 3.76%) in the PolyU database for the\nOpen-World and single biometric case.\n", "date": "2023-07-11", "categories": []}
{"id": "2307.05437", "title": "Improving the Security of Smartwatch Payment with Deep Learning", "abstract": "  Making contactless payments using a smartwatch is increasingly popular, but\nthis payment medium lacks traditional biometric security measures such as\nfacial or fingerprint recognition. In 2022, Sturgess et al. proposed WatchAuth,\na system for authenticating smartwatch payments using the physical gesture of\nreaching towards a payment terminal. While effective, the system requires the\nuser to undergo a burdensome enrolment period to achieve acceptable error\nlevels. In this dissertation, we explore whether applications of deep learning\ncan reduce the number of gestures a user must provide to enrol into an\nauthentication system for smartwatch payment. We firstly construct a\ndeep-learned authentication system that outperforms the current\nstate-of-the-art, including in a scenario where the target user has provided a\nlimited number of gestures. We then develop a regularised autoencoder model for\ngenerating synthetic user-specific gestures. We show that using these gestures\nin training improves classification ability for an authentication system.\nThrough this technique we can reduce the number of gestures required to enrol a\nuser into a WatchAuth-like system without negatively impacting its error rates.\n", "date": "2023-07-11", "categories": []}
{"id": "2209.04547", "title": "Defend Data Poisoning Attacks on Voice Authentication", "abstract": "  With the advances in deep learning, speaker verification has achieved very\nhigh accuracy and is gaining popularity as a type of biometric authentication\noption in many scenes of our daily life, especially the growing market of web\nservices. Compared to traditional passwords, \"vocal passwords\" are much more\nconvenient as they relieve people from memorizing different passwords. However,\nnew machine learning attacks are putting these voice authentication systems at\nrisk. Without a strong security guarantee, attackers could access legitimate\nusers' web accounts by fooling the deep neural network (DNN) based voice\nrecognition models. In this paper, we demonstrate an easy-to-implement data\npoisoning attack to the voice authentication system, which can hardly be\ncaptured by existing defense mechanisms. Thus, we propose a more robust defense\nmethod, called Guardian, which is a convolutional neural network-based\ndiscriminator. The Guardian discriminator integrates a series of novel\ntechniques including bias reduction, input augmentation, and ensemble learning.\nOur approach is able to distinguish about 95% of attacked accounts from normal\naccounts, which is much more effective than existing approaches with only 60%\naccuracy.\n", "date": "2022-09-09", "categories": []}
{"id": "2307.01663", "title": "Exploring Transformers for On-Line Handwritten Signature Verification", "abstract": "  The application of mobile biometrics as a user-friendly authentication method\nhas increased in the last years. Recent studies have proposed novel behavioral\nbiometric recognition systems based on Transformers, which currently outperform\nthe state of the art in several application scenarios. On-line handwritten\nsignature verification aims to verify the identity of subjects, based on their\nbiometric signatures acquired using electronic devices such as tablets or\nsmartphones. This paper investigates the suitability of architectures based on\nrecent Transformers for on-line signature verification. In particular, four\ndifferent configurations are studied, two of them rely on the Vanilla\nTransformer encoder, and the two others have been successfully applied to the\ntasks of gait and activity recognition. We evaluate the four proposed\nconfigurations according to the experimental protocol proposed in the\nSVC-onGoing competition. The results obtained in our experiments are promising,\nand promote the use of Transformers for on-line signature verification.\n", "date": "2023-07-04", "categories": []}
{"id": "2307.02679", "title": "A Study on the Impact of Face Image Quality on Face Recognition in the\n  Wild", "abstract": "  Deep learning has received increasing interests in face recognition recently.\nLarge quantities of deep learning methods have been proposed to handle various\nproblems appeared in face recognition. Quite a lot deep methods claimed that\nthey have gained or even surpassed human-level face verification performance in\ncertain databases. As we know, face image quality poses a great challenge to\ntraditional face recognition methods, e.g. model-driven methods with\nhand-crafted features. However, a little research focus on the impact of face\nimage quality on deep learning methods, and even human performance. Therefore,\nwe raise a question: Is face image quality still one of the challenges for deep\nlearning based face recognition, especially in unconstrained condition. Based\non this, we further investigate this problem on human level. In this paper, we\npartition face images into three different quality sets to evaluate the\nperformance of deep learning methods on cross-quality face images in the wild,\nand then design a human face verification experiment on these cross-quality\ndata. The result indicates that quality issue still needs to be studied\nthoroughly in deep learning, human own better capability in building the\nrelations between different face images with large quality gaps, and saying\ndeep learning method surpasses human-level is too optimistic.\n", "date": "2023-07-05", "categories": []}
{"id": "2307.01845", "title": "Deep Features for Contactless Fingerprint Presentation Attack Detection:\n  Can They Be Generalized?", "abstract": "  The rapid evolution of high-end smartphones with advanced high-resolution\ncameras has resulted in contactless capture of fingerprint biometrics that are\nmore reliable and suitable for verification. Similar to other biometric\nsystems, contactless fingerprint-verification systems are vulnerable to\npresentation attacks. In this paper, we present a comparative study on the\ngeneralizability of seven different pre-trained Convolutional Neural Networks\n(CNN) and a Vision Transformer (ViT) to reliably detect presentation attacks.\nExtensive experiments were carried out on publicly available smartphone-based\npresentation attack datasets using four different Presentation Attack\nInstruments (PAI). The detection performance of the eighth deep feature\ntechnique was evaluated using the leave-one-out protocol to benchmark the\ngeneralization performance for unseen PAI. The obtained results indicated the\nbest generalization performance with the ResNet50 CNN.\n", "date": "2023-07-04", "categories": []}
{"id": "2210.08710", "title": "Handling Label Uncertainty for Camera Incremental Person\n  Re-Identification", "abstract": "  Incremental learning for person re-identification (ReID) aims to develop\nmodels that can be trained with a continuous data stream, which is a more\npractical setting for real-world applications. However, the existing\nincremental ReID methods make two strong assumptions that the cameras are fixed\nand the new-emerging data is class-disjoint from previous classes. This is\nunrealistic as previously observed pedestrians may re-appear and be captured\nagain by new cameras. In this paper, we investigate person ReID in an\nunexplored scenario named Camera Incremental Person ReID (CIPR), which advances\nexisting lifelong person ReID by taking into account the class overlap issue.\nSpecifically, new data collected from new cameras may probably contain an\nunknown proportion of identities seen before. This subsequently leads to the\nlack of cross-camera annotations for new data due to privacy concerns. To\naddress these challenges, we propose a novel framework ExtendOVA. First, to\nhandle the class overlap issue, we introduce an instance-wise seen-class\nidentification module to discover previously seen identities at the instance\nlevel. Then, we propose a criterion for selecting confident ID-wise candidates\nand also devise an early learning regularization term to correct noise issues\nin pseudo labels. Furthermore, to compensate for the lack of previous data, we\nresort prototypical memory bank to create surrogate features, along with a\ncross-camera distillation loss to further retain the inter-camera relationship.\nThe comprehensive experimental results on multiple benchmarks show that\nExtendOVA significantly outperforms the state-of-the-arts with remarkable\nadvantages.\n", "date": "2022-10-17", "categories": []}
{"id": "2307.00578", "title": "TinySiamese Network for Biometric Analysis", "abstract": "  Biometric recognition is the process of verifying or classifying human\ncharacteristics in images or videos. It is a complex task that requires machine\nlearning algorithms, including convolutional neural networks (CNNs) and Siamese\nnetworks. Besides, there are several limitations to consider when using these\nalgorithms for image verification and classification tasks. In fact, training\nmay be computationally intensive, requiring specialized hardware and\nsignificant computational resources to train and deploy. Moreover, it\nnecessitates a large amount of labeled data, which can be time-consuming and\ncostly to obtain. The main advantage of the proposed TinySiamese compared to\nthe standard Siamese is that it does not require the whole CNN for training. In\nfact, using a pre-trained CNN as a feature extractor and the TinySiamese to\nlearn the extracted features gave almost the same performance and efficiency as\nthe standard Siamese for biometric verification. In this way, the TinySiamese\nsolves the problems of memory and computational time with a small number of\nlayers which did not exceed 7. It can be run under low-power machines which\npossess a normal GPU and cannot allocate a large RAM space. Using TinySiamese\nwith only 8 GO of memory, the matching time decreased by 76.78% on the B2F\n(Biometric images of Fingerprints and Faces), FVC2000, FVC2002 and FVC2004\nwhile the training time for 10 epochs went down by approximately 93.14% on the\nB2F, FVC2002, THDD-part1 and CASIA-B datasets. The accuracy of the fingerprint,\ngait (NM-angle 180 degree) and face verification tasks was better than the\naccuracy of a standard Siamese by 0.87%, 20.24% and 3.85% respectively.\nTinySiamese achieved comparable accuracy with related works for the fingerprint\nand gait classification tasks.\n", "date": "2023-07-02", "categories": []}
{"id": "2306.17200", "title": "Residual Feature Pyramid Network for Enhancement of Vascular Patterns", "abstract": "  The accuracy of finger vein recognition systems gets degraded due to low and\nuneven contrast between veins and surroundings, often resulting in poor\ndetection of vein patterns. We propose a finger-vein enhancement technique,\nResFPN (Residual Feature Pyramid Network), as a generic preprocessing method\nagnostic to the recognition pipeline. A bottom-up pyramidal architecture using\nthe novel Structure Detection block (SDBlock) facilitates extraction of veins\nof varied widths. Using a feature aggregation module (FAM), we combine these\nvein-structures, and train the proposed ResFPN for detection of veins across\nscales. With enhanced presentations, our experiments indicate a reduction upto\n5% in the average recognition errors for commonly used recognition pipeline\nover two publicly available datasets. These improvements are persistent even in\ncross-dataset scenario where the dataset used to train the ResFPN is different\nfrom the one used for recognition.\n", "date": "2023-06-29", "categories": []}
{"id": "2306.13907", "title": "Person Recognition using Facial Micro-Expressions with Deep Learning", "abstract": "  This study investigates the efficacy of facial micro-expressions as a soft\nbiometric for enhancing person recognition, aiming to broaden the understanding\nof the subject and its potential applications. We propose a deep learning\napproach designed to capture spatial semantics and motion at a fine temporal\nresolution. Experiments on three widely-used micro-expression databases\ndemonstrate a notable increase in identification accuracy compared to existing\nbenchmarks, highlighting the potential of integrating facial micro-expressions\nfor improved person recognition across various fields.\n", "date": "2023-06-24", "categories": []}
{"id": "2306.14313", "title": "A Closer Look at Geometric Temporal Dynamics for Face Anti-Spoofing", "abstract": "  Face anti-spoofing (FAS) is indispensable for a face recognition system. Many\ntexture-driven countermeasures were developed against presentation attacks\n(PAs), but the performance against unseen domains or unseen spoofing types is\nstill unsatisfactory. Instead of exhaustively collecting all the spoofing\nvariations and making binary decisions of live/spoof, we offer a new\nperspective on the FAS task to distinguish between normal and abnormal\nmovements of live and spoof presentations. We propose Geometry-Aware\nInteraction Network (GAIN), which exploits dense facial landmarks with\nspatio-temporal graph convolutional network (ST-GCN) to establish a more\ninterpretable and modularized FAS model. Additionally, with our cross-attention\nfeature interaction mechanism, GAIN can be easily integrated with other\nexisting methods to significantly boost performance. Our approach achieves\nstate-of-the-art performance in the standard intra- and cross-dataset\nevaluations. Moreover, our model outperforms state-of-the-art methods by a\nlarge margin in the cross-dataset cross-type protocol on CASIA-SURF 3DMask\n(+10.26% higher AUC score), exhibiting strong robustness against domain shifts\nand unseen spoofing types.\n", "date": "2023-06-25", "categories": []}
{"id": "2306.14662", "title": "Cross Architecture Distillation for Face Recognition", "abstract": "  Transformers have emerged as the superior choice for face recognition tasks,\nbut their insufficient platform acceleration hinders their application on\nmobile devices. In contrast, Convolutional Neural Networks (CNNs) capitalize on\nhardware-compatible acceleration libraries. Consequently, it has become\nindispensable to preserve the distillation efficacy when transferring knowledge\nfrom a Transformer-based teacher model to a CNN-based student model, known as\nCross-Architecture Knowledge Distillation (CAKD). Despite its potential, the\ndeployment of CAKD in face recognition encounters two challenges: 1) the\nteacher and student share disparate spatial information for each pixel,\nobstructing the alignment of feature space, and 2) the teacher network is not\ntrained in the role of a teacher, lacking proficiency in handling\ndistillation-specific knowledge. To surmount these two constraints, 1) we first\nintroduce a Unified Receptive Fields Mapping module (URFM) that maps pixel\nfeatures of the teacher and student into local features with unified receptive\nfields, thereby synchronizing the pixel-wise spatial information of teacher and\nstudent. Subsequently, 2) we develop an Adaptable Prompting Teacher network\n(APT) that integrates prompts into the teacher, enabling it to manage\ndistillation-specific knowledge while preserving the model's discriminative\ncapacity. Extensive experiments on popular face benchmarks and two large-scale\nverification sets demonstrate the superiority of our method.\n", "date": "2023-06-26", "categories": []}
{"id": "2306.12572", "title": "Uniqueness of Iris Pattern Based on AR Model", "abstract": "  The assessment of iris uniqueness plays a crucial role in analyzing the\ncapabilities and limitations of iris recognition systems. Among the various\nmethodologies proposed, Daugman's approach to iris uniqueness stands out as one\nof the most widely accepted. According to Daugman, uniqueness refers to the\niris recognition system's ability to enroll an increasing number of classes\nwhile maintaining a near-zero probability of collision between new and enrolled\nclasses. Daugman's approach involves creating distinct IrisCode templates for\neach iris class within the system and evaluating the sustainable population\nunder a fixed Hamming distance between codewords. In our previous work [23], we\nutilized Rate-Distortion Theory (as it pertains to the limits of\nerror-correction codes) to establish boundaries for the maximum possible\npopulation of iris classes supported by Daugman's IrisCode, given the\nconstraint of a fixed Hamming distance between codewords. Building upon that\nresearch, we propose a novel methodology to evaluate the scalability of an iris\nrecognition system, while also measuring iris quality. We achieve this by\nemploying a sphere-packing bound for Gaussian codewords and adopting a approach\nsimilar to Daugman's, which utilizes relative entropy as a distance measure\nbetween iris classes. To demonstrate the efficacy of our methodology, we\nillustrate its application on two small datasets of iris images. We determine\nthe sustainable maximum population for each dataset based on the quality of the\nimages. By providing these illustrations, we aim to assist researchers in\ncomprehending the limitations inherent in their recognition systems, depending\non the quality of their iris databases.\n", "date": "2023-06-21", "categories": []}
{"id": "2306.10919", "title": "Fairness Index Measures to Evaluate Bias in Biometric Recognition", "abstract": "  The demographic disparity of biometric systems has led to serious concerns\nregarding their societal impact as well as applicability of such systems in\nprivate and public domains. A quantitative evaluation of demographic fairness\nis an important step towards understanding, assessment, and mitigation of\ndemographic bias in biometric applications. While few, existing fairness\nmeasures are based on post-decision data (such as verification accuracy) of\nbiometric systems, we discuss how pre-decision data (score distributions)\nprovide useful insights towards demographic fairness. In this paper, we\nintroduce multiple measures, based on the statistical characteristics of score\ndistributions, for the evaluation of demographic fairness of a generic\nbiometric verification system. We also propose different variants for each\nfairness measure depending on how the contribution from constituent demographic\ngroups needs to be combined towards the final measure. In each case, the\nbehavior of the measure has been illustrated numerically and graphically on\nsynthetic data. The demographic imbalance in benchmarking datasets is often\noverlooked during fairness assessment. We provide a novel weighing strategy to\nreduce the effect of such imbalance through a non-linear function of sample\nsizes of demographic groups. The proposed measures are independent of the\nbiometric modality, and thus, applicable across commonly used biometric\nmodalities (e.g., face, fingerprint, etc.).\n", "date": "2023-06-19", "categories": []}
{"id": "2306.08303", "title": "Pedestrian Recognition with Radar Data-Enhanced Deep Learning Approach\n  Based on Micro-Doppler Signatures", "abstract": "  As a hot topic in recent years, the ability of pedestrians identification\nbased on radar micro-Doppler signatures is limited by the lack of adequate\ntraining data. In this paper, we propose a data-enhanced multi-characteristic\nlearning (DEMCL) model with data enhancement (DE) module and\nmulti-characteristic learning (MCL) module to learn more complementary\npedestrian micro-Doppler (m-D) signatures. In DE module, a range-Doppler\ngenerative adversarial network (RDGAN) is proposed to enhance free walking\ndatasets, and MCL module with multi-scale convolution neural network (MCNN) and\nradial basis function neural network (RBFNN) is trained to learn m-D signatures\nextracted from enhanced datasets. Experimental results show that our model is\n3.33% to 10.24% more accurate than other studies and has a short run time of\n0.9324 seconds on a 25-minute walking dataset.\n", "date": "2023-06-14", "categories": []}
{"id": "2306.08549", "title": "An Exploratory Study of Masked Face Recognition with Machine Learning\n  Algorithms", "abstract": "  Automated face recognition is a widely adopted machine learning technology\nfor contactless identification of people in various processes such as automated\nborder control, secure login to electronic devices, community surveillance,\ntracking school attendance, workplace clock in and clock out. Using face masks\nhave become crucial in our daily life with the recent world-wide COVID-19\npandemic. The use of face masks causes the performance of conventional face\nrecognition technologies to degrade considerably. The effect of mask-wearing in\nface recognition is yet an understudied issue. In this paper, we address this\nissue by evaluating the performance of a number of face recognition models\nwhich are tested by identifying masked and unmasked face images. We use six\nconventional machine learning algorithms, which are SVC, KNN, LDA, DT, LR and\nNB, to find out the ones which perform best, besides the ones which poorly\nperform, in the presence of masked face images. Local Binary Pattern (LBP) is\nutilized as the feature extraction operator. We generated and used synthesized\nmasked face images. We prepared unmasked, masked, and half-masked training\ndatasets and evaluated the face recognition performance against both masked and\nunmasked images to present a broad view of this crucial problem. We believe\nthat our study is unique in elaborating the mask-aware facial recognition with\nalmost all possible scenarios including half_masked-to-masked and\nhalf_masked-to-unmasked besides evaluating a larger number of conventional\nmachine learning algorithms compared the other studies in the literature.\n", "date": "2023-06-14", "categories": []}
{"id": "2306.07045", "title": "Data-Driven Bilateral Generalized Two-Dimensional Quaternion Principal\n  Component Analysis with Application to Color Face Recognition", "abstract": "  A new data-driven bilateral generalized two-dimensional quaternion principal\ncomponent analysis (BiG2DQPCA) is presented to extract the features of matrix\nsamples from both row and column directions. This general framework directly\nworks on the 2D color images without vectorizing and well preserves the spatial\nand color information, which makes it flexible to fit various real-world\napplications. A generalized ridge regression model of BiG2DQPCA is firstly\nproposed with orthogonality constrains on aimed features. Applying the\ndeflation technique and the framework of minorization-maximization, a new\nquaternion optimization algorithm is proposed to compute the optimal features\nof BiG2DQPCA and a closed-form solution is obtained at each iteration. A new\napproach based on BiG2DQPCA is presented for color face recognition and image\nreconstruction with a new data-driven weighting technique. Sufficient numerical\nexperiments are implemented on practical color face databases and indicate the\nsuperiority of BiG2DQPCA over the state-of-the-art methods in terms of\nrecognition accuracies and rates of image reconstruction.\n", "date": "2023-06-12", "categories": []}
{"id": "2207.03132", "title": "Style Interleaved Learning for Generalizable Person Re-identification", "abstract": "  Domain generalization (DG) for person re-identification (ReID) is a\nchallenging problem, as access to target domain data is not permitted during\nthe training process. Most existing DG ReID methods update the feature\nextractor and classifier parameters based on the same features. This common\npractice causes the model to overfit to existing feature styles in the source\ndomain, resulting in sub-optimal generalization ability on target domains. To\nsolve this problem, we propose a novel style interleaved learning (IL)\nframework. Unlike conventional learning strategies, IL incorporates two forward\npropagations and one backward propagation for each iteration. We employ the\nfeatures of interleaved styles to update the feature extractor and classifiers\nusing different forward propagations, which helps to prevent the model from\noverfitting to certain domain styles. To generate interleaved feature styles,\nwe further propose a new feature stylization approach. It produces a wide range\nof meaningful styles that are both different and independent from the original\nstyles in the source domain, which caters to the IL methodology. Extensive\nexperimental results show that our model not only consistently outperforms\nstate-of-the-art methods on large-scale benchmarks for DG ReID, but also has\nclear advantages in computational efficiency. The code is available at\nhttps://github.com/WentaoTan/Interleaved-Learning.\n", "date": "2022-07-07", "categories": []}
{"id": "2306.03993", "title": "Real-Time Online Unsupervised Domain Adaptation for Real-World Person\n  Re-identification", "abstract": "  Following the popularity of Unsupervised Domain Adaptation (UDA) in person\nre-identification, the recently proposed setting of Online Unsupervised Domain\nAdaptation (OUDA) attempts to bridge the gap towards practical applications by\nintroducing a consideration of streaming data. However, this still falls short\nof truly representing real-world applications. This paper defines the setting\nof Real-world Real-time Online Unsupervised Domain Adaptation (R$^2$OUDA) for\nPerson Re-identification. The R$^2$OUDA setting sets the stage for true\nreal-world real-time OUDA, bringing to light four major limitations found in\nreal-world applications that are often neglected in current research: system\ngenerated person images, subset distribution selection, time-based data stream\nsegmentation, and a segment-based time constraint. To address all aspects of\nthis new R$^2$OUDA setting, this paper further proposes Real-World Real-Time\nOnline Streaming Mutual Mean-Teaching (R$^2$MMT), a novel multi-camera system\nfor real-world person re-identification. Taking a popular person\nre-identification dataset, R$^2$MMT was used to construct over 100 data subsets\nand train more than 3000 models, exploring the breadth of the R$^2$OUDA setting\nto understand the training time and accuracy trade-offs and limitations for\nreal-world applications. R$^2$MMT, a real-world system able to respect the\nstrict constraints of the proposed R$^2$OUDA setting, achieves accuracies\nwithin 0.1% of comparable OUDA methods that cannot be applied directly to\nreal-world applications.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.04000", "title": "A Quality Aware Sample-to-Sample Comparison for Face Recognition", "abstract": "  Currently available face datasets mainly consist of a large number of\nhigh-quality and a small number of low-quality samples. As a result, a Face\nRecognition (FR) network fails to learn the distribution of low-quality samples\nsince they are less frequent during training (underrepresented). Moreover,\ncurrent state-of-the-art FR training paradigms are based on the\nsample-to-center comparison (i.e., Softmax-based classifier), which results in\na lack of uniformity between train and test metrics. This work integrates a\nquality-aware learning process at the sample level into the classification\ntraining paradigm (QAFace). In this regard, Softmax centers are adaptively\nguided to pay more attention to low-quality samples by using a quality-aware\nfunction. Accordingly, QAFace adds a quality-based adjustment to the updating\nprocedure of the Softmax-based classifier to improve the performance on the\nunderrepresented low-quality samples. Our method adaptively finds and assigns\nmore attention to the recognizable low-quality samples in the training\ndatasets. In addition, QAFace ignores the unrecognizable low-quality samples\nusing the feature magnitude as a proxy for quality. As a result, QAFace\nprevents class centers from getting distracted from the optimal direction. The\nproposed method is superior to the state-of-the-art algorithms in extensive\nexperimental results on the CFP-FP, LFW, CPLFW, CALFW, AgeDB, IJB-B, and IJB-C\ndatasets.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.03577", "title": "An Open Patch Generator based Fingerprint Presentation Attack Detection\n  using Generative Adversarial Network", "abstract": "  The low-cost, user-friendly, and convenient nature of Automatic Fingerprint\nRecognition Systems (AFRS) makes them suitable for a wide range of\napplications. This spreading use of AFRS also makes them vulnerable to various\nsecurity threats. Presentation Attack (PA) or spoofing is one of the threats\nwhich is caused by presenting a spoof of a genuine fingerprint to the sensor of\nAFRS. Fingerprint Presentation Attack Detection (FPAD) is a countermeasure\nintended to protect AFRS against fake or spoof fingerprints created using\nvarious fabrication materials. In this paper, we have proposed a Convolutional\nNeural Network (CNN) based technique that uses a Generative Adversarial Network\n(GAN) to augment the dataset with spoof samples generated from the proposed\nOpen Patch Generator (OPG). This OPG is capable of generating realistic\nfingerprint samples which have no resemblance to the existing spoof fingerprint\nsamples generated with other materials. The augmented dataset is fed to the\nDenseNet classifier which helps in increasing the performance of the\nPresentation Attack Detection (PAD) module for the various real-world attacks\npossible with unknown spoof materials. Experimental evaluations of the proposed\napproach are carried out on the Liveness Detection (LivDet) 2015, 2017, and\n2019 competition databases. An overall accuracy of 96.20\\%, 94.97\\%, and\n92.90\\% has been achieved on the LivDet 2015, 2017, and 2019 databases,\nrespectively under the LivDet protocol scenarios. The performance of the\nproposed PAD model is also validated in the cross-material and cross-sensor\nattack paradigm which further exhibits its capability to be used under\nreal-world attack scenarios.\n", "date": "2023-06-06", "categories": []}
{"id": "2306.01449", "title": "SASMU: boost the performance of generalized recognition model using\n  synthetic face dataset", "abstract": "  Nowadays, deploying a robust face recognition product becomes easy with the\ndevelopment of face recognition techniques for decades. Not only profile image\nverification but also the state-of-the-art method can handle the in-the-wild\nimage almost perfectly. However, the concern of privacy issues raise rapidly\nsince mainstream research results are powered by tons of web-crawled data,\nwhich faces the privacy invasion issue. The community tries to escape this\npredicament completely by training the face recognition model with synthetic\ndata but faces severe domain gap issues, which still need to access real images\nand identity labels to fine-tune the model. In this paper, we propose SASMU, a\nsimple, novel, and effective method for face recognition using a synthetic\ndataset. Our proposed method consists of spatial data augmentation (SA) and\nspectrum mixup (SMU). We first analyze the existing synthetic datasets for\ndeveloping a face recognition system. Then, we reveal that heavy data\naugmentation is helpful for boosting performance when using synthetic data. By\nanalyzing the previous frequency mixup studies, we proposed a novel method for\ndomain generalization. Extensive experimental results have demonstrated the\neffectiveness of SASMU, achieving state-of-the-art performance on several\ncommon benchmarks, such as LFW, AgeDB-30, CA-LFW, CFP-FP, and CP-LFW.\n", "date": "2023-06-02", "categories": []}
{"id": "2306.00231", "title": "A Universal Latent Fingerprint Enhancer Using Transformers", "abstract": "  Forensic science heavily relies on analyzing latent fingerprints, which are\ncrucial for criminal investigations. However, various challenges, such as\nbackground noise, overlapping prints, and contamination, make the\nidentification process difficult. Moreover, limited access to real crime scene\nand laboratory-generated databases hinders the development of efficient\nrecognition algorithms. This study aims to develop a fast method, which we call\nULPrint, to enhance various latent fingerprint types, including those obtained\nfrom real crime scenes and laboratory-created samples, to boost fingerprint\nrecognition system performance. In closed-set identification accuracy\nexperiments, the enhanced image was able to improve the performance of the\nMSU-AFIS from 61.56\\% to 75.19\\% in the NIST SD27 database, from 67.63\\% to\n77.02\\% in the MSP Latent database, and from 46.90\\% to 52.12\\% in the NIST\nSD302 database. Our contributions include (1) the development of a two-step\nlatent fingerprint enhancement method that combines Ridge Segmentation with\nUNet and Mix Visual Transformer (MiT) SegFormer-B5 encoder architecture, (2)\nthe implementation of multiple dilated convolutions in the UNet architecture to\ncapture intricate, non-local patterns better and enhance ridge segmentation,\nand (3) the guided blending of the predicted ridge mask with the latent\nfingerprint. This novel approach, ULPrint, streamlines the enhancement process,\naddressing challenges across diverse latent fingerprint types to improve\nforensic investigations and criminal justice outcomes.\n", "date": "2023-05-31", "categories": []}
{"id": "2306.00272", "title": "Accelerated Fingerprint Enhancement: A GPU-Optimized Mixed Architecture\n  Approach", "abstract": "  This document presents a preliminary approach to latent fingerprint\nenhancement, fundamentally designed around a mixed Unet architecture. It\ncombines the capabilities of the Resnet-101 network and Unet encoder, aiming to\nform a potentially powerful composite. This combination, enhanced with\nattention mechanisms and forward skip connections, is intended to optimize the\nenhancement of ridge and minutiae features in fingerprints. One innovative\nelement of this approach includes a novel Fingerprint Enhancement Gabor layer,\nspecifically designed for GPU computations. This illustrates how modern\ncomputational resources might be harnessed to expedite enhancement. Given its\npotential functionality as either a CNN or Transformer layer, this Gabor layer\ncould offer improved agility and processing speed to the system. However, it is\nimportant to note that this approach is still in the early stages of\ndevelopment and has not yet been fully validated through rigorous experiments.\nAs such, it may require additional time and testing to establish its robustness\nand usability in the field of latent fingerprint enhancement. This includes\nimprovements in processing speed, enhancement adaptability with distinct latent\nfingerprint types, and full validation in experimental approaches such as\nopen-set (identification 1:N) and open-set validation, fingerprint quality\nevaluation, among others.\n", "date": "2023-06-01", "categories": []}
{"id": "2212.13075", "title": "TypeFormer: Transformers for Mobile Keystroke Biometrics", "abstract": "  The broad usage of mobile devices nowadays, the sensitiveness of the\ninformation contained in them, and the shortcomings of current mobile user\nauthentication methods are calling for novel, secure, and unobtrusive solutions\nto verify the users' identity. In this article, we propose TypeFormer, a novel\nTransformer architecture to model free-text keystroke dynamics performed on\nmobile devices for the purpose of user authentication. The proposed model\nconsists in Temporal and Channel Modules enclosing two Long Short-Term Memory\n(LSTM) recurrent layers, Gaussian Range Encoding (GRE), a multi-head\nSelf-Attention mechanism, and a Block-Recurrent structure. Experimenting on one\nof the largest public databases to date, the Aalto mobile keystroke database,\nTypeFormer outperforms current state-of-the-art systems achieving Equal Error\nRate (EER) values of 3.25% using only 5 enrolment sessions of 50 keystrokes\neach. In such way, we contribute to reducing the traditional performance gap of\nthe challenging mobile free-text scenario with respect to its desktop and\nfixed-text counterparts. Additionally, we analyse the behaviour of the model\nwith different experimental configurations such as the length of the keystroke\nsequences and the amount of enrolment sessions, showing margin for improvement\nwith more enrolment data. Finally, a cross-database evaluation is carried out,\ndemonstrating the robustness of the features extracted by TypeFormer in\ncomparison with existing approaches.\n", "date": "2022-12-26", "categories": []}
{"id": "2305.16443", "title": "Human-Machine Comparison for Cross-Race Face Verification: Race Bias at\n  the Upper Limits of Performance?", "abstract": "  Face recognition algorithms perform more accurately than humans in some\ncases, though humans and machines both show race-based accuracy differences. As\nalgorithms continue to improve, it is important to continually assess their\nrace bias relative to humans. We constructed a challenging test of 'cross-race'\nface verification and used it to compare humans and two state-of-the-art face\nrecognition systems. Pairs of same- and different-identity faces of White and\nBlack individuals were selected to be difficult for humans and an open-source\nimplementation of the ArcFace face recognition algorithm from 2019 (5). Human\nparticipants (54 Black; 51 White) judged whether face pairs showed the same\nidentity or different identities on a 7-point Likert-type scale. Two\ntop-performing face recognition systems from the Face Recognition Vendor\nTest-ongoing performed the same test (7). By design, the test proved\nchallenging for humans as a group, who performed above chance, but far less\nthan perfect. Both state-of-the-art face recognition systems scored perfectly\n(no errors), consequently with equal accuracy for both races. We conclude that\nstate-of-the-art systems for identity verification between two frontal face\nimages of Black and White individuals can surpass the general population.\nWhether this result generalizes to challenging in-the-wild images is a pressing\nconcern for deploying face recognition systems in unconstrained environments.\n", "date": "2023-05-25", "categories": []}
{"id": "2305.19962", "title": "GANDiffFace: Controllable Generation of Synthetic Datasets for Face\n  Recognition with Realistic Variations", "abstract": "  Face recognition systems have significantly advanced in recent years, driven\nby the availability of large-scale datasets. However, several issues have\nrecently came up, including privacy concerns that have led to the\ndiscontinuation of well-established public datasets. Synthetic datasets have\nemerged as a solution, even though current synthesis methods present other\ndrawbacks such as limited intra-class variations, lack of realism, and unfair\nrepresentation of demographic groups. This study introduces GANDiffFace, a\nnovel framework for the generation of synthetic datasets for face recognition\nthat combines the power of Generative Adversarial Networks (GANs) and Diffusion\nmodels to overcome the limitations of existing synthetic datasets. In\nGANDiffFace, we first propose the use of GANs to synthesize highly realistic\nidentities and meet target demographic distributions. Subsequently, we\nfine-tune Diffusion models with the images generated with GANs, synthesizing\nmultiple images of the same identity with a variety of accessories, poses,\nexpressions, and contexts. We generate multiple synthetic datasets by changing\nGANDiffFace settings, and compare their mated and non-mated score distributions\nwith the distributions provided by popular real-world datasets for face\nrecognition, i.e. VGG2 and IJB-C. Our results show the feasibility of the\nproposed GANDiffFace, in particular the use of Diffusion models to enhance the\n(limited) intra-class variations provided by GANs towards the level of\nreal-world datasets.\n", "date": "2023-05-31", "categories": []}
{"id": "2305.19160", "title": "Recognizing People by Body Shape Using Deep Networks of Images and Words", "abstract": "  Common and important applications of person identification occur at distances\nand viewpoints in which the face is not visible or is not sufficiently resolved\nto be useful. We examine body shape as a biometric across distance and\nviewpoint variation. We propose an approach that combines standard object\nclassification networks with representations based on linguistic (word-based)\ndescriptions of bodies. Algorithms with and without linguistic training were\ncompared on their ability to identify people from body shape in images captured\nacross a large range of distances/views (close-range, 100m, 200m, 270m, 300m,\n370m, 400m, 490m, 500m, 600m, and at elevated pitch in images taken by an\nunmanned aerial vehicle [UAV]). Accuracy, as measured by identity-match ranking\nand false accept errors in an open-set test, was surprisingly good. For\nidentity-ranking, linguistic models were more accurate for close-range images,\nwhereas non-linguistic models fared better at intermediary distances. Fusion of\nthe linguistic and non-linguistic embeddings improved performance at all, but\nthe farthest distance. Although the non-linguistic model yielded fewer false\naccepts at all distances, fusion of the linguistic and non-linguistic models\ndecreased false accepts for all, but the UAV images. We conclude that\nlinguistic and non-linguistic representations of body shape can offer\ncomplementary identity information for bodies that can improve identification\nin applications of interest.\n", "date": "2023-05-30", "categories": []}
{"id": "2208.05615", "title": "FIGO: Enhanced Fingerprint Identification Approach Using GAN and One\n  Shot Learning Techniques", "abstract": "  Fingerprint evidence plays an important role in a criminal investigation for\nthe identification of individuals. Although various techniques have been\nproposed for fingerprint classification and feature extraction, automated\nfingerprint identification of fingerprints is still in its earliest stage. The\nperformance of traditional \\textit{Automatic Fingerprint Identification System}\n(AFIS) depends on the presence of valid minutiae points and still requires\nhuman expert assistance in feature extraction and identification stages. Based\non this motivation, we propose a Fingerprint Identification approach based on\nGenerative adversarial network and One-shot learning techniques (FIGO). Our\nsolution contains two components: fingerprint enhancement tier and fingerprint\nidentification tier. First, we propose a Pix2Pix model to transform low-quality\nfingerprint images to a higher level of fingerprint images pixel by pixel\ndirectly in the fingerprint enhancement tier. With the proposed enhancement\nalgorithm, the fingerprint identification model's performance is significantly\nimproved. Furthermore, we develop another existing solution based on Gabor\nfilters as a benchmark to compare with the proposed model by observing the\nfingerprint device's recognition accuracy. Experimental results show that our\nproposed Pix2pix model has better support than the baseline approach for\nfingerprint identification. Second, we construct a fully automated fingerprint\nfeature extraction model using a one-shot learning approach to differentiate\neach fingerprint from the others in the fingerprint identification process. Two\ntwin convolutional neural networks (CNNs) with shared weights and parameters\nare used to obtain the feature vectors in this process. Using the proposed\nmethod, we demonstrate that it is possible to learn necessary information from\nonly one training sample with high accuracy.\n", "date": "2022-08-11", "categories": []}
{"id": "2305.15762", "title": "Dynamic Enhancement Network for Partial Multi-modality Person\n  Re-identification", "abstract": "  Many existing multi-modality studies are based on the assumption of modality\nintegrity. However, the problem of missing arbitrary modalities is very common\nin real life, and this problem is less studied, but actually important in the\ntask of multi-modality person re-identification (Re-ID). To this end, we design\na novel dynamic enhancement network (DENet), which allows missing arbitrary\nmodalities while maintaining the representation ability of multiple modalities,\nfor partial multi-modality person Re-ID. To be specific, the multi-modal\nrepresentation of the RGB, near-infrared (NIR) and thermal-infrared (TIR)\nimages is learned by three branches, in which the information of missing\nmodalities is recovered by the feature transformation module. Since the missing\nstate might be changeable, we design a dynamic enhancement module, which\ndynamically enhances modality features according to the missing state in an\nadaptive manner, to improve the multi-modality representation. Extensive\nexperiments on multi-modality person Re-ID dataset RGBNT201 and vehicle Re-ID\ndataset RGBNT100 comparing to the state-of-the-art methods verify the\neffectiveness of our method in complex and changeable environments.\n", "date": "2023-05-25", "categories": []}
{"id": "2305.14601", "title": "FaceFusion: Exploiting Full Spectrum of Multiple Datasets", "abstract": "  The size of training dataset is known to be among the most dominating aspects\nof training high-performance face recognition embedding model. Building a large\ndataset from scratch could be cumbersome and time-intensive, while combining\nmultiple already-built datasets poses the risk of introducing large amount of\nlabel noise. We present a novel training method, named FaceFusion. It creates a\nfused view of different datasets that is untainted by identity conflicts, while\nconcurrently training an embedding network using the view in an end-to-end\nfashion. Using the unified view of combined datasets enables the embedding\nnetwork to be trained against the entire spectrum of the datasets, leading to a\nnoticeable performance boost. Extensive experiments confirm superiority of our\nmethod, whose performance in public evaluation datasets surpasses not only that\nof using a single training dataset, but also that of previously known methods\nunder various training circumstances.\n", "date": "2023-05-24", "categories": []}
{"id": "2305.13605", "title": "Adaptive Face Recognition Using Adversarial Information Network", "abstract": "  In many real-world applications, face recognition models often degenerate\nwhen training data (referred to as source domain) are different from testing\ndata (referred to as target domain). To alleviate this mismatch caused by some\nfactors like pose and skin tone, the utilization of pseudo-labels generated by\nclustering algorithms is an effective way in unsupervised domain adaptation.\nHowever, they always miss some hard positive samples. Supervision on\npseudo-labeled samples attracts them towards their prototypes and would cause\nan intra-domain gap between pseudo-labeled samples and the remaining unlabeled\nsamples within target domain, which results in the lack of discrimination in\nface recognition. In this paper, considering the particularity of face\nrecognition, we propose a novel adversarial information network (AIN) to\naddress it. First, a novel adversarial mutual information (MI) loss is proposed\nto alternately minimize MI with respect to the target classifier and maximize\nMI with respect to the feature extractor. By this min-max manner, the positions\nof target prototypes are adaptively modified which makes unlabeled images\nclustered more easily such that intra-domain gap can be mitigated. Second, to\nassist adversarial MI loss, we utilize a graph convolution network to predict\nlinkage likelihoods between target data and generate pseudo-labels. It\nleverages valuable information in the context of nodes and can achieve more\nreliable results. The proposed method is evaluated under two scenarios, i.e.,\ndomain adaptation across poses and image conditions, and domain adaptation\nacross faces with different skin tones. Extensive experiments show that AIN\nsuccessfully improves cross-domain generalization and offers a new\nstate-of-the-art on RFW dataset.\n", "date": "2023-05-23", "categories": []}
{"id": "2305.10079", "title": "Face Recognition Using Synthetic Face Data", "abstract": "  In the field of deep learning applied to face recognition, securing\nlarge-scale, high-quality datasets is vital for attaining precise and reliable\nresults. However, amassing significant volumes of high-quality real data faces\nhurdles such as time limitations, financial burdens, and privacy issues.\nFurthermore, prevalent datasets are often impaired by racial biases and\nannotation inaccuracies. In this paper, we underscore the promising application\nof synthetic data, generated through rendering digital faces via our computer\ngraphics pipeline, in achieving competitive results with the state-of-the-art\non synthetic data across multiple benchmark datasets. By finetuning the\nmodel,we obtain results that rival those achieved when training with hundreds\nof thousands of real images (98.7% on LFW [1]). We further investigate the\ncontribution of adding intra-class variance factors (e.g., makeup, accessories,\nhaircuts) on model performance. Finally, we reveal the sensitivity of\npre-trained face recognition models to alternating specific parts of the face\nby leveraging the granular control capability in our platform.\n", "date": "2023-05-17", "categories": []}
{"id": "2305.09482", "title": "Your Identity is Your Behavior -- Continuous User Authentication based\n  on Machine Learning and Touch Dynamics", "abstract": "  The aim of this research paper is to look into the use of continuous\nauthentication with mobile touch dynamics, using three different algorithms:\nNeural Network, Extreme Gradient Boosting, and Support Vector Machine. Mobile\ndevices are constantly increasing in popularity in the world, today smartphone\nsubscriptions have surpassed 6 billion. Mobile touch dynamics refer to the\ndistinct patterns of how a user interacts with their mobile device, this\nincludes factors such as touch pressure, swipe speed, and touch duration.\nContinuous authentication refers to the process of continuously verifying a\nuser's identity while they are using a device, rather than just at the initial\nlogin. This research used a dataset of touch dynamics collected from 40\nsubjects using the LG V30+. The participants played four mobile games, PUBG,\nDiep.io, Slither, and Minecraft, for 10 minutes each game. The three algorithms\nwere trained and tested on the extracted dataset, and their performance was\nevaluated based on metrics such as accuracy, precision, false negative rate,\nand false positive rate. The results of the research showed that all three\nalgorithms were able to effectively classify users based on their individual\ntouch dynamics, with accuracy ranging from 80% to 95%. The Neural Network\nalgorithm performed the best, achieving the highest accuracy and precision\nscores, followed closely by XGBoost and SVC. The data shows that continuous\nauthentication using mobile touch dynamics has the potential to be a useful\nmethod for enhancing security and reducing the risk of unauthorized access to\npersonal devices. This research also notes the importance of choosing the\ncorrect algorithm for a given dataset and use case, as different algorithms may\nhave varying levels of performance depending on the specific task.\n", "date": "2023-04-24", "categories": []}
{"id": "2305.07495", "title": "Gallery Sampling for Robust and Fast Face Identification", "abstract": "  Deep learning methods have been achieved brilliant results in face\nrecognition. One of the important tasks to improve the performance is to\ncollect and label images as many as possible. However, labeling identities and\nchecking qualities of large image data are difficult task and mistakes cannot\nbe avoided in processing large data. Previous works have been trying to deal\nwith the problem only in training domain, however it can cause much serious\nproblem if the mistakes are in gallery data of face identification. We proposed\ngallery data sampling methods which are robust to outliers including wrong\nlabeled, low quality, and less-informative images and reduce searching time.\nThe proposed sampling-by-pruning and sampling-by-generating methods\nsignificantly improved face identification performance on our 5.4M web image\ndataset of celebrities. The proposed method achieved 0.0975 in terms of FNIR at\nFPIR=0.01, while conventional method showed 0.3891. The average number of\nfeature vectors for each individual gallery was reduced to 17.1 from 115.9 and\nit can provide much faster search. We also made experiments on public datasets\nand our method achieved 0.1314 and 0.0668 FNIRs at FPIR=0.01 on the\nCASIA-WebFace and MS1MV2, while the convectional method did 0.5446, and 0.1327,\nrespectively.\n", "date": "2023-05-12", "categories": []}
{"id": "2305.07602", "title": "ViT Unified: Joint Fingerprint Recognition and Presentation Attack\n  Detection", "abstract": "  A secure fingerprint recognition system must contain both a presentation\nattack (i.e., spoof) detection and recognition module in order to protect users\nagainst unwanted access by malicious users. Traditionally, these tasks would be\ncarried out by two independent systems; however, recent studies have\ndemonstrated the potential to have one unified system architecture in order to\nreduce the computational burdens on the system, while maintaining high\naccuracy. In this work, we leverage a vision transformer architecture for joint\nspoof detection and matching and report competitive results with\nstate-of-the-art (SOTA) models for both a sequential system (two ViT models\noperating independently) and a unified architecture (a single ViT model for\nboth tasks). ViT models are particularly well suited for this task as the ViT's\nglobal embedding encodes features useful for recognition, whereas the\nindividual, local embeddings are useful for spoof detection. We demonstrate the\ncapability of our unified model to achieve an average integrated matching (IM)\naccuracy of 98.87% across LivDet 2013 and 2015 CrossMatch sensors. This is\ncomparable to IM accuracy of 98.95% of our sequential dual-ViT system, but with\n~50% of the parameters and ~58% of the latency.\n", "date": "2023-05-12", "categories": []}
{"id": "2210.08178", "title": "Is Face Recognition Safe from Realizable Attacks?", "abstract": "  Face recognition is a popular form of biometric authentication and due to its\nwidespread use, attacks have become more common as well. Recent studies show\nthat Face Recognition Systems are vulnerable to attacks and can lead to\nerroneous identification of faces. Interestingly, most of these attacks are\nwhite-box, or they are manipulating facial images in ways that are not\nphysically realizable. In this paper, we propose an attack scheme where the\nattacker can generate realistic synthesized face images with subtle\nperturbations and physically realize that onto his face to attack black-box\nface recognition systems. Comprehensive experiments and analyses show that\nsubtle perturbations realized on attackers face can create successful attacks\non state-of-the-art face recognition systems in black-box settings. Our study\nexposes the underlying vulnerability posed by the Face Recognition Systems\nagainst realizable black-box attacks.\n", "date": "2022-10-15", "categories": []}
{"id": "2305.06307", "title": "Analysis of Adversarial Image Manipulations", "abstract": "  As virtual and physical identity grow increasingly intertwined, the\nimportance of privacy and security in the online sphere becomes paramount. In\nrecent years, multiple news stories have emerged of private companies scraping\nweb content and doing research with or selling the data. Images uploaded online\ncan be scraped without users' consent or knowledge. Users of social media\nplatforms whose images are scraped may be at risk of being identified in other\nuploaded images or in real-world identification situations. This paper\ninvestigates how simple, accessible image manipulation techniques affect the\naccuracy of facial recognition software in identifying an individual's various\nface images based on one unique image.\n", "date": "2023-05-10", "categories": []}
{"id": "2305.05161", "title": "Child Palm-ID: Contactless Palmprint Recognition for Children", "abstract": "  Effective distribution of nutritional and healthcare aid for children,\nparticularly infants and toddlers, in some of the least developed and most\nimpoverished countries of the world, is a major problem due to the lack of\nreliable identification documents. Biometric authentication technology has been\ninvestigated to address child recognition in the absence of reliable ID\ndocuments. We present a mobile-based contactless palmprint recognition system,\ncalled Child Palm-ID, which meets the requirements of usability, hygiene, cost,\nand accuracy for child recognition. Using a contactless child palmprint\ndatabase, Child-PalmDB1, consisting of 19,158 images from 1,020 unique palms\n(in the age range of 6 mos. to 48 mos.), we report a TAR=94.11% @ FAR=0.1%. The\nproposed Child Palm-ID system is also able to recognize adults, achieving a\nTAR=99.4% on the CASIA contactless palmprint database and a TAR=100% on the\nCOEP contactless adult palmprint database, both @ FAR=0.1%. These accuracies\nare competitive with the SOTA provided by COTS systems. Despite these high\naccuracies, we show that the TAR for time-separated child-palmprints is only\n78.1% @ FAR=0.1%.\n", "date": "2023-05-09", "categories": []}
{"id": "2301.13384", "title": "GaitSADA: Self-Aligned Domain Adaptation for mmWave Gait Recognition", "abstract": "  mmWave radar-based gait recognition is a novel user identification method\nthat captures human gait biometrics from mmWave radar return signals. This\ntechnology offers privacy protection and is resilient to weather and lighting\nconditions. However, its generalization performance is yet unknown and limits\nits practical deployment. To address this problem, in this paper, a\nnon-synthetic dataset is collected and analyzed to reveal the presence of\nspatial and temporal domain shifts in mmWave gait biometric data, which\nsignificantly impacts identification accuracy. To mitigate this issue, a novel\nself-aligned domain adaptation method called GaitSADA is proposed. GaitSADA\nimproves system generalization performance by using a two-stage semi-supervised\nmodel training approach. The first stage employs semi-supervised contrastive\nlearning to learn a compact gait representation from both source and target\ndomain data, aligning source-target domain distributions implicitly. The second\nstage uses semi-supervised consistency training with centroid alignment to\nfurther close source-target domain gap by pseudo-labelling the target-domain\nsamples, clustering together the samples belonging to the same class but from\ndifferent domains, and pushing the class centroid close to the weight vector of\neach class. Experiments show that GaitSADA outperforms representative domain\nadaptation methods with an improvement ranging from 15.41\\% to 26.32\\% on\naverage accuracy in low data regimes. Code and dataset will be available at\nhttps://exitudio.github.io/GaitSADA\n", "date": "2023-01-31", "categories": []}
{"id": "2305.04426", "title": "Improving 2D face recognition via fine-level facial depth generation and\n  RGB-D complementary feature learning", "abstract": "  Face recognition in complex scenes suffers severe challenges coming from\nperturbations such as pose deformation, ill illumination, partial occlusion.\nSome methods utilize depth estimation to obtain depth corresponding to RGB to\nimprove the accuracy of face recognition. However, the depth generated by them\nsuffer from image blur, which introduces noise in subsequent RGB-D face\nrecognition tasks. In addition, existing RGB-D face recognition methods are\nunable to fully extract complementary features. In this paper, we propose a\nfine-grained facial depth generation network and an improved multimodal\ncomplementary feature learning network. Extensive experiments on the Lock3DFace\ndataset and the IIIT-D dataset show that the proposed FFDGNet and I MCFLNet can\nimprove the accuracy of RGB-D face recognition while achieving the\nstate-of-the-art performance.\n", "date": "2023-05-08", "categories": []}
{"id": "2304.05753", "title": "Wild Face Anti-Spoofing Challenge 2023: Benchmark and Results", "abstract": "  Face anti-spoofing (FAS) is an essential mechanism for safeguarding the\nintegrity of automated face recognition systems. Despite substantial\nadvancements, the generalization of existing approaches to real-world\napplications remains challenging. This limitation can be attributed to the\nscarcity and lack of diversity in publicly available FAS datasets, which often\nleads to overfitting during training or saturation during testing. In terms of\nquantity, the number of spoof subjects is a critical determinant. Most datasets\ncomprise fewer than 2,000 subjects. With regard to diversity, the majority of\ndatasets consist of spoof samples collected in controlled environments using\nrepetitive, mechanical processes. This data collection methodology results in\nhomogenized samples and a dearth of scenario diversity. To address these\nshortcomings, we introduce the Wild Face Anti-Spoofing (WFAS) dataset, a\nlarge-scale, diverse FAS dataset collected in unconstrained settings. Our\ndataset encompasses 853,729 images of 321,751 spoof subjects and 529,571 images\nof 148,169 live subjects, representing a substantial increase in quantity.\nMoreover, our dataset incorporates spoof data obtained from the internet,\nspanning a wide array of scenarios and various commercial sensors, including 17\npresentation attacks (PAs) that encompass both 2D and 3D forms. This novel data\ncollection strategy markedly enhances FAS data diversity. Leveraging the WFAS\ndataset and Protocol 1 (Known-Type), we host the Wild Face Anti-Spoofing\nChallenge at the CVPR2023 workshop. Additionally, we meticulously evaluate\nrepresentative methods using Protocol 1 and Protocol 2 (Unknown-Type). Through\nan in-depth examination of the challenge outcomes and benchmark baselines, we\nprovide insightful analyses and propose potential avenues for future research.\nThe dataset is released under Insightface.\n", "date": "2023-04-12", "categories": []}
{"id": "2305.03264", "title": "Robust Face Morphing Attack Detection Using Fusion of Multiple Features\n  and Classification Techniques", "abstract": "  Face Recognition System (FRS) are shown to be vulnerable to morphed images of\nnewborns. Detecting morphing attacks stemming from face images of newborn is\nimportant to avoid unwanted consequences, both for security and society. In\nthis paper, we present a new reference-based/Differential Morphing Attack\nDetection (MAD) method to detect newborn morphing images using Wavelet\nScattering Network (WSN). We propose a two-layer WSN with 250 $\\times$ 250\npixels and six rotations of wavelets per layer, resulting in 577 paths. The\nproposed approach is validated on a dataset of 852 bona fide images and 2460\nmorphing images constructed using face images of 42 unique newborns. The\nobtained results indicate a gain of over 10\\% in detection accuracy over other\nexisting D-MAD techniques.\n", "date": "2023-05-05", "categories": []}
{"id": "2305.02745", "title": "Age-Invariant Face Embedding using the Wasserstein Distance", "abstract": "  In this work, we study face verification in datasets where images of the same\nindividuals exhibit significant age differences. This poses a major challenge\nfor current face recognition and verification techniques. To address this\nissue, we propose a novel approach that utilizes multitask learning and a\nWasserstein distance discriminator to disentangle age and identity embeddings\nof facial images. Our approach employs multitask learning with a Wasserstein\ndistance discriminator that minimizes the mutual information between the age\nand identity embeddings by minimizing the Jensen-Shannon divergence. This\nimproves the encoding of age and identity information in face images and\nenhances the performance of face verification in age-variant datasets. We\nevaluate the effectiveness of our approach using multiple age-variant face\ndatasets and demonstrate its superiority over state-of-the-art methods in terms\nof face verification accuracy.\n", "date": "2023-05-04", "categories": []}
{"id": "2211.06761", "title": "Few-Shot Learning for Biometric Verification", "abstract": "  In machine learning applications, it is common practice to feed as much\ninformation as possible. In most cases, the model can handle large data sets\nthat allow to predict more accurately. In the presence of data scarcity, a\nFew-Shot learning (FSL) approach aims to build more accurate algorithms with\nlimited training data. We propose a novel end-to-end lightweight architecture\nthat verifies biometric data by producing competitive results as compared to\nstate-of-the-art accuracies through Few-Shot learning methods. The dense layers\nadd to the complexity of state-of-the-art deep learning models which inhibits\nthem to be used in low-power applications. In presented approach, a shallow\nnetwork is coupled with a conventional machine learning technique that exploits\nhand-crafted features to verify biometric images from multi-modal sources such\nas signatures, periocular region, iris, face, fingerprints etc. We introduce a\nself-estimated threshold that strictly monitors False Acceptance Rate (FAR)\nwhile generalizing its results hence eliminating user-defined thresholds from\nROC curves that are likely to be biased on local data distribution. This hybrid\nmodel benefits from few-shot learning to make up for scarcity of data in\nbiometric use-cases. We have conducted extensive experimentation with commonly\nused biometric datasets. The obtained results provided an effective solution\nfor biometric verification systems.\n", "date": "2022-11-12", "categories": []}
{"id": "2305.01021", "title": "Synthetic Data for Face Recognition: Current State and Future Prospects", "abstract": "  Over the past years, deep learning capabilities and the availability of\nlarge-scale training datasets advanced rapidly, leading to breakthroughs in\nface recognition accuracy. However, these technologies are foreseen to face a\nmajor challenge in the next years due to the legal and ethical concerns about\nusing authentic biometric data in AI model training and evaluation along with\nincreasingly utilizing data-hungry state-of-the-art deep learning models. With\nthe recent advances in deep generative models and their success in generating\nrealistic and high-resolution synthetic image data, privacy-friendly synthetic\ndata has been recently proposed as an alternative to privacy-sensitive\nauthentic data to overcome the challenges of using authentic data in face\nrecognition development. This work aims at providing a clear and structured\npicture of the use-cases taxonomy of synthetic face data in face recognition\nalong with the recent emerging advances of face recognition models developed on\nthe bases of synthetic data. We also discuss the challenges facing the use of\nsynthetic data in face recognition development and several future prospects of\nsynthetic data in the domain of face recognition.\n", "date": "2023-05-01", "categories": []}
{"id": "2305.01294", "title": "Differential Newborn Face Morphing Attack Detection using Wavelet\n  Scatter Network", "abstract": "  Face Recognition System (FRS) are shown to be vulnerable to morphed images of\nnewborns. Detecting morphing attacks stemming from face images of newborn is\nimportant to avoid unwanted consequences, both for security and society. In\nthis paper, we present a new reference-based/Differential Morphing Attack\nDetection (MAD) method to detect newborn morphing images using Wavelet\nScattering Network (WSN). We propose a two-layer WSN with 250 $\\times$ 250\npixels and six rotations of wavelets per layer, resulting in 577 paths. The\nproposed approach is validated on a dataset of 852 bona fide images and 2460\nmorphing images constructed using face images of 42 unique newborns. The\nobtained results indicate a gain of over 10\\% in detection accuracy over other\nexisting D-MAD techniques.\n", "date": "2023-05-02", "categories": []}
{"id": "2305.00320", "title": "Fusion for Visual-Infrared Person ReID in Real-World Surveillance Using\n  Corrupted Multimodal Data", "abstract": "  Visible-infrared person re-identification (V-I ReID) seeks to match images of\nindividuals captured over a distributed network of RGB and IR cameras. The task\nis challenging due to the significant differences between V and I modalities,\nespecially under real-world conditions, where images are corrupted by, e.g,\nblur, noise, and weather. Indeed, state-of-art V-I ReID models cannot leverage\ncorrupted modality information to sustain a high level of accuracy. In this\npaper, we propose an efficient model for multimodal V-I ReID -- named\nMultimodal Middle Stream Fusion (MMSF) -- that preserves modality-specific\nknowledge for improved robustness to corrupted multimodal images. In addition,\nthree state-of-art attention-based multimodal fusion models are adapted to\naddress corrupted multimodal data in V-I ReID, allowing to dynamically balance\neach modality importance. Recently, evaluation protocols have been proposed to\nassess the robustness of ReID models under challenging real-world scenarios.\nHowever, these protocols are limited to unimodal V settings. For realistic\nevaluation of multimodal (and cross-modal) V-I person ReID models, we propose\nnew challenging corrupted datasets for scenarios where V and I cameras are\nco-located (CL) and not co-located (NCL). Finally, the benefits of our Masking\nand Local Multimodal Data Augmentation (ML-MDA) strategy are explored to\nimprove the robustness of ReID models to multimodal corruption. Our experiments\non clean and corrupted versions of the SYSU-MM01, RegDB, and ThermalWORLD\ndatasets indicate the multimodal V-I ReID models that are more likely to\nperform well in real-world operational conditions. In particular, our ML-MDA is\nan important strategy for a V-I person ReID system to sustain high accuracy and\nrobustness when processing corrupted multimodal images. Also, our multimodal\nReID model MMSF outperforms every method under CL and NCL camera scenarios.\n", "date": "2023-04-29", "categories": []}
{"id": "2305.00552", "title": "Deep Learning-based Spatio Temporal Facial Feature Visual Speech\n  Recognition", "abstract": "  In low-resource computing contexts, such as smartphones and other tiny\ndevices, Both deep learning and machine learning are being used in a lot of\nidentification systems. as authentication techniques. The transparent,\ncontactless, and non-invasive nature of these face recognition technologies\ndriven by AI has led to their meteoric rise in popularity in recent years.\nWhile they are mostly successful, there are still methods to get inside without\npermission by utilising things like pictures, masks, glasses, etc. In this\nresearch, we present an alternate authentication process that makes use of both\nfacial recognition and the individual's distinctive temporal facial feature\nmotions while they speak a password. Because the suggested methodology allows\nfor a password to be specified in any language, it is not limited by language.\nThe suggested model attained an accuracy of 96.1% when tested on the\nindustry-standard MIRACL-VC1 dataset, demonstrating its efficacy as a reliable\nand powerful solution. In addition to being data-efficient, the suggested\ntechnique shows promising outcomes with as little as 10 positive video examples\nfor training the model. The effectiveness of the network's training is further\nproved via comparisons with other combined facial recognition and lip reading\nmodels.\n", "date": "2023-04-30", "categories": []}
{"id": "2305.00817", "title": "Racial Bias within Face Recognition: A Survey", "abstract": "  Facial recognition is one of the most academically studied and industrially\ndeveloped areas within computer vision where we readily find associated\napplications deployed globally. This widespread adoption has uncovered\nsignificant performance variation across subjects of different racial profiles\nleading to focused research attention on racial bias within face recognition\nspanning both current causation and future potential solutions. In support,\nthis study provides an extensive taxonomic review of research on racial bias\nwithin face recognition exploring every aspect and stage of the face\nrecognition processing pipeline. Firstly, we discuss the problem definition of\nracial bias, starting with race definition, grouping strategies, and the\nsocietal implications of using race or race-related groupings. Secondly, we\ndivide the common face recognition processing pipeline into four stages: image\nacquisition, face localisation, face representation, face verification and\nidentification, and review the relevant corresponding literature associated\nwith each stage. The overall aim is to provide comprehensive coverage of the\nracial bias problem with respect to each and every stage of the face\nrecognition processing pipeline whilst also highlighting the potential pitfalls\nand limitations of contemporary mitigation strategies that need to be\nconsidered within future research endeavours or commercial applications alike.\n", "date": "2023-05-01", "categories": []}
{"id": "2304.14795", "title": "Semi-Supervised RF Fingerprinting with Consistency-Based Regularization", "abstract": "  As a promising non-password authentication technology, radio frequency (RF)\nfingerprinting can greatly improve wireless security. Recent work has shown\nthat RF fingerprinting based on deep learning can significantly outperform\nconventional approaches. The superiority, however, is mainly attributed to\nsupervised learning using a large amount of labeled data, and it significantly\ndegrades if only limited labeled data is available, making many existing\nalgorithms lack practicability. Considering that it is often easier to obtain\nenough unlabeled data in practice with minimal resources, we leverage deep\nsemi-supervised learning for RF fingerprinting, which largely relies on a\ncomposite data augmentation scheme designed for radio signals, combined with\ntwo popular techniques: consistency-based regularization and pseudo-labeling.\nExperimental results on both simulated and real-world datasets demonstrate that\nour proposed method for semi-supervised RF fingerprinting is far superior to\nother competing ones, and it can achieve remarkable performance almost close to\nthat of fully supervised learning with a very limited number of examples.\n", "date": "2023-04-28", "categories": []}
{"id": "2304.14122", "title": "Deeply-Coupled Convolution-Transformer with Spatial-temporal\n  Complementary Learning for Video-based Person Re-identification", "abstract": "  Advanced deep Convolutional Neural Networks (CNNs) have shown great success\nin video-based person Re-Identification (Re-ID). However, they usually focus on\nthe most obvious regions of persons with a limited global representation\nability. Recently, it witnesses that Transformers explore the inter-patch\nrelations with global observations for performance improvements. In this work,\nwe take both sides and propose a novel spatial-temporal complementary learning\nframework named Deeply-Coupled Convolution-Transformer (DCCT) for\nhigh-performance video-based person Re-ID. Firstly, we couple CNNs and\nTransformers to extract two kinds of visual features and experimentally verify\ntheir complementarity. Further, in spatial, we propose a Complementary Content\nAttention (CCA) to take advantages of the coupled structure and guide\nindependent features for spatial complementary learning. In temporal, a\nHierarchical Temporal Aggregation (HTA) is proposed to progressively capture\nthe inter-frame dependencies and encode temporal information. Besides, a gated\nattention is utilized to deliver aggregated temporal information into the CNN\nand Transformer branches for temporal complementary learning. Finally, we\nintroduce a self-distillation training strategy to transfer the superior\nspatial-temporal knowledge to backbone networks for higher accuracy and more\nefficiency. In this way, two kinds of typical features from same videos are\nintegrated mechanically for more informative representations. Extensive\nexperiments on four public Re-ID benchmarks demonstrate that our framework\ncould attain better performances than most state-of-the-art methods.\n", "date": "2023-04-27", "categories": []}
{"id": "2304.14321", "title": "Person Re-ID through Unsupervised Hypergraph Rank Selection and Fusion", "abstract": "  Person Re-ID has been gaining a lot of attention and nowadays is of\nfundamental importance in many camera surveillance applications. The task\nconsists of identifying individuals across multiple cameras that have no\noverlapping views. Most of the approaches require labeled data, which is not\nalways available, given the huge amount of demanded data and the difficulty of\nmanually assigning a class for each individual. Recently, studies have shown\nthat re-ranking methods are capable of achieving significant gains, especially\nin the absence of labeled data. Besides that, the fusion of feature extractors\nand multiple-source training is another promising research direction not\nextensively exploited. We aim to fill this gap through a manifold rank\naggregation approach capable of exploiting the complementarity of different\nperson Re-ID rankers. In this work, we perform a completely unsupervised\nselection and fusion of diverse ranked lists obtained from multiple and diverse\nfeature extractors. Among the contributions, this work proposes a query\nperformance prediction measure that models the relationship among images\nconsidering a hypergraph structure and does not require the use of any labeled\ndata. Expressive gains were obtained in four datasets commonly used for person\nRe-ID. We achieved results competitive to the state-of-the-art in most of the\nscenarios.\n", "date": "2023-04-27", "categories": []}
{"id": "2304.13015", "title": "Flickr-PAD: New Face High-Resolution Presentation Attack Detection\n  Database", "abstract": "  Nowadays, Presentation Attack Detection is a very active research area.\nSeveral databases are constituted in the state-of-the-art using images\nextracted from videos. One of the main problems identified is that many\ndatabases present a low-quality, small image size and do not represent an\noperational scenario in a real remote biometric system. Currently, these images\nare captured from smartphones with high-quality and bigger resolutions. In\norder to increase the diversity of image quality, this work presents a new PAD\ndatabase based on open-access Flickr images called: \"Flickr-PAD\". Our new\nhand-made database shows high-quality printed and screen scenarios. This will\nhelp researchers to compare new approaches to existing algorithms on a wider\ndatabase. This database will be available for other researchers. A\nleave-one-out protocol was used to train and evaluate three PAD models based on\nMobileNet-V3 (small and large) and EfficientNet-B0. The best result was reached\nwith MobileNet-V3 large with BPCER10 of 7.08% and BPCER20 of 11.15%.\n", "date": "2023-04-25", "categories": []}
{"id": "2304.13021", "title": "Face Feature Visualisation of Single Morphing Attack Detection", "abstract": "  This paper proposes an explainable visualisation of different face feature\nextraction algorithms that enable the detection of bona fide and morphing\nimages for single morphing attack detection. The feature extraction is based on\nraw image, shape, texture, frequency and compression. This visualisation may\nhelp to develop a Graphical User Interface for border policies and specifically\nfor border guard personnel that have to investigate details of suspect images.\nA Random forest classifier was trained in a leave-one-out protocol on three\nlandmarks-based face morphing methods and a StyleGAN-based morphing method for\nwhich morphed images are available in the FRLL database. For morphing attack\ndetection, the Discrete Cosine-Transformation-based method obtained the best\nresults for synthetic images and BSIF for landmark-based image features.\n", "date": "2023-04-25", "categories": []}
{"id": "2304.11668", "title": "CoReFace: Sample-Guided Contrastive Regularization for Deep Face\n  Recognition", "abstract": "  The discriminability of feature representation is the key to open-set face\nrecognition. Previous methods rely on the learnable weights of the\nclassification layer that represent the identities. However, the evaluation\nprocess learns no identity representation and drops the classifier from\ntraining. This inconsistency could confuse the feature encoder in understanding\nthe evaluation goal and hinder the effect of identity-based methods. To\nalleviate the above problem, we propose a novel approach namely Contrastive\nRegularization for Face recognition (CoReFace) to apply image-level\nregularization in feature representation learning. Specifically, we employ\nsample-guided contrastive learning to regularize the training with the\nimage-image relationship directly, which is consistent with the evaluation\nprocess. To integrate contrastive learning into face recognition, we augment\nembeddings instead of images to avoid the image quality degradation. Then, we\npropose a novel contrastive loss for the representation distribution by\nincorporating an adaptive margin and a supervised contrastive mask to generate\nsteady loss values and avoid the collision with the classification supervision\nsignal. Finally, we discover and solve the semantically repetitive signal\nproblem in contrastive learning by exploring new pair coupling protocols.\nExtensive experiments demonstrate the efficacy and efficiency of our CoReFace\nwhich is highly competitive with the state-of-the-art approaches.\n", "date": "2023-04-23", "categories": []}
{"id": "2304.11685", "title": "Child Face Recognition at Scale: Synthetic Data Generation and\n  Performance Benchmark", "abstract": "  We address the need for a large-scale database of children's faces by using\ngenerative adversarial networks (GANs) and face age progression (FAP) models to\nsynthesize a realistic dataset referred to as HDA-SynChildFaces. To this end,\nwe proposed a processing pipeline that initially utilizes StyleGAN3 to sample\nadult subjects, which are subsequently progressed to children of varying ages\nusing InterFaceGAN. Intra-subject variations, such as facial expression and\npose, are created by further manipulating the subjects in their latent space.\nAdditionally, the presented pipeline allows to evenly distribute the races of\nsubjects, allowing to generate a balanced and fair dataset with respect to race\ndistribution. The created HDA-SynChildFaces consists of 1,652 subjects and a\ntotal of 188,832 images, each subject being present at various ages and with\nmany different intra-subject variations. Subsequently, we evaluates the\nperformance of various facial recognition systems on the generated database and\ncompare the results of adults and children at different ages. The study reveals\nthat children consistently perform worse than adults, on all tested systems,\nand the degradation in performance is proportional to age. Additionally, our\nstudy uncovers some biases in the recognition systems, with Asian and Black\nsubjects and females performing worse than White and Latino Hispanic subjects\nand males.\n", "date": "2023-04-23", "categories": []}
{"id": "2304.11725", "title": "Analysis of Recent Trends in Face Recognition Systems", "abstract": "  With the tremendous advancements in face recognition technology, face\nmodality has been widely recognized as a significant biometric identifier in\nestablishing a person's identity rather than any other biometric trait like\nfingerprints that require contact sensors. However, due to inter-class\nsimilarities and intra-class variations, face recognition systems generate\nfalse match and false non-match errors respectively. Recent research focuses on\nimproving the robustness of extracted features and the pre-processing\nalgorithms to enhance recognition accuracy. Since face recognition has been\nextensively used for several applications ranging from law enforcement to\nsurveillance systems, the accuracy and performance of face recognition must be\nthe finest. In this paper various face recognition systems are discussed and\nanalysed like RPRV, LWKPCA, SVM Model, LTrP based SPM and a deep learning\nframework for recognising images from CCTV. All these face recognition methods,\ntheir implementations and performance evaluations are compared to derive the\nbest outcome for future developmental works.\n", "date": "2023-04-23", "categories": []}
{"id": "2211.12483", "title": "PIC-Score: Probabilistic Interpretable Comparison Score for Optimal\n  Matching Confidence in Single- and Multi-Biometric (Face) Recognition", "abstract": "  In the context of biometrics, matching confidence refers to the confidence\nthat a given matching decision is correct. Since many biometric systems operate\nin critical decision-making processes, such as in forensics investigations,\naccurately and reliably stating the matching confidence becomes of high\nimportance. Previous works on biometric confidence estimation can well\ndifferentiate between high and low confidence, but lack interpretability.\nTherefore, they do not provide accurate probabilistic estimates of the\ncorrectness of a decision. In this work, we propose a probabilistic\ninterpretable comparison (PIC) score that accurately reflects the probability\nthat the score originates from samples of the same identity. We prove that the\nproposed approach provides optimal matching confidence. Contrary to other\napproaches, it can also optimally combine multiple samples in a joint PIC score\nwhich further increases the recognition and confidence estimation performance.\nIn the experiments, the proposed PIC approach is compared against all biometric\nconfidence estimation methods available on four publicly available databases\nand five state-of-the-art face recognition systems. The results demonstrate\nthat PIC has a significantly more accurate probabilistic interpretation than\nsimilar approaches and is highly effective for multi-biometric recognition. The\ncode is publicly-available.\n", "date": "2022-11-22", "categories": []}
{"id": "2304.10066", "title": "Recognizability Embedding Enhancement for Very Low-Resolution Face\n  Recognition and Quality Estimation", "abstract": "  Very low-resolution face recognition (VLRFR) poses unique challenges, such as\ntiny regions of interest and poor resolution due to extreme standoff distance\nor wide viewing angle of the acquisition devices. In this paper, we study\nprincipled approaches to elevate the recognizability of a face in the embedding\nspace instead of the visual quality. We first formulate a robust learning-based\nface recognizability measure, namely recognizability index (RI), based on two\ncriteria: (i) proximity of each face embedding against the unrecognizable faces\ncluster center and (ii) closeness of each face embedding against its positive\nand negative class prototypes. We then devise an index diversion loss to push\nthe hard-to-recognize face embedding with low RI away from unrecognizable faces\ncluster to boost the RI, which reflects better recognizability. Additionally, a\nperceptibility attention mechanism is introduced to attend to the most\nrecognizable face regions, which offers better explanatory and discriminative\ntraits for embedding learning. Our proposed model is trained end-to-end and\nsimultaneously serves recognizability-aware embedding learning and face quality\nestimation. To address VLRFR, our extensive evaluations on three challenging\nlow-resolution datasets and face quality assessment demonstrate the superiority\nof the proposed model over the state-of-the-art methods.\n", "date": "2023-04-20", "categories": []}
{"id": "2206.01881", "title": "Face Recognition Accuracy Across Demographics: Shining a Light Into the\n  Problem", "abstract": "  We explore varying face recognition accuracy across demographic groups as a\nphenomenon partly caused by differences in face illumination. We observe that\nfor a common operational scenario with controlled image acquisition, there is a\nlarge difference in face region brightness between African-American and\nCaucasian, and also a smaller difference between male and female. We show that\nimpostor image pairs with both faces under-exposed, or both overexposed, have\nan increased false match rate (FMR). Conversely, image pairs with strongly\ndifferent face brightness have a decreased similarity measure. We propose a\nbrightness information metric to measure variation in brightness in the face\nand show that face brightness that is too low or too high has reduced\ninformation in the face region, providing a cause for the lower accuracy. Based\non this, for operational scenarios with controlled image acquisition,\nillumination should be adjusted for each individual to obtain appropriate face\nimage brightness. This is the first work that we are aware of to explore how\nthe level of brightness of the skin region in a pair of face images (rather\nthan a single image) impacts face recognition accuracy, and to evaluate this as\na systematic factor causing unequal accuracy across demographics. The code is\nat https://github.com/HaiyuWu/FaceBrightness.\n", "date": "2022-06-04", "categories": []}
{"id": "2212.10108", "title": "Efficient aggregation of face embeddings for decentralized face\n  recognition deployments (extended version)", "abstract": "  Biometrics are one of the most privacy-sensitive data. Ubiquitous\nauthentication systems with a focus on privacy favor decentralized approaches\nas they reduce potential attack vectors, both on a technical and organizational\nlevel. The gold standard is to let the user be in control of where their own\ndata is stored, which consequently leads to a high variety of devices used.\nMoreover, in comparison with a centralized system, designs with higher end-user\nfreedom often incur additional network overhead. Therefore, when using face\nrecognition for biometric authentication, an efficient way to compare faces is\nimportant in practical deployments, because it reduces both network and\nhardware requirements that are essential to encourage device diversity. This\npaper proposes an efficient way to aggregate embeddings used for face\nrecognition based on an extensive analysis on different datasets and the use of\ndifferent aggregation strategies. As part of this analysis, a new dataset has\nbeen collected, which is available for research purposes. Our proposed method\nsupports the construction of massively scalable, decentralized face recognition\nsystems with a focus on both privacy and long-term usability.\n", "date": "2022-12-20", "categories": []}
{"id": "2304.07522", "title": "ID2image: Leakage of non-ID information into face descriptors and\n  inversion from descriptors to images", "abstract": "  Embedding a face image to a descriptor vector using a deep CNN is a widely\nused technique in face recognition. Via several possible training strategies,\nsuch embeddings are supposed to capture only identity information. Information\nabout the environment (such as background and lighting) or changeable aspects\nof the face (such as pose, expression, presence of glasses, hat etc.) should be\ndiscarded since they are not useful for recognition. In this paper, we present\na surprising result that this is not the case. We show that non-ID attributes,\nas well as landmark positions and the image histogram can be recovered from the\nID embedding of state-of-the-art face embedding networks (VGGFace2 and\nArcFace). In fact, these non-ID attributes can be predicted from ID embeddings\nwith similar accuracy to a prediction from the original image. Going further,\nwe present an optimisation strategy that uses a generative model (specifically\nStyleGAN2 for faces) to recover images from an ID embedding. We show\nphotorealistic inversion from ID embedding to face image in which not only is\nthe ID realistically reconstructed but the pose, lighting and\nbackground/apparel to some extent as well.\n", "date": "2023-04-15", "categories": []}
{"id": "2304.07580", "title": "Surveillance Face Presentation Attack Detection Challenge", "abstract": "  Face Anti-spoofing (FAS) is essential to secure face recognition systems from\nvarious physical attacks. However, most of the studies lacked consideration of\nlong-distance scenarios. Specifically, compared with FAS in traditional scenes\nsuch as phone unlocking, face payment, and self-service security inspection,\nFAS in long-distance such as station squares, parks, and self-service\nsupermarkets are equally important, but it has not been sufficiently explored\nyet. In order to fill this gap in the FAS community, we collect a large-scale\nSurveillance High-Fidelity Mask (SuHiFiMask). SuHiFiMask contains $10,195$\nvideos from $101$ subjects of different age groups, which are collected by $7$\nmainstream surveillance cameras. Based on this dataset and protocol-$3$ for\nevaluating the robustness of the algorithm under quality changes, we organized\na face presentation attack detection challenge in surveillance scenarios. It\nattracted 180 teams for the development phase with a total of 37 teams\nqualifying for the final round. The organization team re-verified and re-ran\nthe submitted code and used the results as the final ranking. In this paper, we\npresent an overview of the challenge, including an introduction to the dataset\nused, the definition of the protocol, the evaluation metrics, and the\nannouncement of the competition results. Finally, we present the top-ranked\nalgorithms and the research ideas provided by the competition for attack\ndetection in long-range surveillance scenarios.\n", "date": "2023-04-15", "categories": []}
{"id": "2304.07721", "title": "A Novel end-to-end Framework for Occluded Pixel Reconstruction with\n  Spatio-temporal Features for Improved Person Re-identification", "abstract": "  Person re-identification is vital for monitoring and tracking crowd movement\nto enhance public security. However, re-identification in the presence of\nocclusion substantially reduces the performance of existing systems and is a\nchallenging area. In this work, we propose a plausible solution to this problem\nby developing effective occlusion detection and reconstruction framework for\nRGB images/videos consisting of Deep Neural Networks. Specifically, a CNN-based\nocclusion detection model classifies individual input frames, followed by a\nConv-LSTM and Autoencoder to reconstruct the occluded pixels corresponding to\nthe occluded frames for sequential (video) and non-sequential (image) data,\nrespectively. The quality of the reconstructed RGB frames is further refined\nand fine-tuned using a Conditional Generative Adversarial Network (cGAN). Our\nmethod is evaluated on four well-known public data sets of the domain, and the\nqualitative reconstruction results are indeed appealing. Quantitative\nevaluation in terms of re-identification accuracy of the Siamese network showed\nan exceptional Rank-1 accuracy after occluded pixel reconstruction on various\ndatasets. A comparative analysis with state-of-the-art approaches also\ndemonstrates the robustness of our work for use in real-life surveillance\nsystems.\n", "date": "2023-04-16", "categories": []}
{"id": "2304.07822", "title": "A Random-patch based Defense Strategy Against Physical Attacks for Face\n  Recognition Systems", "abstract": "  The physical attack has been regarded as a kind of threat against real-world\ncomputer vision systems. Still, many existing defense methods are only useful\nfor small perturbations attacks and can't detect physical attacks effectively.\nIn this paper, we propose a random-patch based defense strategy to robustly\ndetect physical attacks for Face Recognition System (FRS). Different from\nmainstream defense methods which focus on building complex deep neural networks\n(DNN) to achieve high recognition rate on attacks, we introduce a patch based\ndefense strategy to a standard DNN aiming to obtain robust detection models.\nExtensive experimental results on the employed datasets show the superiority of\nthe proposed defense method on detecting white-box attacks and adaptive attacks\nwhich attack both FRS and the defense method. Additionally, due to the\nsimpleness yet robustness of our method, it can be easily applied to the real\nworld face recognition system and extended to other defense methods to boost\nthe detection performance.\n", "date": "2023-04-16", "categories": []}
{"id": "2304.06723", "title": "Introduction to Presentation Attack Detection in Fingerprint Biometrics", "abstract": "  This chapter provides an introduction to Presentation Attack Detection (PAD)\nin fingerprint biometrics, also coined anti-spoofing, describes early\ndevelopments in this field, and briefly summarizes recent trends and open\nissues.\n", "date": "2023-02-14", "categories": []}
{"id": "2304.07060", "title": "DCFace: Synthetic Face Generation with Dual Condition Diffusion Model", "abstract": "  Generating synthetic datasets for training face recognition models is\nchallenging because dataset generation entails more than creating high fidelity\nimages. It involves generating multiple images of same subjects under different\nfactors (\\textit{e.g.}, variations in pose, illumination, expression, aging and\nocclusion) which follows the real image conditional distribution. Previous\nworks have studied the generation of synthetic datasets using GAN or 3D models.\nIn this work, we approach the problem from the aspect of combining subject\nappearance (ID) and external factor (style) conditions. These two conditions\nprovide a direct way to control the inter-class and intra-class variations. To\nthis end, we propose a Dual Condition Face Generator (DCFace) based on a\ndiffusion model. Our novel Patch-wise style extractor and Time-step dependent\nID loss enables DCFace to consistently produce face images of the same subject\nunder different styles with precise control. Face recognition models trained on\nsynthetic images from the proposed DCFace provide higher verification\naccuracies compared to previous works by $6.11\\%$ on average in $4$ out of $5$\ntest datasets, LFW, CFP-FP, CPLFW, AgeDB and CALFW. Code is available at\nhttps://github.com/mk-minchul/dcface\n", "date": "2023-04-14", "categories": []}
{"id": "2304.07175", "title": "Exploring Causes of Demographic Variations In Face Recognition Accuracy", "abstract": "  In recent years, media reports have called out bias and racism in face\nrecognition technology. We review experimental results exploring several\nspeculated causes for asymmetric cross-demographic performance. We consider\naccuracy differences as represented by variations in non-mated (impostor) and /\nor mated (genuine) distributions for 1-to-1 face matching. Possible causes\nexplored include differences in skin tone, face size and shape, imbalance in\nnumber of identities and images in the training data, and amount of face\nvisible in the test data (\"face pixels\"). We find that demographic differences\nin face pixel information of the test images appear to most directly impact the\nresultant differences in face recognition accuracy.\n", "date": "2023-04-14", "categories": []}
{"id": "2304.06118", "title": "Explanation of Face Recognition via Saliency Maps", "abstract": "  Despite the significant progress in face recognition in the past years, they\nare often treated as \"black boxes\" and have been criticized for lacking\nexplainability. It becomes increasingly important to understand the\ncharacteristics and decisions of deep face recognition systems to make them\nmore acceptable to the public. Explainable face recognition (XFR) refers to the\nproblem of interpreting why the recognition model matches a probe face with one\nidentity over others. Recent studies have explored use of visual saliency maps\nas an explanation, but they often lack a deeper analysis in the context of face\nrecognition. This paper starts by proposing a rigorous definition of\nexplainable face recognition (XFR) which focuses on the decision-making process\nof the deep recognition model. Following the new definition, a similarity-based\nRISE algorithm (S-RISE) is then introduced to produce high-quality visual\nsaliency maps. Furthermore, an evaluation approach is proposed to\nsystematically validate the reliability and accuracy of general visual\nsaliency-based XFR methods.\n", "date": "2023-04-12", "categories": []}
{"id": "2304.06249", "title": "Set-Based Face Recognition Beyond Disentanglement: Burstiness\n  Suppression With Variance Vocabulary", "abstract": "  Set-based face recognition (SFR) aims to recognize the face sets in the\nunconstrained scenario, where the appearance of same identity may change\ndramatically with extreme variances (e.g., illumination, pose, expression). We\nargue that the two crucial issues in SFR, the face quality and burstiness, are\nboth identity-irrelevant and variance-relevant. The quality and burstiness\nassessment are interfered with by the entanglement of identity, and the face\nrecognition is interfered with by the entanglement of variance. Thus we propose\nto separate the identity features with the variance features in a\nlight-weighted set-based disentanglement framework. Beyond disentanglement, the\nvariance features are fully utilized to indicate face quality and burstiness in\na set, rather than being discarded after training. To suppress face burstiness\nin the sets, we propose a vocabulary-based burst suppression (VBS) method which\nquantizes faces with a reference vocabulary. With interword and intra-word\nnormalization operations on the assignment scores, the face burtisness degrees\nare appropriately estimated. The extensive illustrations and experiments\ndemonstrate the effect of the disentanglement framework with VBS, which gets\nnew state-of-the-art on the SFR benchmarks. The code will be released at\nhttps://github.com/Liubinggunzu/set_burstiness.\n", "date": "2023-04-13", "categories": []}
{"id": "2304.05554", "title": "Learning Transferable Pedestrian Representation from Multimodal\n  Information Supervision", "abstract": "  Recent researches on unsupervised person re-identification~(reID) have\ndemonstrated that pre-training on unlabeled person images achieves superior\nperformance on downstream reID tasks than pre-training on ImageNet. However,\nthose pre-trained methods are specifically designed for reID and suffer\nflexible adaption to other pedestrian analysis tasks. In this paper, we propose\nVAL-PAT, a novel framework that learns transferable representations to enhance\nvarious pedestrian analysis tasks with multimodal information. To train our\nframework, we introduce three learning objectives, \\emph{i.e.,} self-supervised\ncontrastive learning, image-text contrastive learning and multi-attribute\nclassification. The self-supervised contrastive learning facilitates the\nlearning of the intrinsic pedestrian properties, while the image-text\ncontrastive learning guides the model to focus on the appearance information of\npedestrians.Meanwhile, multi-attribute classification encourages the model to\nrecognize attributes to excavate fine-grained pedestrian information. We first\nperform pre-training on LUPerson-TA dataset, where each image contains text and\nattribute annotations, and then transfer the learned representations to various\ndownstream tasks, including person reID, person attribute recognition and\ntext-based person search. Extensive experiments demonstrate that our framework\nfacilitates the learning of general pedestrian representations and thus leads\nto promising results on various pedestrian analysis tasks.\n", "date": "2023-04-12", "categories": []}
{"id": "2304.05561", "title": "On the Adversarial Inversion of Deep Biometric Representations", "abstract": "  Biometric authentication service providers often claim that it is not\npossible to reverse-engineer a user's raw biometric sample, such as a\nfingerprint or a face image, from its mathematical (feature-space)\nrepresentation. In this paper, we investigate this claim on the specific\nexample of deep neural network (DNN) embeddings. Inversion of DNN embeddings\nhas been investigated for explaining deep image representations or synthesizing\nnormalized images. Existing studies leverage full access to all layers of the\noriginal model, as well as all possible information on the original dataset.\nFor the biometric authentication use case, we need to investigate this under\nadversarial settings where an attacker has access to a feature-space\nrepresentation but no direct access to the exact original dataset nor the\noriginal learned model. Instead, we assume varying degree of attacker's\nbackground knowledge about the distribution of the dataset as well as the\noriginal learned model (architecture and training process). In these cases, we\nshow that the attacker can exploit off-the-shelf DNN models and public\ndatasets, to mimic the behaviour of the original learned model to varying\ndegrees of success, based only on the obtained representation and attacker's\nprior knowledge. We propose a two-pronged attack that first infers the original\nDNN by exploiting the model footprint on the embedding, and then reconstructs\nthe raw data by using the inferred model. We show the practicality of the\nattack on popular DNNs trained for two prominent biometric modalities, face and\nfingerprint recognition. The attack can effectively infer the original\nrecognition model (mean accuracy 83\\% for faces, 86\\% for fingerprints), and\ncan craft effective biometric reconstructions that are successfully\nauthenticated with 1-vs-1 authentication accuracy of up to 92\\% for some\nmodels.\n", "date": "2023-04-12", "categories": []}
{"id": "2207.13394", "title": "BeCAPTCHA-Type: Biometric Keystroke Data Generation for Improved Bot\n  Detection", "abstract": "  This work proposes a data driven learning model for the synthesis of\nkeystroke biometric data. The proposed method is compared with two statistical\napproaches based on Universal and User-dependent models. These approaches are\nvalidated on the bot detection task, using the keystroke synthetic data to\nimprove the training process of keystroke-based bot detection systems. Our\nexperimental framework considers a dataset with 136 million keystroke events\nfrom 168 thousand subjects. We have analyzed the performance of the three\nsynthesis approaches through qualitative and quantitative experiments.\nDifferent bot detectors are considered based on several supervised classifiers\n(Support Vector Machine, Random Forest, Gaussian Naive Bayes and a Long\nShort-Term Memory network) and a learning framework including human and\nsynthetic samples. The experiments demonstrate the realism of the synthetic\nsamples. The classification results suggest that in scenarios with large\nlabeled data, these synthetic samples can be detected with high accuracy.\nHowever, in few-shot learning scenarios it represents an important challenge.\nFurthermore, these results show the great potential of the presented models.\n", "date": "2022-07-27", "categories": []}
{"id": "2303.04681", "title": "Enhancing Low-resolution Face Recognition with Feature Similarity\n  Knowledge Distillation", "abstract": "  In this study, we introduce a feature knowledge distillation framework to\nimprove low-resolution (LR) face recognition performance using knowledge\nobtained from high-resolution (HR) images. The proposed framework transfers\ninformative features from an HR-trained network to an LR-trained network by\nreducing the distance between them. A cosine similarity measure was employed as\na distance metric to effectively align the HR and LR features. This approach\ndiffers from conventional knowledge distillation frameworks, which use the L_p\ndistance metrics and offer the advantage of converging well when reducing the\ndistance between features of different resolutions. Our framework achieved a 3%\nimprovement over the previous state-of-the-art method on the AgeDB-30 benchmark\nwithout bells and whistles, while maintaining a strong performance on HR\nimages. The effectiveness of cosine similarity as a distance metric was\nvalidated through statistical analysis, making our approach a promising\nsolution for real-world applications in which LR images are frequently\nencountered. The code and pretrained models are publicly available on\nhttps://github.com/gist-ailab/feature-similarity-KD.\n", "date": "2023-03-08", "categories": []}
{"id": "2304.04205", "title": "Shape-Erased Feature Learning for Visible-Infrared Person\n  Re-Identification", "abstract": "  Due to the modality gap between visible and infrared images with high visual\nambiguity, learning \\textbf{diverse} modality-shared semantic concepts for\nvisible-infrared person re-identification (VI-ReID) remains a challenging\nproblem. Body shape is one of the significant modality-shared cues for VI-ReID.\nTo dig more diverse modality-shared cues, we expect that erasing\nbody-shape-related semantic concepts in the learned features can force the ReID\nmodel to extract more and other modality-shared features for identification. To\nthis end, we propose shape-erased feature learning paradigm that decorrelates\nmodality-shared features in two orthogonal subspaces. Jointly learning\nshape-related feature in one subspace and shape-erased features in the\northogonal complement achieves a conditional mutual information maximization\nbetween shape-erased feature and identity discarding body shape information,\nthus enhancing the diversity of the learned representation explicitly.\nExtensive experiments on SYSU-MM01, RegDB, and HITSZ-VCM datasets demonstrate\nthe effectiveness of our method.\n", "date": "2023-04-09", "categories": []}
{"id": "2304.04462", "title": "Grouped Knowledge Distillation for Deep Face Recognition", "abstract": "  Compared with the feature-based distillation methods, logits distillation can\nliberalize the requirements of consistent feature dimension between teacher and\nstudent networks, while the performance is deemed inferior in face recognition.\nOne major challenge is that the light-weight student network has difficulty\nfitting the target logits due to its low model capacity, which is attributed to\nthe significant number of identities in face recognition. Therefore, we seek to\nprobe the target logits to extract the primary knowledge related to face\nidentity, and discard the others, to make the distillation more achievable for\nthe student network. Specifically, there is a tail group with near-zero values\nin the prediction, containing minor knowledge for distillation. To provide a\nclear perspective of its impact, we first partition the logits into two groups,\ni.e., Primary Group and Secondary Group, according to the cumulative\nprobability of the softened prediction. Then, we reorganize the Knowledge\nDistillation (KD) loss of grouped logits into three parts, i.e., Primary-KD,\nSecondary-KD, and Binary-KD. Primary-KD refers to distilling the primary\nknowledge from the teacher, Secondary-KD aims to refine minor knowledge but\nincreases the difficulty of distillation, and Binary-KD ensures the consistency\nof knowledge distribution between teacher and student. We experimentally found\nthat (1) Primary-KD and Binary-KD are indispensable for KD, and (2)\nSecondary-KD is the culprit restricting KD at the bottleneck. Therefore, we\npropose a Grouped Knowledge Distillation (GKD) that retains the Primary-KD and\nBinary-KD but omits Secondary-KD in the ultimate KD loss calculation. Extensive\nexperimental results on popular face recognition benchmarks demonstrate the\nsuperiority of proposed GKD over state-of-the-art methods.\n", "date": "2023-04-10", "categories": []}
{"id": "2304.02284", "title": "Gradient Attention Balance Network: Mitigating Face Recognition Racial\n  Bias via Gradient Attention", "abstract": "  Although face recognition has made impressive progress in recent years, we\nignore the racial bias of the recognition system when we pursue a high level of\naccuracy. Previous work found that for different races, face recognition\nnetworks focus on different facial regions, and the sensitive regions of\ndarker-skinned people are much smaller. Based on this discovery, we propose a\nnew de-bias method based on gradient attention, called Gradient Attention\nBalance Network (GABN). Specifically, we use the gradient attention map (GAM)\nof the face recognition network to track the sensitive facial regions and make\nthe GAMs of different races tend to be consistent through adversarial learning.\nThis method mitigates the bias by making the network focus on similar facial\nregions. In addition, we also use masks to erase the Top-N sensitive facial\nregions, forcing the network to allocate its attention to a larger facial\nregion. This method expands the sensitive region of darker-skinned people and\nfurther reduces the gap between GAM of darker-skinned people and GAM of\nCaucasians. Extensive experiments show that GABN successfully mitigates racial\nbias in face recognition and learns more balanced performance for people of\ndifferent races.\n", "date": "2023-04-05", "categories": []}
{"id": "2304.01537", "title": "PartMix: Regularization Strategy to Learn Part Discovery for\n  Visible-Infrared Person Re-identification", "abstract": "  Modern data augmentation using a mixture-based technique can regularize the\nmodels from overfitting to the training data in various computer vision\napplications, but a proper data augmentation technique tailored for the\npart-based Visible-Infrared person Re-IDentification (VI-ReID) models remains\nunexplored. In this paper, we present a novel data augmentation technique,\ndubbed PartMix, that synthesizes the augmented samples by mixing the part\ndescriptors across the modalities to improve the performance of part-based\nVI-ReID models. Especially, we synthesize the positive and negative samples\nwithin the same and across different identities and regularize the backbone\nmodel through contrastive learning. In addition, we also present an\nentropy-based mining strategy to weaken the adverse impact of unreliable\npositive and negative samples. When incorporated into existing part-based\nVI-ReID model, PartMix consistently boosts the performance. We conduct\nexperiments to demonstrate the effectiveness of our PartMix over the existing\nVI-ReID methods and provide ablation studies.\n", "date": "2023-04-04", "categories": []}
{"id": "2304.01585", "title": "Multi-Channel Time-Series Person and Soft-Biometric Identification", "abstract": "  Multi-channel time-series datasets are popular in the context of human\nactivity recognition (HAR). On-body device (OBD) recordings of human movements\nare often preferred for HAR applications not only for their reliability but as\nan approach for identity protection, e.g., in industrial settings.\nContradictory, the gait activity is a biometric, as the cyclic movement is\ndistinctive and collectable. In addition, the gait cycle has proven to contain\nsoft-biometric information of human groups, such as age and height. Though\ngeneral human movements have not been considered a biometric, they might\ncontain identity information. This work investigates person and soft-biometrics\nidentification from OBD recordings of humans performing different activities\nusing deep architectures. Furthermore, we propose the use of attribute\nrepresentation for soft-biometric identification. We evaluate the method on\nfour datasets of multi-channel time-series HAR, measuring the performance of a\nperson and soft-biometrics identification and its relation concerning performed\nactivities. We find that person identification is not limited to gait activity.\nThe impact of activities on the identification performance was found to be\ntraining and dataset specific. Soft-biometric based attribute representation\nshows promising results and emphasis the necessity of larger datasets.\n", "date": "2023-04-04", "categories": []}
{"id": "2304.01842", "title": "Evaluating Synthetic Pre-Training for Handwriting Processing Tasks", "abstract": "  In this work, we explore massive pre-training on synthetic word images for\nenhancing the performance on four benchmark downstream handwriting analysis\ntasks. To this end, we build a large synthetic dataset of word images rendered\nin several handwriting fonts, which offers a complete supervision signal. We\nuse it to train a simple convolutional neural network (ConvNet) with a fully\nsupervised objective. The vector representations of the images obtained from\nthe pre-trained ConvNet can then be considered as encodings of the handwriting\nstyle. We exploit such representations for Writer Retrieval, Writer\nIdentification, Writer Verification, and Writer Classification and demonstrate\nthat our pre-training strategy allows extracting rich representations of the\nwriters' style that enable the aforementioned tasks with competitive results\nwith respect to task-specific State-of-the-Art approaches.\n", "date": "2023-04-04", "categories": []}
{"id": "2304.00979", "title": "A Latent Fingerprint in the Wild Database", "abstract": "  Latent fingerprints are among the most important and widely used evidence in\ncrime scenes, digital forensics and law enforcement worldwide. Despite the\nnumber of advancements reported in recent works, we note that significant open\nissues such as independent benchmarking and lack of large-scale evaluation\ndatabases for improving the algorithms are inadequately addressed. The\navailable databases are mostly of semi-public nature, lack of acquisition in\nthe wild environment, and post-processing pipelines. Moreover, they do not\nrepresent a realistic capture scenario similar to real crime scenes, to\nbenchmark the robustness of the algorithms. Further, existing databases for\nlatent fingerprint recognition do not have a large number of unique\nsubjects/fingerprint instances or do not provide ground truth/reference\nfingerprint images to conduct a cross-comparison against the latent. In this\npaper, we introduce a new wild large-scale latent fingerprint database that\nincludes five different acquisition scenarios: reference fingerprints from (1)\noptical and (2) capacitive sensors, (3) smartphone fingerprints, latent\nfingerprints captured from (4) wall surface, (5) Ipad surface, and (6)\naluminium foil surface. The new database consists of 1,318 unique fingerprint\ninstances captured in all above mentioned settings. A total of 2,636 reference\nfingerprints from optical and capacitive sensors, 1,318 fingerphotos from\nsmartphones, and 9,224 latent fingerprints from each of the 132 subjects were\nprovided in this work. The dataset is constructed considering various age\ngroups, equal representations of genders and backgrounds. In addition, we\nprovide an extensive set of analysis of various subset evaluations to highlight\nopen challenges for future directions in latent fingerprint recognition\nresearch.\n", "date": "2023-04-03", "categories": []}
{"id": "2303.14481", "title": "Diverse Embedding Expansion Network and Low-Light Cross-Modality\n  Benchmark for Visible-Infrared Person Re-identification", "abstract": "  For the visible-infrared person re-identification (VIReID) task, one of the\nmajor challenges is the modality gaps between visible (VIS) and infrared (IR)\nimages. However, the training samples are usually limited, while the modality\ngaps are too large, which leads that the existing methods cannot effectively\nmine diverse cross-modality clues. To handle this limitation, we propose a\nnovel augmentation network in the embedding space, called diverse embedding\nexpansion network (DEEN). The proposed DEEN can effectively generate diverse\nembeddings to learn the informative feature representations and reduce the\nmodality discrepancy between the VIS and IR images. Moreover, the VIReID model\nmay be seriously affected by drastic illumination changes, while all the\nexisting VIReID datasets are captured under sufficient illumination without\nsignificant light changes. Thus, we provide a low-light cross-modality (LLCM)\ndataset, which contains 46,767 bounding boxes of 1,064 identities captured by 9\nRGB/IR cameras. Extensive experiments on the SYSU-MM01, RegDB and LLCM datasets\nshow the superiority of the proposed DEEN over several other state-of-the-art\nmethods. The code and dataset are released at: https://github.com/ZYK100/LLCM\n", "date": "2023-03-25", "categories": []}
{"id": "2303.14626", "title": "MRCN: A Novel Modality Restitution and Compensation Network for\n  Visible-Infrared Person Re-identification", "abstract": "  Visible-infrared person re-identification (VI-ReID), which aims to search\nidentities across different spectra, is a challenging task due to large\ncross-modality discrepancy between visible and infrared images. The key to\nreduce the discrepancy is to filter out identity-irrelevant interference and\neffectively learn modality-invariant person representations. In this paper, we\npropose a novel Modality Restitution and Compensation Network (MRCN) to narrow\nthe gap between the two modalities. Specifically, we first reduce the modality\ndiscrepancy by using two Instance Normalization (IN) layers. Next, to reduce\nthe influence of IN layers on removing discriminative information and to reduce\nmodality differences, we propose a Modality Restitution Module (MRM) and a\nModality Compensation Module (MCM) to respectively distill modality-irrelevant\nand modality-relevant features from the removed information. Then, the\nmodality-irrelevant features are used to restitute to the normalized visible\nand infrared features, while the modality-relevant features are used to\ncompensate for the features of the other modality. Furthermore, to better\ndisentangle the modality-relevant features and the modality-irrelevant\nfeatures, we propose a novel Center-Quadruplet Causal (CQC) loss to encourage\nthe network to effectively learn the modality-relevant features and the\nmodality-irrelevant features. Extensive experiments are conducted to validate\nthe superiority of our method on the challenging SYSU-MM01 and RegDB datasets.\nMore remarkably, our method achieves 95.1% in terms of Rank-1 and 89.2% in\nterms of mAP on the RegDB dataset.\n", "date": "2023-03-26", "categories": []}
{"id": "2303.13814", "title": "Multimodal Adaptive Fusion of Face and Gait Features using Keyless\n  attention based Deep Neural Networks for Human Identification", "abstract": "  Biometrics plays a significant role in vision-based surveillance\napplications. Soft biometrics such as gait is widely used with face in\nsurveillance tasks like person recognition and re-identification. Nevertheless,\nin practical scenarios, classical fusion techniques respond poorly to changes\nin individual users and in the external environment. To this end, we propose a\nnovel adaptive multi-biometric fusion strategy for the dynamic incorporation of\ngait and face biometric cues by leveraging keyless attention deep neural\nnetworks. Various external factors such as viewpoint and distance to the\ncamera, are investigated in this study. Extensive experiments have shown\nsuperior performanceof the proposed model compared with the state-of-the-art\nmodel.\n", "date": "2023-03-24", "categories": []}
{"id": "2303.14004", "title": "Vulnerability of Face Morphing Attacks: A Case Study on Lookalike and\n  Identical Twins", "abstract": "  Face morphing attacks have emerged as a potential threat, particularly in\nautomatic border control scenarios. Morphing attacks permit more than one\nindividual to use travel documents that can be used to cross borders using\nautomatic border control gates. The potential for morphing attacks depends on\nthe selection of data subjects (accomplice and malicious actors). This work\ninvestigates lookalike and identical twins as the source of face morphing\ngeneration. We present a systematic study on benchmarking the vulnerability of\nFace Recognition Systems (FRS) to lookalike and identical twin morphing images.\nTherefore, we constructed new face morphing datasets using 16 pairs of\nidentical twin and lookalike data subjects. Morphing images from lookalike and\nidentical twins are generated using a landmark-based method. Extensive\nexperiments are carried out to benchmark the attack potential of lookalike and\nidentical twins. Furthermore, experiments are designed to provide insights into\nthe impact of vulnerability with normal face morphing compared with lookalike\nand identical twin face morphing.\n", "date": "2023-03-24", "categories": []}
{"id": "2303.12742", "title": "Empirical Assessment of End-to-End Iris Recognition System Capacity", "abstract": "  Iris is an established modality in biometric recognition applications\nincluding consumer electronics, e-commerce, border security, forensics, and\nde-duplication of identity at a national scale. In light of the expanding usage\nof biometric recognition, identity clash (when templates from two different\npeople match) is an imperative factor of consideration for a system's\ndeployment. This study explores system capacity estimation by empirically\nestimating the constrained capacity of an end-to-end iris recognition system\n(NIR systems with Daugman-based feature extraction) operating at an acceptable\nerror rate i.e. the number of subjects a system can resolve before encountering\nan error. We study the impact of six system parameters on an iris recognition\nsystem's constrained capacity -- number of enrolled identities, image quality,\ntemplate dimension, random feature elimination, filter resolution, and system\noperating point. In our assessment, we analyzed 13.2 million comparisons from\n5158 unique identities for each of 24 different system configurations. This\nwork provides a framework to better understand iris recognition system capacity\nas a function of biometric system configurations beyond the operating point,\nfor large-scale applications.\n", "date": "2023-03-20", "categories": []}
{"id": "2303.13158", "title": "Improvement of Color Image Analysis Using a New Hybrid Face Recognition\n  Algorithm based on Discrete Wavelets and Chebyshev Polynomials", "abstract": "  This work is unique in the use of discrete wavelets that were built from or\nderived from Chebyshev polynomials of the second and third kind, filter the\nDiscrete Second Chebyshev Wavelets Transform (DSCWT), and derive two effective\nfilters. The Filter Discrete Third Chebyshev Wavelets Transform (FDTCWT) is\nused in the process of analyzing color images and removing noise and impurities\nthat accompany the image, as well as because of the large amount of data that\nmakes up the image as it is taken. These data are massive, making it difficult\nto deal with each other during transmission. However to address this issue, the\nimage compression technique is used, with the image not losing information due\nto the readings that were obtained, and the results were satisfactory. Mean\nSquare Error (MSE), Peak Signal Noise Ratio (PSNR), Bit Per Pixel (BPP), and\nCompression Ratio (CR) Coronavirus is the initial treatment, while the\nprocessing stage is done with network training for Convolutional Neural\nNetworks (CNN) with Discrete Second Chebeshev Wavelets Convolutional Neural\nNetwork (DSCWCNN) and Discrete Third Chebeshev Wavelets Convolutional Neural\nNetwork (DTCWCNN) to create an efficient algorithm for face recognition, and\nthe best results were achieved in accuracy and in the least amount of time. Two\nsamples of color images that were made or implemented were used. The proposed\ntheory was obtained with fast and good results; the results are evident shown\nin the tables below.\n", "date": "2023-03-23", "categories": []}
{"id": "2211.06597", "title": "OpenGait: Revisiting Gait Recognition Toward Better Practicality", "abstract": "  Gait recognition is one of the most critical long-distance identification\ntechnologies and increasingly gains popularity in both research and industry\ncommunities. Despite the significant progress made in indoor datasets, much\nevidence shows that gait recognition techniques perform poorly in the wild.\nMore importantly, we also find that some conclusions drawn from indoor datasets\ncannot be generalized to real applications. Therefore, the primary goal of this\npaper is to present a comprehensive benchmark study for better practicality\nrather than only a particular model for better performance. To this end, we\nfirst develop a flexible and efficient gait recognition codebase named\nOpenGait. Based on OpenGait, we deeply revisit the recent development of gait\nrecognition by re-conducting the ablative experiments. Encouragingly,we detect\nsome unperfect parts of certain prior woks, as well as new insights. Inspired\nby these discoveries, we develop a structurally simple, empirically powerful,\nand practically robust baseline model, GaitBase. Experimentally, we\ncomprehensively compare GaitBase with many current gait recognition methods on\nmultiple public datasets, and the results reflect that GaitBase achieves\nsignificantly strong performance in most cases regardless of indoor or outdoor\nsituations. Code is available at https://github.com/ShiqiYu/OpenGait.\n", "date": "2022-11-12", "categories": []}
{"id": "2303.12512", "title": "Sibling-Attack: Rethinking Transferable Adversarial Attacks against Face\n  Recognition", "abstract": "  A hard challenge in developing practical face recognition (FR) attacks is due\nto the black-box nature of the target FR model, i.e., inaccessible gradient and\nparameter information to attackers. While recent research took an important\nstep towards attacking black-box FR models through leveraging transferability,\ntheir performance is still limited, especially against online commercial FR\nsystems that can be pessimistic (e.g., a less than 50% ASR--attack success rate\non average). Motivated by this, we present Sibling-Attack, a new FR attack\ntechnique for the first time explores a novel multi-task perspective (i.e.,\nleveraging extra information from multi-correlated tasks to boost attacking\ntransferability). Intuitively, Sibling-Attack selects a set of tasks correlated\nwith FR and picks the Attribute Recognition (AR) task as the task used in\nSibling-Attack based on theoretical and quantitative analysis. Sibling-Attack\nthen develops an optimization framework that fuses adversarial gradient\ninformation through (1) constraining the cross-task features to be under the\nsame space, (2) a joint-task meta optimization framework that enhances the\ngradient compatibility among tasks, and (3) a cross-task gradient stabilization\nmethod which mitigates the oscillation effect during attacking. Extensive\nexperiments demonstrate that Sibling-Attack outperforms state-of-the-art FR\nattack techniques by a non-trivial margin, boosting ASR by 12.61% and 55.77% on\naverage on state-of-the-art pre-trained FR models and two well-known, widely\nused commercial FR systems.\n", "date": "2023-03-22", "categories": []}
{"id": "2303.12720", "title": "Longitudinal Performance of Iris Recognition in Children: Time Intervals\n  up to Six years", "abstract": "  The temporal stability of iris recognition performance is core to its success\nas a biometric modality. With the expanding horizon of applications for\nchildren, gaps in the knowledge base on the temporal stability of iris\nrecognition performance in children have impacted decision-making during\napplications at the global scale. This report presents the most extensive\nanalysis of longitudinal iris recognition performance in children with data\nfrom the same 230 children over 6.5 years between enrollment and query for ages\n4 to 17 years. Assessment of match scores, statistical modelling of variability\nfactors impacting match scores and in-depth assessment of the root causes of\nthe false rejections concludes no impact on iris recognition performance due to\naging.\n", "date": "2023-03-10", "categories": []}
{"id": "2207.06726", "title": "Octuplet Loss: Make Face Recognition Robust to Image Resolution", "abstract": "  Image resolution, or in general, image quality, plays an essential role in\nthe performance of today's face recognition systems. To address this problem,\nwe propose a novel combination of the popular triplet loss to improve\nrobustness against image resolution via fine-tuning of existing face\nrecognition models. With octuplet loss, we leverage the relationship between\nhigh-resolution images and their synthetically down-sampled variants jointly\nwith their identity labels. Fine-tuning several state-of-the-art approaches\nwith our method proves that we can significantly boost performance for\ncross-resolution (high-to-low resolution) face verification on various datasets\nwithout meaningfully exacerbating the performance on high-to-high resolution\nimages. Our method applied on the FaceTransformer network achieves 95.12% face\nverification accuracy on the challenging XQLFW dataset while reaching 99.73% on\nthe LFW database. Moreover, the low-to-low face verification accuracy benefits\nfrom our method. We release our code to allow seamless integration of the\noctuplet loss into existing frameworks.\n", "date": "2022-07-14", "categories": []}
{"id": "2211.11039", "title": "Deep Composite Face Image Attacks: Generation, Vulnerability and\n  Detection", "abstract": "  Face manipulation attacks have drawn the attention of biometric researchers\nbecause of their vulnerability to Face Recognition Systems (FRS). This paper\nproposes a novel scheme to generate Composite Face Image Attacks (CFIA) based\non facial attributes using Generative Adversarial Networks (GANs). Given the\nface images corresponding to two unique data subjects, the proposed CFIA method\nwill independently generate the segmented facial attributes, then blend them\nusing transparent masks to generate the CFIA samples. We generate $526$ unique\nCFIA combinations of facial attributes for each pair of contributory data\nsubjects. Extensive experiments are carried out on our newly generated CFIA\ndataset consisting of 1000 unique identities with 2000 bona fide samples and\n526000 CFIA samples, thus resulting in an overall 528000 face image samples.\n{{We present a sequence of experiments to benchmark the attack potential of\nCFIA samples using four different automatic FRS}}. We introduced a new metric\nnamed Generalized Morphing Attack Potential (G-MAP) to benchmark the\nvulnerability of generated attacks on FRS effectively. Additional experiments\nare performed on the representative subset of the CFIA dataset to benchmark\nboth perceptual quality and human observer response. Finally, the CFIA\ndetection performance is benchmarked using three different single image based\nface Morphing Attack Detection (MAD) algorithms. The source code of the\nproposed method together with CFIA dataset will be made publicly available:\n\\url{https://github.com/jagmohaniiit/LatentCompositionCode}\n", "date": "2022-11-20", "categories": []}
{"id": "2303.11034", "title": "Internal Structure Attention Network for Fingerprint Presentation Attack\n  Detection from Optical Coherence Tomography", "abstract": "  As a non-invasive optical imaging technique, optical coherence tomography\n(OCT) has proven promising for automatic fingerprint recognition system (AFRS)\napplications. Diverse approaches have been proposed for OCT-based fingerprint\npresentation attack detection (PAD). However, considering the complexity and\nvariety of PA samples, it is extremely challenging to increase the\ngeneralization ability with the limited PA dataset. To solve the challenge,\nthis paper presents a novel supervised learning-based PAD method, denoted as\nISAPAD, which applies prior knowledge to guide network training and enhance the\ngeneralization ability. The proposed dual-branch architecture can not only\nlearns global features from the OCT image, but also concentrate on layered\nstructure feature which comes from the internal structure attention module\n(ISAM). The simple yet effective ISAM enables the proposed network to obtain\nlayered segmentation features belonging only to Bonafide from noisy OCT volume\ndata directly. Combined with effective training strategies and PAD score\ngeneration rules, ISAPAD obtains optimal PAD performance in limited training\ndata. Domain generalization experiments and visualization analysis validate the\neffectiveness of the proposed method for OCT PAD.\n", "date": "2023-03-20", "categories": []}
{"id": "2303.08514", "title": "Deep Learning for Iris Recognition: A Review", "abstract": "  Iris recognition is a secure biometric technology known for its stability and\nprivacy. With no two irises being identical and little change throughout a\nperson's lifetime, iris recognition is considered more reliable and less\nsusceptible to external factors than other biometric recognition methods.\nUnlike traditional machine learning-based iris recognition methods, deep\nlearning technology does not rely on feature engineering and boasts excellent\nperformance. This paper collects 120 relevant papers to summarize the\ndevelopment of iris recognition based on deep learning. We first introduce the\nbackground of iris recognition and the motivation and contribution of this\nsurvey. Then, we present the common datasets widely used in iris recognition.\nAfter that, we summarize the key tasks involved in the process of iris\nrecognition based on deep learning technology, including identification,\nsegmentation, presentation attack detection, and localization. Finally, we\ndiscuss the challenges and potential development of iris recognition. This\nreview provides a comprehensive sight of the research of iris recognition based\non deep learning.\n", "date": "2023-03-15", "categories": []}
{"id": "2212.13906", "title": "DiP: Learning Discriminative Implicit Parts for Person Re-Identification", "abstract": "  In person re-identification (ReID) tasks, many works explore the learning of\npart features to improve the performance over global image features. Existing\nmethods explicitly extract part features by either using a hand-designed image\ndivision or keypoints obtained with external visual systems. In this work, we\npropose to learn Discriminative implicit Parts (DiPs) which are decoupled from\nexplicit body parts. Therefore, DiPs can learn to extract any discriminative\nfeatures that can benefit in distinguishing identities, which is beyond\npredefined body parts (such as accessories). Moreover, we propose a novel\nimplicit position to give a geometric interpretation for each DiP. The implicit\nposition can also serve as a learning signal to encourage DiPs to be more\nposition-equivariant with the identity in the image. Lastly, an additional DiP\nweighting is introduced to handle the invisible or occluded situation and\nfurther improve the feature representation of DiPs. Extensive experiments show\nthat the proposed method achieves state-of-the-art performance on multiple\nperson ReID benchmarks.\n", "date": "2022-12-24", "categories": []}
{"id": "2303.06330", "title": "PRSNet: A Masked Self-Supervised Learning Pedestrian Re-Identification\n  Method", "abstract": "  In recent years, self-supervised learning has attracted widespread academic\ndebate and addressed many of the key issues of computer vision. The present\nresearch focus is on how to construct a good agent task that allows for\nimproved network learning of advanced semantic information on images so that\nmodel reasoning is accelerated during pre-training of the current task. In\norder to solve the problem that existing feature extraction networks are\npre-trained on the ImageNet dataset and cannot extract the fine-grained\ninformation in pedestrian images well, and the existing pre-task of contrast\nself-supervised learning may destroy the original properties of pedestrian\nimages, this paper designs a pre-task of mask reconstruction to obtain a\npre-training model with strong robustness and uses it for the pedestrian\nre-identification task. The training optimization of the network is performed\nby improving the triplet loss based on the centroid, and the mask image is\nadded as an additional sample to the loss calculation, so that the network can\nbetter cope with the pedestrian matching in practical applications after the\ntraining is completed. This method achieves about 5% higher mAP on Marker1501\nand CUHK03 data than existing self-supervised learning pedestrian\nre-identification methods, and about 1% higher for Rank1, and ablation\nexperiments are conducted to demonstrate the feasibility of this method. Our\nmodel code is located at https://github.com/ZJieX/prsnet.\n", "date": "2023-03-11", "categories": []}
{"id": "2303.05459", "title": "Presentation Attack Detection with Advanced CNN Models for\n  Noncontact-based Fingerprint Systems", "abstract": "  Touch-based fingerprint biometrics is one of the most popular biometric\nmodalities with applications in several fields. Problems associated with\ntouch-based techniques such as the presence of latent fingerprints and hygiene\nissues due to many people touching the same surface motivated the community to\nlook for non-contact-based solutions. For the last few years, contactless\nfingerprint systems are on the rise and in demand because of the ability to\nturn any device with a camera into a fingerprint reader. Yet, before we can\nfully utilize the benefit of noncontact-based methods, the biometric community\nneeds to resolve a few concerns such as the resiliency of the system against\npresentation attacks. One of the major obstacles is the limited publicly\navailable data sets with inadequate spoof and live data. In this publication,\nwe have developed a Presentation attack detection (PAD) dataset of more than\n7500 four-finger images and more than 14,000 manually segmented\nsingle-fingertip images, and 10,000 synthetic fingertips (deepfakes). The PAD\ndataset was collected from six different Presentation Attack Instruments (PAI)\nof three different difficulty levels according to FIDO protocols, with five\ndifferent types of PAI materials, and different smartphone cameras with manual\nfocusing. We have utilized DenseNet-121 and NasNetMobile models and our\nproposed dataset to develop PAD algorithms and achieved PAD accuracy of Attack\npresentation classification error rate (APCER) 0.14\\% and Bonafide presentation\nclassification error rate (BPCER) 0.18\\%. We have also reported the test\nresults of the models against unseen spoof types to replicate uncertain\nreal-world testing scenarios.\n", "date": "2023-03-09", "categories": []}
{"id": "2303.02715", "title": "Deep Learning in the Field of Biometric Template Protection: An Overview", "abstract": "  Today, deep learning represents the most popular and successful form of\nmachine learning. Deep learning has revolutionised the field of pattern\nrecognition, including biometric recognition. Biometric systems utilising deep\nlearning have been shown to achieve auspicious recognition accuracy, surpassing\nhuman performance. Apart from said breakthrough advances in terms of biometric\nperformance, the use of deep learning was reported to impact different\ncovariates of biometrics such as algorithmic fairness, vulnerability to\nattacks, or template protection. Technologies of biometric template protection\nare designed to enable a secure and privacy-preserving deployment of\nbiometrics. In the recent past, deep learning techniques have been frequently\napplied in biometric template protection systems for various purposes. This\nwork provides an overview of how advances in deep learning take influence on\nthe field of biometric template protection. The interrelation between improved\nbiometric performance rates and security in biometric template protection is\nelaborated. Further, the use of deep learning for obtaining feature\nrepresentations that are suitable for biometric template protection is\ndiscussed. Novel methods that apply deep learning to achieve various goals of\nbiometric template protection are surveyed along with deep learning-based\nattacks.\n", "date": "2023-03-05", "categories": []}
{"id": "2302.14335", "title": "DC-Former: Diverse and Compact Transformer for Person Re-Identification", "abstract": "  In person re-identification (re-ID) task, it is still challenging to learn\ndiscriminative representation by deep learning, due to limited data. Generally\nspeaking, the model will get better performance when increasing the amount of\ndata. The addition of similar classes strengthens the ability of the classifier\nto identify similar identities, thereby improving the discrimination of\nrepresentation. In this paper, we propose a Diverse and Compact Transformer\n(DC-Former) that can achieve a similar effect by splitting embedding space into\nmultiple diverse and compact subspaces. Compact embedding subspace helps model\nlearn more robust and discriminative embedding to identify similar classes. And\nthe fusion of these diverse embeddings containing more fine-grained information\ncan further improve the effect of re-ID. Specifically, multiple class tokens\nare used in vision transformer to represent multiple embedding spaces. Then, a\nself-diverse constraint (SDC) is applied to these spaces to push them away from\neach other, which makes each embedding space diverse and compact. Further, a\ndynamic weight controller(DWC) is further designed for balancing the relative\nimportance among them during training. The experimental results of our method\nare promising, which surpass previous state-of-the-art methods on several\ncommonly used person re-ID benchmarks.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.14574", "title": "A Little Bit Attention Is All You Need for Person Re-Identification", "abstract": "  Person re-identification plays a key role in applications where a mobile\nrobot needs to track its users over a long period of time, even if they are\npartially unobserved for some time, in order to follow them or be available on\ndemand. In this context, deep-learning based real-time feature extraction on a\nmobile robot is often performed on special-purpose devices whose computational\nresources are shared for multiple tasks. Therefore, the inference speed has to\nbe taken into account. In contrast, person re-identification is often improved\nby architectural changes that come at the cost of significantly slowing down\ninference. Attention blocks are one such example. We will show that some\nwell-performing attention blocks used in the state of the art are subject to\ninference costs that are far too high to justify their use for mobile robotic\napplications. As a consequence, we propose an attention block that only\nslightly affects the inference speed while keeping up with much deeper networks\nor more complex attention blocks in terms of re-identification accuracy. We\nperform extensive neural architecture search to derive rules at which locations\nthis attention block should be integrated into the architecture in order to\nachieve the best trade-off between speed and accuracy. Finally, we confirm that\nthe best performing configuration on a re-identification benchmark also\nperforms well on an indoor robotic dataset.\n", "date": "2023-02-28", "categories": []}
{"id": "2302.13049", "title": "CASIA-Iris-Africa: A Large-scale African Iris Image Database", "abstract": "  Iris biometrics is a phenotypic biometric trait that has proven to be\nagnostic to human natural physiological changes. Research on iris biometrics\nhas progressed tremendously, partly due to publicly available iris databases.\nVarious databases have been available to researchers that address pressing iris\nbiometric challenges such as constraint, mobile, multispectral, synthetics,\nlong-distance, contact lenses, liveness detection, etc. However, these\ndatabases mostly contain subjects of Caucasian and Asian docents with very few\nAfricans. Despite many investigative studies on racial bias in face biometrics,\nvery few studies on iris biometrics have been published, mainly due to the lack\nof racially diverse large-scale databases containing sufficient iris samples of\nAfricans in the public domain. Furthermore, most of these databases contain a\nrelatively small number of subjects and labelled images. This paper proposes a\nlarge-scale African database named CASIA-Iris-Africa that can be used as a\ncomplementary database for the iris recognition community to mediate the effect\nof racial biases on Africans. The database contains 28,717 images of 1023\nAfrican subjects (2046 iris classes) with age, gender, and ethnicity attributes\nthat can be useful in demographically sensitive studies of Africans. Sets of\nspecific application protocols are incorporated with the database to ensure the\ndatabase's variability and scalability. Performance results of some open-source\nSOTA algorithms on the database are presented, which will serve as baseline\nperformances. The relatively poor performances of the baseline algorithms on\nthe proposed database despite better performance on other databases prove that\nracial biases exist in these iris recognition algorithms. The database will be\nmade available on our website: http://www.idealtest.org.\n", "date": "2023-02-25", "categories": []}
{"id": "2302.13286", "title": "Benchmarking of Cancelable Biometrics for Deep Templates", "abstract": "  In this paper, we benchmark several cancelable biometrics (CB) schemes on\ndifferent biometric characteristics. We consider BioHashing, Multi-Layer\nPerceptron (MLP) Hashing, Bloom Filters, and two schemes based on\nIndex-of-Maximum (IoM) Hashing (i.e., IoM-URP and IoM-GRP). In addition to the\nmentioned CB schemes, we introduce a CB scheme (as a baseline) based on\nuser-specific random transformations followed by binarization. We evaluate the\nunlinkability, irreversibility, and recognition performance (which are the\nrequired criteria by the ISO/IEC 24745 standard) of these CB schemes on deep\nlearning based templates extracted from different physiological and behavioral\nbiometric characteristics including face, voice, finger vein, and iris. In\naddition, we provide an open-source implementation of all the experiments\npresented to facilitate the reproducibility of our results.\n", "date": "2023-02-26", "categories": []}
{"id": "2302.13902", "title": "Language identification as improvement for lip-based biometric visual\n  systems", "abstract": "  Language has always been one of humanity's defining characteristics. Visual\nLanguage Identification (VLI) is a relatively new field of research that is\ncomplex and largely understudied. In this paper, we present a preliminary study\nin which we use linguistic information as a soft biometric trait to enhance the\nperformance of a visual (auditory-free) identification system based on lip\nmovement. We report a significant improvement in the identification performance\nof the proposed visual system as a result of the integration of these data\nusing a score-based fusion strategy. Methods of Deep and Machine Learning are\nconsidered and evaluated. To the experimentation purposes, the dataset called\nlaBial Articulation for the proBlem of the spokEn Language rEcognition\n(BABELE), consisting of eight different languages, has been created. It\nincludes a collection of different features of which the spoken language\nrepresents the most relevant, while each sample is also manually labelled with\ngender and age of the subjects.\n", "date": "2023-02-27", "categories": []}
{"id": "2302.12593", "title": "Effect of Lossy Compression Algorithms on Face Image Quality and\n  Recognition", "abstract": "  Lossy face image compression can degrade the image quality and the utility\nfor the purpose of face recognition. This work investigates the effect of lossy\nimage compression on a state-of-the-art face recognition model, and on multiple\nface image quality assessment models. The analysis is conducted over a range of\nspecific image target sizes. Four compression types are considered, namely\nJPEG, JPEG 2000, downscaled PNG, and notably the new JPEG XL format. Frontal\ncolor images from the ColorFERET database were used in a Region Of Interest\n(ROI) variant and a portrait variant. We primarily conclude that JPEG XL allows\nfor superior mean and worst case face recognition performance especially at\nlower target sizes, below approximately 5kB for the ROI variant, while there\nappears to be no critical advantage among the compression types at higher\ntarget sizes. Quality assessments from modern models correlate well overall\nwith the compression effect on face recognition performance.\n", "date": "2023-02-24", "categories": []}
{"id": "2302.09973", "title": "Advanced Image Quality Assessment for Hand- and Fingervein Biometrics", "abstract": "  Natural Scene Statistics commonly used in non-reference image quality\nmeasures and a deep learning based quality assessment approach are proposed as\nbiometric quality indicators for vasculature images. While NIQE and BRISQUE if\ntrained on common images with usual distortions do not work well for assessing\nvasculature pattern samples' quality, their variants being trained on high and\nlow quality vasculature sample data behave as expected from a biometric quality\nestimator in most cases (deviations from the overall trend occur for certain\ndatasets or feature extraction methods). The proposed deep learning based\nquality metric is capable of assigning the correct quality class to the\nvaculature pattern samples in most cases, independent of finger or hand vein\npatterns being assessed. The experiments were conducted on a total of 13\npublicly available finger and hand vein datasets and involve three distinct\ntemplate representations (two of them especially designed for vascular\nbiometrics). The proposed (trained) quality measures are compared to a several\nclassical quality metrics, with their achieved results underlining their\npromising behaviour.\n", "date": "2023-02-20", "categories": []}
{"id": "2302.10423", "title": "Criminal Investigation Tracker with Suspect Prediction using Machine\n  Learning", "abstract": "  An automated approach to identifying offenders in Sri Lanka would be better\nthan the current system. Obtaining information from eyewitnesses is one of the\nless reliable approaches and procedures still in use today. Automated criminal\nidentification has the ability to save lives, notwithstanding Sri Lankan\nculture's lack of awareness of the issue. Using cutting-edge technology like\nbiometrics to finish this task would be the most accurate strategy. The most\nnotable outcomes will be obtained by applying fingerprint and face recognition\nas biometric techniques. The main responsibilities will be image optimization\nand criminality. CCTV footage may be used to identify a person's fingerprint,\nidentify a person's face, and identify crimes involving weapons. Additionally,\nwe unveil a notification system and condense the police report to Additionally,\nto make it simpler for police officers to understand the essential points of\nthe crime, we develop a notification system and condense the police report.\nAdditionally, if an incident involving a weapon is detected, an automated\nnotice of the crime with all the relevant facts is sent to the closest police\nstation. The summarization of the police report is what makes this the most\noriginal. In order to improve the efficacy of the overall image, the system\nwill quickly and precisely identify the full crime scene, identify, and\nrecognize the suspects using their faces and fingerprints, and detect firearms.\nThis study provides a novel approach for crime prediction based on real-world\ndata, and criminality incorporation. A crime or occurrence should be reported\nto the appropriate agencies, and the suggested web application should be\nimproved further to offer a workable channel of communication.\n", "date": "2023-02-21", "categories": []}
{"id": "2211.13735", "title": "Explainable Model-Agnostic Similarity and Confidence in Face\n  Verification", "abstract": "  Recently, face recognition systems have demonstrated remarkable performances\nand thus gained a vital role in our daily life. They already surpass human face\nverification accountability in many scenarios. However, they lack explanations\nfor their predictions. Compared to human operators, typical face recognition\nnetwork system generate only binary decisions without further explanation and\ninsights into those decisions. This work focuses on explanations for face\nrecognition systems, vital for developers and operators. First, we introduce a\nconfidence score for those systems based on facial feature distances between\ntwo input images and the distribution of distances across a dataset. Secondly,\nwe establish a novel visualization approach to obtain more meaningful\npredictions from a face recognition system, which maps the distance deviation\nbased on a systematic occlusion of images. The result is blended with the\noriginal images and highlights similar and dissimilar facial regions. Lastly,\nwe calculate confidence scores and explanation maps for several\nstate-of-the-art face verification datasets and release the results on a web\nplatform. We optimize the platform for a user-friendly interaction and hope to\nfurther improve the understanding of machine learning decisions. The source\ncode is available on GitHub, and the web platform is publicly available at\nhttp://explainable-face-verification.ey.r.appspot.com.\n", "date": "2022-11-24", "categories": []}
{"id": "2302.09053", "title": "OTB-morph: One-Time Biometrics via Morphing", "abstract": "  Cancelable biometrics are a group of techniques to transform the input\nbiometric to an irreversible feature intentionally using a transformation\nfunction and usually a key in order to provide security and privacy in\nbiometric recognition systems. This transformation is repeatable enabling\nsubsequent biometric comparisons. This paper is introducing a new idea to\nexploit as a transformation function for cancelable biometrics aimed at\nprotecting the templates against iterative optimization attacks. Our proposed\nscheme is based on time-varying keys (random biometrics in our case) and\nmorphing transformations. An experimental implementation of the proposed scheme\nis given for face biometrics. The results confirm that the proposed approach is\nable to withstand against leakage attacks while improving the recognition\nperformance.\n", "date": "2023-02-17", "categories": []}
{"id": "2210.10864", "title": "Cluster and Aggregate: Face Recognition with Large Probe Set", "abstract": "  Feature fusion plays a crucial role in unconstrained face recognition where\ninputs (probes) comprise of a set of $N$ low quality images whose individual\nqualities vary. Advances in attention and recurrent modules have led to feature\nfusion that can model the relationship among the images in the input set.\nHowever, attention mechanisms cannot scale to large $N$ due to their quadratic\ncomplexity and recurrent modules suffer from input order sensitivity. We\npropose a two-stage feature fusion paradigm, Cluster and Aggregate, that can\nboth scale to large $N$ and maintain the ability to perform sequential\ninference with order invariance. Specifically, Cluster stage is a linear\nassignment of $N$ inputs to $M$ global cluster centers, and Aggregation stage\nis a fusion over $M$ clustered features. The clustered features play an\nintegral role when the inputs are sequential as they can serve as a\nsummarization of past features. By leveraging the order-invariance of\nincremental averaging operation, we design an update rule that achieves\nbatch-order invariance, which guarantees that the contributions of early image\nin the sequence do not diminish as time steps increase. Experiments on IJB-B\nand IJB-S benchmark datasets show the superiority of the proposed two-stage\nparadigm in unconstrained face recognition. Code and pretrained models are\navailable in https://github.com/mk-minchul/caface\n", "date": "2022-10-19", "categories": []}
{"id": "2302.08320", "title": "Introduction to Presentation Attacks in Signature Biometrics and Recent\n  Advances", "abstract": "  Applications based on biometric authentication have received a lot of\ninterest in the last years due to the breathtaking results obtained using\npersonal traits such as face or fingerprint. However, it is important not to\nforget that these biometric systems have to withstand different types of\npossible attacks. This chapter carries out an analysis of different\nPresentation Attack (PA) scenarios for on-line handwritten signature\nverification. The main contributions of this chapter are: i) an updated\noverview of representative methods for Presentation Attack Detection (PAD) in\nsignature biometrics; ii) a description of the different levels of PAs existing\nin on-line signature verification regarding the amount of information available\nto the impostor, as well as the training, effort, and ability to perform the\nforgeries; and iii) an evaluation of the system performance in signature\nbiometrics under different scenarios considering recent publicly available\nsignature databases, DeepSignDB and SVC2021_EvalDB. This work is in line with\nrecent efforts in the Common Criteria standardization community towards\nsecurity evaluation of biometric systems.\n", "date": "2023-02-16", "categories": []}
{"id": "2302.08498", "title": "An Omnidirectional Approach to Touch-based Continuous Authentication", "abstract": "  This paper focuses on how touch interactions on smartphones can provide a\ncontinuous user authentication service through behaviour captured by a\ntouchscreen. While efforts are made to advance touch-based behavioural\nauthentication, researchers often focus on gathering data, tuning classifiers,\nand enhancing performance by evaluating touch interactions in a sequence rather\nthan independently. However, such systems only work by providing data\nrepresenting distinct behavioural traits. The typical approach separates\nbehaviour into touch directions and creates multiple user profiles. This work\npresents an omnidirectional approach which outperforms the traditional method\nindependent of the touch direction - depending on optimal behavioural features\nand a balanced training set. Thus, we evaluate five behavioural feature sets\nusing the conventional approach against our direction-agnostic method while\ntesting several classifiers, including an Extra-Tree and Gradient Boosting\nClassifier, which is often overlooked. Results show that in comparison with the\ntraditional, an Extra-Trees classifier and the proposed approach are superior\nwhen combining strokes. However, the performance depends on the applied feature\nset. We find that the TouchAlytics feature set outperforms others when using\nour approach when combining three or more strokes. Finally, we highlight the\nimportance of reporting the mean area under the curve and equal error rate for\nsingle-stroke performance and varying the sequence of strokes separately.\n", "date": "2023-01-13", "categories": []}
{"id": "2302.07245", "title": "WSD: Wild Selfie Dataset for Face Recognition in Selfie Images", "abstract": "  With the rise of handy smart phones in the recent years, the trend of\ncapturing selfie images is observed. Hence efficient approaches are required to\nbe developed for recognising faces in selfie images. Due to the short distance\nbetween the camera and face in selfie images, and the different visual effects\noffered by the selfie apps, face recognition becomes more challenging with\nexisting approaches. A dataset is needed to be developed to encourage the study\nto recognize faces in selfie images. In order to alleviate this problem and to\nfacilitate the research on selfie face images, we develop a challenging Wild\nSelfie Dataset (WSD) where the images are captured from the selfie cameras of\ndifferent smart phones, unlike existing datasets where most of the images are\ncaptured in controlled environment. The WSD dataset contains 45,424 images from\n42 individuals (i.e., 24 female and 18 male subjects), which are divided into\n40,862 training and 4,562 test images. The average number of images per subject\nis 1,082 with minimum and maximum number of images for any subject are 518 and\n2,634, respectively. The proposed dataset consists of several challenges,\nincluding but not limited to augmented reality filtering, mirrored images,\nocclusion, illumination, scale, expressions, view-point, aspect ratio, blur,\npartial faces, rotation, and alignment. We compare the proposed dataset with\nexisting benchmark datasets in terms of different characteristics. The\ncomplexity of WSD dataset is also observed experimentally, where the\nperformance of the existing state-of-the-art face recognition methods is poor\non WSD dataset, compared to the existing datasets. Hence, the proposed WSD\ndataset opens up new challenges in the area of face recognition and can be\nbeneficial to the community to study the specific challenges related to selfie\nimages and develop improved methods for face recognition in selfie images.\n", "date": "2023-02-14", "categories": []}
{"id": "2211.14498", "title": "The Impact of Racial Distribution in Training Data on Face Recognition\n  Bias: A Closer Look", "abstract": "  Face recognition algorithms, when used in the real world, can be very useful,\nbut they can also be dangerous when biased toward certain demographics. So, it\nis essential to understand how these algorithms are trained and what factors\naffect their accuracy and fairness to build better ones. In this study, we shed\nsome light on the effect of racial distribution in the training data on the\nperformance of face recognition models. We conduct 16 different experiments\nwith varying racial distributions of faces in the training data. We analyze\nthese trained models using accuracy metrics, clustering metrics, UMAP\nprojections, face quality, and decision thresholds. We show that a uniform\ndistribution of races in the training datasets alone does not guarantee\nbias-free face recognition algorithms and how factors like face image quality\nplay a crucial role. We also study the correlation between the clustering\nmetrics and bias to understand whether clustering is a good indicator of bias.\nFinally, we introduce a metric called racial gradation to study the inter and\nintra race correlation in facial features and how they affect the learning\nability of the face recognition models. With this study, we try to bring more\nunderstanding to an essential element of face recognition training, the data. A\nbetter understanding of the impact of training data on the bias of face\nrecognition algorithms will aid in creating better datasets and, in turn,\nbetter face recognition systems.\n", "date": "2022-11-26", "categories": []}
{"id": "2302.05576", "title": "Sketch Less Face Image Retrieval: A New Challenge", "abstract": "  In some specific scenarios, face sketch was used to identify a person.\nHowever, drawing a complete face sketch often needs skills and takes time,\nwhich hinder its widespread applicability in the practice. In this study, we\nproposed a new task named sketch less face image retrieval (SLFIR), in which\nthe retrieval was carried out at each stroke and aim to retrieve the target\nface photo using a partial sketch with as few strokes as possible (see Fig.1).\nFirstly, we developed a method to generate the data of sketch with drawing\nprocess, and opened such dataset; Secondly, we proposed a two-stage method as\nthe baseline for SLFIR that (1) A triplet network, was first adopt to learn the\njoint embedding space shared between the complete sketch and its target face\nphoto; (2) Regarding the sketch drawing episode as a sequence, we designed a\nLSTM module to optimize the representation of the incomplete face sketch.\nExperiments indicate that the new framework can finish the retrieval using a\npartial or pool drawing sketch.\n", "date": "2023-02-11", "categories": []}
{"id": "2302.05753", "title": "DaliID: Distortion-Adaptive Learned Invariance for Identification Models", "abstract": "  In unconstrained scenarios, face recognition and person re-identification are\nsubject to distortions such as motion blur, atmospheric turbulence, or\nupsampling artifacts. To improve robustness in these scenarios, we propose a\nmethodology called Distortion-Adaptive Learned Invariance for Identification\n(DaliID) models. We contend that distortion augmentations, which degrade image\nquality, can be successfully leveraged to a greater degree than has been shown\nin the literature. Aided by an adaptive weighting schedule, a novel distortion\naugmentation is applied at severe levels during training. This training\nstrategy increases feature-level invariance to distortions and decreases domain\nshift to unconstrained scenarios. At inference, we use a magnitude-weighted\nfusion of features from parallel models to retain robustness across the range\nof images. DaliID models achieve state-of-the-art (SOTA) for both face\nrecognition and person re-identification on seven benchmark datasets, including\nIJB-S, TinyFace, DeepChange, and MSMT17. Additionally, we provide recaptured\nevaluation data at a distance of 750+ meters and further validate on real\nlong-distance face imagery.\n", "date": "2023-02-11", "categories": []}
{"id": "2302.03657", "title": "Toward Face Biometric De-identification using Adversarial Examples", "abstract": "  The remarkable success of face recognition (FR) has endangered the privacy of\ninternet users particularly in social media. Recently, researchers turned to\nuse adversarial examples as a countermeasure. In this paper, we assess the\neffectiveness of using two widely known adversarial methods (BIM and ILLC) for\nde-identifying personal images. We discovered, unlike previous claims in the\nliterature, that it is not easy to get a high protection success rate\n(suppressing identification rate) with imperceptible adversarial perturbation\nto the human visual system. Finally, we found out that the transferability of\nadversarial examples is highly affected by the training parameters of the\nnetwork with which they are generated.\n", "date": "2023-02-07", "categories": []}
{"id": "2302.02075", "title": "X-ReID: Cross-Instance Transformer for Identity-Level Person\n  Re-Identification", "abstract": "  Currently, most existing person re-identification methods use Instance-Level\nfeatures, which are extracted only from a single image. However, these\nInstance-Level features can easily ignore the discriminative information due to\nthe appearance of each identity varies greatly in different images. Thus, it is\nnecessary to exploit Identity-Level features, which can be shared across\ndifferent images of each identity. In this paper, we propose to promote\nInstance-Level features to Identity-Level features by employing cross-attention\nto incorporate information from one image to another of the same identity, thus\nmore unified and discriminative pedestrian information can be obtained. We\npropose a novel training framework named X-ReID. Specifically, a Cross\nIntra-Identity Instances module (IntraX) fuses different intra-identity\ninstances to transfer Identity-Level knowledge and make Instance-Level features\nmore compact. A Cross Inter-Identity Instances module (InterX) involves hard\npositive and hard negative instances to improve the attention response to the\nsame identity instead of different identity, which minimizes intra-identity\nvariation and maximizes inter-identity variation. Extensive experiments on\nbenchmark datasets show the superiority of our method over existing works.\nParticularly, on the challenging MSMT17, our proposed method gains 1.1% mAP\nimprovements when compared to the second place.\n", "date": "2023-02-04", "categories": []}
{"id": "2302.01512", "title": "Spectral Aware Softmax for Visible-Infrared Person Re-Identification", "abstract": "  Visible-infrared person re-identification (VI-ReID) aims to match specific\npedestrian images from different modalities. Although suffering an extra\nmodality discrepancy, existing methods still follow the softmax loss training\nparadigm, which is widely used in single-modality classification tasks. The\nsoftmax loss lacks an explicit penalty for the apparent modality gap, which\nadversely limits the performance upper bound of the VI-ReID task. In this\npaper, we propose the spectral-aware softmax (SA-Softmax) loss, which can fully\nexplore the embedding space with the modality information and has clear\ninterpretability. Specifically, SA-Softmax loss utilizes an asynchronous\noptimization strategy based on the modality prototype instead of the\nsynchronous optimization based on the identity prototype in the original\nsoftmax loss. To encourage a high overlapping between two modalities,\nSA-Softmax optimizes each sample by the prototype from another spectrum. Based\non the observation and analysis of SA-Softmax, we modify the SA-Softmax with\nthe Feature Mask and Absolute-Similarity Term to alleviate the ambiguous\noptimization during model training. Extensive experimental evaluations\nconducted on RegDB and SYSU-MM01 demonstrate the superior performance of the\nSA-Softmax over the state-of-the-art methods in such a cross-modality\ncondition.\n", "date": "2023-02-03", "categories": []}
{"id": "2302.01751", "title": "Motion ID: Human Authentication Approach", "abstract": "  We introduce a novel approach to user authentication called Motion ID. The\nmethod employs motion sensing provided by inertial measurement units (IMUs),\nusing it to verify the persons identity via short time series of IMU data\ncaptured by the mobile device. The paper presents two labeled datasets with\nunlock events: the first features IMU measurements, provided by six users who\ncontinuously collected data on six different smartphones for a period of 12\nweeks. The second one contains 50 hours of IMU data for one specific motion\npattern, provided by 101 users. Moreover, we present a two-stage user\nauthentication process that employs motion pattern identification and user\nverification and is based on data preprocessing and machine learning. The\nResults section details the assessment of the method proposed, comparing it\nwith existing biometric authentication methods and the Android biometric\nstandard. The method has demonstrated high accuracy, indicating that it could\nbe successfully used in combination with existing methods. Furthermore, the\nmethod exhibits significant promise as a standalone solution. We provide the\ndatasets to the scholarly community and share our project code.\n", "date": "2023-01-25", "categories": []}
{"id": "2302.00884", "title": "Exploring Invariant Representation for Visible-Infrared Person\n  Re-Identification", "abstract": "  Cross-spectral person re-identification, which aims to associate identities\nto pedestrians across different spectra, faces a main challenge of the modality\ndiscrepancy. In this paper, we address the problem from both image-level and\nfeature-level in an end-to-end hybrid learning framework named robust feature\nmining network (RFM). In particular, we observe that the reflective intensity\nof the same surface in photos shot in different wavelengths could be\ntransformed using a linear model. Besides, we show the variable linear factor\nacross the different surfaces is the main culprit which initiates the modality\ndiscrepancy. We integrate such a reflection observation into an image-level\ndata augmentation by proposing the linear transformation generator (LTG).\nMoreover, at the feature level, we introduce a cross-center loss to explore a\nmore compact intra-class distribution and modality-aware spatial attention to\ntake advantage of textured regions more efficiently. Experiment results on two\nstandard cross-spectral person re-identification datasets, i.e., RegDB and\nSYSU-MM01, have demonstrated state-of-the-art performance.\n", "date": "2023-02-02", "categories": []}
{"id": "2301.12046", "title": "Semantic Adversarial Attacks on Face Recognition through Significant\n  Attributes", "abstract": "  Face recognition is known to be vulnerable to adversarial face images.\nExisting works craft face adversarial images by indiscriminately changing a\nsingle attribute without being aware of the intrinsic attributes of the images.\nTo this end, we propose a new Semantic Adversarial Attack called SAA-StarGAN\nthat tampers with the significant facial attributes for each image. We predict\nthe most significant attributes by applying the cosine similarity or\nprobability score. The probability score method is based on training a Face\nVerification model for an attribute prediction task to obtain a class\nprobability score for each attribute. The prediction process will help craft\nadversarial face images more easily and efficiently, as well as improve the\nadversarial transferability. Then, we change the most significant facial\nattributes, with either one or more of the facial attributes for impersonation\nand dodging attacks in white-box and black-box settings. Experimental results\nshow that our method could generate diverse and realistic adversarial face\nimages meanwhile avoid affecting human perception of the face recognition.\nSAA-StarGAN achieves an 80.5% attack success rate against black-box models,\noutperforming existing methods by 35.5% under the impersonation attack.\nConcerning the black-box setting, SAA-StarGAN achieves high attack success\nrates on various models. The experiments confirm that predicting the most\nimportant attributes significantly affects the success of adversarial attacks\nin both white-box and black-box settings and could enhance the transferability\nof the crafted adversarial examples.\n", "date": "2023-01-28", "categories": []}
{"id": "2301.06678", "title": "Feature-based Image Matching for Identifying Individual K\\=ak\\=a", "abstract": "  This report investigates an unsupervised, feature-based image matching\npipeline for the novel application of identifying individual k\\=ak\\=a. Applied\nwith a similarity network for clustering, this addresses a weakness of current\nsupervised approaches to identifying individual birds which struggle to handle\nthe introduction of new individuals to the population. Our approach uses object\nlocalisation to locate k\\=ak\\=a within images and then extracts local features\nthat are invariant to rotation and scale. These features are matched between\nimages with nearest neighbour matching techniques and mismatch removal to\nproduce a similarity score for image match comparison. The results show that\nmatches obtained via the image matching pipeline achieve high accuracy of true\nmatches. We conclude that feature-based image matching could be used with a\nsimilarity network to provide a viable alternative to existing supervised\napproaches.\n", "date": "2023-01-17", "categories": []}
{"id": "2301.09461", "title": "Study on the identification limits of craniofacial superimposition", "abstract": "  Craniofacial Superimposition involves the superimposition of an image of a\nskull with a number of ante-mortem face images of an individual and the\nanalysis of their morphological correspondence. Despite being used for one\ncentury, it is not yet a mature and fully accepted technique due to the absence\nof solid scientific approaches, significant reliability studies, and\ninternational standards. In this paper we present a comprehensive\nexperimentation on the limitations of Craniofacial Superimposition as a\nforensic identification technique. The study involves different experiments\nover more than 1 Million comparisons performed by a landmark-based automatic\n3D/2D superimposition method. The total sample analyzed consists of 320\nsubjects and 29 craniofacial landmarks.\n", "date": "2023-01-23", "categories": []}
{"id": "2301.09542", "title": "Improving Presentation Attack Detection for ID Cards on Remote\n  Verification Systems", "abstract": "  In this paper, an updated two-stage, end-to-end Presentation Attack Detection\nmethod for remote biometric verification systems of ID cards, based on\nMobileNetV2, is presented. Several presentation attack species such as printed,\ndisplay, composite (based on cropped and spliced areas), plastic (PVC), and\nsynthetic ID card images using different capture sources are used. This\nproposal was developed using a database consisting of 190.000 real case Chilean\nID card images with the support of a third-party company. Also, a new framework\ncalled PyPAD, used to estimate multi-class metrics compliant with the ISO/IEC\n30107-3 standard was developed, and will be made available for research\npurposes. Our method is trained on two convolutional neural networks\nseparately, reaching BPCER\\textsubscript{100} scores on ID cards attacks of\n1.69\\% and 2.36\\% respectively. The two-stage method using both models together\ncan reach a BPCER\\textsubscript{100} score of 0.92\\%.\n", "date": "2023-01-23", "categories": []}
{"id": "2301.03045", "title": "Seamless Multimodal Biometrics for Continuous Personalised Wellbeing\n  Monitoring", "abstract": "  Artificially intelligent perception is increasingly present in the lives of\nevery one of us. Vehicles are no exception, (...) In the near future, pattern\nrecognition will have an even stronger role in vehicles, as self-driving cars\nwill require automated ways to understand what is happening around (and within)\nthem and act accordingly. (...) This doctoral work focused on advancing\nin-vehicle sensing through the research of novel computer vision and pattern\nrecognition methodologies for both biometrics and wellbeing monitoring. The\nmain focus has been on electrocardiogram (ECG) biometrics, a trait well-known\nfor its potential for seamless driver monitoring. Major efforts were devoted to\nachieving improved performance in identification and identity verification in\noff-the-person scenarios, well-known for increased noise and variability. Here,\nend-to-end deep learning ECG biometric solutions were proposed and important\ntopics were addressed such as cross-database and long-term performance,\nwaveform relevance through explainability, and interlead conversion. Face\nbiometrics, a natural complement to the ECG in seamless unconstrained\nscenarios, was also studied in this work. The open challenges of masked face\nrecognition and interpretability in biometrics were tackled in an effort to\nevolve towards algorithms that are more transparent, trustworthy, and robust to\nsignificant occlusions. Within the topic of wellbeing monitoring, improved\nsolutions to multimodal emotion recognition in groups of people and\nactivity/violence recognition in in-vehicle scenarios were proposed. At last,\nwe also proposed a novel way to learn template security within end-to-end\nmodels, dismissing additional separate encryption processes, and a\nself-supervised learning approach tailored to sequential data, in order to\nensure data security and optimal performance. (...)\n", "date": "2023-01-08", "categories": []}
{"id": "2301.07315", "title": "Face Recognition in the age of CLIP & Billion image datasets", "abstract": "  CLIP (Contrastive Language-Image Pre-training) models developed by OpenAI\nhave achieved outstanding results on various image recognition and retrieval\ntasks, displaying strong zero-shot performance. This means that they are able\nto perform effectively on tasks for which they have not been explicitly\ntrained. Inspired by the success of OpenAI CLIP, a new publicly available\ndataset called LAION-5B was collected which resulted in the development of open\nViT-H/14, ViT-G/14 models that outperform the OpenAI L/14 model. The LAION-5B\ndataset also released an approximate nearest neighbor index, with a web\ninterface for search & subset creation.\n  In this paper, we evaluate the performance of various CLIP models as\nzero-shot face recognizers. Our findings show that CLIP models perform well on\nface recognition tasks, but increasing the size of the CLIP model does not\nnecessarily lead to improved accuracy. Additionally, we investigate the\nrobustness of CLIP models against data poisoning attacks by testing their\nperformance on poisoned data. Through this analysis, we aim to understand the\npotential consequences and misuse of search engines built using CLIP models,\nwhich could potentially function as unintentional face recognition engines.\n", "date": "2023-01-18", "categories": []}
{"id": "2301.07320", "title": "Robust Knowledge Adaptation for Federated Unsupervised Person ReID", "abstract": "  Person Re-identification (ReID) has been extensively studied in recent years\ndue to the increasing demand in public security. However, collecting and\ndealing with sensitive personal data raises privacy concerns. Therefore,\nfederated learning has been explored for Person ReID, which aims to share\nminimal sensitive data between different parties (clients). However, existing\nfederated learning based person ReID methods generally rely on laborious and\ntime-consuming data annotations and it is difficult to guarantee cross-domain\nconsistency. Thus, in this work, a federated unsupervised cluster-contrastive\n(FedUCC) learning method is proposed for Person ReID. FedUCC introduces a\nthree-stage modelling strategy following a coarse-to-fine manner. In detail,\ngeneric knowledge, specialized knowledge and patch knowledge are discovered\nusing a deep neural network. This enables the sharing of mutual knowledge among\nclients while retaining local domain-specific knowledge based on the kinds of\nnetwork layers and their parameters. Comprehensive experiments on 8 public\nbenchmark datasets demonstrate the state-of-the-art performance of our proposed\nmethod.\n", "date": "2023-01-18", "categories": []}
{"id": "2301.05776", "title": "Young Labeled Faces in the Wild (YLFW): A Dataset for Children Faces\n  Recognition", "abstract": "  Face recognition has achieved outstanding performance in the last decade with\nthe development of deep learning techniques.\n  Nowadays, the challenges in face recognition are related to specific\nscenarios, for instance, the performance under diverse image quality, the\nrobustness for aging and edge cases of person age (children and elders),\ndistinguishing of related identities.\n  In this set of problems, recognizing children's faces is one of the most\nsensitive and important. One of the reasons for this problem is the existing\nbias towards adults in existing face datasets.\n  In this work, we present a benchmark dataset for children's face recognition,\nwhich is compiled similarly to the famous face recognition benchmarks LFW,\nCALFW, CPLFW, XQLFW and AgeDB.\n  We also present a development dataset (separated into train and test parts)\nfor adapting face recognition models for face images of children.\n  The proposed data is balanced for African, Asian, Caucasian, and Indian\nraces. To the best of our knowledge, this is the first standartized data tool\nset for benchmarking and the largest collection for development for children's\nface recognition. Several face recognition experiments are presented to\ndemonstrate the performance of the proposed data tool set.\n", "date": "2023-01-13", "categories": []}
{"id": "2301.06882", "title": "Multi-Biometric Fuzzy Vault based on Face and Fingerprints", "abstract": "  The fuzzy vault scheme has been established as cryptographic primitive\nsuitable for privacy-preserving biometric authentication. To improve accuracy\nand privacy protection, biometric information of multiple characteristics can\nbe fused at feature level prior to locking it in a fuzzy vault. We construct a\nmulti-biometric fuzzy vault based on face and multiple fingerprints. On a\nmulti-biometric database constructed from the FRGCv2 face and the MCYT-100\nfingerprint databases, a perfect recognition accuracy is achieved at a false\naccept security above 30 bits. Further, we provide a formalisation of\nfeature-level fusion in multi-biometric fuzzy vaults, on the basis of which\nrelevant security issues are elaborated. Said security issues, for which we\ndefine countermeasures, are commonly ignored and may impair the overall\nsystem's security.\n", "date": "2023-01-17", "categories": []}
{"id": "2301.04402", "title": "Secure access system using signature verification over tablet PC", "abstract": "  Low-cost portable devices capable of capturing signature signals are being\nincreasingly used. Additionally, the social and legal acceptance of the written\nsignature for authentication purposes is opening a range of new applications.\nWe describe a highly versatile and scalable prototype for Web-based secure\naccess using signature verification. The proposed architecture can be easily\nextended to work with different kinds of sensors and large-scale databases.\nSeveral remarks are also given on security and privacy of network-based\nsignature verification.\n", "date": "2023-01-11", "categories": []}
{"id": "2301.01879", "title": "Learning Feature Recovery Transformer for Occluded Person\n  Re-identification", "abstract": "  One major issue that challenges person re-identification (Re-ID) is the\nubiquitous occlusion over the captured persons. There are two main challenges\nfor the occluded person Re-ID problem, i.e., the interference of noise during\nfeature matching and the loss of pedestrian information brought by the\nocclusions. In this paper, we propose a new approach called Feature Recovery\nTransformer (FRT) to address the two challenges simultaneously, which mainly\nconsists of visibility graph matching and feature recovery transformer. To\nreduce the interference of the noise during feature matching, we mainly focus\non visible regions that appear in both images and develop a visibility graph to\ncalculate the similarity. In terms of the second challenge, based on the\ndeveloped graph similarity, for each query image, we propose a recovery\ntransformer that exploits the feature sets of its $k$-nearest neighbors in the\ngallery to recover the complete features. Extensive experiments across\ndifferent person Re-ID datasets, including occluded, partial and holistic\ndatasets, demonstrate the effectiveness of FRT. Specifically, FRT significantly\noutperforms state-of-the-art results by at least 6.2\\% Rank-1 accuracy and\n7.2\\% mAP scores on the challenging Occluded-Duke dataset. The code is\navailable at https://github.com/xbq1994/Feature-Recovery-Transformer.\n", "date": "2023-01-05", "categories": []}
{"id": "2301.03966", "title": "AdvBiom: Adversarial Attacks on Biometric Matchers", "abstract": "  With the advent of deep learning models, face recognition systems have\nachieved impressive recognition rates. The workhorses behind this success are\nConvolutional Neural Networks (CNNs) and the availability of large training\ndatasets. However, we show that small human-imperceptible changes to face\nsamples can evade most prevailing face recognition systems. Even more alarming\nis the fact that the same generator can be extended to other traits in the\nfuture. In this work, we present how such a generator can be trained and also\nextended to other biometric modalities, such as fingerprint recognition\nsystems.\n", "date": "2023-01-10", "categories": []}
{"id": "2301.02934", "title": "Advancing 3D finger knuckle recognition via deep feature learning", "abstract": "  Contactless 3D finger knuckle patterns have emerged as an effective biometric\nidentifier due to its discriminativeness, visibility from a distance, and\nconvenience. Recent research has developed a deep feature collaboration network\nwhich simultaneously incorporates intermediate features from deep neural\nnetworks with multiple scales. However, this approach results in a large\nfeature dimension, and the trained classification layer is required for\ncomparing probe samples, which limits the introduction of new classes. This\npaper advances this approach by investigating the possibility of learning a\ndiscriminative feature vector with the least possible dimension for\nrepresenting 3D finger knuckle images. Experimental results are presented using\na publicly available 3D finger knuckle images database with comparisons to\npopular deep learning architectures and the state-of-the-art 3D finger knuckle\nrecognition methods. The proposed approach offers outperforming results in\nclassification and identification tasks under the more practical feature\ncomparison scenario, i.e., using the extracted deep feature instead of the\ntrained classification layer for comparing probe samples. More importantly,\nthis approach can offer 99% reduction in the size of feature templates, which\nis highly attractive for deploying biometric systems in the real world.\nExperiments are also performed using other two public biometric databases with\nsimilar patterns to ascertain the effectiveness and generalizability of our\nproposed approach.\n", "date": "2023-01-07", "categories": []}
{"id": "2301.01922", "title": "Open-Set Face Identification on Few-Shot Gallery by Fine-Tuning", "abstract": "  In this paper, we focus on addressing the open-set face identification\nproblem on a few-shot gallery by fine-tuning. The problem assumes a realistic\nscenario for face identification, where only a small number of face images is\ngiven for enrollment and any unknown identity must be rejected during\nidentification. We observe that face recognition models pretrained on a large\ndataset and naively fine-tuned models perform poorly for this task. Motivated\nby this issue, we propose an effective fine-tuning scheme with classifier\nweight imprinting and exclusive BatchNorm layer tuning. For further improvement\nof rejection accuracy on unknown identities, we propose a novel matcher called\nNeighborhood Aware Cosine (NAC) that computes similarity based on neighborhood\ninformation. We validate the effectiveness of the proposed schemes thoroughly\non large-scale face benchmarks across different convolutional neural network\narchitectures. The source code for this project is available at:\nhttps://github.com/1ho0jin1/OSFI-by-FineTuning\n", "date": "2023-01-05", "categories": []}
{"id": "2301.00975", "title": "Surveillance Face Anti-spoofing", "abstract": "  Face Anti-spoofing (FAS) is essential to secure face recognition systems from\nvarious physical attacks. However, recent research generally focuses on\nshort-distance applications (i.e., phone unlocking) while lacking consideration\nof long-distance scenes (i.e., surveillance security checks). In order to\npromote relevant research and fill this gap in the community, we collect a\nlarge-scale Surveillance High-Fidelity Mask (SuHiFiMask) dataset captured under\n40 surveillance scenes, which has 101 subjects from different age groups with\n232 3D attacks (high-fidelity masks), 200 2D attacks (posters, portraits, and\nscreens), and 2 adversarial attacks. In this scene, low image resolution and\nnoise interference are new challenges faced in surveillance FAS. Together with\nthe SuHiFiMask dataset, we propose a Contrastive Quality-Invariance Learning\n(CQIL) network to alleviate the performance degradation caused by image quality\nfrom three aspects: (1) An Image Quality Variable module (IQV) is introduced to\nrecover image information associated with discrimination by combining the\nsuper-resolution network. (2) Using generated sample pairs to simulate quality\nvariance distributions to help contrastive learning strategies obtain robust\nfeature representation under quality variation. (3) A Separate Quality Network\n(SQN) is designed to learn discriminative features independent of image\nquality. Finally, a large number of experiments verify the quality of the\nSuHiFiMask dataset and the superiority of the proposed CQIL.\n", "date": "2023-01-03", "categories": []}
{"id": "2301.00531", "title": "Multi-Stage Spatio-Temporal Aggregation Transformer for Video Person\n  Re-identification", "abstract": "  In recent years, the Transformer architecture has shown its superiority in\nthe video-based person re-identification task. Inspired by video representation\nlearning, these methods mainly focus on designing modules to extract\ninformative spatial and temporal features. However, they are still limited in\nextracting local attributes and global identity information, which are critical\nfor the person re-identification task. In this paper, we propose a novel\nMulti-Stage Spatial-Temporal Aggregation Transformer (MSTAT) with two novel\ndesigned proxy embedding modules to address the above issue. Specifically,\nMSTAT consists of three stages to encode the attribute-associated, the\nidentity-associated, and the attribute-identity-associated information from the\nvideo clips, respectively, achieving the holistic perception of the input\nperson. We combine the outputs of all the stages for the final identification.\nIn practice, to save the computational cost, the Spatial-Temporal Aggregation\n(STA) modules are first adopted in each stage to conduct the self-attention\noperations along the spatial and temporal dimensions separately. We further\nintroduce the Attribute-Aware and Identity-Aware Proxy embedding modules (AAP\nand IAP) to extract the informative and discriminative feature representations\nat different stages. All of them are realized by employing newly designed\nself-attention operations with specific meanings. Moreover, temporal patch\nshuffling is also introduced to further improve the robustness of the model.\nExtensive experimental results demonstrate the effectiveness of the proposed\nmodules in extracting the informative and discriminative information from the\nvideos, and illustrate the MSTAT can achieve state-of-the-art accuracies on\nvarious standard benchmarks.\n", "date": "2023-01-02", "categories": []}
{"id": "2211.00933", "title": "Deep Multimodal Fusion for Generalizable Person Re-identification", "abstract": "  Person re-identification plays a significant role in realistic scenarios due\nto its various applications in public security and video surveillance.\nRecently, leveraging the supervised or semi-unsupervised learning paradigms,\nwhich benefits from the large-scale datasets and strong computing performance,\nhas achieved a competitive performance on a specific target domain. However,\nwhen Re-ID models are directly deployed in a new domain without target samples,\nthey always suffer from considerable performance degradation and poor domain\ngeneralization. To address this challenge, we propose a Deep Multimodal Fusion\nnetwork to elaborate rich semantic knowledge for assisting in representation\nlearning during the pre-training. Importantly, a multimodal fusion strategy is\nintroduced to translate the features of different modalities into the common\nspace, which can significantly boost generalization capability of Re-ID model.\nAs for the fine-tuning stage, a realistic dataset is adopted to fine-tune the\npre-trained model for better distribution alignment with real-world data.\nComprehensive experiments on benchmarks demonstrate that our method can\nsignificantly outperform previous domain generalization or meta-learning\nmethods with a clear margin. Our source code will also be publicly available at\nhttps://github.com/JeremyXSC/DMF.\n", "date": "2022-11-02", "categories": []}
{"id": "2212.14107", "title": "Joint Discriminative and Metric Embedding Learning for Person\n  Re-Identification", "abstract": "  Person re-identification is a challenging task because of the high\nintra-class variance induced by the unrestricted nuisance factors of variations\nsuch as pose, illumination, viewpoint, background, and sensor noise. Recent\napproaches postulate that powerful architectures have the capacity to learn\nfeature representations invariant to nuisance factors, by training them with\nlosses that minimize intra-class variance and maximize inter-class separation,\nwithout modeling nuisance factors explicitly. The dominant approaches use\neither a discriminative loss with margin, like the softmax loss with the\nadditive angular margin, or a metric learning loss, like the triplet loss with\nbatch hard mining of triplets. Since the softmax imposes feature normalization,\nit limits the gradient flow supervising the feature embedding. We address this\nby joining the losses and leveraging the triplet loss as a proxy for the\nmissing gradients. We further improve invariance to nuisance factors by adding\nthe discriminative task of predicting attributes. Our extensive evaluation\nhighlights that when only a holistic representation is learned, we consistently\noutperform the state-of-the-art on the three most challenging datasets. Such\nrepresentations are easier to deploy in practical systems. Finally, we found\nthat joining the losses removes the requirement for having a margin in the\nsoftmax loss while increasing performance.\n", "date": "2022-12-28", "categories": []}
{"id": "2212.14110", "title": "Learning Representations for Masked Facial Recovery", "abstract": "  The pandemic of these very recent years has led to a dramatic increase in\npeople wearing protective masks in public venues. This poses obvious challenges\nto the pervasive use of face recognition technology that now is suffering a\ndecline in performance. One way to address the problem is to revert to face\nrecovery methods as a preprocessing step. Current approaches to face\nreconstruction and manipulation leverage the ability to model the face\nmanifold, but tend to be generic. We introduce a method that is specific for\nthe recovery of the face image from an image of the same individual wearing a\nmask. We do so by designing a specialized GAN inversion method, based on an\nappropriate set of losses for learning an unmasking encoder. With extensive\nexperiments, we show that the approach is effective at unmasking face images.\nIn addition, we also show that the identity information is preserved\nsufficiently well to improve face verification performance based on several\nface recognition benchmark datasets.\n", "date": "2022-12-28", "categories": []}
{"id": "2212.12733", "title": "Artificial Pupil Dilation for Data Augmentation in Iris Semantic\n  Segmentation", "abstract": "  Biometrics is the science of identifying an individual based on their\nintrinsic anatomical or behavioural characteristics, such as fingerprints,\nface, iris, gait, and voice. Iris recognition is one of the most successful\nmethods because it exploits the rich texture of the human iris, which is unique\neven for twins and does not degrade with age. Modern approaches to iris\nrecognition utilize deep learning to segment the valid portion of the iris from\nthe rest of the eye, so it can then be encoded, stored and compared. This paper\naims to improve the accuracy of iris semantic segmentation systems by\nintroducing a novel data augmentation technique. Our method can transform an\niris image with a certain dilation level into any desired dilation level, thus\naugmenting the variability and number of training examples from a small\ndataset. The proposed method is fast and does not require training. The results\nindicate that our data augmentation method can improve segmentation accuracy up\nto 15% for images with high pupil dilation, which creates a more reliable iris\nrecognition pipeline, even under extreme dilation.\n", "date": "2022-12-24", "categories": []}
{"id": "2212.13038", "title": "A Survey of Face Recognition", "abstract": "  Recent years witnessed the breakthrough of face recognition with deep\nconvolutional neural networks. Dozens of papers in the field of FR are\npublished every year. Some of them were applied in the industrial community and\nplayed an important role in human life such as device unlock, mobile payment,\nand so on. This paper provides an introduction to face recognition, including\nits history, pipeline, algorithms based on conventional manually designed\nfeatures or deep learning, mainstream training, evaluation datasets, and\nrelated applications. We have analyzed and compared state-of-the-art works as\nmany as possible, and also carefully designed a set of experiments to find the\neffect of backbone size and data distribution. This survey is a material of the\ntutorial named The Practical Face Recognition Technology in the Industrial\nWorld in the FG2023.\n", "date": "2022-12-26", "categories": []}
{"id": "2212.10812", "title": "Secure and Privacy Preserving Proxy Biometrics Identities", "abstract": "  With large-scale adaption to biometric based applications, security and\nprivacy of biometrics is utmost important especially when operating in\nunsupervised online mode. This work proposes a novel approach for generating\nnew artificial fingerprints also called proxy fingerprints that are natural\nlooking, non-invertible, revocable and privacy preserving. These proxy\nbiometrics can be generated from original ones only with the help of a\nuser-specific key. Instead of using the original fingerprint, these proxy\ntemplates can be used anywhere with same convenience. The manuscripts walks\nthrough an interesting way in which proxy fingerprints of different types can\nbe generated and how they can be combined with use-specific keys to provide\nrevocability and cancelability in case of compromise. Using the proposed\napproach a proxy dataset is generated from samples belonging to Anguli\nfingerprint database. Matching experiments were performed on the new set which\nis 5 times larger than the original, and it was found that their performance is\nat par with 0 FAR and 0 FRR in the stolen key, safe key scenarios. Other\nparameters on revocability and diversity are also analyzed for protection\nperformance.\n", "date": "2022-12-21", "categories": []}
{"id": "2212.09981", "title": "Benchmarking person re-identification datasets and approaches for\n  practical real-world implementations", "abstract": "  Recently, Person Re-Identification (Re-ID) has received a lot of attention.\nLarge datasets containing labeled images of various individuals have been\nreleased, allowing researchers to develop and test many successful approaches.\nHowever, when such Re-ID models are deployed in new cities or environments, the\ntask of searching for people within a network of security cameras is likely to\nface an important domain shift, thus resulting in decreased performance.\nIndeed, while most public datasets were collected in a limited geographic area,\nimages from a new city present different features (e.g., people's ethnicity and\nclothing style, weather, architecture, etc.). In addition, the whole frames of\nthe video streams must be converted into cropped images of people using\npedestrian detection models, which behave differently from the human annotators\nwho created the dataset used for training. To better understand the extent of\nthis issue, this paper introduces a complete methodology to evaluate Re-ID\napproaches and training datasets with respect to their suitability for\nunsupervised deployment for live operations. This method is used to benchmark\nfour Re-ID approaches on three datasets, providing insight and guidelines that\ncan help to design better Re-ID pipelines in the future.\n", "date": "2022-12-20", "categories": []}
{"id": "2212.08969", "title": "A Brief Survey on Person Recognition at a Distance", "abstract": "  Person recognition at a distance entails recognizing the identity of an\nindividual appearing in images or videos collected by long-range imaging\nsystems such as drones or surveillance cameras. Despite recent advances in deep\nconvolutional neural networks (DCNNs), this remains challenging. Images or\nvideos collected by long-range cameras often suffer from atmospheric\nturbulence, blur, low-resolution, unconstrained poses, and poor illumination.\nIn this paper, we provide a brief survey of recent advances in person\nrecognition at a distance. In particular, we review recent work in\nmulti-spectral face verification, person re-identification, and gait-based\nanalysis techniques. Furthermore, we discuss the merits and drawbacks of\nexisting approaches and identify important, yet under explored challenges for\ndeploying remote person recognition systems in-the-wild.\n", "date": "2022-12-17", "categories": []}
{"id": "2212.09042", "title": "Gait Recognition Using 3-D Human Body Shape Inference", "abstract": "  Gait recognition, which identifies individuals based on their walking\npatterns, is an important biometric technique since it can be observed from a\ndistance and does not require the subject's cooperation. Recognizing a person's\ngait is difficult because of the appearance variants in human silhouette\nsequences produced by varying viewing angles, carrying objects, and clothing.\nRecent research has produced a number of ways for coping with these variants.\nIn this paper, we present the usage of inferring 3-D body shapes distilled from\nlimited images, which are, in principle, invariant to the specified variants.\nInference of 3-D shape is a difficult task, especially when only silhouettes\nare provided in a dataset. We provide a method for learning 3-D body inference\nfrom silhouettes by transferring knowledge from 3-D shape prior from RGB\nphotos. We use our method on multiple existing state-of-the-art gait baselines\nand obtain consistent improvements for gait identification on two public\ndatasets, CASIA-B and OUMVLP, on several variants and settings, including a new\nsetting of novel views not seen during training.\n", "date": "2022-12-18", "categories": []}
{"id": "2212.09498", "title": "Feature Disentanglement Learning with Switching and Aggregation for\n  Video-based Person Re-Identification", "abstract": "  In video person re-identification (Re-ID), the network must consistently\nextract features of the target person from successive frames. Existing methods\ntend to focus only on how to use temporal information, which often leads to\nnetworks being fooled by similar appearances and same backgrounds. In this\npaper, we propose a Disentanglement and Switching and Aggregation Network\n(DSANet), which segregates the features representing identity and features\nbased on camera characteristics, and pays more attention to ID information. We\nalso introduce an auxiliary task that utilizes a new pair of features created\nthrough switching and aggregation to increase the network's capability for\nvarious camera scenarios. Furthermore, we devise a Target Localization Module\n(TLM) that extracts robust features against a change in the position of the\ntarget according to the frame flow and a Frame Weight Generation (FWG) that\nreflects temporal information in the final representation. Various loss\nfunctions for disentanglement learning are designed so that each component of\nthe network can cooperate while satisfactorily performing its own role.\nQuantitative and qualitative results from extensive experiments demonstrate the\nsuperiority of DSANet over state-of-the-art methods on three benchmark\ndatasets.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.08445", "title": "Conditional Generative Adversarial Network for keystroke presentation\n  attack", "abstract": "  Cybersecurity is a crucial step in data protection to ensure user security\nand personal data privacy. In this sense, many companies have started to\ncontrol and restrict access to their data using authentication systems.\nHowever, these traditional authentication methods, are not enough for ensuring\ndata protection, and for this reason, behavioral biometrics have gained\nimportance. Despite their promising results and the wide range of applications,\nbiometric systems have shown to be vulnerable to malicious attacks, such as\nPresentation Attacks. For this reason, in this work, we propose to study a new\napproach aiming to deploy a presentation attack towards a keystroke\nauthentication system. Our idea is to use Conditional Generative Adversarial\nNetworks (cGAN) for generating synthetic keystroke data that can be used for\nimpersonating an authorized user. These synthetic data are generated following\ntwo different real use cases, one in which the order of the typed words is\nknown (ordered dynamic) and the other in which this order is unknown\n(no-ordered dynamic). Finally, both keystroke dynamics (ordered and no-ordered)\nare validated using an external keystroke authentication system. Results\nindicate that the cGAN can effectively generate keystroke dynamics patterns\nthat can be used for deceiving keystroke authentication systems.\n", "date": "2022-12-16", "categories": []}
{"id": "2212.07575", "title": "Evaluation of direct attacks to fingerprint verification systems", "abstract": "  The vulnerabilities of fingerprint-based recognition systems to direct\nattacks with and without the cooperation of the user are studied. Two different\nsystems, one minutiae-based and one ridge feature-based, are evaluated on a\ndatabase of real and fake fingerprints. Based on the fingerprint images quality\nand on the results achieved on different operational scenarios, we obtain a\nnumber of statistically significant observations regarding the robustness of\nthe systems.\n", "date": "2022-12-15", "categories": []}
{"id": "2212.07299", "title": "Child PalmID: Contactless Palmprint Recognition", "abstract": "  Developing and least developed countries face the dire challenge of ensuring\nthat each child in their country receives required doses of vaccination,\nadequate nutrition and proper medication. International agencies such as\nUNICEF, WHO and WFP, among other organizations, strive to find innovative\nsolutions to determine which child has received the benefits and which have\nnot. Biometric recognition systems have been sought out to help solve this\nproblem. To that end, this report establishes a baseline accuracy of a\ncommercial contactless palmprint recognition system that may be deployed for\nrecognizing children in the age group of one to five years old. On a database\nof contactless palmprint images of one thousand unique palms from 500 children,\nwe establish SOTA authentication accuracy of 90.85% @ FAR of 0.01%, rank-1\nidentification accuracy of 99.0% (closed set), and FPIR=0.01 @ FNIR=0.3 for\nopen-set identification using PalmMobile SDK from Armatura.\n", "date": "2022-12-14", "categories": []}
{"id": "2212.05884", "title": "Finger-NestNet: Interpretable Fingerphoto Verification on Smartphone\n  using Deep Nested Residual Network", "abstract": "  Fingerphoto images captured using a smartphone are successfully used to\nverify the individuals that have enabled several applications. This work\npresents a novel algorithm for fingerphoto verification using a nested residual\nblock: Finger-NestNet. The proposed Finger-NestNet architecture is designed\nwith three consecutive convolution blocks followed by a series of nested\nresidual blocks to achieve reliable fingerphoto verification. This paper also\npresents the interpretability of the proposed method using four different\nvisualization techniques that can shed light on the critical regions in the\nfingerphoto biometrics that can contribute to the reliable verification\nperformance of the proposed method. Extensive experiments are performed on the\nfingerphoto dataset comprised of 196 unique fingers collected from 52 unique\ndata subjects using an iPhone6S. Experimental results indicate the improved\nverification of the proposed method compared to six different existing methods\nwith EER = 1.15%.\n", "date": "2022-12-09", "categories": []}
{"id": "2212.04712", "title": "Occluded Person Re-Identification via Relational Adaptive Feature\n  Correction Learning", "abstract": "  Occluded person re-identification (Re-ID) in images captured by multiple\ncameras is challenging because the target person is occluded by pedestrians or\nobjects, especially in crowded scenes. In addition to the processes performed\nduring holistic person Re-ID, occluded person Re-ID involves the removal of\nobstacles and the detection of partially visible body parts. Most existing\nmethods utilize the off-the-shelf pose or parsing networks as pseudo labels,\nwhich are prone to error. To address these issues, we propose a novel Occlusion\nCorrection Network (OCNet) that corrects features through relational-weight\nlearning and obtains diverse and representative features without using external\nnetworks. In addition, we present a simple concept of a center feature in order\nto provide an intuitive solution to pedestrian occlusion scenarios.\nFurthermore, we suggest the idea of Separation Loss (SL) for focusing on\ndifferent parts between global features and part features. We conduct extensive\nexperiments on five challenging benchmark datasets for occluded and holistic\nRe-ID tasks to demonstrate that our method achieves superior performance to\nstate-of-the-art methods especially on occluded scene.\n", "date": "2022-12-09", "categories": []}
{"id": "2210.12008", "title": "End-to-End Context-Aided Unicity Matching for Person Re-identification", "abstract": "  Most existing person re-identification methods compute the matching relations\nbetween person images across camera views based on the ranking of the pairwise\nsimilarities. This matching strategy with the lack of the global viewpoint and\nthe context's consideration inevitably leads to ambiguous matching results and\nsub-optimal performance. Based on a natural assumption that images belonging to\nthe same person identity should not match with images belonging to multiple\ndifferent person identities across views, called the unicity of person matching\non the identity level, we propose an end-to-end person unicity matching\narchitecture for learning and refining the person matching relations. First, we\nadopt the image samples' contextual information in feature space to generate\nthe initial soft matching results by using graph neural networks. Secondly, we\nutilize the samples' global context relationship to refine the soft matching\nresults and reach the matching unicity through bipartite graph matching. Given\nfull consideration to real-world person re-identification applications, we\nachieve the unicity matching in both one-shot and multi-shot settings of person\nre-identification and further develop a fast version of the unicity matching\nwithout losing the performance. The proposed method is evaluated on five public\nbenchmarks, including four multi-shot datasets MSMT17, DukeMTMC, Market1501,\nCUHK03, and a one-shot dataset VIPeR. Experimental results show the superiority\nof the proposed method on performance and efficiency.\n", "date": "2022-10-20", "categories": []}
{"id": "2212.03680", "title": "Face Presentation Attack Detection", "abstract": "  Face recognition technology has been widely used in daily interactive\napplications such as checking-in and mobile payment due to its convenience and\nhigh accuracy. However, its vulnerability to presentation attacks (PAs) limits\nits reliable use in ultra-secure applicational scenarios. A presentation attack\nis first defined in ISO standard as: a presentation to the biometric data\ncapture subsystem with the goal of interfering with the operation of the\nbiometric system. Specifically, PAs range from simple 2D print, replay and more\nsophisticated 3D masks and partial masks. To defend the face recognition\nsystems against PAs, both academia and industry have paid extensive attention\nto developing face presentation attack detection (PAD) technology (or namely\n`face anti-spoofing (FAS)').\n", "date": "2022-12-07", "categories": []}
{"id": "2211.09210", "title": "edBB-Demo: Biometrics and Behavior Analysis for Online Educational\n  Platforms", "abstract": "  We present edBB-Demo, a demonstrator of an AI-powered research platform for\nstudent monitoring in remote education. The edBB platform aims to study the\nchallenges associated to user recognition and behavior understanding in digital\nplatforms. This platform has been developed for data collection, acquiring\nsignals from a variety of sensors including keyboard, mouse, webcam,\nmicrophone, smartwatch, and an Electroencephalography band. The information\ncaptured from the sensors during the student sessions is modelled in a\nmultimodal learning framework. The demonstrator includes: i) Biometric user\nauthentication in an unsupervised environment; ii) Human action recognition\nbased on remote video analysis; iii) Heart rate estimation from webcam video;\nand iv) Attention level estimation from facial expression analysis.\n", "date": "2022-11-16", "categories": []}
{"id": "2211.13897", "title": "AFR-Net: Attention-Driven Fingerprint Recognition Network", "abstract": "  The use of vision transformers (ViT) in computer vision is increasing due to\nlimited inductive biases (e.g., locality, weight sharing, etc.) and increased\nscalability compared to other deep learning methods. This has led to some\ninitial studies on the use of ViT for biometric recognition, including\nfingerprint recognition. In this work, we improve on these initial studies for\ntransformers in fingerprint recognition by i.) evaluating additional\nattention-based architectures, ii.) scaling to larger and more diverse training\nand evaluation datasets, and iii.) combining the complimentary representations\nof attention-based and CNN-based embeddings for improved state-of-the-art\n(SOTA) fingerprint recognition (both authentication and identification). Our\ncombined architecture, AFR-Net (Attention-Driven Fingerprint Recognition\nNetwork), outperforms several baseline transformer and CNN-based models,\nincluding a SOTA commercial fingerprint system, Verifinger v12.3, across\nintra-sensor, cross-sensor, and latent to rolled fingerprint matching datasets.\nAdditionally, we propose a realignment strategy using local embeddings\nextracted from intermediate feature maps within the networks to refine the\nglobal embeddings in low certainty situations, which boosts the overall\nrecognition accuracy significantly across each of the models. This realignment\nstrategy requires no additional training and can be applied as a wrapper to any\nexisting deep learning network (including attention-based, CNN-based, or both)\nto boost its performance.\n", "date": "2022-11-25", "categories": []}
{"id": "2212.01906", "title": "Combining multiple matchers for fingerprint verification: A case study\n  in biosecure network of excellence", "abstract": "  We report on experiments for the fingerprint modality conducted during the\nFirst BioSecure Residential Workshop. Two reference systems for fingerprint\nverification have been tested together with two additional non-reference\nsystems. These systems follow different approaches of fingerprint processing\nand are discussed in detail. Fusion experiments I volving different\ncombinations of the available systems are presented. The experimental results\nshow that the best recognition strategy involves both minutiae-based and\ncorrelation-based measurements. Regarding the fusion experiments, the best\nrelative improvement is obtained when fusing systems that are based on\nheterogeneous strategies for feature extraction and/or matching. The best\ncombinations of two/three/four systems always include the best individual\nsystems whereas the best verification performance is obtained when combining\nall the available systems.\n", "date": "2022-12-04", "categories": []}
{"id": "2212.02398", "title": "Generalizable Person Re-Identification via Viewpoint Alignment and\n  Fusion", "abstract": "  In the current person Re-identification (ReID) methods, most domain\ngeneralization works focus on dealing with style differences between domains\nwhile largely ignoring unpredictable camera view change, which we identify as\nanother major factor leading to a poor generalization of ReID methods. To\ntackle the viewpoint change, this work proposes to use a 3D dense pose\nestimation model and a texture mapping module to map the pedestrian images to\ncanonical view images. Due to the imperfection of the texture mapping module,\nthe canonical view images may lose the discriminative detail clues from the\noriginal images, and thus directly using them for ReID will inevitably result\nin poor performance. To handle this issue, we propose to fuse the original\nimage and canonical view image via a transformer-based module. The key insight\nof this design is that the cross-attention mechanism in the transformer could\nbe an ideal solution to align the discriminative texture clues from the\noriginal image with the canonical view image, which could compensate for the\nlow-quality texture information of the canonical view image. Through extensive\nexperiments, we show that our method can lead to superior performance over the\nexisting approaches in various evaluation settings.\n", "date": "2022-12-05", "categories": []}
{"id": "2212.01015", "title": "Improving Training and Inference of Face Recognition Models via Random\n  Temperature Scaling", "abstract": "  Data uncertainty is commonly observed in the images for face recognition\n(FR). However, deep learning algorithms often make predictions with high\nconfidence even for uncertain or irrelevant inputs. Intuitively, FR algorithms\ncan benefit from both the estimation of uncertainty and the detection of\nout-of-distribution (OOD) samples. Taking a probabilistic view of the current\nclassification model, the temperature scalar is exactly the scale of\nuncertainty noise implicitly added in the softmax function. Meanwhile, the\nuncertainty of images in a dataset should follow a prior distribution. Based on\nthe observation, a unified framework for uncertainty modeling and FR, Random\nTemperature Scaling (RTS), is proposed to learn a reliable FR algorithm. The\nbenefits of RTS are two-fold. (1) In the training phase, it can adjust the\nlearning strength of clean and noisy samples for stability and accuracy. (2) In\nthe test phase, it can provide a score of confidence to detect uncertain,\nlow-quality and even OOD samples, without training on extra labels. Extensive\nexperiments on FR benchmarks demonstrate that the magnitude of variance in RTS,\nwhich serves as an OOD detection metric, is closely related to the uncertainty\nof the input image. RTS can achieve top performance on both the FR and OOD\ndetection tasks. Moreover, the model trained with RTS can perform robustly on\ndatasets with noise. The proposed module is light-weight and only adds\nnegligible computation cost to the model.\n", "date": "2022-12-02", "categories": []}
{"id": "2212.00057", "title": "Part-based Face Recognition with Vision Transformers", "abstract": "  Holistic methods using CNNs and margin-based losses have dominated research\non face recognition. In this work, we depart from this setting in two ways: (a)\nwe employ the Vision Transformer as an architecture for training a very strong\nbaseline for face recognition, simply called fViT, which already surpasses most\nstate-of-the-art face recognition methods. (b) Secondly, we capitalize on the\nTransformer's inherent property to process information (visual tokens)\nextracted from irregular grids to devise a pipeline for face recognition which\nis reminiscent of part-based face recognition methods. Our pipeline, called\npart fViT, simply comprises a lightweight network to predict the coordinates of\nfacial landmarks followed by the Vision Transformer operating on patches\nextracted from the predicted landmarks, and it is trained end-to-end with no\nlandmark supervision. By learning to extract discriminative patches, our\npart-based Transformer further boosts the accuracy of our Vision Transformer\nbaseline achieving state-of-the-art accuracy on several face recognition\nbenchmarks.\n", "date": "2022-11-30", "categories": []}
{"id": "2211.07582", "title": "AttenFace: A Real Time Attendance System using Face Recognition", "abstract": "  The current approach to marking attendance in colleges is tedious and time\nconsuming. I propose AttenFace, a standalone system to analyze, track and grant\nattendance in real time using face recognition. Using snapshots of class from\nlive camera feed, the system identifies students and marks them as present in a\nclass based on their presence in multiple snapshots taken throughout the class\nduration. Face recognition for each class is performed independently and in\nparallel, ensuring that the system scales with number of concurrent classes.\nFurther, the separation of the face recognition server from the back-end server\nfor attendance calculation allows the face recognition module to be integrated\nwith existing attendance tracking software like Moodle. The face recognition\nalgorithm runs at 10 minute intervals on classroom snapshots, significantly\nreducing computation compared to direct processing of live camera feed. This\nmethod also provides students the flexibility to leave class for a short\nduration (such as for a phone call) without losing attendance for that class.\nAttendance is granted to a student if he remains in class for a number of\nsnapshots above a certain threshold. The system is fully automatic and requires\nno professor intervention or any form of manual attendance or even camera\nset-up, since the back-end directly interfaces with in-class cameras. AttenFace\nis a first-of-its-kind one-stop solution for face-recognition-enabled\nattendance in educational institutions that prevents proxy, handling all\naspects from students checking attendance to professors deciding their own\nattendance policy, to college administration enforcing default attendance\nrules.\n", "date": "2022-11-14", "categories": []}
{"id": "2211.16739", "title": "Quasi Non-Negative Quaternion Matrix Factorization with Application to\n  Color Face Recognition", "abstract": "  To address the non-negativity dropout problem of quaternion models, a novel\nquasi non-negative quaternion matrix factorization (QNQMF) model is presented\nfor color image processing. To implement QNQMF, the quaternion projected\ngradient algorithm and the quaternion alternating direction method of\nmultipliers are proposed via formulating QNQMF as the non-convex constraint\nquaternion optimization problems. Some properties of the proposed algorithms\nare studied. The numerical experiments on the color image reconstruction show\nthat these algorithms encoded on the quaternion perform better than these\nalgorithms encoded on the red, green and blue channels. Furthermore, we apply\nthe proposed algorithms to the color face recognition. Numerical results\nindicate that the accuracy rate of face recognition on the quaternion model is\nbetter than on the red, green and blue channels of color image as well as\nsingle channel of gray level images for the same data, when large facial\nexpressions and shooting angle variations are presented.\n", "date": "2022-11-30", "categories": []}
{"id": "2211.15225", "title": "Meet-in-the-middle: Multi-scale upsampling and matching for\n  cross-resolution face recognition", "abstract": "  In this paper, we aim to address the large domain gap between high-resolution\nface images, e.g., from professional portrait photography, and low-quality\nsurveillance images, e.g., from security cameras. Establishing an identity\nmatch between disparate sources like this is a classical surveillance face\nidentification scenario, which continues to be a challenging problem for modern\nface recognition techniques. To that end, we propose a method that combines\nface super-resolution, resolution matching, and multi-scale template\naccumulation to reliably recognize faces from long-range surveillance footage,\nincluding from low quality sources. The proposed approach does not require\ntraining or fine-tuning on the target dataset of real surveillance images.\nExtensive experiments show that our proposed method is able to outperform even\nexisting methods fine-tuned to the SCFace dataset.\n", "date": "2022-11-28", "categories": []}
{"id": "2211.16201", "title": "Lifelong Person Re-Identification via Knowledge Refreshing and\n  Consolidation", "abstract": "  Lifelong person re-identification (LReID) is in significant demand for\nreal-world development as a large amount of ReID data is captured from diverse\nlocations over time and cannot be accessed at once inherently. However, a key\nchallenge for LReID is how to incrementally preserve old knowledge and\ngradually add new capabilities to the system. Unlike most existing LReID\nmethods, which mainly focus on dealing with catastrophic forgetting, our focus\nis on a more challenging problem, which is, not only trying to reduce the\nforgetting on old tasks but also aiming to improve the model performance on\nboth new and old tasks during the lifelong learning process. Inspired by the\nbiological process of human cognition where the somatosensory neocortex and the\nhippocampus work together in memory consolidation, we formulated a model called\nKnowledge Refreshing and Consolidation (KRC) that achieves both positive\nforward and backward transfer. More specifically, a knowledge refreshing scheme\nis incorporated with the knowledge rehearsal mechanism to enable bi-directional\nknowledge transfer by introducing a dynamic memory model and an adaptive\nworking model. Moreover, a knowledge consolidation scheme operating on the dual\nspace further improves model stability over the long term. Extensive\nevaluations show KRC's superiority over the state-of-the-art LReID methods on\nchallenging pedestrian benchmarks.\n", "date": "2022-11-29", "categories": []}
{"id": "2211.14443", "title": "Siamese based Neural Network for Offline Writer Identification on word\n  level data", "abstract": "  Handwriting recognition is one of the desirable attributes of document\ncomprehension and analysis. It is concerned with the documents writing style\nand characteristics that distinguish the authors. The diversity of text images,\nnotably in images with varying handwriting, makes the process of learning good\nfeatures difficult in cases where little data is available. In this paper, we\npropose a novel scheme to identify the author of a document based on the input\nword image. Our method is text independent and does not impose any constraint\non the size of the input image under examination. To begin with, we detect\ncrucial components in handwriting and extract regions surrounding them using\nScale Invariant Feature Transform (SIFT). These patches are designed to capture\nindividual writing features (including allographs, characters, or combinations\nof characters) that are likely to be unique for an individual writer. These\nfeatures are then passed through a deep Convolutional Neural Network (CNN) in\nwhich the weights are learned by applying the concept of Similarity learning\nusing Siamese network. Siamese network enhances the discrimination power of CNN\nby mapping similarity between different pairs of input image. Features learned\nat different scales of the extracted SIFT key-points are encoded using Sparse\nPCA, each components of the Sparse PCA is assigned a saliency score signifying\nits level of significance in discriminating different writers effectively.\nFinally, the weighted Sparse PCA corresponding to each SIFT key-points is\ncombined to arrive at a final classification score for each writer. The\nproposed algorithm was evaluated on two publicly available databases (namely\nIAM and CVL) and is able to achieve promising result, when compared with other\ndeep learning based algorithm.\n", "date": "2022-11-17", "categories": []}
{"id": "2211.14716", "title": "Fingerprint Pore Detection: A Survey", "abstract": "  This work presents the first survey on fingerprint pore detection. The survey\nprovides a general overview of the field and discusses methods, datasets, and\nevaluation protocols. We also present a baseline method inspired on the\nstate-of-the-art that implements a customizable Fully Convolutional Network,\nwhose hyperparameters were tuned to achieve optimal pore detection rates.\nFinally, we also reimplementated three other approaches proposed in the\nliterature for evaluation purposes. We have made the source code of (1) the\nbaseline method, (2) the reimplemented approaches, and (3) the training and\nevaluation processes for two different datasets available to the public to\nattract more researchers to the field and to facilitate future comparisons\nunder the same conditions. The code is available in the following repository:\nhttps://github.com/azimIbragimov/Fingerprint-Pore-Detection-A-Survey\n", "date": "2022-11-27", "categories": []}
{"id": "2210.08836", "title": "MSDS: A Large-Scale Chinese Signature and Token Digit String Dataset for\n  Handwriting Verification", "abstract": "  Although online handwriting verification has made great progress recently,\nthe verification performances are still far behind the real usage owing to the\nsmall scale of the datasets as well as the limited biometric mediums.\nTherefore, this paper proposes a new handwriting verification benchmark dataset\nnamed Multimodal Signature and Digit String (MSDS), which consists of two\nsubsets: MSDS-ChS (Chinese Signatures) and MSDS-TDS (Token Digit Strings),\ncontributed by 402 users, with 20 genuine samples and 20 skilled forgeries per\nuser per subset. MSDS-ChS consists of handwritten Chinese signatures, which, to\nthe best of our knowledge, is the largest publicly available Chinese signature\ndataset for handwriting verification, at least eight times larger than existing\nonline datasets. Meanwhile, MSDS-TDS consists of handwritten Token Digit\nStrings, i.e, the actual phone numbers of users, which have not been explored\nyet. Extensive experiments with different baselines are respectively conducted\nfor MSDS-ChS and MSDS-TDS. Surprisingly, verification performances of\nstate-of-the-art methods on MSDS-TDS are generally better than those on\nMSDS-ChS, which indicates that the handwritten Token Digit String could be a\nmore effective biometric than handwritten Chinese signature. This is a\npromising discovery that could inspire us to explore new biometric traits. The\nMSDS dataset is available at https://github.com/HCIILAB/MSDS.\n", "date": "2022-10-17", "categories": []}
{"id": "2211.13554", "title": "Quality-Based Conditional Processing in Multi-Biometrics: Application to\n  Sensor Interoperability", "abstract": "  As biometric technology is increasingly deployed, it will be common to\nreplace parts of operational systems with newer designs. The cost and\ninconvenience of reacquiring enrolled users when a new vendor solution is\nincorporated makes this approach difficult and many applications will require\nto deal with information from different sources regularly. These\ninteroperability problems can dramatically affect the performance of biometric\nsystems and thus, they need to be overcome. Here, we describe and evaluate the\nATVS-UAM fusion approach submitted to the quality-based evaluation of the 2007\nBioSecure Multimodal Evaluation Campaign, whose aim was to compare fusion\nalgorithms when biometric signals were generated using several biometric\ndevices in mismatched conditions. Quality measures from the raw biometric data\nare available to allow system adjustment to changing quality conditions due to\ndevice changes. This system adjustment is referred to as quality-based\nconditional processing. The proposed fusion approach is based on linear\nlogistic regression, in which fused scores tend to be log-likelihood-ratios.\nThis allows the easy and efficient combination of matching scores from\ndifferent devices assuming low dependence among modalities. In our system,\nquality information is used to switch between different system modules\ndepending on the data source (the sensor in our case) and to reject channels\nwith low quality data during the fusion. We compare our fusion approach to a\nset of rule-based fusion schemes over normalized scores. Results show that the\nproposed approach outperforms all the rule-based fusion schemes. We also show\nthat with the quality-based channel rejection scheme, an overall improvement of\n25% in the equal error rate is obtained.\n", "date": "2022-11-24", "categories": []}
{"id": "2211.11925", "title": "Multimodal Data Augmentation for Visual-Infrared Person ReID with\n  Corrupted Data", "abstract": "  The re-identification (ReID) of individuals over a complex network of cameras\nis a challenging task, especially under real-world surveillance conditions.\nSeveral deep learning models have been proposed for visible-infrared (V-I)\nperson ReID to recognize individuals from images captured using RGB and IR\ncameras. However, performance may decline considerably if RGB and IR images\ncaptured at test time are corrupted (e.g., noise, blur, and weather\nconditions). Although various data augmentation (DA) methods have been explored\nto improve the generalization capacity, these are not adapted for V-I person\nReID. In this paper, a specialized DA strategy is proposed to address this\nmultimodal setting. Given both the V and I modalities, this strategy allows to\ndiminish the impact of corruption on the accuracy of deep person ReID models.\nCorruption may be modality-specific, and an additional modality often provides\ncomplementary information. Our multimodal DA strategy is designed specifically\nto encourage modality collaboration and reinforce generalization capability.\nFor instance, punctual masking of modalities forces the model to select the\ninformative modality. Local DA is also explored for advanced selection of\nfeatures within and among modalities. The impact of training baseline fusion\nmodels for V-I person ReID using the proposed multimodal DA strategy is\nassessed on corrupted versions of the SYSU-MM01, RegDB, and ThermalWORLD\ndatasets in terms of complexity and efficiency. Results indicate that using our\nstrategy provides V-I ReID models the ability to exploit both shared and\nindividual modality knowledge so they can outperform models trained with no or\nunimodal DA. GitHub code: https://github.com/art2611/ML-MDA.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.12280", "title": "Transformer Based Multi-Grained Features for Unsupervised Person\n  Re-Identification", "abstract": "  Multi-grained features extracted from convolutional neural networks (CNNs)\nhave demonstrated their strong discrimination ability in supervised person\nre-identification (Re-ID) tasks. Inspired by them, this work investigates the\nway of extracting multi-grained features from a pure transformer network to\naddress the unsupervised Re-ID problem that is label-free but much more\nchallenging. To this end, we build a dual-branch network architecture based\nupon a modified Vision Transformer (ViT). The local tokens output in each\nbranch are reshaped and then uniformly partitioned into multiple stripes to\ngenerate part-level features, while the global tokens of two branches are\naveraged to produce a global feature. Further, based upon offline-online\nassociated camera-aware proxies (O2CAP) that is a top-performing unsupervised\nRe-ID method, we define offline and online contrastive learning losses with\nrespect to both global and part-level features to conduct unsupervised\nlearning. Extensive experiments on three person Re-ID datasets show that the\nproposed method outperforms state-of-the-art unsupervised methods by a\nconsiderable margin, greatly mitigating the gap to supervised counterparts.\nCode will be available soon at https://github.com/RikoLi/WACV23-workshop-TMGF.\n", "date": "2022-11-22", "categories": []}
{"id": "2211.05410", "title": "Robust Smart Home Face Recognition under Starving Federated Data", "abstract": "  Over the past few years, the field of adversarial attack received numerous\nattention from various researchers with the help of successful attack success\nrate against well-known deep neural networks that were acknowledged to achieve\nhigh classification ability in various tasks. However, majority of the\nexperiments were completed under a single model, which we believe it may not be\nan ideal case in a real-life situation. In this paper, we introduce a novel\nfederated adversarial training method for smart home face recognition, named\nFLATS, where we observed some interesting findings that may not be easily\nnoticed in a traditional adversarial attack to federated learning experiments.\nBy applying different variations to the hyperparameters, we have spotted that\nour method can make the global model to be robust given a starving federated\nenvironment. Our code can be found on https://github.com/jcroh0508/FLATS.\n", "date": "2022-11-10", "categories": []}
{"id": "2211.08565", "title": "Using Auxiliary Information for Person Re-Identification -- A Tutorial\n  Overview", "abstract": "  Person re-identification (re-id) is a pivotal task within an intelligent\nsurveillance pipeline and there exist numerous re-id frameworks that achieve\nsatisfactory performance in challenging benchmarks. However, these systems\nstruggle to generate acceptable results when there are significant differences\nbetween the camera views, illumination conditions, or occlusions. This result\ncan be attributed to the deficiency that exists within many recently proposed\nre-id pipelines where they are predominately driven by appearance-based\nfeatures and little attention is paid to other auxiliary information that could\naid the re-id. In this paper, we systematically review the current\nState-Of-The-Art (SOTA) methods in both uni-modal and multimodal person re-id.\nExtending beyond a conceptual framework, we illustrate how the existing SOTA\nmethods can be extended to support these additional auxiliary information and\nquantitatively evaluate the utility of such auxiliary feature information,\nranging from logos printed on the objects carried by the subject or printed on\nthe clothes worn by the subject, through to his or her behavioural\ntrajectories. To the best of our knowledge, this is the first work that\nexplores the fusion of multiple information to generate a more discriminant\nperson descriptor and the principal aim of this paper is to provide a thorough\ntheoretical analysis regarding the implementation of such a framework. In\naddition, using model interpretation techniques, we validate the contributions\nfrom different combinations of the auxiliary information versus the original\nfeatures that the SOTA person re-id models extract. We outline the limitations\nof the proposed approaches and propose future research directions that could be\npursued to advance the area of multi-modal person re-id.\n", "date": "2022-11-15", "categories": []}
{"id": "2211.03242", "title": "Fast Key Points Detection and Matching for Tree-Structured Images", "abstract": "  This paper offers a new authentication algorithm based on image matching of\nnano-resolution visual identifiers with tree-shaped patterns. The algorithm\nincludes image-to-tree conversion by greedy extraction of the fractal pattern\nskeleton along with a custom-built graph matching algorithm that is robust\nagainst imaging artifacts such as scaling, rotation, scratch, and illumination\nchange. The proposed algorithm is applicable to a variety of tree-structured\nimage matching, but our focus is on dendrites, recently-developed visual\nidentifiers. Dendrites are entropy rich and unclonable with existing 2D and 3D\nprinters due to their natural randomness, nano-resolution granularity, and 3D\nfacets, making them an appropriate choice for security applications such as\nsupply chain trace and tracking. The proposed algorithm improves upon graph\nmatching with standard image descriptors. For instance, image inconsistency due\nto the camera sensor noise may cause unexpected feature extraction leading to\ninaccurate tree conversion and authentication failure. Also, previous tree\nextraction algorithms are prohibitively slow hindering their scalability to\nlarge systems. In this paper, we fix the current issues of [1] and accelerate\nthe key points extraction up to 10-times faster by implementing a new skeleton\nextraction method, a new key points searching algorithm, as well as an\noptimized key point matching algorithm. Using minimum enclosing circle and\ncenter points, make the algorithm robust to the choice of pattern shape. In\ncontrast to [1] our algorithm handles general graphs with loop connections,\ntherefore is applicable to a wider range of applications such as transportation\nmap analysis, fingerprints, and retina vessel imaging.\n", "date": "2022-11-07", "categories": []}
{"id": "2211.07371", "title": "Unsupervised Face Recognition using Unlabeled Synthetic Data", "abstract": "  Over the past years, the main research innovations in face recognition\nfocused on training deep neural networks on large-scale identity-labeled\ndatasets using variations of multi-class classification losses. However, many\nof these datasets are retreated by their creators due to increased privacy and\nethical concerns. Very recently, privacy-friendly synthetic data has been\nproposed as an alternative to privacy-sensitive authentic data to comply with\nprivacy regulations and to ensure the continuity of face recognition research.\nIn this paper, we propose an unsupervised face recognition model based on\nunlabeled synthetic data (USynthFace). Our proposed USynthFace learns to\nmaximize the similarity between two augmented images of the same synthetic\ninstance. We enable this by a large set of geometric and color transformations\nin addition to GAN-based augmentation that contributes to the USynthFace model\ntraining. We also conduct numerous empirical studies on different components of\nour USynthFace. With the proposed set of augmentation operations, we proved the\neffectiveness of our USynthFace in achieving relatively high recognition\naccuracies using unlabeled synthetic data.\n", "date": "2022-11-14", "categories": []}
{"id": "2211.05970", "title": "Palm Vein Recognition via Multi-task Loss Function and Attention Layer", "abstract": "  With the improvement of arithmetic power and algorithm accuracy of personal\ndevices, biological features are increasingly widely used in personal\nidentification, and palm vein recognition has rich extractable features and has\nbeen widely studied in recent years. However, traditional recognition methods\nare poorly robust and susceptible to environmental influences such as\nreflections and noise. In this paper, a convolutional neural network based on\nVGG-16 transfer learning fused attention mechanism is used as the feature\nextraction network on the infrared palm vein dataset. The palm vein\nclassification task is first trained using palmprint classification methods,\nfollowed by matching using a similarity function, in which we propose the\nmulti-task loss function to improve the accuracy of the matching task. In order\nto verify the robustness of the model, some experiments were carried out on\ndatasets from different sources. Then, we used K-means clustering to determine\nthe adaptive matching threshold and finally achieved an accuracy rate of 98.89%\non prediction set. At the same time, the matching is with high efficiency which\ntakes an average of 0.13 seconds per palm vein pair, and that means our method\ncan be adopted in practice.\n", "date": "2022-11-11", "categories": []}
{"id": "2211.06408", "title": "Physically-Based Face Rendering for NIR-VIS Face Recognition", "abstract": "  Near infrared (NIR) to Visible (VIS) face matching is challenging due to the\nsignificant domain gaps as well as a lack of sufficient data for cross-modality\nmodel training. To overcome this problem, we propose a novel method for paired\nNIR-VIS facial image generation. Specifically, we reconstruct 3D face shape and\nreflectance from a large 2D facial dataset and introduce a novel method of\ntransforming the VIS reflectance to NIR reflectance. We then use a\nphysically-based renderer to generate a vast, high-resolution and\nphotorealistic dataset consisting of various poses and identities in the NIR\nand VIS spectra. Moreover, to facilitate the identity feature learning, we\npropose an IDentity-based Maximum Mean Discrepancy (ID-MMD) loss, which not\nonly reduces the modality gap between NIR and VIS images at the domain level\nbut encourages the network to focus on the identity features instead of facial\ndetails, such as poses and accessories. Extensive experiments conducted on four\nchallenging NIR-VIS face recognition benchmarks demonstrate that the proposed\nmethod can achieve comparable performance with the state-of-the-art (SOTA)\nmethods without requiring any existing NIR-VIS face recognition datasets. With\nslightly fine-tuning on the target NIR-VIS face recognition datasets, our\nmethod can significantly surpass the SOTA performance. Code and pretrained\nmodels are released under the insightface\n(https://github.com/deepinsight/insightface/tree/master/recognition).\n", "date": "2022-11-11", "categories": []}
{"id": "2211.04656", "title": "MEVID: Multi-view Extended Videos with Identities for Video Person\n  Re-Identification", "abstract": "  In this paper, we present the Multi-view Extended Videos with Identities\n(MEVID) dataset for large-scale, video person re-identification (ReID) in the\nwild. To our knowledge, MEVID represents the most-varied video person ReID\ndataset, spanning an extensive indoor and outdoor environment across nine\nunique dates in a 73-day window, various camera viewpoints, and entity clothing\nchanges. Specifically, we label the identities of 158 unique people wearing 598\noutfits taken from 8, 092 tracklets, average length of about 590 frames, seen\nin 33 camera views from the very large-scale MEVA person activities dataset.\nWhile other datasets have more unique identities, MEVID emphasizes a richer set\nof information about each individual, such as: 4 outfits/identity vs. 2\noutfits/identity in CCVID, 33 viewpoints across 17 locations vs. 6 in 5\nsimulated locations for MTA, and 10 million frames vs. 3 million for LS-VID.\nBeing based on the MEVA video dataset, we also inherit data that is\nintentionally demographically balanced to the continental United States. To\naccelerate the annotation process, we developed a semi-automatic annotation\nframework and GUI that combines state-of-the-art real-time models for object\ndetection, pose estimation, person ReID, and multi-object tracking. We evaluate\nseveral state-of-the-art methods on MEVID challenge problems and\ncomprehensively quantify their robustness in terms of changes of outfit, scale,\nand background location. Our quantitative analysis on the realistic, unique\naspects of MEVID shows that there are significant remaining challenges in video\nperson ReID and indicates important directions for future research.\n", "date": "2022-11-09", "categories": []}
{"id": "2211.05544", "title": "Near-infrared and visible-light periocular recognition with Gabor\n  features using frequency-adaptive automatic eye detection", "abstract": "  Periocular recognition has gained attention recently due to demands of\nincreased robustness of face or iris in less controlled scenarios. We present a\nnew system for eye detection based on complex symmetry filters, which has the\nadvantage of not needing training. Also, separability of the filters allows\nfaster detection via one-dimensional convolutions. This system is used as input\nto a periocular algorithm based on retinotopic sampling grids and Gabor\nspectrum decomposition. The evaluation framework is composed of six databases\nacquired both with near-infrared and visible sensors. The experimental setup is\ncomplemented with four iris matchers, used for fusion experiments. The eye\ndetection system presented shows very high accuracy with near-infrared data,\nand a reasonable good accuracy with one visible database. Regarding the\nperiocular system, it exhibits great robustness to small errors in locating the\neye centre, as well as to scale changes of the input image. The density of the\nsampling grid can also be reduced without sacrificing accuracy. Lastly, despite\nthe poorer performance of the iris matchers with visible data, fusion with the\nperiocular system can provide an improvement of more than 20%. The six\ndatabases used have been manually annotated, with the annotation made publicly\navailable.\n", "date": "2022-11-10", "categories": []}
{"id": "2211.05629", "title": "Haven't I Seen You Before? Assessing Identity Leakage in Synthetic\n  Irises", "abstract": "  Generative Adversarial Networks (GANs) have proven to be a preferred method\nof synthesizing fake images of objects, such as faces, animals, and\nautomobiles. It is not surprising these models can also generate ISO-compliant,\nyet synthetic iris images, which can be used to augment training data for iris\nmatchers and liveness detectors. In this work, we trained one of the most\nrecent GAN models (StyleGAN3) to generate fake iris images with two primary\ngoals: (i) to understand the GAN's ability to produce \"never-before-seen\"\nirises, and (ii) to investigate the phenomenon of identity leakage as a\nfunction of the GAN's training time. Previous work has shown that personal\nbiometric data can inadvertently flow from training data into synthetic\nsamples, raising a privacy concern for subjects who accidentally appear in the\ntraining dataset. This paper presents analysis for three different iris\nmatchers at varying points in the GAN training process to diagnose where and\nwhen authentic training samples are in jeopardy of leaking through the\ngenerative process. Our results show that while most synthetic samples do not\nshow signs of identity leakage, a handful of generated samples match authentic\n(training) samples nearly perfectly, with consensus across all matchers. In\norder to prioritize privacy, security, and trust in the machine learning model\ndevelopment process, the research community must strike a delicate balance\nbetween the benefits of using synthetic data and the corresponding threats\nagainst privacy from potential identity leakage.\n", "date": "2022-11-03", "categories": []}
{"id": "2211.05662", "title": "Warmup and Transfer Knowledge-Based Federated Learning Approach for IoT\n  Continuous Authentication", "abstract": "  Continuous behavioural authentication methods add a unique layer of security\nby allowing individuals to verify their unique identity when accessing a\ndevice. Maintaining session authenticity is now feasible by monitoring users'\nbehaviour while interacting with a mobile or Internet of Things (IoT) device,\nmaking credential theft and session hijacking ineffective. Such a technique is\nmade possible by integrating the power of artificial intelligence and Machine\nLearning (ML). Most of the literature focuses on training machine learning for\nthe user by transmitting their data to an external server, subject to private\nuser data exposure to threats. In this paper, we propose a novel Federated\nLearning (FL) approach that protects the anonymity of user data and maintains\nthe security of his data. We present a warmup approach that provides a\nsignificant accuracy increase. In addition, we leverage the transfer learning\ntechnique based on feature extraction to boost the models' performance. Our\nextensive experiments based on four datasets: MNIST, FEMNIST, CIFAR-10 and\nUMDAA-02-FD, show a significant increase in user authentication accuracy while\nmaintaining user privacy and data security.\n", "date": "2022-11-10", "categories": []}
{"id": "2211.04884", "title": "Composite Fixed-Length Ordered Features for Palmprint Template\n  Protection with Diminished Performance Loss", "abstract": "  Palmprint recognition has become more and more popular due to its advantages\nover other biometric modalities such as fingerprint, in that it is larger in\narea, richer in information and able to work at a distance. However, the issue\nof palmprint privacy and security (especially palmprint template protection)\nremains under-studied. Among the very few research works, most of them only use\nthe directional and orientation features of the palmprint with transformation\nprocessing, yielding unsatisfactory protection and identification performance.\nThus, this paper proposes a palmprint template protection-oriented operator\nthat has a fixed length and is ordered in nature, by fusing point features and\norientation features. Firstly, double orientations are extracted with more\naccuracy based on MFRAT. Then key points of SURF are extracted and converted to\nbe fixed-length and ordered features. Finally, composite features that fuse up\nthe double orientations and SURF points are transformed using the irreversible\ntransformation of IOM to generate the revocable palmprint template. Experiments\nshow that the EER after irreversible transformation on the PolyU and CASIA\ndatabases are 0.17% and 0.19% respectively, and the absolute precision loss is\n0.08% and 0.07%, respectively, which proves the advantage of our method.\n", "date": "2022-11-09", "categories": []}
{"id": "2209.09574", "title": "Sampling Agnostic Feature Representation for Long-Term Person\n  Re-identification", "abstract": "  Person re-identification is a problem of identifying individuals across\nnon-overlapping cameras. Although remarkable progress has been made in the\nre-identification problem, it is still a challenging problem due to appearance\nvariations of the same person as well as other people of similar appearance.\nSome prior works solved the issues by separating features of positive samples\nfrom features of negative ones. However, the performances of existing models\nconsiderably depend on the characteristics and statistics of the samples used\nfor training. Thus, we propose a novel framework named sampling independent\nrobust feature representation network (SirNet) that learns disentangled feature\nembedding from randomly chosen samples. A carefully designed sampling\nindependent maximum discrepancy loss is introduced to model samples of the same\nperson as a cluster. As a result, the proposed framework can generate\nadditional hard negatives/positives using the learned features, which results\nin better discriminability from other identities. Extensive experimental\nresults on large-scale benchmark datasets verify that the proposed model is\nmore effective than prior state-of-the-art models.\n", "date": "2022-09-20", "categories": []}
{"id": "2211.03663", "title": "Generalizable Re-Identification from Videos with Cycle Association", "abstract": "  In this paper, we are interested in learning a generalizable person\nre-identification (re-ID) representation from unlabeled videos. Compared with\n1) the popular unsupervised re-ID setting where the training and test sets are\ntypically under the same domain, and 2) the popular domain generalization (DG)\nre-ID setting where the training samples are labeled, our novel scenario\ncombines their key challenges: the training samples are unlabeled, and\ncollected form various domains which do no align with the test domain. In other\nwords, we aim to learn a representation in an unsupervised manner and directly\nuse the learned representation for re-ID in novel domains. To fulfill this\ngoal, we make two main contributions: First, we propose Cycle Association\n(CycAs), a scalable self-supervised learning method for re-ID with low training\ncomplexity; and second, we construct a large-scale unlabeled re-ID dataset\nnamed LMP-video, tailored for the proposed method. Specifically, CycAs learns\nre-ID features by enforcing cycle consistency of instance association between\ntemporally successive video frame pairs, and the training cost is merely linear\nto the data size, making large-scale training possible. On the other hand, the\nLMP-video dataset is extremely large, containing 50 million unlabeled person\nimages cropped from over 10K Youtube videos, therefore is sufficient to serve\nas fertile soil for self-supervised learning. Trained on LMP-video, we show\nthat CycAs learns good generalization towards novel domains. The achieved\nresults sometimes even outperform supervised domain generalizable models.\nRemarkably, CycAs achieves 82.2% Rank-1 on Market-1501 and 49.0% Rank-1 on\nMSMT17 with zero human annotation, surpassing state-of-the-art supervised DG\nre-ID methods. Moreover, we also demonstrate the superiority of CycAs under the\ncanonical unsupervised re-ID and the pretrain-and-finetune scenarios.\n", "date": "2022-11-07", "categories": []}
{"id": "2210.09835", "title": "When Age-Invariant Face Recognition Meets Face Age Synthesis: A\n  Multi-Task Learning Framework and A New Benchmark", "abstract": "  To minimize the impact of age variation on face recognition, age-invariant\nface recognition (AIFR) extracts identity-related discriminative features by\nminimizing the correlation between identity- and age-related features while\nface age synthesis (FAS) eliminates age variation by converting the faces in\ndifferent age groups to the same group. However, AIFR lacks visual results for\nmodel interpretation and FAS compromises downstream recognition due to\nartifacts. Therefore, we propose a unified, multi-task framework to jointly\nhandle these two tasks, termed MTLFace, which can learn the age-invariant\nidentity-related representation for face recognition while achieving pleasing\nface synthesis for model interpretation. Specifically, we propose an\nattention-based feature decomposition to decompose the mixed face features into\ntwo uncorrelated components -- identity- and age-related features -- in a\nspatially constrained way. Unlike the conventional one-hot encoding that\nachieves group-level FAS, we propose a novel identity conditional module to\nachieve identity-level FAS, which can improve the age smoothness of synthesized\nfaces through a weight-sharing strategy. Benefiting from the proposed\nmulti-task framework, we then leverage those high-quality synthesized faces\nfrom FAS to further boost AIFR via a novel selective fine-tuning strategy.\nFurthermore, to advance both AIFR and FAS, we collect and release a large\ncross-age face dataset with age and gender annotations, and a new benchmark\nspecifically designed for tracing long-missing children. Extensive experimental\nresults on five benchmark cross-age datasets demonstrate that MTLFace yields\nsuperior performance for both AIFR and FAS. We further validate MTLFace on two\npopular general face recognition datasets, obtaining competitive performance on\nface recognition in the wild. Code is available at\nhttp://hzzone.github.io/MTLFace.\n", "date": "2022-10-17", "categories": []}
{"id": "2211.02952", "title": "A Comparative Analysis of the Face Recognition Methods in Video\n  Surveillance Scenarios", "abstract": "  Facial recognition is fundamental for a wide variety of security systems\noperating in real-time applications. In video surveillance based face\nrecognition, face images are typically captured over multiple frames in\nuncontrolled conditions; where head pose, illumination, shadowing, motion blur\nand focus change over the sequence. We can generalize that the three\nfundamental operations involved in the facial recognition tasks: face\ndetection, face alignment and face recognition. This study presents comparative\nbenchmark tables for the state-of-art face recognition methods by testing them\nwith same backbone architecture in order to focus only on the face recognition\nsolution instead of network architecture. For this purpose, we constructed a\nvideo surveillance dataset of face IDs that has high age variance, intra-class\nvariance (face make-up, beard, etc.) with native surveillance facial imagery\ndata for evaluation. On the other hand, this work discovers the best\nrecognition methods for different conditions like non-masked faces, masked\nfaces, and faces with glasses.\n", "date": "2022-11-05", "categories": []}
{"id": "2211.03626", "title": "Camera Alignment and Weighted Contrastive Learning for Domain Adaptation\n  in Video Person ReID", "abstract": "  Systems for person re-identification (ReID) can achieve a high accuracy when\ntrained on large fully-labeled image datasets. However, the domain shift\ntypically associated with diverse operational capture conditions (e.g., camera\nviewpoints and lighting) may translate to a significant decline in performance.\nThis paper focuses on unsupervised domain adaptation (UDA) for video-based ReID\n- a relevant scenario that is less explored in the literature. In this\nscenario, the ReID model must adapt to a complex target domain defined by a\nnetwork of diverse video cameras based on tracklet information. State-of-art\nmethods cluster unlabeled target data, yet domain shifts across target cameras\n(sub-domains) can lead to poor initialization of clustering methods that\npropagates noise across epochs, thus preventing the ReID model to accurately\nassociate samples of same identity. In this paper, an UDA method is introduced\nfor video person ReID that leverages knowledge on video tracklets, and on the\ndistribution of frames captured over target cameras to improve the performance\nof CNN backbones trained using pseudo-labels. Our method relies on an\nadversarial approach, where a camera-discriminator network is introduced to\nextract discriminant camera-independent representations, facilitating the\nsubsequent clustering. In addition, a weighted contrastive loss is proposed to\nleverage the confidence of clusters, and mitigate the risk of incorrect\nidentity associations. Experimental results obtained on three challenging\nvideo-based person ReID datasets - PRID2011, iLIDS-VID, and MARS - indicate\nthat our proposed method can outperform related state-of-the-art methods. Our\ncode is available at: \\url{https://github.com/dmekhazni/CAWCL-ReID}\n", "date": "2022-11-07", "categories": []}
{"id": "2211.02622", "title": "PhysioGait: Context-Aware Physiological Context Modeling for Person\n  Re-identification Attack on Wearable Sensing", "abstract": "  Person re-identification is a critical privacy breach in publicly shared\nhealthcare data. We investigate the possibility of a new type of privacy threat\non publicly shared privacy insensitive large scale wearable sensing data. In\nthis paper, we investigate user specific biometric signatures in terms of two\ncontextual biometric traits, physiological (photoplethysmography and\nelectrodermal activity) and physical (accelerometer) contexts. In this regard,\nwe propose PhysioGait, a context-aware physiological signal model that consists\nof a Multi-Modal Siamese Convolutional Neural Network (mmSNN) which learns the\nspatial and temporal information individually and performs sensor fusion in a\nSiamese cost with the objective of predicting a person's identity. We evaluated\nPhysioGait attack model using 4 real-time collected datasets (3-data under IRB\n#HP-00064387 and one publicly available data) and two combined datasets\nachieving 89% - 93% accuracy of re-identifying persons.\n", "date": "2022-10-30", "categories": []}
{"id": "2211.01917", "title": "Expanding Accurate Person Recognition to New Altitudes and Ranges: The\n  BRIAR Dataset", "abstract": "  Face recognition technology has advanced significantly in recent years due\nlargely to the availability of large and increasingly complex training datasets\nfor use in deep learning models. These datasets, however, typically comprise\nimages scraped from news sites or social media platforms and, therefore, have\nlimited utility in more advanced security, forensics, and military\napplications. These applications require lower resolution, longer ranges, and\nelevated viewpoints. To meet these critical needs, we collected and curated the\nfirst and second subsets of a large multi-modal biometric dataset designed for\nuse in the research and development (R&D) of biometric recognition technologies\nunder extremely challenging conditions. Thus far, the dataset includes more\nthan 350,000 still images and over 1,300 hours of video footage of\napproximately 1,000 subjects. To collect this data, we used Nikon DSLR cameras,\na variety of commercial surveillance cameras, specialized long-rage R&D\ncameras, and Group 1 and Group 2 UAV platforms. The goal is to support the\ndevelopment of algorithms capable of accurately recognizing people at ranges up\nto 1,000 m and from high angles of elevation. These advances will include\nimprovements to the state of the art in face recognition and will support new\nresearch in the area of whole-body recognition using methods based on gait and\nanthropometry. This paper describes methods used to collect and curate the\ndataset, and the dataset's characteristics at the current stage.\n", "date": "2022-11-03", "categories": []}
{"id": "2209.06172", "title": "Comparative analysis of segmentation and generative models for\n  fingerprint retrieval task", "abstract": "  Biometric Authentication like Fingerprints has become an integral part of the\nmodern technology for authentication and verification of users. It is pervasive\nin more ways than most of us are aware of. However, these fingerprint images\ndeteriorate in quality if the fingers are dirty, wet, injured or when sensors\nmalfunction. Therefore, extricating the original fingerprint by removing the\nnoise and inpainting it to restructure the image is crucial for its\nauthentication. Hence, this paper proposes a deep learning approach to address\nthese issues using Generative (GAN) and Segmentation models. Qualitative and\nQuantitative comparison has been done between pix2pixGAN and cycleGAN\n(generative models) as well as U-net (segmentation model). To train the model,\nwe created our own dataset NFD - Noisy Fingerprint Dataset meticulously with\ndifferent backgrounds along with scratches in some images to make it more\nrealistic and robust. In our research, the u-net model performed better than\nthe GAN networks\n", "date": "2022-09-13", "categories": []}
{"id": "2211.00098", "title": "Synthetic ID Card Image Generation for Improving Presentation Attack\n  Detection", "abstract": "  Currently, it is ever more common to access online services for activities\nwhich formerly required physical attendance. From banking operations to visa\napplications, a significant number of processes have been digitised, especially\nsince the advent of the COVID-19 pandemic, requiring remote biometric\nauthentication of the user. On the downside, some subjects intend to interfere\nwith the normal operation of remote systems for personal profit by using fake\nidentity documents, such as passports and ID cards. Deep learning solutions to\ndetect such frauds have been presented in the literature. However, due to\nprivacy concerns and the sensitive nature of personal identity documents,\ndeveloping a dataset with the necessary number of examples for training deep\nneural networks is challenging. This work explores three methods for\nsynthetically generating ID card images to increase the amount of data while\ntraining fraud-detection networks. These methods include computer vision\nalgorithms and Generative Adversarial Networks. Our results indicate that\ndatabases can be supplemented with synthetic images without any loss in\nperformance for the print/scan Presentation Attack Instrument Species (PAIS)\nand a loss in performance of 1% for the screen capture PAIS.\n", "date": "2022-10-31", "categories": []}
{"id": "2211.00129", "title": "Is Facial Recognition Biased at Near-Infrared Spectrum As Well?", "abstract": "  Published academic research and media articles suggest face recognition is\nbiased across demographics. Specifically, unequal performance is obtained for\nwomen, dark-skinned people, and older adults. However, these published studies\nhave examined the bias of facial recognition in the visible spectrum (VIS).\nFactors such as facial makeup, facial hair, skin color, and illumination\nvariation have been attributed to the bias of this technology at the VIS. The\nnear-infrared (NIR) spectrum offers an advantage over the VIS in terms of\nrobustness to factors such as illumination changes, facial makeup, and skin\ncolor. Therefore, it is worthwhile to investigate the bias of facial\nrecognition at the near-infrared spectrum (NIR). This first study investigates\nthe bias of the face recognition systems at the NIR spectrum. To this aim, two\npopular NIR facial image datasets namely, CASIA-Face-Africa and Notre-Dame-NIVL\nconsisting of African and Caucasian subjects, respectively, are used to\ninvestigate the bias of facial recognition technology across gender and race.\nInterestingly, experimental results suggest equitable face recognition\nperformance across gender and race at the NIR spectrum.\n", "date": "2022-10-31", "categories": []}
{"id": "2211.00312", "title": "HDNet: Hierarchical Dynamic Network for Gait Recognition using\n  Millimeter-Wave Radar", "abstract": "  Gait recognition is widely used in diversified practical applications.\nCurrently, the most prevalent approach is to recognize human gait from RGB\nimages, owing to the progress of computer vision technologies. Nevertheless,\nthe perception capability of RGB cameras deteriorates in rough circumstances,\nand visual surveillance may cause privacy invasion. Due to the robustness and\nnon-invasive feature of millimeter wave (mmWave) radar, radar-based gait\nrecognition has attracted increasing attention in recent years. In this\nresearch, we propose a Hierarchical Dynamic Network (HDNet) for gait\nrecognition using mmWave radar. In order to explore more dynamic information,\nwe propose point flow as a novel point clouds descriptor. We also devise a\ndynamic frame sampling module to promote the efficiency of computation without\ndeteriorating performance noticeably. To prove the superiority of our methods,\nwe perform extensive experiments on two public mmWave radar-based gait\nrecognition datasets, and the results demonstrate that our model is superior to\nexisting state-of-the-art methods.\n", "date": "2022-11-01", "categories": []}
{"id": "2210.15510", "title": "Fusion-based Few-Shot Morphing Attack Detection and Fingerprinting", "abstract": "  The vulnerability of face recognition systems to morphing attacks has posed a\nserious security threat due to the wide adoption of face biometrics in the real\nworld. Most existing morphing attack detection (MAD) methods require a large\namount of training data and have only been tested on a few predefined attack\nmodels. The lack of good generalization properties, especially in view of the\ngrowing interest in developing novel morphing attacks, is a critical limitation\nwith existing MAD research. To address this issue, we propose to extend MAD\nfrom supervised learning to few-shot learning and from binary detection to\nmulticlass fingerprinting in this paper. Our technical contributions include:\n1) We propose a fusion-based few-shot learning (FSL) method to learn\ndiscriminative features that can generalize to unseen morphing attack types\nfrom predefined presentation attacks; 2) The proposed FSL based on the fusion\nof the PRNU model and Noiseprint network is extended from binary MAD to\nmulticlass morphing attack fingerprinting (MAF). 3) We have collected a\nlarge-scale database, which contains five face datasets and eight different\nmorphing algorithms, to benchmark the proposed few-shot MAF (FS-MAF) method.\nExtensive experimental results show the outstanding performance of our\nfusion-based FS-MAF. The code and data will be publicly available at\nhttps://github.com/nz0001na/mad maf.\n", "date": "2022-10-27", "categories": []}
{"id": "2209.04093", "title": "Learning Audio-Visual embedding for Person Verification in the Wild", "abstract": "  It has already been observed that audio-visual embedding is more robust than\nuni-modality embedding for person verification. Here, we proposed a novel\naudio-visual strategy that considers aggregators from a fusion perspective.\nFirst, we introduced weight-enhanced attentive statistics pooling for the first\ntime in face verification. We find that a strong correlation exists between\nmodalities during pooling, so joint attentive pooling is proposed which\ncontains cycle consistency to learn the implicit inter-frame weight. Finally,\neach modality is fused with a gated attention mechanism to gain robust\naudio-visual embedding. All the proposed models are trained on the VoxCeleb2\ndev dataset and the best system obtains 0.18%, 0.27%, and 0.49% EER on three\nofficial trial lists of VoxCeleb1 respectively, which is to our knowledge the\nbest-published results for person verification.\n", "date": "2022-09-09", "categories": []}
{"id": "2210.13994", "title": "Minutiae-Guided Fingerprint Embeddings via Vision Transformers", "abstract": "  Minutiae matching has long dominated the field of fingerprint recognition.\nHowever, deep networks can be used to extract fixed-length embeddings from\nfingerprints. To date, the few studies that have explored the use of CNN\narchitectures to extract such embeddings have shown extreme promise. Inspired\nby these early works, we propose the first use of a Vision Transformer (ViT) to\nlearn a discriminative fixed-length fingerprint embedding. We further\ndemonstrate that by guiding the ViT to focus in on local, minutiae related\nfeatures, we can boost the recognition performance. Finally, we show that by\nfusing embeddings learned by CNNs and ViTs we can reach near parity with a\ncommercial state-of-the-art (SOTA) matcher. In particular, we obtain a\nTAR=94.23% @ FAR=0.1% on the NIST SD 302 public-domain dataset, compared to a\nSOTA commercial matcher which obtains TAR=96.71% @ FAR=0.1%. Additionally, our\nfixed-length embeddings can be matched orders of magnitude faster than the\ncommercial system (2.5 million matches/second compared to 50K matches/second).\nWe make our code and models publicly available to encourage further research on\nthis topic: https://github.com/tba.\n", "date": "2022-10-25", "categories": []}
{"id": "2210.01585", "title": "How Image Generation Helps Visible-to-Infrared Person Re-Identification?", "abstract": "  Compared to visible-to-visible (V2V) person re-identification (ReID), the\nvisible-to-infrared (V2I) person ReID task is more challenging due to the lack\nof sufficient training samples and the large cross-modality discrepancy.\n  To this end, we propose Flow2Flow, a unified framework that could jointly\nachieve training sample expansion and cross-modality image generation for V2I\nperson ReID.\n  Specifically, Flow2Flow learns bijective transformations from both the\nvisible image domain and the infrared domain to a shared isotropic Gaussian\ndomain with an invertible visible flow-based generator and an infrared one,\nrespectively.\n  With Flow2Flow, we are able to generate pseudo training samples by the\ntransformation from latent Gaussian noises to visible or infrared images, and\ngenerate cross-modality images by transformations from existing-modality images\nto latent Gaussian noises to missing-modality images.\n  For the purpose of identity alignment and modality alignment of generated\nimages, we develop adversarial training strategies to train Flow2Flow.\n  Specifically, we design an image encoder and a modality discriminator for\neach modality.\n  The image encoder encourages the generated images to be similar to real\nimages of the same identity via identity adversarial training, and the modality\ndiscriminator makes the generated images modal-indistinguishable from real\nimages via modality adversarial training.\n  Experimental results on SYSU-MM01 and RegDB demonstrate that both training\nsample expansion and cross-modality image generation can significantly improve\nV2I ReID accuracy.\n", "date": "2022-10-04", "categories": []}
{"id": "2210.13999", "title": "Unsupervised domain-adaptive person re-identification with multi-camera\n  constraints", "abstract": "  Person re-identification is a key technology for analyzing video-based human\nbehavior; however, its application is still challenging in practical situations\ndue to the performance degradation for domains different from those in the\ntraining data. Here, we propose an environment-constrained adaptive network for\nreducing the domain gap. This network refines pseudo-labels estimated via a\nself-training scheme by imposing multi-camera constraints. The proposed method\nincorporates person-pair information without person identity labels obtained\nfrom the environment into the model training. In addition, we develop a method\nthat appropriately selects a person from the pair that contributes to the\nperformance improvement. We evaluate the performance of the network using\npublic and private datasets and confirm the performance surpasses\nstate-of-the-art methods in domains with overlapping camera views. To the best\nof our knowledge, this is the first study on domain-adaptive learning with\nmulti-camera constraints that can be obtained in real environments.\n", "date": "2022-10-25", "categories": []}
{"id": "2210.10972", "title": "A Multimodal Sensor Fusion Framework Robust to Missing Modalities for\n  Person Recognition", "abstract": "  Utilizing the sensor characteristics of the audio, visible camera, and\nthermal camera, the robustness of person recognition can be enhanced. Existing\nmultimodal person recognition frameworks are primarily formulated assuming that\nmultimodal data is always available. In this paper, we propose a novel trimodal\nsensor fusion framework using the audio, visible, and thermal camera, which\naddresses the missing modality problem. In the framework, a novel deep latent\nembedding framework, termed the AVTNet, is proposed to learn multiple latent\nembeddings. Also, a novel loss function, termed missing modality loss, accounts\nfor possible missing modalities based on the triplet loss calculation while\nlearning the individual latent embeddings. Additionally, a joint latent\nembedding utilizing the trimodal data is learnt using the multi-head attention\ntransformer, which assigns attention weights to the different modalities. The\ndifferent latent embeddings are subsequently used to train a deep neural\nnetwork. The proposed framework is validated on the Speaking Faces dataset. A\ncomparative analysis with baseline algorithms shows that the proposed framework\nsignificantly increases the person recognition accuracy while accounting for\nmissing modalities.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.12964", "title": "Non-Contrastive Learning-based Behavioural Biometrics for Smart IoT\n  Devices", "abstract": "  Behaviour biometrics are being explored as a viable alternative to overcome\nthe limitations of traditional authentication methods such as passwords and\nstatic biometrics. Also, they are being considered as a viable authentication\nmethod for IoT devices such as smart headsets with AR/VR capabilities,\nwearables, and erables, that do not have a large form factor or the ability to\nseamlessly interact with the user. Recent behavioural biometric solutions use\ndeep learning models that require large amounts of annotated training data.\nCollecting such volumes of behaviour biometrics data raises privacy and\nusability concerns. To this end, we propose using SimSiam-based non-contrastive\nself-supervised learning to improve the label efficiency of behavioural\nbiometric systems. The key idea is to use large volumes of unlabelled (and\nanonymised) data to build good feature extractors that can be subsequently used\nin supervised settings. Using two EEG datasets, we show that at lower amounts\nof labelled data, non-contrastive learning performs 4%-11% more than\nconventional methods such as supervised learning and data augmentation. We also\nshow that, in general, self-supervised learning methods perform better than\nother baselines. Finally, through careful experimentation, we show various\nmodifications that can be incorporated into the non-contrastive learning\nprocess to archive high performance.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.13101", "title": "Towards an efficient Iris Recognition System on Embedded Devices", "abstract": "  Iris Recognition (IR) is one of the market's most reliable and accurate\nbiometric systems. Today, it is challenging to build NIR-capturing devices\nunder the premise of hardware price reduction. Commercial NIR sensors are\nprotected from modification. The process of building a new device is not\ntrivial because it is required to start from scratch with the process of\ncapturing images with quality, calibrating operational distances, and building\nlightweight software such as eyes/iris detectors and segmentation sub-systems.\nIn light of such challenges, this work aims to develop and implement iris\nrecognition software in an embedding system and calibrate NIR in a contactless\nbinocular setup. We evaluate and contrast speed versus performance obtained\nwith two embedded computers and infrared cameras. Further, a lightweight\nsegmenter sub-system called \"Unet_xxs\" is proposed, which can be used for iris\nsemantic segmentation under restricted memory resources.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.13125", "title": "Iris super-resolution using CNNs: is photo-realism important to iris\n  recognition?", "abstract": "  The use of low-resolution images adopting more relaxed acquisition conditions\nsuch as mobile phones and surveillance videos is becoming increasingly common\nin iris recognition nowadays. Concurrently, a great variety of single image\nsuper-resolution techniques are emerging, especially with the use of\nconvolutional neural networks (CNNs). The main objective of these methods is to\ntry to recover finer texture details generating more photo-realistic images\nbased on the optimisation of an objective function depending basically on the\nCNN architecture and training approach. In this work, the authors explore\nsingle image super-resolution using CNNs for iris recognition. For this, they\ntest different CNN architectures and use different training databases,\nvalidating their approach on a database of 1.872 near infrared iris images and\non a mobile phone image database. They also use quality assessment, visual\nresults and recognition experiments to verify if the photo-realism provided by\nthe CNNs which have already proven to be effective for natural images can\nreflect in a better recognition rate for iris recognition. The results show\nthat using deeper architectures trained with texture databases that provide a\nbalance between edge preservation and the smoothness of the method can lead to\ngood results in the iris recognition process.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.13129", "title": "Facial Soft Biometrics for Recognition in the Wild: Recent Works,\n  Annotation, and COTS Evaluation", "abstract": "  The role of soft biometrics to enhance person recognition systems in\nunconstrained scenarios has not been extensively studied. Here, we explore the\nutility of the following modalities: gender, ethnicity, age, glasses, beard,\nand moustache. We consider two assumptions: 1) manual estimation of soft\nbiometrics and 2) automatic estimation from two commercial off-the-shelf\nsystems (COTS). All experiments are reported using the labeled faces in the\nwild (LFW) database. First, we study the discrimination capabilities of soft\nbiometrics standalone. Then, experiments are carried out fusing soft biometrics\nwith two state-of-the-art face recognition systems based on deep learning. We\nobserve that soft biometrics is a valuable complement to the face modality in\nunconstrained scenarios, with relative improvements up to 40%/15% in the\nverification performance when using manual/automatic soft biometrics\nestimation. Results are reproducible as we make public our manual annotations\nand COTS outputs of soft biometrics over LFW, as well as the face recognition\nscores.\n", "date": "2022-10-24", "categories": []}
{"id": "2210.11939", "title": "Automatic Cattle Identification using YOLOv5 and Mosaic Augmentation: A\n  Comparative Analysis", "abstract": "  You Only Look Once (YOLO) is a single-stage object detection model popular\nfor real-time object detection, accuracy, and speed. This paper investigates\nthe YOLOv5 model to identify cattle in the yards. The current solution to\ncattle identification includes radio-frequency identification (RFID) tags. The\nproblem occurs when the RFID tag is lost or damaged. A biometric solution\nidentifies the cattle and helps to assign the lost or damaged tag or replace\nthe RFID-based system. Muzzle patterns in cattle are unique biometric solutions\nlike a fingerprint in humans. This paper aims to present our recent research in\nutilizing five popular object detection models, looking at the architecture of\nYOLOv5, investigating the performance of eight backbones with the YOLOv5 model,\nand the influence of mosaic augmentation in YOLOv5 by experimental results on\nthe available cattle muzzle images. Finally, we concluded with the excellent\npotential of using YOLOv5 in automatic cattle identification. Our experiments\nshow YOLOv5 with transformer performed best with mean Average Precision (mAP)\n0.5 (the average of AP when the IoU is greater than 50%) of 0.995, and mAP\n0.5:0.95 (the average of AP from 50% to 95% IoU with an interval of 5%) of\n0.9366. In addition, our experiments show the increase in accuracy of the model\nby using mosaic augmentation in all backbones used in our experiments.\nMoreover, we can also detect cattle with partial muzzle images.\n", "date": "2022-10-21", "categories": []}
{"id": "2210.11129", "title": "Super-Resolution and Image Re-projection for Iris Recognition", "abstract": "  Several recent works have addressed the ability of deep learning to disclose\nrich, hierarchical and discriminative models for the most diverse purposes.\nSpecifically in the super-resolution field, Convolutional Neural Networks\n(CNNs) using different deep learning approaches attempt to recover realistic\ntexture and fine grained details from low resolution images. In this work we\nexplore the viability of these approaches for iris Super-Resolution (SR) in an\niris recognition environment. For this, we test different architectures with\nand without a so called image re-projection to reduce artifacts applying it to\ndifferent iris databases to verify the viability of the different CNNs for iris\nsuper-resolution. Results show that CNNs and image re-projection can improve\nthe results specially for the accuracy of recognition systems using a complete\ndifferent training database performing the transfer learning successfully.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.11135", "title": "On-line signature verification using Tablet PC", "abstract": "  On-line signature verification for Tablet PC devices is studied. The on-line\nsignature verification algorithm presented by the authors at the First\nInternational Signature Verification Competition (SVC 2004) is adapted to work\nin Tablet PC environments. An example prototype of securing access and securing\ndocument application using this Tablet PC system is also reported. Two\ndifferent commercial Tablet PCs are evaluated, including information of\ninterest for signature verification systems such as sampling and pressure\nstatistics. Authentication performance experiments are reported considering\nboth random and skilled forgeries by using a new database with over 3000\nsignatures.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.11139", "title": "Sensor interoperability and fusion in signature verification: A case\n  study using tablet PC", "abstract": "  Several works related to information fusion for signature verification have\nbeen presented. However, few works have focused on sensor fusion and sensor\ninteroperability. In this paper, these two topics are evaluated for signature\nverification using two different commercial Tablet PCs. An enrolment strategy\nusing signatures from the two Tablet PCs is also proposed. Authentication\nperformance experiments are reported by using a database with over 3000\nsignatures.\n", "date": "2022-10-20", "categories": []}
{"id": "2210.09153", "title": "Face Pasting Attack", "abstract": "  Cujo AI and Adversa AI hosted the MLSec face recognition challenge. The goal\nwas to attack a black box face recognition model with targeted attacks. The\nmodel returned the confidence of the target class and a stealthiness score. For\nan attack to be considered successful the target class has to have the highest\nconfidence among all classes and the stealthiness has to be at least 0.5. In\nour approach we paste the face of a target into a source image. By utilizing\nposition, scaling, rotation and transparency attributes we reached 3rd place.\nOur approach took approximately 200 queries per attack for the final highest\nscore and about ~7.7 queries minimum for a successful attack. The code is\navailable at https://github.com/bunni90/FacePastingAttack .\n", "date": "2022-10-17", "categories": []}
{"id": "2210.10354", "title": "Stating Comparison Score Uncertainty and Verification Decision\n  Confidence Towards Transparent Face Recognition", "abstract": "  Face Recognition (FR) is increasingly used in critical verification decisions\nand thus, there is a need for assessing the trustworthiness of such decisions.\nThe confidence of a decision is often based on the overall performance of the\nmodel or on the image quality. We propose to propagate model uncertainties to\nscores and decisions in an effort to increase the transparency of verification\ndecisions. This work presents two contributions. First, we propose an approach\nto estimate the uncertainty of face comparison scores. Second, we introduce a\nconfidence measure of the system's decision to provide insights into the\nverification decision. The suitability of the comparison scores uncertainties\nand the verification decision confidences have been experimentally proven on\nthree face recognition models on two datasets.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.10403", "title": "Segmentation-free Direct Iris Localization Networks", "abstract": "  This paper proposes an efficient iris localization method without using iris\nsegmentation and circle fitting. Conventional iris localization methods first\nextract iris regions by using semantic segmentation methods such as U-Net.\nAfterward, the inner and outer iris circles are localized using the traditional\ncircle fitting algorithm. However, this approach requires high-resolution\nencoder-decoder networks for iris segmentation, so it causes computational\ncosts to be high. In addition, traditional circle fitting tends to be sensitive\nto noise in input images and fitting parameters, causing the iris recognition\nperformance to be poor. To solve these problems, we propose an iris\nlocalization network (ILN), that can directly localize pupil and iris circles\nwith eyelid points from a low-resolution iris image. We also introduce a pupil\nrefinement network (PRN) to improve the accuracy of pupil localization.\nExperimental results show that the combination of ILN and PRN works in 34.5 ms\nfor one iris image on a CPU, and its localization performance outperforms\nconventional iris segmentation methods. In addition, generalized evaluation\nresults show that the proposed method has higher robustness for datasets in\ndifferent domain than other segmentation methods. Furthermore, we also confirm\nthat the proposed ILN and PRN improve the iris recognition accuracy.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.10409", "title": "Domain generalization Person Re-identification on Attention-aware\n  multi-operation strategery", "abstract": "  Domain generalization person re-identification (DG Re-ID) aims to directly\ndeploy a model trained on the source domain to the unseen target domain with\ngood generalization, which is a challenging problem and has practical value in\na real-world deployment. In the existing DG Re-ID methods, invariant operations\nare effective in extracting domain generalization features, and Instance\nNormalization (IN) or Batch Normalization (BN) is used to alleviate the bias to\nunseen domains. Due to domain-specific information being used to capture\ndiscriminability of the individual source domain, the generalized ability for\nunseen domains is unsatisfactory. To address this problem, an Attention-aware\nMulti-operation Strategery (AMS) for DG Re-ID is proposed to extract more\ngeneralized features. We investigate invariant operations and construct a\nmulti-operation module based on IN and group whitening (GW) to extract\ndomain-invariant feature representations. Furthermore, we analyze different\ndomain-invariant characteristics, and apply spatial attention to the IN\noperation and channel attention to the GW operation to enhance the\ndomain-invariant features. The proposed AMS module can be used as a\nplug-and-play module to incorporate into existing network architectures.\nExtensive experimental results show that AMS can effectively enhance the\nmodel's generalization ability to unseen domains and significantly improves the\nrecognition performance in DG Re-ID on three protocols with ten datasets.\n", "date": "2022-10-19", "categories": []}
{"id": "2210.10667", "title": "Analysis of Master Vein Attacks on Finger Vein Recognition Systems", "abstract": "  Finger vein recognition (FVR) systems have been commercially used, especially\nin ATMs, for customer verification. Thus, it is essential to measure their\nrobustness against various attack methods, especially when a hand-crafted FVR\nsystem is used without any countermeasure methods. In this paper, we are the\nfirst in the literature to introduce master vein attacks in which we craft a\nvein-looking image so that it can falsely match with as many identities as\npossible by the FVR systems. We present two methods for generating master veins\nfor use in attacking these systems. The first uses an adaptation of the latent\nvariable evolution algorithm with a proposed generative model (a multi-stage\ncombination of beta-VAE and WGAN-GP models). The second uses an adversarial\nmachine learning attack method to attack a strong surrogate CNN-based\nrecognition system. The two methods can be easily combined to boost their\nattack ability. Experimental results demonstrated that the proposed methods\nalone and together achieved false acceptance rates up to 73.29% and 88.79%,\nrespectively, against Miura's hand-crafted FVR system. We also point out that\nMiura's system is easily compromised by non-vein-looking samples generated by a\nWGAN-GP model with false acceptance rates up to 94.21%. The results raise the\nalarm about the robustness of such systems and suggest that master vein attacks\nshould be considered an important security measure.\n", "date": "2022-10-18", "categories": []}
{"id": "2207.08980", "title": "DeformIrisNet: An Identity-Preserving Model of Iris Texture Deformation", "abstract": "  Nonlinear iris texture deformations due to pupil size variations are one of\nthe main factors responsible for within-class variance of genuine comparison\nscores in iris recognition. In dominant approaches to iris recognition, the\nsize of a ring-shaped iris region is linearly scaled to a canonical rectangle,\nused further in encoding and matching. However, the biological complexity of\nthe iris sphincter and dilator muscles causes the movements of iris features to\nbe nonlinear in a function of pupil size, and not solely organized along radial\npaths. Alternatively to the existing theoretical models based on the\nbiomechanics of iris musculature, in this paper we propose a novel deep\nautoencoder-based model that can effectively learn complex movements of iris\ntexture features directly from the data. The proposed model takes two inputs,\n(a) an ISO-compliant near-infrared iris image with initial pupil size, and (b)\nthe binary mask defining the target shape of the iris. The model makes all the\nnecessary nonlinear deformations to the iris texture to match the shape of the\niris in an image (a) with the shape provided by the target mask (b). The\nidentity-preservation component of the loss function helps the model in finding\ndeformations that preserve identity and not only the visual realism of the\ngenerated samples. We also demonstrate two immediate applications of this\nmodel: better compensation for iris texture deformations in iris recognition\nalgorithms, compared to linear models, and the creation of a generative\nalgorithm that can aid human forensic examiners, who may need to compare iris\nimages with a large difference in pupil dilation. We offer the source codes and\nmodel weights available along with this paper.\n", "date": "2022-07-18", "categories": []}
{"id": "2210.09765", "title": "Very Low-Resolution Iris Recognition Via Eigen-Patch Super-Resolution\n  and Matcher Fusion", "abstract": "  Current research in iris recognition is moving towards enabling more relaxed\nacquisition conditions. This has effects on the quality of acquired images,\nwith low resolution being a predominant issue. Here, we evaluate a\nsuper-resolution algorithm used to reconstruct iris images based on\nEigen-transformation of local image patches. Each patch is reconstructed\nseparately, allowing better quality of enhanced images by preserving local\ninformation. Contrast enhancement is used to improve the reconstruction\nquality, while matcher fusion has been adopted to improve iris recognition\nperformance. We validate the system using a database of 1,872 near-infrared\niris images. The presented approach is superior to bilinear or bicubic\ninterpolation, especially at lower resolutions, and the fusion of the two\nsystems pushes the EER to below 5% for down-sampling factors up to a image size\nof only 13x13.\n", "date": "2022-10-18", "categories": []}
{"id": "2210.09778", "title": "Compact multi-scale periocular recognition using SAFE features", "abstract": "  In this paper, we present a new approach for periocular recognition based on\nthe Symmetry Assessment by Feature Expansion (SAFE) descriptor, which encodes\nthe presence of various symmetric curve families around image key points. We\nuse the sclera center as single key point for feature extraction, highlighting\nthe object-like identity properties that concentrates to this unique point of\nthe eye. As it is demonstrated, such discriminative properties can be encoded\nwith a reduced set of symmetric curves. Experiments are done with a database of\nperiocular images captured with a digital camera. We test our system against\nreference periocular features, achieving top performance with a considerably\nsmaller feature vector (given by the use of a single key point). All the\nsystems tested also show a nearly steady correlation between acquisition\ndistance and performance, and they are also able to cope well when enrolment\nand test images are not captured at the same distance. Fusion experiments among\nthe available systems are also provided.\n", "date": "2022-10-18", "categories": []}
{"id": "2210.09843", "title": "BIOWISH: Biometric Recognition using Wearable Inertial Sensors detecting\n  Heart Activity", "abstract": "  Wearable devices are increasingly used, thanks to the wide set of\napplications that can be deployed exploiting their ability to monitor physical\nactivity and health-related parameters. Their usage has been recently proposed\nto perform biometric recognition, leveraging on the uniqueness of the recorded\ntraits to generate discriminative identifiers. Most of the studies conducted on\nthis topic have considered signals derived from cardiac activity, detecting it\nmainly using electrical measurements thorugh electrocardiography, or optical\nrecordings employing photoplethysmography. In this paper we instead propose a\nBIOmetric recognition approach using Wearable Inertial Sensors detecting Heart\nactivity (BIOWISH). In more detail, we investigate the feasibility of\nexploiting mechanical measurements obtained through seismocardiography and\ngyrocardiography to recognize a person. Several feature extractors and\nclassifiers, including deep learning techniques relying on transfer learning\nand siamese training, are employed to derive distinctive characteristics from\nthe considered signals, and differentiate between legitimate and impostor\nsubjects. An multi-session database, comprising acquisitions taken from\nsubjects performing different activities, is employed to perform experimental\ntests simulating a verification system. The obtained results testify that\nidentifiers derived from measurements of chest vibrations, collected by\nwearable inertial sensors, could be employed to guarantee high recognition\nperformance, even when considering short-time recordings.\n", "date": "2022-10-18", "categories": []}
{"id": "2210.05866", "title": "Deep Learning for Iris Recognition: A Survey", "abstract": "  In this survey, we provide a comprehensive review of more than 200 papers,\ntechnical reports, and GitHub repositories published over the last 10 years on\nthe recent developments of deep learning techniques for iris recognition,\ncovering broad topics on algorithm designs, open-source tools, open challenges,\nand emerging research. First, we conduct a comprehensive analysis of deep\nlearning techniques developed for two main sub-tasks in iris biometrics:\nsegmentation and recognition. Second, we focus on deep learning techniques for\nthe robustness of iris recognition systems against presentation attacks and via\nhuman-machine pairing. Third, we delve deep into deep learning techniques for\nforensic application, especially in post-mortem iris recognition. Fourth, we\nreview open-source resources and tools in deep learning techniques for iris\nrecognition. Finally, we highlight the technical challenges, emerging research\ntrends, and outlook for the future of deep learning in iris recognition.\n", "date": "2022-10-12", "categories": []}
{"id": "2210.02018", "title": "InterFace:Adjustable Angular Margin Inter-class Loss for Deep Face\n  Recognition", "abstract": "  In the field of face recognition, it is always a hot research topic to\nimprove the loss solution to make the face features extracted by the network\nhave greater discriminative power. Research works in recent years has improved\nthe discriminative power of the face model by normalizing softmax to the cosine\nspace step by step and then adding a fixed penalty margin to reduce the\nintra-class distance to increase the inter-class distance. Although a great\ndeal of previous work has been done to optimize the boundary penalty to improve\nthe discriminative power of the model, adding a fixed margin penalty to the\ndepth feature and the corresponding weight is not consistent with the pattern\nof data in the real scenario. To address this issue, in this paper, we propose\na novel loss function, InterFace, releasing the constraint of adding a margin\npenalty only between the depth feature and the corresponding weight to push the\nseparability of classes by adding corresponding margin penalties between the\ndepth features and all weights. To illustrate the advantages of InterFace over\na fixed penalty margin, we explained geometrically and comparisons on a set of\nmainstream benchmarks. From a wider perspective, our InterFace has advanced the\nstate-of-the-art face recognition performance on five out of thirteen\nmainstream benchmarks. All training codes, pre-trained models, and training\nlogs, are publicly released\n\\footnote{$https://github.com/iamsangmeng/InterFace$}.\n", "date": "2022-10-05", "categories": []}
{"id": "2210.04050", "title": "Multi-Modal Human Authentication Using Silhouettes, Gait and RGB", "abstract": "  Whole-body-based human authentication is a promising approach for remote\nbiometrics scenarios. Current literature focuses on either body recognition\nbased on RGB images or gait recognition based on body shapes and walking\npatterns; both have their advantages and drawbacks. In this work, we propose\nDual-Modal Ensemble (DME), which combines both RGB and silhouette data to\nachieve more robust performances for indoor and outdoor whole-body based\nrecognition. Within DME, we propose GaitPattern, which is inspired by the\ndouble helical gait pattern used in traditional gait analysis. The GaitPattern\ncontributes to robust identification performance over a large range of viewing\nangles. Extensive experimental results on the CASIA-B dataset demonstrate that\nthe proposed method outperforms state-of-the-art recognition systems. We also\nprovide experimental results using the newly collected BRIAR dataset.\n", "date": "2022-10-08", "categories": []}
{"id": "2210.04567", "title": "BoundaryFace: A mining framework with noise label self-correction for\n  Face Recognition", "abstract": "  Face recognition has made tremendous progress in recent years due to the\nadvances in loss functions and the explosive growth in training sets size. A\nproperly designed loss is seen as key to extract discriminative features for\nclassification. Several margin-based losses have been proposed as alternatives\nof softmax loss in face recognition. However, two issues remain to consider: 1)\nThey overlook the importance of hard sample mining for discriminative learning.\n2) Label noise ubiquitously exists in large-scale datasets, which can seriously\ndamage the model's performance. In this paper, starting from the perspective of\ndecision boundary, we propose a novel mining framework that focuses on the\nrelationship between a sample's ground truth class center and its nearest\nnegative class center. Specifically, a closed-set noise label self-correction\nmodule is put forward, making this framework work well on datasets containing a\nlot of label noise. The proposed method consistently outperforms SOTA methods\nin various face recognition benchmarks. Training code has been released at\nhttps://github.com/SWJTU-3DVision/BoundaryFace.\n", "date": "2022-10-10", "categories": []}
{"id": "2210.04816", "title": "Ensemble Learning using Transformers and Convolutional Networks for\n  Masked Face Recognition", "abstract": "  Wearing a face mask is one of the adjustments we had to follow to reduce the\nspread of the coronavirus. Having our faces covered by masks constantly has\ndriven the need to understand and investigate how this behavior affects the\nrecognition capability of face recognition systems. Current face recognition\nsystems have extremely high accuracy when dealing with unconstrained general\nface recognition cases but do not generalize well with occluded masked faces.\nIn this work, we propose a system for masked face recognition. The proposed\nsystem comprises two Convolutional Neural Network (CNN) models and two\nTransformer models. The CNN models have been fine-tuned on FaceNet pre-trained\nmodel. We ensemble the predictions of the four models using the majority voting\ntechnique to identify the person with the mask. The proposed system has been\nevaluated on a synthetically masked LFW dataset created in this work. The best\naccuracy is obtained using the ensembled models with an accuracy of 92%. This\nrecognition rate outperformed the accuracy of other models and it shows the\ncorrectness and robustness of the proposed model for recognizing masked faces.\nThe code and data are available at https://github.com/Hamzah-Luqman/MFR\n", "date": "2022-10-10", "categories": []}
{"id": "2210.03592", "title": "Specialized Re-Ranking: A Novel Retrieval-Verification Framework for\n  Cloth Changing Person Re-Identification", "abstract": "  Cloth changing person re-identification(Re-ID) can work under more\ncomplicated scenarios with higher security than normal Re-ID and biometric\ntechniques and is therefore extremely valuable in applications. Meanwhile,\nhigher flexibility in appearance always leads to more similar-looking confusing\nimages, which is the weakness of the widely used retrieval methods. In this\nwork, we shed light on how to handle these similar images. Specifically, we\npropose a novel retrieval-verification framework. Given an image, the retrieval\nmodule can search for similar images quickly. Our proposed verification network\nwill then compare the input image and the candidate images by contrasting those\nlocal details and give a similarity score. An innovative ranking strategy is\nalso introduced to take a good balance between retrieval and verification\nresults. Comprehensive experiments are conducted to show the effectiveness of\nour framework and its capability in improving the state-of-the-art methods\nremarkably on both synthetic and realistic datasets.\n", "date": "2022-10-07", "categories": []}
{"id": "2210.02505", "title": "A novel non-linear transformation based multi-user identification\n  algorithm for fixed text keystroke behavioral dynamics", "abstract": "  In this paper, we propose a new technique to uniquely classify and identify\nmultiple users accessing a single application using keystroke dynamics. This\nproblem is usually encountered when multiple users have legitimate access to\nshared computers and accounts, where, at times, one user can inadvertently be\nlogged in on another user's account. Since the login processes are usually\nbypassed at this stage, we rely on keystroke dynamics in order to tell users\napart. Our algorithm uses the quantile transform and techniques from\nlocalization to classify and identify users. Specifically, we use an algorithm\nknown as ordinal Unfolding based Localization (UNLOC), which uses only ordinal\ndata obtained from comparing distance proxies, by \"locating\" users in a reduced\nPCA/Kernel-PCA/t-SNE space based on their typing patterns. Our results are\nvalidated with the help of benchmark keystroke datasets and show that our\nalgorithm outperforms other methods.\n", "date": "2022-10-05", "categories": []}
{"id": "2210.02579", "title": "DigiFace-1M: 1 Million Digital Face Images for Face Recognition", "abstract": "  State-of-the-art face recognition models show impressive accuracy, achieving\nover 99.8% on Labeled Faces in the Wild (LFW) dataset. Such models are trained\non large-scale datasets that contain millions of real human face images\ncollected from the internet. Web-crawled face images are severely biased (in\nterms of race, lighting, make-up, etc) and often contain label noise. More\nimportantly, the face images are collected without explicit consent, raising\nethical concerns. To avoid such problems, we introduce a large-scale synthetic\ndataset for face recognition, obtained by rendering digital faces using a\ncomputer graphics pipeline. We first demonstrate that aggressive data\naugmentation can significantly reduce the synthetic-to-real domain gap. Having\nfull control over the rendering pipeline, we also study how each attribute\n(e.g., variation in facial pose, accessories and textures) affects the\naccuracy. Compared to SynFace, a recent method trained on GAN-generated\nsynthetic faces, we reduce the error rate on LFW by 52.5% (accuracy from 91.93%\nto 96.17%). By fine-tuning the network on a smaller number of real face images\nthat could reasonably be obtained with consent, we achieve accuracy that is\ncomparable to the methods trained on millions of real face images.\n", "date": "2022-10-05", "categories": []}
{"id": "2210.03072", "title": "IJCB 2022 Mobile Behavioral Biometrics Competition (MobileB2C)", "abstract": "  This paper describes the experimental framework and results of the IJCB 2022\nMobile Behavioral Biometrics Competition (MobileB2C). The aim of MobileB2C is\nbenchmarking mobile user authentication systems based on behavioral biometric\ntraits transparently acquired by mobile devices during ordinary Human-Computer\nInteraction (HCI), using a novel public database, BehavePassDB, and a standard\nexperimental protocol. The competition is divided into four tasks corresponding\nto typical user activities: keystroke, text reading, gallery swiping, and\ntapping. The data are composed of touchscreen data and several background\nsensor data simultaneously acquired. \"Random\" (different users with different\ndevices) and \"skilled\" (different user on the same device attempting to imitate\nthe legitimate one) impostor scenarios are considered. The results achieved by\nthe participants show the feasibility of user authentication through behavioral\nbiometrics, although this proves to be a non-trivial challenge. MobileB2C will\nbe established as an on-going competition.\n", "date": "2022-10-06", "categories": []}
{"id": "2206.02502", "title": "BehavePassDB: Public Database for Mobile Behavioral Biometrics and\n  Benchmark Evaluation", "abstract": "  Mobile behavioral biometrics have become a popular topic of research,\nreaching promising results in terms of authentication, exploiting a multimodal\ncombination of touchscreen and background sensor data. However, there is no way\nof knowing whether state-of-the-art classifiers in the literature can\ndistinguish between the notion of user and device. In this article, we present\na new database, BehavePassDB, structured into separate acquisition sessions and\ntasks to mimic the most common aspects of mobile Human-Computer Interaction\n(HCI). BehavePassDB is acquired through a dedicated mobile app installed on the\nsubjects' devices, also including the case of different users on the same\ndevice for evaluation. We propose a standard experimental protocol and\nbenchmark for the research community to perform a fair comparison of novel\napproaches with the state of the art. We propose and evaluate a system based on\nLong-Short Term Memory (LSTM) architecture with triplet loss and modality\nfusion at score level.\n", "date": "2022-06-06", "categories": []}
{"id": "2207.07596", "title": "Mobile Keystroke Biometrics Using Transformers", "abstract": "  Among user authentication methods, behavioural biometrics has proven to be\neffective against identity theft as well as user-friendly and unobtrusive. One\nof the most popular traits in the literature is keystroke dynamics due to the\nlarge deployment of computers and mobile devices in our society. This paper\nfocuses on improving keystroke biometric systems on the free-text scenario.\nThis scenario is characterised as very challenging due to the uncontrolled text\nconditions, the influence of the user's emotional and physical state, and the\nin-use application. To overcome these drawbacks, methods based on deep learning\nsuch as Convolutional Neural Networks (CNNs) and Recurrent Neural Networks\n(RNNs) have been proposed in the literature, outperforming traditional machine\nlearning methods. However, these architectures still have aspects that need to\nbe reviewed and improved. To the best of our knowledge, this is the first study\nthat proposes keystroke biometric systems based on Transformers. The proposed\nTransformer architecture has achieved Equal Error Rate (EER) values of 3.84\\%\nin the popular Aalto mobile keystroke database using only 5 enrolment sessions,\noutperforming by a large margin other state-of-the-art approaches in the\nliterature.\n", "date": "2022-07-15", "categories": []}
{"id": "2209.10077", "title": "Can Shadows Reveal Biometric Information?", "abstract": "  We study the problem of extracting biometric information of individuals by\nlooking at shadows of objects cast on diffuse surfaces. We show that the\nbiometric information leakage from shadows can be sufficient for reliable\nidentity inference under representative scenarios via a maximum likelihood\nanalysis. We then develop a learning-based method that demonstrates this\nphenomenon in real settings, exploiting the subtle cues in the shadows that are\nthe source of the leakage without requiring any labeled real data. In\nparticular, our approach relies on building synthetic scenes composed of 3D\nface models obtained from a single photograph of each identity. We transfer\nwhat we learn from the synthetic data to the real data using domain adaptation\nin a completely unsupervised way. Our model is able to generalize well to the\nreal domain and is robust to several variations in the scenes. We report high\nclassification accuracies in an identity classification task that takes place\nin a scene with unknown geometry and occluding objects.\n", "date": "2022-09-21", "categories": []}
{"id": "2210.01594", "title": "GANTouch: An Attack-Resilient Framework for Touch-based Continuous\n  Authentication System", "abstract": "  Previous studies have shown that commonly studied (vanilla) implementations\nof touch-based continuous authentication systems (V-TCAS) are susceptible to\nactive adversarial attempts. This study presents a novel Generative Adversarial\nNetwork assisted TCAS (G-TCAS) framework and compares it to the V-TCAS under\nthree active adversarial environments viz. Zero-effort, Population, and\nRandom-vector. The Zero-effort environment was implemented in two variations\nviz. Zero-effort (same-dataset) and Zero-effort (cross-dataset). The first\ninvolved a Zero-effort attack from the same dataset, while the second used\nthree different datasets. G-TCAS showed more resilience than V-TCAS under the\nPopulation and Random-vector, the more damaging adversarial scenarios than the\nZero-effort. On average, the increase in the false accept rates (FARs) for\nV-TCAS was much higher (27.5% and 21.5%) than for G-TCAS (14% and 12.5%) for\nPopulation and Random-vector attacks, respectively. Moreover, we performed a\nfairness analysis of TCAS for different genders and found TCAS to be fair\nacross genders. The findings suggest that we should evaluate TCAS under active\nadversarial environments and affirm the usefulness of GANs in the TCAS\npipeline.\n", "date": "2022-10-02", "categories": []}
{"id": "2210.00127", "title": "Robust Person Identification: A WiFi Vision-based Approach", "abstract": "  Person re-identification (Re-ID) has become increasingly important as it\nsupports a wide range of security applications. Traditional person Re-ID mainly\nrelies on optical camera-based systems, which incur several limitations due to\nthe changes in the appearance of people, occlusions, and human poses. In this\nwork, we propose a WiFi vision-based system, 3D-ID, for person Re-ID in 3D\nspace. Our system leverages the advances of WiFi and deep learning to help WiFi\ndevices see, identify, and recognize people. In particular, we leverage\nmultiple antennas on next-generation WiFi devices and 2D AoA estimation of the\nsignal reflections to enable WiFi to visualize a person in the physical\nenvironment. We then leverage deep learning to digitize the visualization of\nthe person into 3D body representation and extract both the static body shape\nand dynamic walking patterns for person Re-ID. Our evaluation results under\nvarious indoor environments show that the 3D-ID system achieves an overall\nrank-1 accuracy of 85.3%. Results also show that our system is resistant to\nvarious attacks. The proposed 3D-ID is thus very promising as it could augment\nor complement camera-based systems.\n", "date": "2022-09-30", "categories": []}
{"id": "2210.00615", "title": "iCTGAN--An Attack Mitigation Technique for Random-vector Attack on\n  Accelerometer-based Gait Authentication Systems", "abstract": "  A recent study showed that commonly (vanilla) studied implementations of\naccelerometer-based gait authentication systems ($v$ABGait) are susceptible to\nrandom-vector attack. The same study proposed a beta noise-assisted\nimplementation ($\\beta$ABGait) to mitigate the attack. In this paper, we assess\nthe effectiveness of the random-vector attack on both $v$ABGait and\n$\\beta$ABGait using three accelerometer-based gait datasets. In addition, we\npropose $i$ABGait, an alternative implementation of ABGait, which uses a\nConditional Tabular Generative Adversarial Network. Then we evaluate\n$i$ABGait's resilience against the traditional zero-effort and random-vector\nattacks. The results show that $i$ABGait mitigates the impact of the\nrandom-vector attack to a reasonable extent and outperforms $\\beta$ABGait in\nmost experimental settings.\n", "date": "2022-10-02", "categories": []}
{"id": "2209.15471", "title": "Two-headed eye-segmentation approach for biometric identification", "abstract": "  Iris-based identification systems are among the most popular approaches for\nperson identification. Such systems require good-quality segmentation modules\nthat ideally identify the regions for different eye components. This paper\nintroduces the new two-headed architecture, where the eye components and\neyelashes are segmented using two separate decoding modules. Moreover, we\ninvestigate various training scenarios by adopting different training losses.\nThanks to the two-headed approach, we were also able to examine the quality of\nthe model with the convex prior, which enforces the convexity of the segmented\nshapes. We conducted an extensive evaluation of various learning scenarios on\nreal-life conditions high-resolution near-infrared iris images.\n", "date": "2022-09-30", "categories": []}
{"id": "2209.14498", "title": "Teaching Where to Look: Attention Similarity Knowledge Distillation for\n  Low Resolution Face Recognition", "abstract": "  Deep learning has achieved outstanding performance for face recognition\nbenchmarks, but performance reduces significantly for low resolution (LR)\nimages. We propose an attention similarity knowledge distillation approach,\nwhich transfers attention maps obtained from a high resolution (HR) network as\na teacher into an LR network as a student to boost LR recognition performance.\nInspired by humans being able to approximate an object's region from an LR\nimage based on prior knowledge obtained from HR images, we designed the\nknowledge distillation loss using the cosine similarity to make the student\nnetwork's attention resemble the teacher network's attention. Experiments on\nvarious LR face related benchmarks confirmed the proposed method generally\nimproved recognition performances on LR settings, outperforming\nstate-of-the-art results by simply transferring well-constructed attention\nmaps. The code and pretrained models are publicly available in the\nhttps://github.com/gist-ailab/teaching-where-to-look.\n", "date": "2022-09-29", "categories": []}
{"id": "2209.14692", "title": "Digital and Physical Face Attacks: Reviewing and One Step Further", "abstract": "  With the rapid progress over the past five years, face authentication has\nbecome the most pervasive biometric recognition method. Thanks to the\nhigh-accuracy recognition performance and user-friendly usage, automatic face\nrecognition (AFR) has exploded into a plethora of practical applications over\ndevice unlocking, checking-in, and financial payment. In spite of the\ntremendous success of face authentication, a variety of face presentation\nattacks (FPA), such as print attacks, replay attacks, and 3D mask attacks, have\nraised pressing mistrust concerns. Besides physical face attacks, face\nvideos/images are vulnerable to a wide variety of digital attack techniques\nlaunched by malicious hackers, causing potential menace to the public at large.\nDue to the unrestricted access to enormous digital face images/videos and\ndisclosed easy-to-use face manipulation tools circulating on the internet,\nnon-expert attackers without any prior professional skills are able to readily\ncreate sophisticated fake faces, leading to numerous dangerous applications\nsuch as financial fraud, impersonation, and identity theft. This survey aims to\nbuild the integrity of face forensics by providing thorough analyses of\nexisting literature and highlighting the issues requiring further attention. In\nthis paper, we first comprehensively survey both physical and digital face\nattack types and datasets. Then, we review the latest and most advanced\nprogress on existing counter-attack methodologies and highlight their current\nlimits. Moreover, we outline possible future research directions for existing\nand upcoming challenges in the face forensics community. Finally, the necessity\nof joint physical and digital face attack detection has been discussed, which\nhas never been studied in previous surveys.\n", "date": "2022-09-29", "categories": []}
{"id": "2209.12208", "title": "A Uniform Representation Learning Method for OCT-based Fingerprint\n  Presentation Attack Detection and Reconstruction", "abstract": "  The technology of optical coherence tomography (OCT) to fingerprint imaging\nopens up a new research potential for fingerprint recognition owing to its\nability to capture depth information of the skin layers. Developing robust and\nhigh security Automated Fingerprint Recognition Systems (AFRSs) are possible if\nthe depth information can be fully utilized. However, in existing studies,\nPresentation Attack Detection (PAD) and subsurface fingerprint reconstruction\nbased on depth information are treated as two independent branches, resulting\nin high computation and complexity of AFRS building.Thus, this paper proposes a\nuniform representation model for OCT-based fingerprint PAD and subsurface\nfingerprint reconstruction. Firstly, we design a novel semantic segmentation\nnetwork which only trained by real finger slices of OCT-based fingerprints to\nextract multiple subsurface structures from those slices (also known as\nB-scans). The latent codes derived from the network are directly used to\neffectively detect the PA since they contain abundant subsurface biological\ninformation, which is independent with PA materials and has strong robustness\nfor unknown PAs. Meanwhile, the segmented subsurface structures are adopted to\nreconstruct multiple subsurface 2D fingerprints. Recognition can be easily\nachieved by using existing mature technologies based on traditional 2D\nfingerprints. Extensive experiments are carried on our own established\ndatabase, which is the largest public OCT-based fingerprint database with 2449\nvolumes. In PAD task, our method can improve 0.33% Acc from the\nstate-of-the-art method. For reconstruction performance, our method achieves\nthe best performance with 0.834 mIOU and 0.937 PA. By comparing with the\nrecognition performance on surface 2D fingerprints, the effectiveness of our\nproposed method on high quality subsurface fingerprint reconstruction is\nfurther proved.\n", "date": "2022-09-25", "categories": []}
{"id": "2209.11582", "title": "Pose-Aided Video-based Person Re-Identification via Recurrent Graph\n  Convolutional Network", "abstract": "  Existing methods for video-based person re-identification (ReID) mainly learn\nthe appearance feature of a given pedestrian via a feature extractor and a\nfeature aggregator.\n  However, the appearance models would fail when different pedestrians have\nsimilar appearances.\n  Considering that different pedestrians have different walking postures and\nbody proportions, we propose to learn the discriminative pose feature beyond\nthe appearance feature for video retrieval.\n  Specifically, we implement a two-branch architecture to separately learn the\nappearance feature and pose feature, and then concatenate them together for\ninference.\n  To learn the pose feature, we first detect the pedestrian pose in each frame\nthrough an off-the-shelf pose detector, and construct a temporal graph using\nthe pose sequence.\n  We then exploit a recurrent graph convolutional network (RGCN) to learn the\nnode embeddings of the temporal pose graph, which devises a global information\npropagation mechanism to simultaneously achieve the neighborhood aggregation of\nintra-frame nodes and message passing among inter-frame graphs.\n  Finally, we propose a dual-attention method consisting of node-attention and\ntime-attention to obtain the temporal graph representation from the node\nembeddings, where the self-attention mechanism is employed to learn the\nimportance of each node and each frame.\n  We verify the proposed method on three video-based ReID datasets, i.e., Mars,\nDukeMTMC and iLIDS-VID, whose experimental results demonstrate that the learned\npose feature can effectively improve the performance of existing appearance\nmodels.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.11584", "title": "Multi-Granularity Graph Pooling for Video-based Person Re-Identification", "abstract": "  The video-based person re-identification (ReID) aims to identify the given\npedestrian video sequence across multiple non-overlapping cameras.\n  To aggregate the temporal and spatial features of the video samples, the\ngraph neural networks (GNNs) are introduced.\n  However, existing graph-based models, like STGCN, perform the\n\\textit{mean}/\\textit{max pooling} on node features to obtain the graph\nrepresentation, which neglect the graph topology and node importance.\n  In this paper, we propose the graph pooling network (GPNet) to learn the\nmulti-granularity graph representation for the video retrieval, where the\n\\textit{graph pooling layer} is implemented to downsample the graph.\n  We first construct a multi-granular graph, whose node features denote image\nembedding learned by backbone, and edges are established between the temporal\nand Euclidean neighborhood nodes.\n  We then implement multiple graph convolutional layers to perform the\nneighborhood aggregation on the graphs.\n  To downsample the graph, we propose a multi-head full attention graph pooling\n(MHFAPool) layer, which integrates the advantages of existing node clustering\nand node selection pooling methods.\n  Specifically, MHFAPool takes the main eigenvector of full attention matrix as\nthe aggregation coefficients to involve the global graph information in each\npooled nodes.\n  Extensive experiments demonstrate that our GPNet achieves the competitive\nresults on four widely-used datasets, i.e., MARS, DukeMTMC-VideoReID, iLIDS-VID\nand PRID-2011.\n", "date": "2022-09-23", "categories": []}
{"id": "2209.11020", "title": "Privacy Attacks Against Biometric Models with Fewer Samples:\n  Incorporating the Output of Multiple Models", "abstract": "  Authentication systems are vulnerable to model inversion attacks where an\nadversary is able to approximate the inverse of a target machine learning\nmodel. Biometric models are a prime candidate for this type of attack. This is\nbecause inverting a biometric model allows the attacker to produce a realistic\nbiometric input to spoof biometric authentication systems.\n  One of the main constraints in conducting a successful model inversion attack\nis the amount of training data required. In this work, we focus on iris and\nfacial biometric systems and propose a new technique that drastically reduces\nthe amount of training data necessary. By leveraging the output of multiple\nmodels, we are able to conduct model inversion attacks with 1/10th the training\nset size of Ahmad and Fuller (IJCB 2020) for iris data and 1/1000th the\ntraining set size of Mai et al. (Pattern Analysis and Machine Intelligence\n2019) for facial data. We denote our new attack technique as structured random\nwith alignment loss. Our attacks are black-box, requiring no knowledge of the\nweights of the target neural network, only the dimension, and values of the\noutput vector.\n  To show the versatility of the alignment loss, we apply our attack framework\nto the task of membership inference (Shokri et al., IEEE S&P 2017) on biometric\ndata. For the iris, membership inference attack against classification networks\nimproves from 52% to 62% accuracy.\n", "date": "2022-09-22", "categories": []}
{"id": "2209.11024", "title": "Google Coral-based edge computing person reidentification using human\n  parsing combined with analytical method", "abstract": "  Person reidentification (re-ID) is becoming one of the most significant\napplication areas of computer vision due to its importance for science and\nsocial security. Due to enormous size and scale of camera systems it is\nbeneficial to develop edge computing re-ID applications where at least part of\nthe analysis could be performed by the cameras. However, conventional re-ID\nrelies heavily on deep learning (DL) computationally demanding models which are\nnot readily applicable for edge computing. In this paper we adapt a recently\nproposed re-ID method that combines DL human parsing with analytical feature\nextraction and ranking schemes to be more suitable for edge computing re-ID.\nFirst, we compare parsers that use ResNet101, ResNet18, MobileNetV2, and OSNet\nbackbones and show that parsing can be performed using compact backbones with\nsufficient accuracy. Second, we transfer parsers to tensor processing unit\n(TPU) of Google Coral Dev Board and show that it can act as a portable edge\ncomputing re-ID station. We also implement the analytical part of re-ID method\non Coral CPU to ensure that it can perform a complete re-ID cycle. For\nquantitative analysis we compare inference speed, parsing masks, and re-ID\naccuracy on GPU and Coral TPU depending on parser backbone. We also discuss\npossible application scenarios of edge computing in re-ID taking into account\nknown limitations mainly related to memory and storage space of portable\ndevices.\n", "date": "2022-09-22", "categories": []}
{"id": "2209.09348", "title": "Visible-Infrared Person Re-Identification Using Privileged Intermediate\n  Information", "abstract": "  Visible-infrared person re-identification (ReID) aims to recognize a same\nperson of interest across a network of RGB and IR cameras. Some deep learning\n(DL) models have directly incorporated both modalities to discriminate persons\nin a joint representation space. However, this cross-modal ReID problem remains\nchallenging due to the large domain shift in data distributions between RGB and\nIR modalities. % This paper introduces a novel approach for a creating\nintermediate virtual domain that acts as bridges between the two main domains\n(i.e., RGB and IR modalities) during training. This intermediate domain is\nconsidered as privileged information (PI) that is unavailable at test time, and\nallows formulating this cross-modal matching task as a problem in learning\nunder privileged information (LUPI). We devised a new method to generate images\nbetween visible and infrared domains that provide additional information to\ntrain a deep ReID model through an intermediate domain adaptation. In\nparticular, by employing color-free and multi-step triplet loss objectives\nduring training, our method provides common feature representation spaces that\nare robust to large visible-infrared domain shifts. % Experimental results on\nchallenging visible-infrared ReID datasets indicate that our proposed approach\nconsistently improves matching accuracy, without any computational overhead at\ntest time. The code is available at:\n\\href{https://github.com/alehdaghi/Cross-Modal-Re-ID-via-LUPI}{https://github.com/alehdaghi/Cross-Modal-Re-ID-via-LUPI}\n", "date": "2022-09-19", "categories": []}
{"id": "2209.08788", "title": "Scale Attention for Learning Deep Face Representation: A Study Against\n  Visual Scale Variation", "abstract": "  Human face images usually appear with wide range of visual scales. The\nexisting face representations pursue the bandwidth of handling scale variation\nvia multi-scale scheme that assembles a finite series of predefined scales.\nSuch multi-shot scheme brings inference burden, and the predefined scales\ninevitably have gap from real data. Instead, learning scale parameters from\ndata, and using them for one-shot feature inference, is a decent solution. To\nthis end, we reform the conv layer by resorting to the scale-space theory, and\nachieve two-fold facilities: 1) the conv layer learns a set of scales from real\ndata distribution, each of which is fulfilled by a conv kernel; 2) the layer\nautomatically highlights the feature at the proper channel and location\ncorresponding to the input pattern scale and its presence. Then, we accomplish\nthe hierarchical scale attention by stacking the reformed layers, building a\nnovel style named SCale AttentioN Conv Neural Network (\\textbf{SCAN-CNN}). We\napply SCAN-CNN to the face recognition task and push the frontier of SOTA\nperformance. The accuracy gain is more evident when the face images are blurry.\nMeanwhile, as a single-shot scheme, the inference is more efficient than\nmulti-shot fusion. A set of tools are made to ensure the fast training of\nSCAN-CNN and zero increase of inference cost compared with the plain CNN.\n", "date": "2022-09-19", "categories": []}
{"id": "2209.08930", "title": "HiMFR: A Hybrid Masked Face Recognition Through Face Inpainting", "abstract": "  To recognize the masked face, one of the possible solutions could be to\nrestore the occluded part of the face first and then apply the face recognition\nmethod. Inspired by the recent image inpainting methods, we propose an\nend-to-end hybrid masked face recognition system, namely HiMFR, consisting of\nthree significant parts: masked face detector, face inpainting, and face\nrecognition. The masked face detector module applies a pretrained Vision\nTransformer (ViT\\_b32) to detect whether faces are covered with masked or not.\nThe inpainting module uses a fine-tune image inpainting model based on a\nGenerative Adversarial Network (GAN) to restore faces. Finally, the hybrid face\nrecognition module based on ViT with an EfficientNetB3 backbone recognizes the\nfaces. We have implemented and evaluated our proposed method on four different\npublicly available datasets: CelebA, SSDMNV2, MAFA, {Pubfig83} with our locally\ncollected small dataset, namely Face5. Comprehensive experimental results show\nthe efficacy of the proposed HiMFR method with competitive performance. Code is\navailable at https://github.com/mdhosen/HiMFR\n", "date": "2022-09-19", "categories": []}
{"id": "2209.07001", "title": "Pose Attention-Guided Profile-to-Frontal Face Recognition", "abstract": "  In recent years, face recognition systems have achieved exceptional success\ndue to promising advances in deep learning architectures. However, they still\nfail to achieve expected accuracy when matching profile images against a\ngallery of frontal images. Current approaches either perform pose normalization\n(i.e., frontalization) or disentangle pose information for face recognition. We\ninstead propose a new approach to utilize pose as an auxiliary information via\nan attention mechanism. In this paper, we hypothesize that pose attended\ninformation using an attention mechanism can guide contextual and distinctive\nfeature extraction from profile faces, which further benefits a better\nrepresentation learning in an embedded domain. To achieve this, first, we\ndesign a unified coupled profile-to-frontal face recognition network. It learns\nthe mapping from faces to a compact embedding subspace via a class-specific\ncontrastive loss. Second, we develop a novel pose attention block (PAB) to\nspecially guide the pose-agnostic feature extraction from profile faces. To be\nmore specific, PAB is designed to explicitly help the network to focus on\nimportant features along both channel and spatial dimension while learning\ndiscriminative yet pose invariant features in an embedding subspace. To\nvalidate the effectiveness of our proposed method, we conduct experiments on\nboth controlled and in the wild benchmarks including Multi-PIE, CFP, IJBC, and\nshow superiority over the state of the arts.\n", "date": "2022-09-15", "categories": []}
{"id": "2209.07220", "title": "Face Shape-Guided Deep Feature Alignment for Face Recognition Robust to\n  Face Misalignment", "abstract": "  For the past decades, face recognition (FR) has been actively studied in\ncomputer vision and pattern recognition society. Recently, due to the advances\nin deep learning, the FR technology shows high performance for most of the\nbenchmark datasets. However, when the FR algorithm is applied to a real-world\nscenario, the performance has been known to be still unsatisfactory. This is\nmainly attributed to the mismatch between training and testing sets. Among such\nmismatches, face misalignment between training and testing faces is one of the\nfactors that hinder successful FR. To address this limitation, we propose a\nface shape-guided deep feature alignment framework for FR robust to the face\nmisalignment. Based on a face shape prior (e.g., face keypoints), we train the\nproposed deep network by introducing alignment processes, i.e., pixel and\nfeature alignments, between well-aligned and misaligned face images. Through\nthe pixel alignment process that decodes the aggregated feature extracted from\na face image and face shape prior, we add the auxiliary task to reconstruct the\nwell-aligned face image. Since the aggregated features are linked to the face\nfeature extraction network as a guide via the feature alignment process, we\ntrain the robust face feature to the face misalignment. Even if the face shape\nestimation is required in the training stage, the additional face alignment\nprocess, which is usually incorporated in the conventional FR pipeline, is not\nnecessarily needed in the testing phase. Through the comparative experiments,\nwe validate the effectiveness of the proposed method for the face misalignment\nwith the FR datasets.\n", "date": "2022-09-15", "categories": []}
{"id": "2209.06452", "title": "TrADe Re-ID -- Live Person Re-Identification using Tracking and Anomaly\n  Detection", "abstract": "  Person Re-Identification (Re-ID) aims to search for a person of interest\n(query) in a network of cameras. In the classic Re-ID setting the query is\nsought in a gallery containing properly cropped images of entire bodies.\nRecently, the live Re-ID setting was introduced to represent the practical\napplication context of Re-ID better. It consists in searching for the query in\nshort videos, containing whole scene frames. The initial live Re-ID baseline\nused a pedestrian detector to build a large search gallery and a classic Re-ID\nmodel to find the query in the gallery. However, the galleries generated were\ntoo large and contained low-quality images, which decreased the live Re-ID\nperformance. Here, we present a new live Re-ID approach called TrADe, to\ngenerate lower high-quality galleries. TrADe first uses a Tracking algorithm to\nidentify sequences of images of the same individual in the gallery. Following,\nan Anomaly Detection model is used to select a single good representative of\neach tracklet. TrADe is validated on the live Re-ID version of the PRID-2011\ndataset and shows significant improvements over the baseline.\n", "date": "2022-09-14", "categories": []}
{"id": "2208.13600", "title": "Towards Robust Face Recognition with Comprehensive Search", "abstract": "  Data cleaning, architecture, and loss function design are important factors\ncontributing to high-performance face recognition. Previously, the research\ncommunity tries to improve the performance of each single aspect but failed to\npresent a unified solution on the joint search of the optimal designs for all\nthree aspects. In this paper, we for the first time identify that these aspects\nare tightly coupled to each other. Optimizing the design of each aspect\nactually greatly limits the performance and biases the algorithmic design.\nSpecifically, we find that the optimal model architecture or loss function is\nclosely coupled with the data cleaning. To eliminate the bias of single-aspect\nresearch and provide an overall understanding of the face recognition model\ndesign, we first carefully design the search space for each aspect, then a\ncomprehensive search method is introduced to jointly search optimal data\ncleaning, architecture, and loss function design. In our framework, we make the\nproposed comprehensive search as flexible as possible, by using an innovative\nreinforcement learning based approach. Extensive experiments on million-level\nface recognition benchmarks demonstrate the effectiveness of our newly-designed\nsearch space for each aspect and the comprehensive search. We outperform expert\nalgorithms developed for each single research track by large margins. More\nimportantly, we analyze the difference between our searched optimal design and\nthe independent design of the single factors. We point out that strong models\ntend to optimize with more difficult training datasets and loss functions. Our\nempirical study can provide guidance in future research towards more robust\nface recognition systems.\n", "date": "2022-08-29", "categories": []}
{"id": "2209.01880", "title": "ScaleFace: Uncertainty-aware Deep Metric Learning", "abstract": "  The performance of modern deep learning-based systems dramatically depends on\nthe quality of input objects. For example, face recognition quality would be\nlower for blurry or corrupted inputs. However, it is hard to predict the\ninfluence of input quality on the resulting accuracy in more complex scenarios.\nWe propose an approach for deep metric learning that allows direct estimation\nof the uncertainty with almost no additional computational cost. The developed\n\\textit{ScaleFace} algorithm uses trainable scale values that modify\nsimilarities in the space of embeddings. These input-dependent scale values\nrepresent a measure of confidence in the recognition result, thus allowing\nuncertainty estimation. We provide comprehensive experiments on face\nrecognition tasks that show the superior performance of ScaleFace compared to\nother uncertainty-aware face recognition approaches. We also extend the results\nto the task of text-to-image retrieval showing that the proposed approach beats\nthe competitors with significant margin.\n", "date": "2022-09-05", "categories": []}
{"id": "2209.04802", "title": "Examining Uniqueness and Permanence of the WAY EEG GAL dataset toward\n  User Authentication", "abstract": "  This study evaluates the discriminating capacity (uniqueness) of the EEG data\nfrom the WAY EEG GAL public dataset to authenticate individuals against one\nanother as well as its permanence. In addition to the EEG data, Luciw et al.\nprovide EMG (Electromyography), and kinematics data for engineers and\nresearchers to utilize WAY EEG GAL for further studies. However, evaluating the\nEMG and kinematics data is outside the scope of this study. The goal of the\nstate-of-the-art is to determine whether EEG data can be utilized to control\nprosthetic devices. On the other hand, this study aims to evaluate the\nseparability of individuals through EEG data to perform user authentication. A\nfeature importance algorithm is utilized to select the best features for each\nuser to authenticate them against all others. The authentication platform\nimplemented for this study is based on Machine Learning models/classifiers. As\nan initial test, two pilot studies are performed using Linear Discriminant\nAnalysis (LDA) and Support Vector Machine (SVM) to observe the learning trends\nof the models by multi-labeling the EEG dataset. Utilizing kNN first as the\nclassifier for user authentication, accuracy around 75% is observed. Thereafter\nto improve the performance both linear and non-linear SVMs are used to perform\nclassification. The overall average accuracies of 85.18% and 86.92% are\nachieved using linear and non-linear SVMs respectively. In addition to\naccuracy, F1 scores are also calculated. The overall average F1 score of 87.51%\nand 88.94% are achieved for linear and non-linear SVMs respectively. Beyond the\noverall performance, high performing individuals with 95.3% accuracy (95.3% F1\nscore) using linear SVM and 97.4% accuracy (97.3% F1 score) using non-linear\nSVM are also observed.\n", "date": "2022-09-11", "categories": []}
{"id": "2209.04909", "title": "Diversity and Novelty MasterPrints: Generating Multiple DeepMasterPrints\n  for Increased User Coverage", "abstract": "  This work expands on previous advancements in genetic fingerprint spoofing\nvia the DeepMasterPrints and introduces Diversity and Novelty MasterPrints.\nThis system uses quality diversity evolutionary algorithms to generate\ndictionaries of artificial prints with a focus on increasing coverage of users\nfrom the dataset. The Diversity MasterPrints focus on generating solution\nprints that match with users not covered by previously found prints, and the\nNovelty MasterPrints explicitly search for prints with more that are farther in\nuser space than previous prints. Our multi-print search methodologies\noutperform the singular DeepMasterPrints in both coverage and generalization\nwhile maintaining quality of the fingerprint image output.\n", "date": "2022-09-11", "categories": []}
{"id": "2209.05047", "title": "Is Synthetic Dataset Reliable for Benchmarking Generalizable Person\n  Re-Identification?", "abstract": "  Recent studies show that models trained on synthetic datasets are able to\nachieve better generalizable person re-identification (GPReID) performance than\nthat trained on public real-world datasets. On the other hand, due to the\nlimitations of real-world person ReID datasets, it would also be important and\ninteresting to use large-scale synthetic datasets as test sets to benchmark\nperson ReID algorithms. Yet this raises a critical question: is synthetic\ndataset reliable for benchmarking generalizable person re-identification? In\nthe literature there is no evidence showing this. To address this, we design a\nmethod called Pairwise Ranking Analysis (PRA) to quantitatively measure the\nranking similarity and perform the statistical test of identical distributions.\nSpecifically, we employ Kendall rank correlation coefficients to evaluate\npairwise similarity values between algorithm rankings on different datasets.\nThen, a non-parametric two-sample Kolmogorov-Smirnov (KS) test is performed for\nthe judgement of whether algorithm ranking correlations between synthetic and\nreal-world datasets and those only between real-world datasets lie in identical\ndistributions. We conduct comprehensive experiments, with ten representative\nalgorithms, three popular real-world person ReID datasets, and three recently\nreleased large-scale synthetic datasets. Through the designed pairwise ranking\nanalysis and comprehensive evaluations, we conclude that a recent large-scale\nsynthetic dataset ClonedPerson can be reliably used to benchmark GPReID,\nstatistically the same as real-world datasets. Therefore, this study guarantees\nthe usage of synthetic datasets for both source training set and target testing\nset, with completely no privacy concerns from real-world surveillance data.\nBesides, the study in this paper might also inspire future designs of synthetic\ndatasets.\n", "date": "2022-09-12", "categories": []}
{"id": "2209.05235", "title": "Style Variable and Irrelevant Learning for Generalizable Person\n  Re-identification", "abstract": "  Recently, due to the poor performance of supervised person re-identification\n(ReID) to an unseen domain, Domain Generalization (DG) person ReID has\nattracted a lot of attention which aims to learn a domain-insensitive model and\ncan resist the influence of domain bias. In this paper, we first verify through\nan experiment that style factors are a vital part of domain bias. Base on this\nconclusion, we propose a Style Variable and Irrelevant Learning (SVIL) method\nto eliminate the effect of style factors on the model. Specifically, we design\na Style Jitter Module (SJM) in SVIL. The SJM module can enrich the style\ndiversity of the specific source domain and reduce the style differences of\nvarious source domains. This leads to the model focusing on identity-relevant\ninformation and being insensitive to the style changes. Besides, we organically\ncombine the SJM module with a meta-learning algorithm, maximizing the benefits\nand further improving the generalization ability of the model. Note that our\nSJM module is plug-and-play and inference cost-free. Extensive experiments\nconfirm the effectiveness of our SVIL and our method outperforms the\nstate-of-the-art methods on DG-ReID benchmarks by a large margin.\n", "date": "2022-09-12", "categories": []}
{"id": "2209.03456", "title": "Information Maximization for Extreme Pose Face Recognition", "abstract": "  In this paper, we seek to draw connections between the frontal and profile\nface images in an abstract embedding space. We exploit this connection using a\ncoupled-encoder network to project frontal/profile face images into a common\nlatent embedding space. The proposed model forces the similarity of\nrepresentations in the embedding space by maximizing the mutual information\nbetween two views of the face. The proposed coupled-encoder benefits from three\ncontributions for matching faces with extreme pose disparities. First, we\nleverage our pose-aware contrastive learning to maximize the mutual information\nbetween frontal and profile representations of identities. Second, a memory\nbuffer, which consists of latent representations accumulated over past\niterations, is integrated into the model so it can refer to relatively much\nmore instances than the mini-batch size. Third, a novel pose-aware adversarial\ndomain adaptation method forces the model to learn an asymmetric mapping from\nprofile to frontal representation. In our framework, the coupled-encoder learns\nto enlarge the margin between the distribution of genuine and imposter faces,\nwhich results in high mutual information between different views of the same\nidentity. The effectiveness of the proposed model is investigated through\nextensive experiments, evaluations, and ablation studies on four benchmark\ndatasets, and comparison with the compelling state-of-the-art algorithms.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.03846", "title": "Transformer based Fingerprint Feature Extraction", "abstract": "  Fingerprint feature extraction is a task that is solved using either a global\nor a local representation. State-of-the-art global approaches use heavy deep\nlearning models to process the full fingerprint image at once, which makes the\ncorresponding approach memory intensive. On the other hand, local approaches\ninvolve minutiae based patch extraction, multiple feature extraction steps and\nan expensive matching stage, which make the corresponding approach time\nintensive. However, both these approaches provide useful and sometimes\nexclusive insights for solving the problem. Using both approaches together for\nextracting fingerprint representations is semantically useful but quite\ninefficient. Our convolutional transformer based approach with an in-built\nminutiae extractor provides a time and memory efficient solution to extract a\nglobal as well as a local representation of the fingerprint. The use of these\nrepresentations along with a smart matching process gives us state-of-the-art\nperformance across multiple databases. The project page can be found at\nhttps://saraansh1999.github.io/global-plus-local-fp-transformer.\n", "date": "2022-09-08", "categories": []}
{"id": "2209.02933", "title": "Facial De-morphing: Extracting Component Faces from a Single Morph", "abstract": "  A face morph is created by strategically combining two or more face images\ncorresponding to multiple identities. The intention is for the morphed image to\nmatch with multiple identities. Current morph attack detection strategies can\ndetect morphs but cannot recover the images or identities used in creating\nthem. The task of deducing the individual face images from a morphed face image\nis known as \\textit{de-morphing}. Existing work in de-morphing assume the\navailability of a reference image pertaining to one identity in order to\nrecover the image of the accomplice - i.e., the other identity. In this work,\nwe propose a novel de-morphing method that can recover images of both\nidentities simultaneously from a single morphed face image without needing a\nreference image or prior information about the morphing process. We propose a\ngenerative adversarial network that achieves single image-based de-morphing\nwith a surprisingly high degree of visual realism and biometric similarity with\nthe original face images. We demonstrate the performance of our method on\nlandmark-based morphs and generative model-based morphs with promising results.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.02941", "title": "Can GAN-induced Attribute Manipulations Impact Face Recognition?", "abstract": "  Impact due to demographic factors such as age, sex, race, etc., has been\nstudied extensively in automated face recognition systems. However, the impact\nof \\textit{digitally modified} demographic and facial attributes on face\nrecognition is relatively under-explored. In this work, we study the effect of\nattribute manipulations induced via generative adversarial networks (GANs) on\nface recognition performance. We conduct experiments on the CelebA dataset by\nintentionally modifying thirteen attributes using AttGAN and STGAN and\nevaluating their impact on two deep learning-based face verification methods,\nArcFace and VGGFace. Our findings indicate that some attribute manipulations\ninvolving eyeglasses and digital alteration of sex cues can significantly\nimpair face recognition by up to 73% and need further analysis.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.03142", "title": "RF Fingerprinting Needs Attention: Multi-task Approach for Real-World\n  WiFi and Bluetooth", "abstract": "  A novel cross-domain attentional multi-task architecture - xDom - for robust\nreal-world wireless radio frequency (RF) fingerprinting is presented in this\nwork. To the best of our knowledge, this is the first time such comprehensive\nattention mechanism is applied to solve RF fingerprinting problem. In this\npaper, we resort to real-world IoT WiFi and Bluetooth (BT) emissions (instead\nof synthetic waveform generation) in a rich multipath and unavoidable\ninterference environment in an indoor experimental testbed. We show the impact\nof the time-frame of capture by including waveforms collected over a span of\nmonths and demonstrate the same time-frame and multiple time-frame\nfingerprinting evaluations. The effectiveness of resorting to a multi-task\narchitecture is also experimentally proven by conducting single-task and\nmulti-task model analyses. Finally, we demonstrate the significant gain in\nperformance achieved with the proposed xDom architecture by benchmarking\nagainst a well-known state-of-the-art model for fingerprinting. Specifically,\nwe report performance improvements by up to 59.3% and 4.91x under single-task\nWiFi and BT fingerprinting respectively, and up to 50.5% increase in\nfingerprinting accuracy under the multi-task setting.\n", "date": "2022-09-07", "categories": []}
{"id": "2209.02368", "title": "Finger Multimodal Feature Fusion and Recognition Based on Channel\n  Spatial Attention", "abstract": "  Due to the instability and limitations of unimodal biometric systems,\nmultimodal systems have attracted more and more attention from researchers.\nHowever, how to exploit the independent and complementary information between\ndifferent modalities remains a key and challenging problem. In this paper, we\npropose a multimodal biometric fusion recognition algorithm based on\nfingerprints and finger veins (Fingerprint Finger Veins-Channel Spatial\nAttention Fusion Module, FPV-CSAFM). Specifically, for each pair of fingerprint\nand finger vein images, we first propose a simple and effective Convolutional\nNeural Network (CNN) to extract features. Then, we build a multimodal feature\nfusion module (Channel Spatial Attention Fusion Module, CSAFM) to fully fuse\nthe complementary information between fingerprints and finger veins. Different\nfrom existing fusion strategies, our fusion method can dynamically adjust the\nfusion weights according to the importance of different modalities in channel\nand spatial dimensions, so as to better combine the information between\ndifferent modalities and improve the overall recognition performance. To\nevaluate the performance of our method, we conduct a series of experiments on\nmultiple public datasets. Experimental results show that the proposed FPV-CSAFM\nachieves excellent recognition performance on three multimodal datasets based\non fingerprints and finger veins.\n", "date": "2022-09-06", "categories": []}
{"id": "2209.02425", "title": "Learning an Ensemble of Deep Fingerprint Representations", "abstract": "  Deep neural networks (DNNs) have shown incredible promise in learning\nfixed-length representations from fingerprints. Since the representation\nlearning is often focused on capturing specific prior knowledge (e.g.,\nminutiae), there is no universal representation that comprehensively\nencapsulates all the discriminatory information available in a fingerprint.\nWhile learning an ensemble of representations can mitigate this problem, two\ncritical challenges need to be addressed: (i) How to extract multiple diverse\nrepresentations from the same fingerprint image? and (ii) How to optimally\nexploit these representations during the matching process? In this work, we\ntrain multiple instances of DeepPrint (a state-of-the-art DNN-based fingerprint\nencoder) on different transformations of the input image to generate an\nensemble of fingerprint embeddings. We also propose a feature fusion technique\nthat distills these multiple representations into a single embedding, which\nfaithfully captures the diversity present in the ensemble without increasing\nthe computational complexity. The proposed approach has been comprehensively\nevaluated on five databases containing rolled, plain, and latent fingerprints\n(NIST SD4, NIST SD14, NIST SD27, NIST SD302, and FVC2004 DB2A) and\nstatistically significant improvements in accuracy have been consistently\ndemonstrated across a range of verification as well as closed- and open-set\nidentification settings. The proposed approach serves as a wrapper capable of\nimproving the accuracy of any DNN-based recognition system.\n", "date": "2022-09-02", "categories": []}
{"id": "2209.01115", "title": "Distilling Facial Knowledge With Teacher-Tasks:\n  Semantic-Segmentation-Features For Pose-Invariant Face-Recognition", "abstract": "  This paper demonstrates a novel approach to improve face-recognition\npose-invariance using semantic-segmentation features. The proposed\nSeg-Distilled-ID network jointly learns identification and\nsemantic-segmentation tasks, where the segmentation task is then \"distilled\"\n(MobileNet encoder). Performance is benchmarked against three state-of-the-art\nencoders on a publicly available data-set emphasizing head-pose variations.\nExperimental evaluations show the Seg-Distilled-ID network shows notable\nrobustness benefits, achieving 99.9% test-accuracy in comparison to 81.6% on\nResNet-101, 96.1% on VGG-19 and 96.3% on InceptionV3. This is achieved using\napproximately one-tenth of the top encoder's inference parameters. These\nresults demonstrate distilling semantic-segmentation features can efficiently\naddress face-recognition pose-invariance.\n", "date": "2022-09-02", "categories": []}
{"id": "2208.13811", "title": "Synthetic Latent Fingerprint Generator", "abstract": "  Given a full fingerprint image (rolled or slap), we present CycleGAN models\nto generate multiple latent impressions of the same identity as the full print.\nOur models can control the degree of distortion, noise, blurriness and\nocclusion in the generated latent print images to obtain Good, Bad and Ugly\nlatent image categories as introduced in the NIST SD27 latent database. The\ncontributions of our work are twofold: (i) demonstrate the similarity of\nsynthetically generated latent fingerprint images to crime scene latents in\nNIST SD27 and MSP databases as evaluated by the NIST NFIQ 2 quality measure and\nROC curves obtained by a SOTA fingerprint matcher, and (ii) use of synthetic\nlatents to augment small-size latent training databases in the public domain to\nimprove the performance of DeepPrint, a SOTA fingerprint matcher designed for\nrolled to rolled fingerprint matching on three latent databases (NIST SD27,\nNIST SD302, and IIITD-SLF). As an example, with synthetic latent data\naugmentation, the Rank-1 retrieval performance of DeepPrint is improved from\n15.50% to 29.07% on challenging NIST SD27 latent database. Our approach for\ngenerating synthetic latent fingerprints can be used to improve the recognition\nperformance of any latent matcher and its individual components (e.g.,\nenhancement, segmentation and feature extraction).\n", "date": "2022-08-29", "categories": []}
{"id": "2208.14326", "title": "GaitFi: Robust Device-Free Human Identification via WiFi and Vision\n  Multimodal Learning", "abstract": "  As an important biomarker for human identification, human gait can be\ncollected at a distance by passive sensors without subject cooperation, which\nplays an essential role in crime prevention, security detection and other human\nidentification applications. At present, most research works are based on\ncameras and computer vision techniques to perform gait recognition. However,\nvision-based methods are not reliable when confronting poor illuminations,\nleading to degrading performances. In this paper, we propose a novel multimodal\ngait recognition method, namely GaitFi, which leverages WiFi signals and videos\nfor human identification. In GaitFi, Channel State Information (CSI) that\nreflects the multi-path propagation of WiFi is collected to capture human\ngaits, while videos are captured by cameras. To learn robust gait information,\nwe propose a Lightweight Residual Convolution Network (LRCN) as the backbone\nnetwork, and further propose the two-stream GaitFi by integrating WiFi and\nvision features for the gait retrieval task. The GaitFi is trained by the\ntriplet loss and classification loss on different levels of features. Extensive\nexperiments are conducted in the real world, which demonstrates that the GaitFi\noutperforms state-of-the-art gait recognition methods based on single WiFi or\ncamera, achieving 94.2% for human identification tasks of 12 subjects.\n", "date": "2022-08-30", "categories": []}
{"id": "2208.13164", "title": "Face Anti-Spoofing from the Perspective of Data Sampling", "abstract": "  Without deploying face anti-spoofing countermeasures, face recognition\nsystems can be spoofed by presenting a printed photo, a video, or a silicon\nmask of a genuine user. Thus, face presentation attack detection (PAD) plays a\nvital role in providing secure facial access to digital devices. Most existing\nvideo-based PAD countermeasures lack the ability to cope with long-range\ntemporal variations in videos. Moreover, the key-frame sampling prior to the\nfeature extraction step has not been widely studied in the face anti-spoofing\ndomain. To mitigate these issues, this paper provides a data sampling approach\nby proposing a video processing scheme that models the long-range temporal\nvariations based on Gaussian Weighting Function. Specifically, the proposed\nscheme encodes the consecutive t frames of video sequences into a single RGB\nimage based on a Gaussian-weighted summation of the t frames. Using simply the\ndata sampling scheme alone, we demonstrate that state-of-the-art performance\ncan be achieved without any bells and whistles in both intra-database and\ninter-database testing scenarios for the three public benchmark datasets;\nnamely, Replay-Attack, MSU-MFSD, and CASIA-FASD. In particular, the proposed\nscheme provides a much lower error (from 15.2% to 6.7% on CASIA-FASD and 5.9%\nto 4.9% on Replay-Attack) compared to baselines in cross-database scenarios.\n", "date": "2022-08-28", "categories": []}
{"id": "2208.13341", "title": "Artificial Neural Networks for Finger Vein Recognition: A Survey", "abstract": "  Finger vein recognition is an emerging biometric recognition technology.\nDifferent from the other biometric features on the body surface, the venous\nvascular tissue of the fingers is buried deep inside the skin. Due to this\nadvantage, finger vein recognition is highly stable and private. They are\nalmost impossible to be stolen and difficult to interfere with by external\nconditions. Unlike the finger vein recognition methods based on traditional\nmachine learning, the artificial neural network technique, especially deep\nlearning, it without relying on feature engineering and have superior\nperformance. To summarize the development of finger vein recognition based on\nartificial neural networks, this paper collects 149 related papers. First, we\nintroduce the background of finger vein recognition and the motivation of this\nsurvey. Then, the development history of artificial neural networks and the\nrepresentative networks on finger vein recognition tasks are introduced. The\npublic datasets that are widely used in finger vein recognition are then\ndescribed. After that, we summarize the related finger vein recognition tasks\nbased on classical neural networks and deep neural networks, respectively.\nFinally, the challenges and potential development directions in finger vein\nrecognition are discussed. To our best knowledge, this paper is the first\ncomprehensive survey focusing on finger vein recognition based on artificial\nneural networks.\n", "date": "2022-08-29", "categories": []}
{"id": "2208.13400", "title": "Towards Explaining Demographic Bias through the Eyes of Face Recognition\n  Models", "abstract": "  Biases inherent in both data and algorithms make the fairness of widespread\nmachine learning (ML)-based decision-making systems less than optimal. To\nimprove the trustfulness of such ML decision systems, it is crucial to be aware\nof the inherent biases in these solutions and to make them more transparent to\nthe public and developers. In this work, we aim at providing a set of\nexplainability tool that analyse the difference in the face recognition models'\nbehaviors when processing different demographic groups. We do that by\nleveraging higher-order statistical information based on activation maps to\nbuild explainability tools that link the FR models' behavior differences to\ncertain facial regions. The experimental results on two datasets and two face\nrecognition models pointed out certain areas of the face where the FR models\nreact differently for certain demographic groups compared to reference groups.\nThe outcome of these analyses interestingly aligns well with the results of\nstudies that analyzed the anthropometric differences and the human judgment\ndifferences on the faces of different demographic groups. This is thus the\nfirst study that specifically tries to explain the biased behavior of FR models\non different demographic groups and link it directly to the spatial facial\nfeatures. The code is publicly available here.\n", "date": "2022-08-29", "categories": []}
{"id": "2208.11814", "title": "Skeleton Prototype Contrastive Learning with Multi-Level Graph Relation\n  Modeling for Unsupervised Person Re-Identification", "abstract": "  Person re-identification (re-ID) via 3D skeletons is an important emerging\ntopic with many merits. Existing solutions rarely explore valuable\nbody-component relations in skeletal structure or motion, and they typically\nlack the ability to learn general representations with unlabeled skeleton data\nfor person re-ID. This paper proposes a generic unsupervised Skeleton Prototype\nContrastive learning paradigm with Multi-level Graph Relation learning\n(SPC-MGR) to learn effective representations from unlabeled skeletons to\nperform person re-ID. Specifically, we first construct unified multi-level\nskeleton graphs to fully model body structure within skeletons. Then we propose\na multi-head structural relation layer to comprehensively capture relations of\nphysically-connected body-component nodes in graphs. A full-level collaborative\nrelation layer is exploited to infer collaboration between motion-related body\nparts at various levels, so as to capture rich body features and recognizable\nwalking patterns. Lastly, we propose a skeleton prototype contrastive learning\nscheme that clusters feature-correlative instances of unlabeled graph\nrepresentations and contrasts their inherent similarity with representative\nskeleton features (\"skeleton prototypes\") to learn discriminative skeleton\nrepresentations for person re-ID. Empirical evaluations show that SPC-MGR\nsignificantly outperforms several state-of-the-art skeleton-based methods, and\nit also achieves highly competitive person re-ID performance for more general\nscenarios.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.11822", "title": "Benchmarking Human Face Similarity Using Identical Twins", "abstract": "  The problem of distinguishing identical twins and non-twin look-alikes in\nautomated facial recognition (FR) applications has become increasingly\nimportant with the widespread adoption of facial biometrics. Due to the high\nfacial similarity of both identical twins and look-alikes, these face pairs\nrepresent the hardest cases presented to facial recognition tools. This work\npresents an application of one of the largest twin datasets compiled to date to\naddress two FR challenges: 1) determining a baseline measure of facial\nsimilarity between identical twins and 2) applying this similarity measure to\ndetermine the impact of doppelgangers, or look-alikes, on FR performance for\nlarge face datasets. The facial similarity measure is determined via a deep\nconvolutional neural network. This network is trained on a tailored\nverification task designed to encourage the network to group together highly\nsimilar face pairs in the embedding space and achieves a test AUC of 0.9799.\nThe proposed network provides a quantitative similarity score for any two given\nfaces and has been applied to large-scale face datasets to identify similar\nface pairs. An additional analysis which correlates the comparison score\nreturned by a facial recognition tool and the similarity score returned by the\nproposed network has also been performed.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.12023", "title": "Identity-Sensitive Knowledge Propagation for Cloth-Changing Person\n  Re-identification", "abstract": "  Cloth-changing person re-identification (CC-ReID), which aims to match person\nidentities under clothing changes, is a new rising research topic in recent\nyears. However, typical biometrics-based CC-ReID methods often require\ncumbersome pose or body part estimators to learn cloth-irrelevant features from\nhuman biometric traits, which comes with high computational costs. Besides, the\nperformance is significantly limited due to the resolution degradation of\nsurveillance images. To address the above limitations, we propose an effective\nIdentity-Sensitive Knowledge Propagation framework (DeSKPro) for CC-ReID.\nSpecifically, a Cloth-irrelevant Spatial Attention module is introduced to\neliminate the distraction of clothing appearance by acquiring knowledge from\nthe human parsing module. To mitigate the resolution degradation issue and mine\nidentity-sensitive cues from human faces, we propose to restore the missing\nfacial details using prior facial knowledge, which is then propagated to a\nsmaller network. After training, the extra computations for human parsing or\nface restoration are no longer required. Extensive experiments show that our\nframework outperforms state-of-the-art methods by a large margin. Our code is\navailable at https://github.com/KimbingNg/DeskPro.\n", "date": "2022-08-25", "categories": []}
{"id": "2208.11440", "title": "Dynamic Template Initialization for Part-Aware Person Re-ID", "abstract": "  Many of the existing Person Re-identification (Re-ID) approaches depend on\nfeature maps which are either partitioned to localize parts of a person or\nreduced to create a global representation. While part localization has shown\nsignificant success, it uses either na{\\i}ve position-based partitions or\nstatic feature templates. These, however, hypothesize the pre-existence of the\nparts in a given image or their positions, ignoring the input image-specific\ninformation which limits their usability in challenging scenarios such as Re-ID\nwith partial occlusions and partial probe images. In this paper, we introduce a\nspatial attention-based Dynamic Part Template Initialization module that\ndynamically generates part-templates using mid-level semantic features at the\nearlier layers of the backbone. Following a self-attention layer, human\npart-level features of the backbone are used to extract the templates of\ndiverse human body parts using a simplified cross-attention scheme which will\nthen be used to identify and collate representations of various human parts\nfrom semantically rich features, increasing the discriminative ability of the\nentire model. We further explore adaptive weighting of part descriptors to\nquantify the absence or occlusion of local attributes and suppress the\ncontribution of the corresponding part descriptors to the matching criteria.\nExtensive experiments on holistic, occluded, and partial Re-ID task benchmarks\ndemonstrate that our proposed architecture is able to achieve competitive\nperformance. Codes will be included in the supplementary material and will be\nmade publicly available.\n", "date": "2022-08-24", "categories": []}
{"id": "2208.11483", "title": "SubFace: Learning with Softmax Approximation for Face Recognition", "abstract": "  The softmax-based loss functions and its variants (e.g., cosface, sphereface,\nand arcface) significantly improve the face recognition performance in wild\nunconstrained scenes. A common practice of these algorithms is to perform\noptimizations on the multiplication between the embedding features and the\nlinear transformation matrix. However in most cases, the dimension of embedding\nfeatures is given based on traditional design experience, and there is\nless-studied on improving performance using the feature itself when giving a\nfixed size. To address this challenge, this paper presents a softmax\napproximation method called SubFace, which employs the subspace feature to\npromote the performance of face recognition. Specifically, we dynamically\nselect the non-overlapping subspace features in each batch during training, and\nthen use the subspace features to approximate full-feature among softmax-based\nloss, so the discriminability of the deep model can be significantly enhanced\nfor face recognition. Comprehensive experiments conducted on benchmark datasets\ndemonstrate that our method can significantly improve the performance of\nvanilla CNN baseline, which strongly proves the effectiveness of subspace\nstrategy with the margin-based loss.\n", "date": "2022-08-24", "categories": []}
{"id": "2208.04712", "title": "Areas of Strategic Visibility: Disability Bias in Biometrics", "abstract": "  This response to the RFI considers the potential for biometrics to help or\nharm disabled people2. Biometrics are already integrated into many aspects of\ndaily life, from airport travel to mobile phone use. Yet many of these systems\nare not accessible to people who experience different kinds of disability\nexclusion . Different personal characteristics may impact any or all of the\nphysical (DNA, fingerprints, face or retina) and behavioral (gesture, gait,\nvoice) characteristics listed in the RFI as examples of biometric signals.\n", "date": "2022-07-14", "categories": []}
{"id": "2208.07841", "title": "OrthoMAD: Morphing Attack Detection Through Orthogonal Identity\n  Disentanglement", "abstract": "  Morphing attacks are one of the many threats that are constantly affecting\ndeep face recognition systems. It consists of selecting two faces from\ndifferent individuals and fusing them into a final image that contains the\nidentity information of both. In this work, we propose a novel regularisation\nterm that takes into account the existent identity information in both and\npromotes the creation of two orthogonal latent vectors. We evaluate our\nproposed method (OrthoMAD) in five different types of morphing in the FRLL\ndataset and evaluate the performance of our model when trained on five distinct\ndatasets. With a small ResNet-18 as the backbone, we achieve state-of-the-art\nresults in the majority of the experiments, and competitive results in the\nothers. The code of this paper will be publicly available.\n", "date": "2022-08-16", "categories": []}
{"id": "2208.10564", "title": "State Of The Art In Open-Set Iris Presentation Attack Detection", "abstract": "  Research in presentation attack detection (PAD) for iris recognition has\nlargely moved beyond evaluation in \"closed-set\" scenarios, to emphasize ability\nto generalize to presentation attack types not present in the training data.\nThis paper offers several contributions to understand and extend the\nstate-of-the-art in open-set iris PAD. First, it describes the most\nauthoritative evaluation to date of iris PAD. We have curated the largest\npublicly-available image dataset for this problem, drawing from 26 benchmarks\npreviously released by various groups, and adding 150,000 images being released\nwith the journal version of this paper, to create a set of 450,000 images\nrepresenting authentic iris and seven types of presentation attack instrument\n(PAI). We formulate a leave-one-PAI-out evaluation protocol, and show that even\nthe best algorithms in the closed-set evaluations exhibit catastrophic failures\non multiple attack types in the open-set scenario. This includes algorithms\nperforming well in the most recent LivDet-Iris 2020 competition, which may come\nfrom the fact that the LivDet-Iris protocol emphasizes sequestered images\nrather than unseen attack types. Second, we evaluate the accuracy of five\nopen-source iris presentation attack algorithms available today, one of which\nis newly-proposed in this paper, and build an ensemble method that beats the\nwinner of the LivDet-Iris 2020 by a substantial margin. This paper demonstrates\nthat closed-set iris PAD, when all PAIs are known during training, is a solved\nproblem, with multiple algorithms showing very high accuracy, while open-set\niris PAD, when evaluated correctly, is far from being solved. The newly-created\ndataset, new open-source algorithms, and evaluation protocol, made publicly\navailable with the journal version of this paper, provide the experimental\nartifacts that researchers can use to measure progress on this important\nproblem.\n", "date": "2022-08-22", "categories": []}
{"id": "2208.11099", "title": "Explaining Bias in Deep Face Recognition via Image Characteristics", "abstract": "  In this paper, we propose a novel explanatory framework aimed to provide a\nbetter understanding of how face recognition models perform as the underlying\ndata characteristics (protected attributes: gender, ethnicity, age;\nnon-protected attributes: facial hair, makeup, accessories, face orientation\nand occlusion, image distortion, emotions) on which they are tested change.\nWith our framework, we evaluate ten state-of-the-art face recognition models,\ncomparing their fairness in terms of security and usability on two data sets,\ninvolving six groups based on gender and ethnicity. We then analyze the impact\nof image characteristics on models performance. Our results show that trends\nappearing in a single-attribute analysis disappear or reverse when\nmulti-attribute groups are considered, and that performance disparities are\nalso related to non-protected attributes. Source code: https://cutt.ly/2XwRLiA.\n", "date": "2022-08-23", "categories": []}
{"id": "2208.08901", "title": "EEG-BBNet: a Hybrid Framework for Brain Biometric using Graph\n  Connectivity", "abstract": "  Brain biometrics based on electroencephalography (EEG) have been used\nincreasingly for personal identification. Traditional machine learning\ntechniques as well as modern day deep learning methods have been applied with\npromising results. In this paper we present EEG-BBNet, a hybrid network which\nintegrates convolutional neural networks (CNN) with graph convolutional neural\nnetworks (GCNN). The benefit of the CNN in automatic feature extraction and the\ncapability of GCNN in learning connectivity between EEG electrodes through\ngraph representation are jointly exploited. We examine various connectivity\nmeasures, namely the Euclidean distance, Pearson's correlation coefficient,\nphase-locked value, phase-lag index, and Rho index. The performance of the\nproposed method is assessed on a benchmark dataset consisting of various\nbrain-computer interface (BCI) tasks and compared to other state-of-the-art\napproaches. We found that our models outperform all baselines in the\nevent-related potential (ERP) task with an average correct recognition rates up\nto 99.26% using intra-session data. EEG-BBNet with Pearson's correlation and\nRHO index provide the best classification results. In addition, our model\ndemonstrates greater adaptability using inter-session and inter-task data. We\nalso investigate the practicality of our proposed model with smaller number of\nelectrodes. Electrode placements over the frontal lobe region appears to be\nmost appropriate with minimal lost in performance.\n", "date": "2022-08-17", "categories": []}
{"id": "2208.08207", "title": "Time flies by: Analyzing the Impact of Face Ageing on the Recognition\n  Performance with Synthetic Data", "abstract": "  The vast progress in synthetic image synthesis enables the generation of\nfacial images in high resolution and photorealism. In biometric applications,\nthe main motivation for using synthetic data is to solve the shortage of\npublicly-available biometric data while reducing privacy risks when processing\nsuch sensitive information. These advantages are exploited in this work by\nsimulating human face ageing with recent face age modification algorithms to\ngenerate mated samples, thereby studying the impact of ageing on the\nperformance of an open-source biometric recognition system. Further, a real\ndataset is used to evaluate the effects of short-term ageing, comparing the\nbiometric performance to the synthetic domain. The main findings indicate that\nshort-term ageing in the range of 1-5 years has only minor effects on the\ngeneral recognition performance. However, the correct verification of mated\nfaces with long-term age differences beyond 20 years poses still a significant\nchallenge and requires further investigation.\n", "date": "2022-08-17", "categories": []}
{"id": "2208.07613", "title": "Does lossy image compression affect racial bias within face recognition?", "abstract": "  Yes - This study investigates the impact of commonplace lossy image\ncompression on face recognition algorithms with regard to the racial\ncharacteristics of the subject. We adopt a recently proposed racial\nphenotype-based bias analysis methodology to measure the effect of varying\nlevels of lossy compression across racial phenotype categories. Additionally,\nwe determine the relationship between chroma-subsampling and race-related\nphenotypes for recognition performance. Prior work investigates the impact of\nlossy JPEG compression algorithm on contemporary face recognition performance.\nHowever, there is a gap in how this impact varies with different race-related\ninter-sectional groups and the cause of this impact. Via an extensive\nexperimental setup, we demonstrate that common lossy image compression\napproaches have a more pronounced negative impact on facial recognition\nperformance for specific racial phenotype categories such as darker skin tones\n(by up to 34.55\\%). Furthermore, removing chroma-subsampling during compression\nimproves the false matching rate (up to 15.95\\%) across all phenotype\ncategories affected by the compression, including darker skin tones, wide\nnoses, big lips, and monolid eye categories. In addition, we outline the\ncharacteristics that may be attributable as the underlying cause of such\nphenomenon for lossy compression algorithms such as JPEG.\n", "date": "2022-08-16", "categories": []}
{"id": "2208.07828", "title": "Learning Facial Liveness Representation for Domain Generalized Face\n  Anti-spoofing", "abstract": "  Face anti-spoofing (FAS) aims at distinguishing face spoof attacks from the\nauthentic ones, which is typically approached by learning proper models for\nperforming the associated classification task. In practice, one would expect\nsuch models to be generalized to FAS in different image domains. Moreover, it\nis not practical to assume that the type of spoof attacks would be known in\nadvance. In this paper, we propose a deep learning model for addressing the\naforementioned domain-generalized face anti-spoofing task. In particular, our\nproposed network is able to disentangle facial liveness representation from the\nirrelevant ones (i.e., facial content and image domain features). The resulting\nliveness representation exhibits sufficient domain invariant properties, and\nthus it can be applied for performing domain-generalized FAS. In our\nexperiments, we conduct experiments on five benchmark datasets with various\nsettings, and we verify that our model performs favorably against\nstate-of-the-art approaches in identifying novel types of spoof attacks in\nunseen image domains.\n", "date": "2022-08-16", "categories": []}
{"id": "2208.02760", "title": "OCFR 2022: Competition on Occluded Face Recognition From Synthetically\n  Generated Structure-Aware Occlusions", "abstract": "  This work summarizes the IJCB Occluded Face Recognition Competition 2022\n(IJCB-OCFR-2022) embraced by the 2022 International Joint Conference on\nBiometrics (IJCB 2022). OCFR-2022 attracted a total of 3 participating teams,\nfrom academia. Eventually, six valid submissions were submitted and then\nevaluated by the organizers. The competition was held to address the challenge\nof face recognition in the presence of severe face occlusions. The participants\nwere free to use any training data and the testing data was built by the\norganisers by synthetically occluding parts of the face images using a\nwell-known dataset. The submitted solutions presented innovations and performed\nvery competitively with the considered baseline. A major output of this\ncompetition is a challenging, realistic, and diverse, and publicly available\noccluded face recognition benchmark with well defined evaluation protocols.\n", "date": "2022-08-04", "categories": []}
{"id": "2208.07241", "title": "HEFT: Homomorphically Encrypted Fusion of Biometric Templates", "abstract": "  This paper proposes a non-interactive end-to-end solution for secure fusion\nand matching of biometric templates using fully homomorphic encryption (FHE).\nGiven a pair of encrypted feature vectors, we perform the following ciphertext\noperations, i) feature concatenation, ii) fusion and dimensionality reduction\nthrough a learned linear projection, iii) scale normalization to unit\n$\\ell_2$-norm, and iv) match score computation. Our method, dubbed HEFT\n(Homomorphically Encrypted Fusion of biometric Templates), is custom-designed\nto overcome the unique constraint imposed by FHE, namely the lack of support\nfor non-arithmetic operations. From an inference perspective, we systematically\nexplore different data packing schemes for computationally efficient linear\nprojection and introduce a polynomial approximation for scale normalization.\nFrom a training perspective, we introduce an FHE-aware algorithm for learning\nthe linear projection matrix to mitigate errors induced by approximate\nnormalization. Experimental evaluation for template fusion and matching of face\nand voice biometrics shows that HEFT (i) improves biometric verification\nperformance by 11.07% and 9.58% AUROC compared to the respective unibiometric\nrepresentations while compressing the feature vectors by a factor of 16 (512D\nto 32D), and (ii) fuses a pair of encrypted feature vectors and computes its\nmatch score against a gallery of size 1024 in 884 ms. Code and data are\navailable at https://github.com/human-analysis/encrypted-biometric-fusion\n", "date": "2022-08-15", "categories": []}
{"id": "2208.07337", "title": "SYN-MAD 2022: Competition on Face Morphing Attack Detection Based on\n  Privacy-aware Synthetic Training Data", "abstract": "  This paper presents a summary of the Competition on Face Morphing Attack\nDetection Based on Privacy-aware Synthetic Training Data (SYN-MAD) held at the\n2022 International Joint Conference on Biometrics (IJCB 2022). The competition\nattracted a total of 12 participating teams, both from academia and industry\nand present in 11 different countries. In the end, seven valid submissions were\nsubmitted by the participating teams and evaluated by the organizers. The\ncompetition was held to present and attract solutions that deal with detecting\nface morphing attacks while protecting people's privacy for ethical and legal\nreasons. To ensure this, the training data was limited to synthetic data\nprovided by the organizers. The submitted solutions presented innovations that\nled to outperforming the considered baseline in many experimental settings. The\nevaluation benchmark is now available at:\nhttps://github.com/marcohuber/SYN-MAD-2022.\n", "date": "2022-08-15", "categories": []}
{"id": "2208.05864", "title": "Face Morphing Attacks and Face Image Quality: The Effect of Morphing and\n  the Unsupervised Attack Detection by Quality", "abstract": "  Morphing attacks are a form of presentation attacks that gathered increasing\nattention in recent years. A morphed image can be successfully verified to\nmultiple identities. This operation, therefore, poses serious security issues\nrelated to the ability of a travel or identity document to be verified to\nbelong to multiple persons. Previous works touched on the issue of the quality\nof morphing attack images, however, with the main goal of quantitatively\nproofing the realistic appearance of the produced morphing attacks. We theorize\nthat the morphing processes might have an effect on both, the perceptual image\nquality and the image utility in face recognition (FR) when compared to bona\nfide samples. Towards investigating this theory, this work provides an\nextensive analysis of the effect of morphing on face image quality, including\nboth general image quality measures and face image utility measures. This\nanalysis is not limited to a single morphing technique, but rather looks at six\ndifferent morphing techniques and five different data sources using ten\ndifferent quality measures. This analysis reveals consistent separability\nbetween the quality scores of morphing attack and bona fide samples measured by\ncertain quality measures. Our study goes further to build on this effect and\ninvestigate the possibility of performing unsupervised morphing attack\ndetection (MAD) based on quality scores. Our study looks intointra and\ninter-dataset detectability to evaluate the generalizability of such a\ndetection concept on different morphing techniques and bona fide sources. Our\nfinal results point out that a set of quality measures, such as MagFace and\nCNNNIQA, can be used to perform unsupervised and generalized MAD with a correct\nclassification accuracy of over 70%.\n", "date": "2022-08-11", "categories": []}
{"id": "2208.04040", "title": "Eight Years of Face Recognition Research: Reproducibility, Achievements\n  and Open Issues", "abstract": "  Automatic face recognition is a research area with high popularity. Many\ndifferent face recognition algorithms have been proposed in the last thirty\nyears of intensive research in the field. With the popularity of deep learning\nand its capability to solve a huge variety of different problems, face\nrecognition researchers have concentrated effort on creating better models\nunder this paradigm. From the year 2015, state-of-the-art face recognition has\nbeen rooted in deep learning models. Despite the availability of large-scale\nand diverse datasets for evaluating the performance of face recognition\nalgorithms, many of the modern datasets just combine different factors that\ninfluence face recognition, such as face pose, occlusion, illumination, facial\nexpression and image quality. When algorithms produce errors on these datasets,\nit is not clear which of the factors has caused this error and, hence, there is\nno guidance in which direction more research is required. This work is a\nfollowup from our previous works developed in 2014 and eventually published in\n2016, showing the impact of various facial aspects on face recognition\nalgorithms. By comparing the current state-of-the-art with the best systems\nfrom the past, we demonstrate that faces under strong occlusions, some types of\nillumination, and strong expressions are problems mastered by deep learning\nalgorithms, whereas recognition with low-resolution images, extreme pose\nvariations, and open-set recognition is still an open problem. To show this, we\nrun a sequence of experiments using six different datasets and five different\nface recognition algorithms in an open-source and reproducible manner. We\nprovide the source code to run all of our experiments, which is easily\nextensible so that utilizing your own deep network in our evaluation is just a\nfew minutes away.\n", "date": "2022-08-08", "categories": []}
{"id": "2208.04352", "title": "Rethinking Robust Representation Learning Under Fine-grained Noisy Faces", "abstract": "  Learning robust feature representation from large-scale noisy faces stands\nout as one of the key challenges in high-performance face recognition. Recent\nattempts have been made to cope with this challenge by alleviating the\nintra-class conflict and inter-class conflict. However, the unconstrained noise\ntype in each conflict still makes it difficult for these algorithms to perform\nwell. To better understand this, we reformulate the noise type of each class in\na more fine-grained manner as N-identities|K^C-clusters. Different types of\nnoisy faces can be generated by adjusting the values of \\nkc. Based on this\nunified formulation, we found that the main barrier behind the noise-robust\nrepresentation learning is the flexibility of the algorithm under different N,\nK, and C. For this potential problem, we propose a new method, named Evolving\nSub-centers Learning~(ESL), to find optimal hyperplanes to accurately describe\nthe latent space of massive noisy faces. More specifically, we initialize M\nsub-centers for each class and ESL encourages it to be automatically aligned to\nN-identities|K^C-clusters faces via producing, merging, and dropping\noperations. Images belonging to the same identity in noisy faces can\neffectively converge to the same sub-center and samples with different\nidentities will be pushed away. We inspect its effectiveness with an elaborate\nablation study on the synthetic noisy dataset with different N, K, and C.\nWithout any bells and whistles, ESL can achieve significant performance gains\nover state-of-the-art methods on large-scale noisy faces\n", "date": "2022-08-08", "categories": []}
{"id": "2208.02991", "title": "Analyzing the Impact of Shape & Context on the Face Recognition\n  Performance of Deep Networks", "abstract": "  In this article, we analyze how changing the underlying 3D shape of the base\nidentity in face images can distort their overall appearance, especially from\nthe perspective of deep face recognition. As done in popular training data\naugmentation schemes, we graphically render real and synthetic face images with\nrandomly chosen or best-fitting 3D face models to generate novel views of the\nbase identity. We compare deep features generated from these images to assess\nthe perturbation these renderings introduce into the original identity. We\nperform this analysis at various degrees of facial yaw with the base identities\nvarying in gender and ethnicity. Additionally, we investigate if adding some\nform of context and background pixels in these rendered images, when used as\ntraining data, further improves the downstream performance of a face\nrecognition model. Our experiments demonstrate the significance of facial shape\nin accurate face matching and underpin the importance of contextual data for\nnetwork training.\n", "date": "2022-08-05", "categories": []}
{"id": "2208.03110", "title": "MorDeephy: Face Morphing Detection Via Fused Classification", "abstract": "  Face morphing attack detection (MAD) is one of the most challenging tasks in\nthe field of face recognition nowadays. In this work, we introduce a novel deep\nlearning strategy for a single image face morphing detection, which implies the\ndiscrimination of morphed face images along with a sophisticated face\nrecognition task in a complex classification scheme. It is directed onto\nlearning the deep facial features, which carry information about the\nauthenticity of these features. Our work also introduces several additional\ncontributions: the public and easy-to-use face morphing detection benchmark and\nthe results of our wild datasets filtering strategy. Our method, which we call\nMorDeephy, achieved the state of the art performance and demonstrated a\nprominent ability for generalising the task of morphing detection to unseen\nscenarios.\n", "date": "2022-08-05", "categories": []}
{"id": "2208.03138", "title": "Human Saliency-Driven Patch-based Matching for Interpretable Post-mortem\n  Iris Recognition", "abstract": "  Forensic iris recognition, as opposed to live iris recognition, is an\nemerging research area that leverages the discriminative power of iris\nbiometrics to aid human examiners in their efforts to identify deceased\npersons. As a machine learning-based technique in a predominantly\nhuman-controlled task, forensic recognition serves as \"back-up\" to human\nexpertise in the task of post-mortem identification. As such, the machine\nlearning model must be (a) interpretable, and (b) post-mortem-specific, to\naccount for changes in decaying eye tissue. In this work, we propose a method\nthat satisfies both requirements, and that approaches the creation of a\npost-mortem-specific feature extractor in a novel way employing human\nperception. We first train a deep learning-based feature detector on\npost-mortem iris images, using annotations of image regions highlighted by\nhumans as salient for their decision making. In effect, the method learns\ninterpretable features directly from humans, rather than purely data-driven\nfeatures. Second, regional iris codes (again, with human-driven filtering\nkernels) are used to pair detected iris patches, which are translated into\npairwise, patch-based comparison scores. In this way, our method presents human\nexaminers with human-understandable visual cues in order to justify the\nidentification decision and corresponding confidence score. When tested on a\ndataset of post-mortem iris images collected from 259 deceased subjects, the\nproposed method places among the three best iris matchers, demonstrating better\nresults than the commercial (non-human-interpretable) VeriEye approach. We\npropose a unique post-mortem iris recognition method trained with human\nsaliency to give fully-interpretable comparison outcomes for use in the context\nof forensic examination, achieving state-of-the-art recognition performance.\n", "date": "2022-08-03", "categories": []}
{"id": "2208.02450", "title": "Learning Modal-Invariant and Temporal-Memory for Video-based\n  Visible-Infrared Person Re-Identification", "abstract": "  Thanks for the cross-modal retrieval techniques, visible-infrared (RGB-IR)\nperson re-identification (Re-ID) is achieved by projecting them into a common\nspace, allowing person Re-ID in 24-hour surveillance systems. However, with\nrespect to the probe-to-gallery, almost all existing RGB-IR based cross-modal\nperson Re-ID methods focus on image-to-image matching, while the video-to-video\nmatching which contains much richer spatial- and temporal-information remains\nunder-explored. In this paper, we primarily study the video-based cross-modal\nperson Re-ID method. To achieve this task, a video-based RGB-IR dataset is\nconstructed, in which 927 valid identities with 463,259 frames and 21,863\ntracklets captured by 12 RGB/IR cameras are collected. Based on our constructed\ndataset, we prove that with the increase of frames in a tracklet, the\nperformance does meet more enhancement, demonstrating the significance of\nvideo-to-video matching in RGB-IR person Re-ID. Additionally, a novel method is\nfurther proposed, which not only projects two modalities to a modal-invariant\nsubspace, but also extracts the temporal-memory for motion-invariant. Thanks to\nthese two strategies, much better results are achieved on our video-based\ncross-modal person Re-ID. The code and dataset are released at:\nhttps://github.com/VCMproject233/MITML.\n", "date": "2022-08-04", "categories": []}
{"id": "2208.00785", "title": "Verification system based on long-range iris and Graph Siamese Neural\n  Networks", "abstract": "  Biometric systems represent valid solutions in tasks like user authentication\nand verification, since they are able to analyze physical and behavioural\nfeatures with high precision. However, especially when physical biometrics are\nused, as is the case of iris recognition, they require specific hardware such\nas retina scanners, sensors, or HD cameras to achieve relevant results. At the\nsame time, they require the users to be very close to the camera to extract\nhigh-resolution information. For this reason, in this work, we propose a novel\napproach that uses long-range (LR) distance images for implementing an iris\nverification system. More specifically, we present a novel methodology for\nconverting LR iris images into graphs and then use Graph Siamese Neural\nNetworks (GSNN) to predict whether two graphs belong to the same person. In\nthis study, we not only describe this methodology but also evaluate how the\nspectral components of these images can be used for improving the graph\nextraction and the final classification task. Results demonstrate the\nsuitability of this approach, encouraging the community to explore graph\napplication in biometric systems.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.14452", "title": "Deep Learning-based Occluded Person Re-identification: A Survey", "abstract": "  Occluded person re-identification (Re-ID) aims at addressing the occlusion\nproblem when retrieving the person of interest across multiple cameras. With\nthe promotion of deep learning technology and the increasing demand for\nintelligent video surveillance, the frequent occlusion in real-world\napplications has made occluded person Re-ID draw considerable interest from\nresearchers. A large number of occluded person Re-ID methods have been proposed\nwhile there are few surveys that focus on occlusion. To fill this gap and help\nboost future research, this paper provides a systematic survey of occluded\nperson Re-ID. Through an in-depth analysis of the occlusion in person Re-ID,\nmost existing methods are found to only consider part of the problems brought\nby occlusion. Therefore, we review occlusion-related person Re-ID methods from\nthe perspective of issues and solutions. We summarize four issues caused by\nocclusion in person Re-ID, i.e., position misalignment, scale misalignment,\nnoisy information, and missing information. The occlusion-related methods\naddressing different issues are then categorized and introduced accordingly.\nAfter that, we summarize and compare the performance of recent occluded person\nRe-ID methods on four popular datasets: Partial-ReID, Partial-iLIDS,\nOccluded-ReID, and Occluded-DukeMTMC. Finally, we provide insights on promising\nfuture research directions.\n", "date": "2022-07-29", "categories": []}
{"id": "2207.14243", "title": "Combining human parsing with analytical feature extraction and ranking\n  schemes for high-generalization person reidentification", "abstract": "  Person reidentification (re-ID) has been receiving increasing attention in\nrecent years due to its importance for both science and society. Machine\nlearning and particularly Deep Learning (DL) has become the main re-id tool\nthat allowed researches to achieve unprecedented accuracy levels on benchmark\ndatasets. However, there is a known problem of poor generalization of DL\nmodels. That is, models trained to achieve high accuracy on one dataset perform\npoorly on other ones and require re-training. To address this issue, we present\na model without trainable parameters which shows great potential for high\ngeneralization. It combines a fully analytical feature extraction and\nsimilarity ranking scheme with DL-based human parsing used to obtain the\ninitial subregion classification. We show that such combination to a high\nextent eliminates the drawbacks of existing analytical methods. We use\ninterpretable color and texture features which have human-readable similarity\nmeasures associated with them. To verify the proposed method we conduct\nexperiments on Market1501 and CUHK03 datasets achieving competitive rank-1\naccuracy comparable with that of DL-models. Most importantly we show that our\nmethod achieves 63.9% and 93.5% rank-1 cross-domain accuracy when applied to\ntransfer learning tasks. It is significantly higher than previously reported\n30-50% transfer accuracy. We discuss the potential ways of adding new features\nto further improve the model. We also show the advantage of interpretable\nfeatures for constructing human-generated queries from verbal description to\nconduct search without a query image.\n", "date": "2022-07-28", "categories": []}
{"id": "2207.13037", "title": "Learning Resolution-Adaptive Representations for Cross-Resolution Person\n  Re-Identification", "abstract": "  The cross-resolution person re-identification (CRReID) problem aims to match\nlow-resolution (LR) query identity images against high resolution (HR) gallery\nimages. It is a challenging and practical problem since the query images often\nsuffer from resolution degradation due to the different capturing conditions\nfrom real-world cameras. To address this problem, state-of-the-art (SOTA)\nsolutions either learn the resolution-invariant representation or adopt\nsuper-resolution (SR) module to recover the missing information from the LR\nquery. This paper explores an alternative SR-free paradigm to directly compare\nHR and LR images via a dynamic metric, which is adaptive to the resolution of a\nquery image. We realize this idea by learning resolution-adaptive\nrepresentations for cross-resolution comparison. Specifically, we propose two\nresolution-adaptive mechanisms. The first one disentangles the\nresolution-specific information into different sub-vectors in the penultimate\nlayer of the deep neural networks, and thus creates a varying-length\nrepresentation. To better extract resolution-dependent information, we further\npropose to learn resolution-adaptive masks for intermediate residual feature\nblocks. A novel progressive learning strategy is proposed to train those masks\nproperly. These two mechanisms are combined to boost the performance of CRReID.\nExperimental results show that the proposed method is superior to existing\napproaches and achieves SOTA performance on multiple CRReID benchmarks.\n", "date": "2022-07-09", "categories": []}
{"id": "2207.12272", "title": "Online Adaptive Personalization for Face Anti-spoofing", "abstract": "  Face authentication systems require a robust anti-spoofing module as they can\nbe deceived by fabricating spoof images of authorized users. Most recent face\nanti-spoofing methods rely on optimized architectures and training objectives\nto alleviate the distribution shift between train and test users. However, in\nreal online scenarios, past data from a user contains valuable information that\ncould be used to alleviate the distribution shift. We thus introduce OAP\n(Online Adaptive Personalization): a lightweight solution which can adapt the\nmodel online using unlabeled data. OAP can be applied on top of most\nanti-spoofing methods without the need to store original biometric images.\nThrough experimental evaluation on the SiW dataset, we show that OAP improves\nrecognition performance of existing methods on both single video setting and\ncontinual setting, where spoof videos are interleaved with live ones to\nsimulate spoofing attacks. We also conduct ablation studies to confirm the\ndesign choices for our solution.\n", "date": "2022-07-04", "categories": []}
{"id": "2207.10807", "title": "A Machine Learning Approach for Driver Identification Based on CAN-BUS\n  Sensor Data", "abstract": "  Driver identification is a momentous field of modern decorated vehicles in\nthe controller area network (CAN-BUS) perspective. Many conventional systems\nare used to identify the driver. One step ahead, most of the researchers use\nsensor data of CAN-BUS but there are some difficulties because of the variation\nof the protocol of different models of vehicle. Our aim is to identify the\ndriver through supervised learning algorithms based on driving behavior\nanalysis. To determine the driver, a driver verification technique is proposed\nthat evaluate driving pattern using the measurement of CAN sensor data. In this\npaper on-board diagnostic (OBD-II) is used to capture the data from the CAN-BUS\nsensor and the sensors are listed under SAE J1979 statement. According to the\nservice of OBD-II, drive identification is possible. However, we have gained\ntwo types of accuracy on a complete data set with 10 drivers and a partial data\nset with two drivers. The accuracy is good with less number of drivers compared\nto the higher number of drivers. We have achieved statistically significant\nresults in terms of accuracy in contrast to the baseline algorithm\n", "date": "2022-07-16", "categories": []}
{"id": "2207.10180", "title": "Controllable and Guided Face Synthesis for Unconstrained Face\n  Recognition", "abstract": "  Although significant advances have been made in face recognition (FR), FR in\nunconstrained environments remains challenging due to the domain gap between\nthe semi-constrained training datasets and unconstrained testing scenarios. To\naddress this problem, we propose a controllable face synthesis model (CFSM)\nthat can mimic the distribution of target datasets in a style latent space.\nCFSM learns a linear subspace with orthogonal bases in the style latent space\nwith precise control over the diversity and degree of synthesis. Furthermore,\nthe pre-trained synthesis model can be guided by the FR model, making the\nresulting images more beneficial for FR model training. Besides, target dataset\ndistributions are characterized by the learned orthogonal bases, which can be\nutilized to measure the distributional similarity among face datasets. Our\napproach yields significant performance gains on unconstrained benchmarks, such\nas IJB-B, IJB-C, TinyFace and IJB-S (+5.76% Rank1).\n", "date": "2022-07-20", "categories": []}
{"id": "2207.09505", "title": "An Efficient Method for Face Quality Assessment on the Edge", "abstract": "  Face recognition applications in practice are composed of two main steps:\nface detection and feature extraction. In a sole vision-based solution, the\nfirst step generates multiple detection for a single identity by ingesting a\ncamera stream. A practical approach on edge devices should prioritize these\ndetection of identities according to their conformity to recognition. In this\nperspective, we propose a face quality score regression by just appending a\nsingle layer to a face landmark detection network. With almost no additional\ncost, face quality scores are obtained by training this single layer to regress\nrecognition scores with surveillance like augmentations. We implemented the\nproposed approach on edge GPUs with all face detection pipeline steps,\nincluding detection, tracking, and alignment. Comprehensive experiments show\nthe proposed approach's efficiency through comparison with SOTA face quality\nregression models on different data sets and real-life scenarios.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.09884", "title": "Negative Samples are at Large: Leveraging Hard-distance Elastic Loss for\n  Re-identification", "abstract": "  We present a Momentum Re-identification (MoReID) framework that can leverage\na very large number of negative samples in training for general\nre-identification task. The design of this framework is inspired by Momentum\nContrast (MoCo), which uses a dictionary to store current and past batches to\nbuild a large set of encoded samples. As we find it less effective to use past\npositive samples which may be highly inconsistent to the encoded feature\nproperty formed with the current positive samples, MoReID is designed to use\nonly a large number of negative samples stored in the dictionary. However, if\nwe train the model using the widely used Triplet loss that uses only one sample\nto represent a set of positive/negative samples, it is hard to effectively\nleverage the enlarged set of negative samples acquired by the MoReID framework.\nTo maximize the advantage of using the scaled-up negative sample set, we newly\nintroduce Hard-distance Elastic loss (HE loss), which is capable of using more\nthan one hard sample to represent a large number of samples. Our experiments\ndemonstrate that a large number of negative samples provided by MoReID\nframework can be utilized at full capacity only with the HE loss, achieving the\nstate-of-the-art accuracy on three re-ID benchmarks, VeRi-776, Market-1501, and\nVeRi-Wild.\n", "date": "2022-07-20", "categories": []}
{"id": "2207.07316", "title": "Privacy-Preserving Face Recognition with Learnable Privacy Budgets in\n  Frequency Domain", "abstract": "  Face recognition technology has been used in many fields due to its high\nrecognition accuracy, including the face unlocking of mobile devices, community\naccess control systems, and city surveillance. As the current high accuracy is\nguaranteed by very deep network structures, facial images often need to be\ntransmitted to third-party servers with high computational power for inference.\nHowever, facial images visually reveal the user's identity information. In this\nprocess, both untrusted service providers and malicious users can significantly\nincrease the risk of a personal privacy breach. Current privacy-preserving\napproaches to face recognition are often accompanied by many side effects, such\nas a significant increase in inference time or a noticeable decrease in\nrecognition accuracy. This paper proposes a privacy-preserving face recognition\nmethod using differential privacy in the frequency domain. Due to the\nutilization of differential privacy, it offers a guarantee of privacy in\ntheory. Meanwhile, the loss of accuracy is very slight. This method first\nconverts the original image to the frequency domain and removes the direct\ncomponent termed DC. Then a privacy budget allocation method can be learned\nbased on the loss of the back-end face recognition network within the\ndifferential privacy framework. Finally, it adds the corresponding noise to the\nfrequency domain features. Our method performs very well with several classical\nface recognition test sets according to the extensive experiments.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.09046", "title": "Dynamic Prototype Mask for Occluded Person Re-Identification", "abstract": "  Although person re-identification has achieved an impressive improvement in\nrecent years, the common occlusion case caused by different obstacles is still\nan unsettled issue in real application scenarios. Existing methods mainly\naddress this issue by employing body clues provided by an extra network to\ndistinguish the visible part. Nevertheless, the inevitable domain gap between\nthe assistant model and the ReID datasets has highly increased the difficulty\nto obtain an effective and efficient model. To escape from the extra\npre-trained networks and achieve an automatic alignment in an end-to-end\ntrainable network, we propose a novel Dynamic Prototype Mask (DPM) based on two\nself-evident prior knowledge. Specifically, we first devise a Hierarchical Mask\nGenerator which utilizes the hierarchical semantic to select the visible\npattern space between the high-quality holistic prototype and the feature\nrepresentation of the occluded input image. Under this condition, the occluded\nrepresentation could be well aligned in a selected subspace spontaneously.\nThen, to enrich the feature representation of the high-quality holistic\nprototype and provide a more complete feature space, we introduce a Head Enrich\nModule to encourage different heads to aggregate different patterns\nrepresentation in the whole image. Extensive experimental evaluations conducted\non occluded and holistic person re-identification benchmarks demonstrate the\nsuperior performance of the DPM over the state-of-the-art methods. The code is\nreleased at https://github.com/stone96123/DPM.\n", "date": "2022-07-19", "categories": []}
{"id": "2207.07896", "title": "Cross Vision-RF Gait Re-identification with Low-cost RGB-D Cameras and\n  mmWave Radars", "abstract": "  Human identification is a key requirement for many applications in everyday\nlife, such as personalized services, automatic surveillance, continuous\nauthentication, and contact tracing during pandemics, etc. This work studies\nthe problem of cross-modal human re-identification (ReID), in response to the\nregular human movements across camera-allowed regions (e.g., streets) and\ncamera-restricted regions (e.g., offices) deployed with heterogeneous sensors.\nBy leveraging the emerging low-cost RGB-D cameras and mmWave radars, we propose\nthe first-of-its-kind vision-RF system for cross-modal multi-person ReID at the\nsame time. Firstly, to address the fundamental inter-modality discrepancy, we\npropose a novel signature synthesis algorithm based on the observed specular\nreflection model of a human body. Secondly, an effective cross-modal deep\nmetric learning model is introduced to deal with interference caused by\nunsynchronized data across radars and cameras. Through extensive experiments in\nboth indoor and outdoor environments, we demonstrate that our proposed system\nis able to achieve ~92.5% top-1 accuracy and ~97.5% top-5 accuracy out of 56\nvolunteers. We also show that our proposed system is able to robustly\nreidentify subjects even when multiple subjects are present in the sensors'\nfield of view.\n", "date": "2022-07-16", "categories": []}
{"id": "2207.07311", "title": "Towards Privacy-Preserving Person Re-identification via Person Identify\n  Shift", "abstract": "  Recently privacy concerns of person re-identification (ReID) raise more and\nmore attention and preserving the privacy of the pedestrian images used by ReID\nmethods become essential. De-identification (DeID) methods alleviate privacy\nissues by removing the identity-related of the ReID data. However, most of the\nexisting DeID methods tend to remove all personal identity-related information\nand compromise the usability of de-identified data on the ReID task. In this\npaper, we aim to develop a technique that can achieve a good trade-off between\nprivacy protection and data usability for person ReID. To achieve this, we\npropose a novel de-identification method designed explicitly for person ReID,\nnamed Person Identify Shift (PIS). PIS removes the absolute identity in a\npedestrian image while preserving the identity relationship between image\npairs. By exploiting the interpolation property of variational auto-encoder,\nPIS shifts each pedestrian image from the current identity to another with a\nnew identity, resulting in images still preserving the relative identities.\nExperimental results show that our method has a better trade-off between\nprivacy-preserving and model performance than existing de-identification\nmethods and can defend against human and model attacks for data privacy.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.07340", "title": "DuetFace: Collaborative Privacy-Preserving Face Recognition via Channel\n  Splitting in the Frequency Domain", "abstract": "  With the wide application of face recognition systems, there is rising\nconcern that original face images could be exposed to malicious intents and\nconsequently cause personal privacy breaches. This paper presents DuetFace, a\nnovel privacy-preserving face recognition method that employs collaborative\ninference in the frequency domain. Starting from a counterintuitive discovery\nthat face recognition can achieve surprisingly good performance with only\nvisually indistinguishable high-frequency channels, this method designs a\ncredible split of frequency channels by their cruciality for visualization and\noperates the server-side model on non-crucial channels. However, the model\ndegrades in its attention to facial features due to the missing visual\ninformation. To compensate, the method introduces a plug-in interactive block\nto allow attention transfer from the client-side by producing a feature mask.\nThe mask is further refined by deriving and overlaying a facial region of\ninterest (ROI). Extensive experiments on multiple datasets validate the\neffectiveness of the proposed method in protecting face images from undesired\nvisual inspection, reconstruction, and identification while maintaining high\ntask availability and performance. Results show that the proposed method\nachieves a comparable recognition accuracy and computation cost to the\nunprotected ArcFace and outperforms the state-of-the-art privacy-preserving\nmethods. The source code is available at\nhttps://github.com/Tencent/TFace/tree/master/recognition/tasks/duetface.\n", "date": "2022-07-15", "categories": []}
{"id": "2207.06478", "title": "Analysis of face detection, face landmarking, and face recognition\n  performance with masked face images", "abstract": "  Face recognition has become an essential task in our lives. However, the\ncurrent COVID-19 pandemic has led to the widespread use of face masks. The\neffect of wearing face masks is currently an understudied issue. The aim of\nthis paper is to analyze face detection, face landmarking, and face recognition\nperformance with masked face images. HOG and CNN face detectors are used for\nface detection in combination with 5-point and 68-point face landmark\npredictors and VGG16 face recognition model is used for face recognition on\nmasked and unmasked images. We found that the performance of face detection,\nface landmarking, and face recognition is negatively impacted by face masks\n", "date": "2022-06-03", "categories": []}
{"id": "2207.05933", "title": "Rapid Person Re-Identification via Sub-space Consistency Regularization", "abstract": "  Person Re-Identification (ReID) matches pedestrians across disjoint cameras.\nExisting ReID methods adopting real-value feature descriptors have achieved\nhigh accuracy, but they are low in efficiency due to the slow Euclidean\ndistance computation as well as complex quick-sort algorithms. Recently, some\nworks propose to yield binary encoded person descriptors which instead only\nrequire fast Hamming distance computation and simple counting-sort algorithms.\nHowever, the performances of such binary encoded descriptors, especially with\nshort code (e.g., 32 and 64 bits), are hardly satisfactory given the sparse\nbinary space. To strike a balance between the model accuracy and efficiency, we\npropose a novel Sub-space Consistency Regularization (SCR) algorithm that can\nspeed up the ReID procedure by $0.25$ times than real-value features under the\nsame dimensions whilst maintaining a competitive accuracy, especially under\nshort codes. SCR transforms real-value features vector (e.g., 2048 float32)\nwith short binary codes (e.g., 64 bits) by first dividing real-value features\nvector into $M$ sub-spaces, each with $C$ clustered centroids. Thus the\ndistance between two samples can be expressed as the summation of the\nrespective distance to the centroids, which can be sped up by offline\ncalculation and maintained via a look-up table. On the other side, these\nreal-value centroids help to achieve significantly higher accuracy than using\nbinary code. Lastly, we convert the distance look-up table to be integer and\napply the counting-sort algorithm to speed up the ranking stage.\n  We also propose a novel consistency regularization with an iterative\nframework. Experimental results on Market-1501 and DukeMTMC-reID show promising\nand exciting results. Under short code, our proposed SCR enjoys\nReal-value-level accuracy and Hashing-level speed.\n", "date": "2022-07-13", "categories": []}
{"id": "2207.06035", "title": "Perturbation Inactivation Based Adversarial Defense for Face Recognition", "abstract": "  Deep learning-based face recognition models are vulnerable to adversarial\nattacks. To curb these attacks, most defense methods aim to improve the\nrobustness of recognition models against adversarial perturbations. However,\nthe generalization capacities of these methods are quite limited. In practice,\nthey are still vulnerable to unseen adversarial attacks. Deep learning models\nare fairly robust to general perturbations, such as Gaussian noises. A\nstraightforward approach is to inactivate the adversarial perturbations so that\nthey can be easily handled as general perturbations. In this paper, a\nplug-and-play adversarial defense method, named perturbation inactivation\n(PIN), is proposed to inactivate adversarial perturbations for adversarial\ndefense. We discover that the perturbations in different subspaces have\ndifferent influences on the recognition model. There should be a subspace,\ncalled the immune space, in which the perturbations have fewer adverse impacts\non the recognition model than in other subspaces. Hence, our method estimates\nthe immune space and inactivates the adversarial perturbations by restricting\nthem to this subspace. The proposed method can be generalized to unseen\nadversarial perturbations since it does not rely on a specific kind of\nadversarial attack method. This approach not only outperforms several\nstate-of-the-art adversarial defense methods but also demonstrates a superior\ngeneralization capacity through exhaustive experiments. Moreover, the proposed\nmethod can be successfully applied to four commercial APIs without additional\ntraining, indicating that it can be easily generalized to existing face\nrecognition systems. The source code is available at\nhttps://github.com/RenMin1991/Perturbation-Inactivate\n", "date": "2022-07-13", "categories": []}
{"id": "2207.05316", "title": "Twin identification over viewpoint change: A deep convolutional neural\n  network surpasses humans", "abstract": "  Deep convolutional neural networks (DCNNs) have achieved human-level accuracy\nin face identification (Phillips et al., 2018), though it is unclear how\naccurately they discriminate highly-similar faces. Here, humans and a DCNN\nperformed a challenging face-identity matching task that included identical\ntwins. Participants (N=87) viewed pairs of face images of three types:\nsame-identity, general imposter pairs (different identities from similar\ndemographic groups), and twin imposter pairs (identical twin siblings). The\ntask was to determine whether the pairs showed the same person or different\npeople. Identity comparisons were tested in three viewpoint-disparity\nconditions: frontal to frontal, frontal to 45-degree profile, and frontal to\n90-degree profile. Accuracy for discriminating matched-identity pairs from\ntwin-imposters and general imposters was assessed in each viewpoint-disparity\ncondition. Humans were more accurate for general-imposter pairs than\ntwin-imposter pairs, and accuracy declined with increased viewpoint disparity\nbetween the images in a pair. A DCNN trained for face identification (Ranjan et\nal., 2018) was tested on the same image pairs presented to humans. Machine\nperformance mirrored the pattern of human accuracy, but with performance at or\nabove all humans in all but one condition. Human and machine similarity scores\nwere compared across all image-pair types. This item-level analysis showed that\nhuman and machine similarity ratings correlated significantly in six of nine\nimage-pair types [range r=0.38 to r=0.63], suggesting general accord between\nthe perception of face similarity by humans and the DCNN. These findings also\ncontribute to our understanding of DCNN performance for discriminating\nhigh-resemblance faces, demonstrate that the DCNN performs at a level at or\nabove humans, and suggest a degree of parity between the features used by\nhumans and the DCNN.\n", "date": "2022-07-12", "categories": []}
{"id": "2207.05447", "title": "On the Effects of Image Quality Degradation on Minutiae- and Ridge-Based\n  Automatic Fingerprint Recognition", "abstract": "  The effect of image quality degradation on the verification performance of\nautomatic fingerprint recognition is investigated. We study the performance of\ntwo fingerprint matchers based on minutiae and ridge information under varying\nfingerprint image quality. The ridge-based system is found to be more robust to\nimage quality degradation than the minutiae-based system for a number of\ndifferent image quality criteria.\n", "date": "2022-07-12", "categories": []}
{"id": "2207.05449", "title": "A review of schemes for fingerprint image quality computation", "abstract": "  Fingerprint image quality affects heavily the performance of fingerprint\nrecognition systems. This paper reviews existing approaches for fingerprint\nimage quality computation. We also implement, test and compare a selection of\nthem using the MCYT database including 9000 fingerprint images. Experimental\nresults show that most of the algorithms behave similarly.\n", "date": "2022-07-12", "categories": []}
{"id": "2207.05658", "title": "Dynamic Gradient Reactivation for Backward Compatible Person\n  Re-identification", "abstract": "  We study the backward compatible problem for person re-identification\n(Re-ID), which aims to constrain the features of an updated new model to be\ncomparable with the existing features from the old model in galleries. Most of\nthe existing works adopt distillation-based methods, which focus on pushing new\nfeatures to imitate the distribution of the old ones. However, the\ndistillation-based methods are intrinsically sub-optimal since it forces the\nnew feature space to imitate the inferior old feature space. To address this\nissue, we propose the Ranking-based Backward Compatible Learning (RBCL), which\ndirectly optimizes the ranking metric between new features and old features.\nDifferent from previous methods, RBCL only pushes the new features to find\nbest-ranking positions in the old feature space instead of strictly alignment,\nand is in line with the ultimate goal of backward retrieval. However, the sharp\nsigmoid function used to make the ranking metric differentiable also incurs the\ngradient vanish issue, therefore stems the ranking refinement during the later\nperiod of training. To address this issue, we propose the Dynamic Gradient\nReactivation (DGR), which can reactivate the suppressed gradients by adding\ndynamic computed constant during forward step. To further help targeting the\nbest-ranking positions, we include the Neighbor Context Agents (NCAs) to\napproximate the entire old feature space during training. Unlike previous works\nwhich only test on the in-domain settings, we make the first attempt to\nintroduce the cross-domain settings (including both supervised and\nunsupervised), which are more meaningful and difficult. The experimental\nresults on all five settings show that the proposed RBCL outperforms previous\nstate-of-the-art methods by large margins under all settings.\n", "date": "2022-07-12", "categories": []}
{"id": "2207.04809", "title": "Fingerprint Liveness Detection Based on Quality Measures", "abstract": "  A new fingerprint parameterization for liveness detection based on quality\nmeasures is presented. The novel feature set is used in a complete liveness\ndetection system and tested on the development set of the LivDET competition,\ncomprising over 4,500 real and fake images acquired with three different\noptical sensors. The proposed solution proves to be robust to the multi-sensor\nscenario, and presents an overall rate of 93% of correctly classified samples.\nFurthermore, the liveness detection method presented has the added advantage\nover previously studied techniques of needing just one image from a finger to\ndecide whether it is real or fake.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.04813", "title": "On the vulnerability of fingerprint verification systems to fake\n  fingerprint attacks", "abstract": "  A new method to generate gummy fingers is presented. A medium-size fake\nfingerprint database is described and two different fingerprint verification\nsystems are evaluated on it. Three different scenarios are considered in the\nexperiments, namely: enrollment and test with real fingerprints, enrollment and\ntest with fake fingerprints, and enrollment with real fingerprints and test\nwith fake fingerprints. Results for an optical and a thermal sweeping sensors\nare given. Both systems are shown to be vulnerable to direct attacks.\n", "date": "2022-07-11", "categories": []}
{"id": "2207.02631", "title": "Context Sensing Attention Network for Video-based Person\n  Re-identification", "abstract": "  Video-based person re-identification (ReID) is challenging due to the\npresence of various interferences in video frames. Recent approaches handle\nthis problem using temporal aggregation strategies. In this work, we propose a\nnovel Context Sensing Attention Network (CSA-Net), which improves both the\nframe feature extraction and temporal aggregation steps. First, we introduce\nthe Context Sensing Channel Attention (CSCA) module, which emphasizes responses\nfrom informative channels for each frame. These informative channels are\nidentified with reference not only to each individual frame, but also to the\ncontent of the entire sequence. Therefore, CSCA explores both the individuality\nof each frame and the global context of the sequence. Second, we propose the\nContrastive Feature Aggregation (CFA) module, which predicts frame weights for\ntemporal aggregation. Here, the weight for each frame is determined in a\ncontrastive manner: i.e., not only by the quality of each individual frame, but\nalso by the average quality of the other frames in a sequence. Therefore, it\neffectively promotes the contribution of relatively good frames. Extensive\nexperimental results on four datasets show that CSA-Net consistently achieves\nstate-of-the-art performance.\n", "date": "2022-07-06", "categories": []}
{"id": "2207.02148", "title": "Deep Learning for Finger Vein Recognition: A Brief Survey of Recent\n  Trend", "abstract": "  Finger vein image recognition technology plays an important role in biometric\nrecognition and has been successfully applied in many fields. Because veins are\nburied beneath the skin tissue, finger vein image recognition has an\nunparalleled advantage, which is not easily disturbed by external factors. This\nreview summarizes 46 papers about deep learning for finger vein image\nrecognition from 2017 to 2021. These papers are summarized according to the\ntasks of deep neural networks. Besides, we present the challenges and potential\ndevelopment directions of finger vein image recognition.\n", "date": "2022-07-05", "categories": []}
{"id": "2207.02179", "title": "Activation Template Matching Loss for Explainable Face Recognition", "abstract": "  Can we construct an explainable face recognition network able to learn a\nfacial part-based feature like eyes, nose, mouth and so forth, without any\nmanual annotation or additionalsion datasets? In this paper, we propose a\ngeneric Explainable Channel Loss (ECLoss) to construct an explainable face\nrecognition network. The explainable network trained with ECLoss can easily\nlearn the facial part-based representation on the target convolutional layer,\nwhere an individual channel can detect a certain face part. Our experiments on\ndozens of datasets show that ECLoss achieves superior explainability metrics,\nand at the same time improves the performance of face verification without face\nalignment. In addition, our visualization results also illustrate the\neffectiveness of the proposed ECLoss.\n", "date": "2022-07-05", "categories": []}
{"id": "2207.00587", "title": "Pair-Relationship Modeling for Latent Fingerprint Recognition", "abstract": "  Latent fingerprints are important for identifying criminal suspects. However,\nrecognizing a latent fingerprint in a collection of reference fingerprints\nremains a challenge. Most, if not all, of existing methods would extract\nrepresentation features of each fingerprint independently and then compare the\nsimilarity of these representation features for recognition in a different\nprocess. Without the supervision of similarity for the feature extraction\nprocess, the extracted representation features are hard to optimally reflect\nthe similarity of the two compared fingerprints which is the base for matching\ndecision making. In this paper, we propose a new scheme that can model the\npair-relationship of two fingerprints directly as the similarity feature for\nrecognition. The pair-relationship is modeled by a hybrid deep network which\ncan handle the difficulties of random sizes and corrupted areas of latent\nfingerprints. Experimental results on two databases show that the proposed\nmethod outperforms the state of the art.\n", "date": "2022-07-02", "categories": []}
{"id": "2207.01204", "title": "Adversarial Pairwise Reverse Attention for Camera Performance Imbalance\n  in Person Re-identification: New Dataset and Metrics", "abstract": "  Existing evaluation metrics for Person Re-Identification (Person ReID) models\nfocus on system-wide performance. However, our studies reveal weaknesses due to\nthe uneven data distributions among cameras and different camera properties\nthat expose the ReID system to exploitation. In this work, we raise the\nlong-ignored ReID problem of camera performance imbalance and collect a\nreal-world privacy-aware dataset from 38 cameras to assist the study of the\nimbalance issue. We propose new metrics to quantify camera performance\nimbalance and further propose the Adversarial Pairwise Reverse Attention (APRA)\nModule to guide the model learning the camera invariant feature with a novel\npairwise attention inversion mechanism.\n", "date": "2022-07-04", "categories": []}
{"id": "2207.00161", "title": "Mitigating Presentation Attack using DCGAN and Deep CNN", "abstract": "  Biometric based authentication is currently playing an essential role over\nconventional authentication system; however, the risk of presentation attacks\nsubsequently rising. Our research aims at identifying the areas where\npresentation attack can be prevented even though adequate biometric image\nsamples of users are limited. Our work focusses on generating photorealistic\nsynthetic images from the real image sets by implementing Deep Convolution\nGenerative Adversarial Net (DCGAN). We have implemented the temporal and\nspatial augmentation during the fake image generation. Our work detects the\npresentation attacks on facial and iris images using our deep CNN, inspired by\nVGGNet [1]. We applied the deep neural net techniques on three different\nbiometric image datasets, namely MICHE I [2], VISOB [3], and UBIPr [4]. The\ndatasets, used in this research, contain images that are captured both in\ncontrolled and uncontrolled environment along with different resolutions and\nsizes. We obtained the best test accuracy of 97% on UBI-Pr [4] Iris datasets.\nFor MICHE-I [2] and VISOB [3] datasets, we achieved the test accuracies of 95%\nand 96% respectively.\n", "date": "2022-06-22", "categories": []}
{"id": "2206.14350", "title": "Convolutional Neural Network Based Partial Face Detection", "abstract": "  Due to the massive explanation of artificial intelligence, machine learning\ntechnology is being used in various areas of our day-to-day life. In the world,\nthere are a lot of scenarios where a simple crime can be prevented before it\nmay even happen or find the person responsible for it. A face is one\ndistinctive feature that we have and can differentiate easily among many other\nspecies. But not just different species, it also plays a significant role in\ndetermining someone from the same species as us, humans. Regarding this\ncritical feature, a single problem occurs most often nowadays. When the camera\nis pointed, it cannot detect a person's face, and it becomes a poor image. On\nthe other hand, where there was a robbery and a security camera installed, the\nrobber's identity is almost indistinguishable due to the low-quality camera.\nBut just making an excellent algorithm to work and detecting a face reduces the\ncost of hardware, and it doesn't cost that much to focus on that area. Facial\nrecognition, widget control, and such can be done by detecting the face\ncorrectly. This study aims to create and enhance a machine learning model that\ncorrectly recognizes faces. Total 627 Data have been collected from different\nBangladeshi people's faces on four angels. In this work, CNN, Harr Cascade,\nCascaded CNN, Deep CNN & MTCNN are these five machine learning approaches\nimplemented to get the best accuracy of our dataset. After creating and running\nthe model, Multi-Task Convolutional Neural Network (MTCNN) achieved 96.2% best\nmodel accuracy with training data rather than other machine learning models.\n", "date": "2022-06-29", "categories": []}
{"id": "2206.09362", "title": "Towards Generalizable Person Re-identification with a Bi-stream\n  Generative Model", "abstract": "  Generalizable person re-identification (re-ID) has attracted growing\nattention due to its powerful adaptation capability in the unseen data domain.\nHowever, existing solutions often neglect either crossing cameras (e.g.,\nillumination and resolution differences) or pedestrian misalignments (e.g.,\nviewpoint and pose discrepancies), which easily leads to poor generalization\ncapability when adapted to the new domain. In this paper, we formulate these\ndifficulties as: 1) Camera-Camera (CC) problem, which denotes the various human\nappearance changes caused by different cameras; 2) Camera-Person (CP) problem,\nwhich indicates the pedestrian misalignments caused by the same identity person\nunder different camera viewpoints or changing pose. To solve the above issues,\nwe propose a Bi-stream Generative Model (BGM) to learn the fine-grained\nrepresentations fused with camera-invariant global feature and\npedestrian-aligned local feature, which contains an encoding network and two\nstream decoding sub-networks. Guided by original pedestrian images, one stream\nis employed to learn a camera-invariant global feature for the CC problem via\nfiltering cross-camera interference factors. For the CP problem, another stream\nlearns a pedestrian-aligned local feature for pedestrian alignment using\ninformation-complete densely semantically aligned part maps. Moreover, a\npart-weighted loss function is presented to reduce the influence of missing\nparts on pedestrian alignment. Extensive experiments demonstrate that our\nmethod outperforms the state-of-the-art methods on the large-scale\ngeneralizable re-ID benchmarks, involving domain generalization setting and\ncross-domain setting.\n", "date": "2022-06-19", "categories": []}
{"id": "2206.12590", "title": "RSTAM: An Effective Black-Box Impersonation Attack on Face Recognition\n  using a Mobile and Compact Printer", "abstract": "  Face recognition has achieved considerable progress in recent years thanks to\nthe development of deep neural networks, but it has recently been discovered\nthat deep neural networks are vulnerable to adversarial examples. This means\nthat face recognition models or systems based on deep neural networks are also\nsusceptible to adversarial examples. However, the existing methods of attacking\nface recognition models or systems with adversarial examples can effectively\ncomplete white-box attacks but not black-box impersonation attacks, physical\nattacks, or convenient attacks, particularly on commercial face recognition\nsystems. In this paper, we propose a new method to attack face recognition\nmodels or systems called RSTAM, which enables an effective black-box\nimpersonation attack using an adversarial mask printed by a mobile and compact\nprinter. First, RSTAM enhances the transferability of the adversarial masks\nthrough our proposed random similarity transformation strategy. Furthermore, we\npropose a random meta-optimization strategy for ensembling several pre-trained\nface models to generate more general adversarial masks. Finally, we conduct\nexperiments on the CelebA-HQ, LFW, Makeup Transfer (MT), and CASIA-FaceV5\ndatasets. The performance of the attacks is also evaluated on state-of-the-art\ncommercial face recognition systems: Face++, Baidu, Aliyun, Tencent, and\nMicrosoft. Extensive experiments show that RSTAM can effectively perform\nblack-box impersonation attacks on face recognition models or systems.\n", "date": "2022-06-25", "categories": []}
{"id": "2206.12885", "title": "FingerGAN: A Constrained Fingerprint Generation Scheme for Latent\n  Fingerprint Enhancement", "abstract": "  Latent fingerprint enhancement is an essential pre-processing step for latent\nfingerprint identification. Most latent fingerprint enhancement methods try to\nrestore corrupted gray ridges/valleys. In this paper, we propose a new method\nthat formulates the latent fingerprint enhancement as a constrained fingerprint\ngeneration problem within a generative adversarial network (GAN) framework. We\nname the proposed network as FingerGAN. It can enforce its generated\nfingerprint (i.e, enhanced latent fingerprint) indistinguishable from the\ncorresponding ground-truth instance in terms of the fingerprint skeleton map\nweighted by minutia locations and the orientation field regularized by the\nFOMFE model. Because minutia is the primary feature for fingerprint recognition\nand minutia can be retrieved directly from the fingerprint skeleton map, we\noffer a holistic framework which can perform latent fingerprint enhancement in\nthe context of directly optimizing minutia information. This will help improve\nlatent fingerprint identification performance significantly. Experimental\nresults on two public latent fingerprint databases demonstrate that our method\noutperforms the state of the arts significantly. The codes will be available\nfor non-commercial purposes from\n\\url{https://github.com/HubYZ/LatentEnhancement}.\n", "date": "2022-06-26", "categories": []}
{"id": "2206.11759", "title": "What makes you, you? Analyzing Recognition by Swapping Face Parts", "abstract": "  Deep learning advanced face recognition to an unprecedented accuracy.\nHowever, understanding how local parts of the face affect the overall\nrecognition performance is still mostly unclear. Among others, face swap has\nbeen experimented to this end, but just for the entire face. In this paper, we\npropose to swap facial parts as a way to disentangle the recognition relevance\nof different face parts, like eyes, nose and mouth. In our method, swapping\nparts from a source face to a target one is performed by fitting a 3D prior,\nwhich establishes dense pixels correspondence between parts, while also\nhandling pose differences. Seamless cloning is then used to obtain smooth\ntransitions between the mapped source regions and the shape and skin tone of\nthe target face. We devised an experimental protocol that allowed us to draw\nsome preliminary conclusions when the swapped images are classified by deep\nnetworks, indicating a prominence of the eyes and eyebrows region. Code\navailable at https://github.com/clferrari/FacePartsSwap\n", "date": "2022-06-23", "categories": []}
{"id": "2206.10465", "title": "An Overview of Privacy-enhancing Technologies in Biometric Recognition", "abstract": "  Privacy-enhancing technologies are technologies that implement fundamental\ndata protection principles. With respect to biometric recognition, different\ntypes of privacy-enhancing technologies have been introduced for protecting\nstored biometric data which are generally classified as sensitive. In this\nregard, various taxonomies and conceptual categorizations have been proposed\nand standardization activities have been carried out. However, these efforts\nhave mainly been devoted to certain sub-categories of privacy-enhancing\ntechnologies and therefore lack generalization. This work provides an overview\nof concepts of privacy-enhancing technologies for biometrics in a unified\nframework. Key aspects and differences between existing concepts are\nhighlighted in detail at each processing step. Fundamental properties and\nlimitations of existing approaches are discussed and related to data protection\ntechniques and principles. Moreover, scenarios and methods for the assessment\nof privacy-enhancing technologies for biometrics are presented. This paper is\nmeant as a point of entry to the field of biometric data protection and is\ndirected towards experienced researchers as well as non-experts.\n", "date": "2022-06-21", "categories": []}
{"id": "2206.10520", "title": "SFace: Privacy-friendly and Accurate Face Recognition using Synthetic\n  Data", "abstract": "  Recent deep face recognition models proposed in the literature utilized\nlarge-scale public datasets such as MS-Celeb-1M and VGGFace2 for training very\ndeep neural networks, achieving state-of-the-art performance on mainstream\nbenchmarks. Recently, many of these datasets, e.g., MS-Celeb-1M and VGGFace2,\nare retracted due to credible privacy and ethical concerns. This motivates this\nwork to propose and investigate the feasibility of using a privacy-friendly\nsynthetically generated face dataset to train face recognition models. Towards\nthis end, we utilize a class-conditional generative adversarial network to\ngenerate class-labeled synthetic face images, namely SFace. To address the\nprivacy aspect of using such data to train a face recognition model, we provide\nextensive evaluation experiments on the identity relation between the synthetic\ndataset and the original authentic dataset used to train the generative model.\nOur reported evaluation proved that associating an identity of the authentic\ndataset to one with the same class label in the synthetic dataset is hardly\npossible. We also propose to train face recognition on our privacy-friendly\ndataset, SFace, using three different learning strategies, multi-class\nclassification, label-free knowledge transfer, and combined learning of\nmulti-class classification and knowledge transfer. The reported evaluation\nresults on five authentic face benchmarks demonstrated that the\nprivacy-friendly synthetic dataset has high potential to be used for training\nface recognition models, achieving, for example, a verification accuracy of\n91.87\\% on LFW using multi-class classification and 99.13\\% using the combined\nlearning strategy.\n", "date": "2022-06-21", "categories": []}
{"id": "2206.10526", "title": "QuantFace: Towards Lightweight Face Recognition by Synthetic Data\n  Low-bit Quantization", "abstract": "  Deep learning-based face recognition models follow the common trend in deep\nneural networks by utilizing full-precision floating-point networks with high\ncomputational costs. Deploying such networks in use-cases constrained by\ncomputational requirements is often infeasible due to the large memory required\nby the full-precision model. Previous compact face recognition approaches\nproposed to design special compact architectures and train them from scratch\nusing real training data, which may not be available in a real-world scenario\ndue to privacy concerns. We present in this work the QuantFace solution based\non low-bit precision format model quantization. QuantFace reduces the required\ncomputational cost of the existing face recognition models without the need for\ndesigning a particular architecture or accessing real training data. QuantFace\nintroduces privacy-friendly synthetic face data to the quantization process to\nmitigate potential privacy concerns and issues related to the accessibility to\nreal training data. Through extensive evaluation experiments on seven\nbenchmarks and four network architectures, we demonstrate that QuantFace can\nsuccessfully reduce the model size up to 5x while maintaining, to a large\ndegree, the verification performance of the full-precision model without\naccessing real training datasets.\n", "date": "2022-06-21", "categories": []}
{"id": "2206.08683", "title": "AggNet: Learning to Aggregate Faces for Group Membership Verification", "abstract": "  In some face recognition applications, we are interested to verify whether an\nindividual is a member of a group, without revealing their identity. Some\nexisting methods, propose a mechanism for quantizing precomputed face\ndescriptors into discrete embeddings and aggregating them into one group\nrepresentation. However, this mechanism is only optimized for a given closed\nset of individuals and needs to learn the group representations from scratch\nevery time the groups are changed. In this paper, we propose a deep\narchitecture that jointly learns face descriptors and the aggregation mechanism\nfor better end-to-end performances. The system can be applied to new groups\nwith individuals never seen before and the scheme easily manages new\nmemberships or membership endings. We show through experiments on multiple\nlarge-scale wild-face datasets, that the proposed method leads to higher\nverification performance compared to other baselines.\n", "date": "2022-06-17", "categories": []}
{"id": "2206.07442", "title": "Predicting Gender via Eye Movements", "abstract": "  In this paper, we report the first stable results on gender prediction via\neye movements. We use a dataset with images of faces as stimuli and with a\nlarge number of 370 participants. Stability has two meanings for us: first that\nwe are able to estimate the standard deviation (SD) of a single prediction\nexperiment (it is around 4.1 %); this is achieved by varying the number of\nparticipants. And second, we are able to provide a mean accuracy with a very\nlow standard error (SEM): our accuracy is 65.2 %, and the SEM is 0.80 %; this\nis achieved through many runs of randomly selecting training and test sets for\nthe prediction. Our study shows that two particular classifiers achieve the\nbest accuracies: Random Forests and Logistic Regression. Our results reconfirm\nprevious findings that females are more biased towards the left eyes of the\nstimuli.\n", "date": "2022-06-15", "categories": []}
{"id": "2206.05842", "title": "Efficiency Comparison of AI classification algorithms for Image\n  Detection and Recognition in Real-time", "abstract": "  Face detection and identification is the most difficult and often used task\nin Artificial Intelligence systems. The goal of this study is to present and\ncompare the results of several face detection and recognition algorithms used\nin the system. This system begins with a training image of a human, then\ncontinues on to the test image, identifying the face, comparing it to the\ntrained face, and finally classifying it using OpenCV classifiers. This\nresearch will discuss the most effective and successful tactics used in the\nsystem, which are implemented using Python, OpenCV, and Matplotlib. It may also\nbe used in locations with CCTV, such as public spaces, shopping malls, and ATM\nbooths.\n", "date": "2022-06-12", "categories": []}
{"id": "2206.04783", "title": "ReFace: Real-time Adversarial Attacks on Face Recognition Systems", "abstract": "  Deep neural network based face recognition models have been shown to be\nvulnerable to adversarial examples. However, many of the past attacks require\nthe adversary to solve an input-dependent optimization problem using gradient\ndescent which makes the attack impractical in real-time. These adversarial\nexamples are also tightly coupled to the attacked model and are not as\nsuccessful in transferring to different models. In this work, we propose\nReFace, a real-time, highly-transferable attack on face recognition models\nbased on Adversarial Transformation Networks (ATNs). ATNs model adversarial\nexample generation as a feed-forward neural network. We find that the white-box\nattack success rate of a pure U-Net ATN falls substantially short of\ngradient-based attacks like PGD on large face recognition datasets. We\ntherefore propose a new architecture for ATNs that closes this gap while\nmaintaining a 10000x speedup over PGD. Furthermore, we find that at a given\nperturbation magnitude, our ATN adversarial perturbations are more effective in\ntransferring to new face recognition models than PGD. ReFace attacks can\nsuccessfully deceive commercial face recognition services in a transfer attack\nsetting and reduce face identification accuracy from 82% to 16.4% for AWS\nSearchFaces API and Azure face verification accuracy from 91% to 50.1%.\n", "date": "2022-06-09", "categories": []}
{"id": "2206.04854", "title": "Heterogeneous Face Recognition via Face Synthesis with\n  Identity-Attribute Disentanglement", "abstract": "  Heterogeneous Face Recognition (HFR) aims to match faces across different\ndomains (e.g., visible to near-infrared images), which has been widely applied\nin authentication and forensics scenarios. However, HFR is a challenging\nproblem because of the large cross-domain discrepancy, limited heterogeneous\ndata pairs, and large variation of facial attributes. To address these\nchallenges, we propose a new HFR method from the perspective of heterogeneous\ndata augmentation, named Face Synthesis with Identity-Attribute Disentanglement\n(FSIAD). Firstly, the identity-attribute disentanglement (IAD) decouples face\nimages into identity-related representations and identity-unrelated\nrepresentations (called attributes), and then decreases the correlation between\nidentities and attributes. Secondly, we devise a face synthesis module (FSM) to\ngenerate a large number of images with stochastic combinations of disentangled\nidentities and attributes for enriching the attribute diversity of synthetic\nimages. Both the original images and the synthetic ones are utilized to train\nthe HFR network for tackling the challenges and improving the performance of\nHFR. Extensive experiments on five HFR databases validate that FSIAD obtains\nsuperior performance than previous HFR approaches. Particularly, FSIAD obtains\n4.8% improvement over state of the art in terms of VR@FAR=0.01% on LAMP-HQ, the\nlargest HFR database so far.\n", "date": "2022-06-10", "categories": []}
{"id": "2206.04867", "title": "The Gender Gap in Face Recognition Accuracy Is a Hairy Problem", "abstract": "  It is broadly accepted that there is a \"gender gap\" in face recognition\naccuracy, with females having higher false match and false non-match rates.\nHowever, relatively little is known about the cause(s) of this gender gap. Even\nthe recent NIST report on demographic effects lists \"analyze cause and effect\"\nunder \"what we did not do\". We first demonstrate that female and male\nhairstyles have important differences that impact face recognition accuracy. In\nparticular, compared to females, male facial hair contributes to creating a\ngreater average difference in appearance between different male faces. We then\ndemonstrate that when the data used to estimate recognition accuracy is\nbalanced across gender for how hairstyles occlude the face, the initially\nobserved gender gap in accuracy largely disappears. We show this result for two\ndifferent matchers, and analyzing images of Caucasians and of\nAfrican-Americans. These results suggest that future research on demographic\nvariation in accuracy should include a check for balanced quality of the test\ndata as part of the problem formulation. To promote reproducible research,\nmatchers, attribute classifiers, and datasets used in this research are/will be\npublicly available.\n", "date": "2022-06-10", "categories": []}
{"id": "2206.04295", "title": "Reconstruct Face from Features Using GAN Generator as a Distribution\n  Constraint", "abstract": "  Face recognition based on the deep convolutional neural networks (CNN) shows\nsuperior accuracy performance attributed to the high discriminative features\nextracted. Yet, the security and privacy of the extracted features from deep\nlearning models (deep features) have been often overlooked. This paper proposes\nthe reconstruction of face images from deep features without accessing the CNN\nnetwork configurations as a constrained optimization problem. Such optimization\nminimizes the distance between the features extracted from the original face\nimage and the reconstructed face image. Instead of directly solving the\noptimization problem in the image space, we innovatively reformulate the\nproblem by looking for a latent vector of a GAN generator, then use it to\ngenerate the face image. The GAN generator serves as a dual role in this novel\nframework, i.e., face distribution constraint of the optimization goal and a\nface generator. On top of the novel optimization task, we also propose an\nattack pipeline to impersonate the target user based on the generated face\nimage. Our results show that the generated face images can achieve a\nstate-of-the-art successful attack rate of 98.0\\% on LFW under type-I attack @\nFAR of 0.1\\%. Our work sheds light on the biometric deployment to meet the\nprivacy-preserving and security policies.\n", "date": "2022-06-09", "categories": []}
{"id": "2206.02325", "title": "Evaluation-oriented Knowledge Distillation for Deep Face Recognition", "abstract": "  Knowledge distillation (KD) is a widely-used technique that utilizes large\nnetworks to improve the performance of compact models. Previous KD approaches\nusually aim to guide the student to mimic the teacher's behavior completely in\nthe representation space. However, such one-to-one corresponding constraints\nmay lead to inflexible knowledge transfer from the teacher to the student,\nespecially those with low model capacities. Inspired by the ultimate goal of KD\nmethods, we propose a novel Evaluation oriented KD method (EKD) for deep face\nrecognition to directly reduce the performance gap between the teacher and\nstudent models during training. Specifically, we adopt the commonly used\nevaluation metrics in face recognition, i.e., False Positive Rate (FPR) and\nTrue Positive Rate (TPR) as the performance indicator. According to the\nevaluation protocol, the critical pair relations that cause the TPR and FPR\ndifference between the teacher and student models are selected. Then, the\ncritical relations in the student are constrained to approximate the\ncorresponding ones in the teacher by a novel rank-based loss function, giving\nmore flexibility to the student with low capacity. Extensive experimental\nresults on popular benchmarks demonstrate the superiority of our EKD over\nstate-of-the-art competitors.\n", "date": "2022-06-06", "categories": []}
{"id": "2206.01441", "title": "Exploring Transformers for Behavioural Biometrics: A Case Study in Gait\n  Recognition", "abstract": "  Biometrics on mobile devices has attracted a lot of attention in recent years\nas it is considered a user-friendly authentication method. This interest has\nalso been motivated by the success of Deep Learning (DL). Architectures based\non Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs)\nhave been established to be convenient for the task, improving the performance\nand robustness in comparison to traditional machine learning techniques.\nHowever, some aspects must still be revisited and improved. To the best of our\nknowledge, this is the first article that intends to explore and propose novel\ngait biometric recognition systems based on Transformers, which currently\nobtain state-of-the-art performance in many applications. Several\nstate-of-the-art architectures (Vanilla, Informer, Autoformer, Block-Recurrent\nTransformer, and THAT) are considered in the experimental framework. In\naddition, new configurations of the Transformers are proposed to further\nincrease the performance. Experiments are carried out using the two popular\npublic databases whuGAIT and OU-ISIR. The results achieved prove the high\nability of the proposed Transformer, outperforming state-of-the-art CNN and RNN\narchitectures.\n", "date": "2022-06-03", "categories": []}
{"id": "2206.01475", "title": "Functional Connectivity Methods for EEG-based Biometrics on a Large,\n  Heterogeneous Dataset", "abstract": "  This study examines the utility of functional connectivity (FC) and\ngraph-based (GB) measures with a support vector machine classifier for use in\nelectroencephalogram (EEG) based biometrics. Although FC-based features have\nbeen used in biometric applications, studies assessing the identification\nalgorithms on heterogeneous and large datasets are scarce. This work\ninvestigates the performance of FC and GB metrics on a dataset of 184 subjects\nformed by pooling three datasets recorded under different protocols and\nacquisition systems. The results demonstrate the higher discriminatory power of\nFC than GB metrics. The identification accuracy increases with higher frequency\nEEG bands, indicating the enhanced uniqueness of the neural signatures in beta\nand gamma bands. Using all the 56 EEG channels common to the three databases,\nthe best identification accuracy of 97.4% is obtained using phase-locking value\n(PLV) based measures extracted from the gamma frequency band. Further, we\ninvestigate the effect of the length of the analysis epoch to determine the\ndata acquisition time required to obtain satisfactory identification accuracy.\nWhen the number of channels is reduced to 21 from 56, there is a marginal\nreduction of 2.4% only in the identification accuracy using PLV features in the\ngamma band. Additional experiments have been conducted to study the effect of\nthe cognitive state of the subject and mismatched train/test conditions on the\nperformance of the system.\n", "date": "2022-06-03", "categories": []}
{"id": "2206.01061", "title": "FV-UPatches: Enhancing Universality in Finger Vein Recognition", "abstract": "  Many deep learning-based models have been introduced in finger vein\nrecognition in recent years. These solutions, however, suffer from data\ndependency and are difficult to achieve model generalization. To address this\nproblem, we are inspired by the idea of domain adaptation and propose a\nuniversal learning-based framework, which achieves generalization while\ntraining with limited data. To reduce differences between data distributions, a\ncompressed U-Net is introduced as a domain mapper to map the raw region of\ninterest image onto a target domain. The concentrated target domain is a\nunified feature space for the subsequent matching, in which a local descriptor\nmodel SOSNet is employed to embed patches into descriptors measuring the\nsimilarity of matching pairs. In the proposed framework, the domain mapper is\nan approximation to a specific extraction function thus the training is only a\none-time effort with limited data. Moreover, the local descriptor model can be\ntrained to be representative enough based on a public dataset of\nnon-finger-vein images. The whole pipeline enables the framework to be well\ngeneralized, making it possible to enhance universality and helps to reduce\ncosts of data collection, tuning and retraining. The comparable experimental\nresults to state-of-the-art (SOTA) performance in five public datasets prove\nthe effectiveness of the proposed framework. Furthermore, the framework shows\napplication potential in other vein-based biometric recognition as well.\n", "date": "2022-06-02", "categories": []}
{"id": "2206.00580", "title": "Dog nose print matching with dual global descriptor based on Contrastive\n  Learning", "abstract": "  Recent studies in biometric-based identification tasks have shown that deep\nlearning methods can achieve better performance. These methods generally\nextract the global features as descriptor to represent the original image.\nNonetheless, it does not perform well for biometric identification under\nfine-grained tasks. The main reason is that the single image descriptor\ncontains insufficient information to represent image. In this paper, we present\na dual global descriptor model, which combines multiple global descriptors to\nexploit multi level image features. Moreover, we utilize a contrastive loss to\nenlarge the distance between image representations of confusing classes. The\nproposed framework achieves the top2 on the CVPR2022 Biometrics Workshop Pet\nBiometric Challenge. The source code and trained models are publicly available\nat: https://github.com/flyingsheepbin/pet-biometrics\n", "date": "2022-06-01", "categories": []}
{"id": "2606.21900", "title": "Continuous Behavioral Authentication via Multi-Expert BERT Log Analysis for Secure Data Sharing", "abstract": "Continuous authentication for mobile and zero-trust systems requires nonintrusive evidence confirming the enrolled user-device context remains valid after initial login. This paper presents a BERT log analysis framework for continuous behavioral authentication using Android system logs. The proposed pipeline parses logcat streams into event templates and dynamic variables, pre-trains a domain-adapted BERT encoder on Android log syntax, and fine-tunes three expert models for network/device identity, battery-transition timing, and Wi-Fi topology. The expert confidence scores are fused through a log-space transformation and a 5-nearest-neighbor distance classifier to generate a normality score that is provided to a Policy Decision Point (PDP) for risk-aware access control. Experiments on normal traces, controlled anomaly injections, and benign Wi-Fi perturbations indicate that multi-expert BERT log analysis can detect semantic, battery-timing, and topology deviations in the evaluated setting while maintaining sub-1% False Positive Rate (FPR). The results suggest that Android system logs are a practical sensor-free signal for continuous authentication and user-device context assurance.", "date": "2026-06-20", "categories": []}
{"id": "2606.22029", "title": "Topological summaries of fingerprint ridge patterns carry identity information", "abstract": "Fingerprints are the most widely deployed biometric. Verifying whether two impressions come from the same finger typically relies on minutiae, small landmarks such as skin ridge endings and bifurcations. These landmarks are extracted through a multi-stage pipeline of image enhancement, skeletonization, minutiae detection, and alignment. We investigate an alternative: using topological data analysis to represent the full pattern of skin ridges and valleys directly, bypassing minutiae detection and the downstream matching pipeline. We apply persistent homology, a topological tool that tracks how loops in the ridge pattern form and fill in across spatial scales, producing multi-scale summaries of ridge geometry. We develop and compare a range of verification methods on a standard benchmark dataset, FVC2000 DB1. Even the simplest topological summaries, with no trained parameters, substantially outperform geometry-only baselines. A trained method achieves an AUC of 0.91, while an optimal-transport method excels at the strictest false-accept thresholds, suggesting they capture different aspects of the ridge pattern. Fusing these two approaches yields the best performance at every low false-accept threshold we examine. Our results establish that these topological summaries capture substantial fingerprint identity information, far more effective for verification than raw pixel-level geometry. Because the entire pipeline is openly specified, it offers a transparent complement to minutiae-based systems, and we provide a modular framework for constructing, evaluating, and combining topological verification methods.", "date": "2026-06-20", "categories": []}
{"id": "2606.22986", "title": "Subject-Level Unknown-Identity Identification from Leap Motion Controller 2 Hand Landmarks", "abstract": "This work studies subject recognition from Leap Motion Controller 2 (LMC2) hand landmark data under a subject-level unknown-identity identification protocol on the Multi View Leap2 Hand Pose (ML2HP) dataset. Using only the landmark modality, we retain the original geometric representation and enrich it with fingertip-to-palm distances and palm-normalized inter-finger angular descriptors. Evaluation is performed under a Leave-One-Subject-Out (LOSO) protocol in which, for each outer fold, one subject is excluded from the enrolled set and treated as unknown at test time. To avoid tuning on the true outer unknown subject, the unknown-rejection threshold is selected in an inner validation step by temporarily withholding one enrolled subject from the inner gallery and using it only for threshold estimation. We compare a tree ensemble baseline with two neural alternatives: a learned embedding baseline based on centroid matching and cosine-similarity-based rejection, and an MLP+OpenMax model, which represents a more established open-set recognition approach. Under this evaluation setup, Extra Trees remains the strongest overall method, indicating that the main challenge on this benchmark is not enrolled-subject discrimination alone, but robust score separation between known and unknown probes. The results support the feasibility of compact, interpretable landmark-based descriptors for contactless hand-based unknown-subject rejection and identification on a small-cohort dataset.", "date": "2026-06-22", "categories": []}
{"id": "2606.23230", "title": "Privacy-Preserving Person Re-Identification from Temporal Sequences with Transformer and Hungarian Optimization", "abstract": "Person re-identification (Re-ID) is a crucial task in surveillance and human behavior analysis, often used in public spaces such as transport hubs. Traditional RGB-based Re-ID methods raise privacy concerns and are highly sensitive to lighting variations and occlusion. In this paper, we propose a novel Re-ID approach that leverages depth images, which inherently obscures facial and other identifiable features, making it a privacy-preserving solution. Our method addresses the association problem between multiple views of individuals by applying the Hungarian algorithm, optimizing the matching process through minimization of the global cost across the distance matrix. We further enhance the approach by introducing temporal sequences of frames as input to a Transformer encoder architecture, which exploits both RGB and depth modalities. This architecture captures dynamic movement patterns, improving feature extraction and re-identification accuracy. Additionally, we employ batch hard triplet loss to enhance discriminative feature learning by focusing on the hardest samples. We evaluate both depth-only and RGB-D models on several top-view datasets, including TVPR2, GODPR, and BIWI RGBD-ID. Our results demonstrate that depth-only re-identification can achieve competitive performance compared to state-of-the-art methods, as measured by standard metrics such as Cumulative Matching Characteristics (CMC) and Mean Average Precision (mAP), while prioritizing privacy preservation.", "date": "2026-06-22", "categories": []}
{"id": "2606.24586", "title": "EERLoss: A Novel Loss Function for Training Deep Biometric Models. A Case Study in Keystroke Dynamics", "abstract": "Deep learning approaches to biometric verification are commonly trained by optimizing indirect objectives, creating a misalignment between the optimization process and the primary evaluation metric, typically the Equal Error Rate (EER). This paper introduces EERLoss: a subdifferentiable, arbitrarily accurate approximation to EER for training deep biometric models. Furthermore, this framework has the potential to be adapted to optimize any specific operating point on the DET curve, enhancing its generalizability. To validate this approach, EERLoss is evaluated on a particularly demanding behavioral biometric modality: keystroke dynamics verification. This task is characterized by its high intra-class and low inter-class variability. Experiments are conducted on the large-scale KVC-onGoing benchmark, incorporating data from over 185,000 subjects across different scenarios. A comprehensive ablation study initially demonstrates the superiority of EERLoss in comparison to existing state-of-the-art loss functions. It also converges substantially faster compared to other losses, reducing the overall training cost. Additionally, a comparison is made between the proposed loss and the KVC-winning architecture by re-training it with EERLoss, demonstrating that the proposed approach significantly outperforms the original SoTA, achieving a relative EER reduction of up to approx. 30\\%. This improvement on a challenging, large-scale benchmark validates the effectiveness of EERLoss as a task-aligned training objective specifically suited for high-variance biometric traits.", "date": "2026-06-23", "categories": []}
{"id": "2606.30321", "title": "Optimizing Image Preparation and Compression for Face Recognition within 1024 Bytes", "abstract": "ICAO-compliant machine readable travel documents enable automated biometric face verification. The biometric reference is stored on an RFID chip included in form of a JPEG or JPEG 2000 compressed facial image. In contrast, temporary travel documents lack of machine readability, which excludes the owner from such automated processes. This disadvantage could be solved by equipping such documents with 2D barcodes. This technology offers a resource-saving alternative to expensive RFID chips, while still offering machine readability and fast issuing processes. However, this solution introduces the challenge of storing the face images at significantly smaller storage capacities, creating the need for reducing the file size of the included facial image to a maximum of 1024 bytes. This study examines preprocessing steps and compression configurations, using JPEG, JPEG 2000, JPEG XL, JPEG AI, HEIF, AVIF, and WebP for image compression to this target size, while still preserving as much face recognition performance as possible. While the reference sample must always comply with ICAO specifications, the individual samples may or may not meet these requirements, depending on the application. This work optimizes compression steps for both of these prerequisites. It is shown that the recently standardised JPEG AI, when using optimized settings, provides the best face recognition performance, in particular when the comparison includes only images with high face image quality. AVIF and WebP also provide good results. The losses caused by the strong lossy compression are comparatively small. For the comparison of ICAO-compliant face images only, converting the images to grayscale proves to be a helpful preprocessing step, whereas for comparisons involving less suitable samples, preserving color is preferable. In addition, smoothing and resizing the images beforehand also turns out to be beneficial.", "date": "2026-06-29", "categories": []}
{"id": "2606.31664", "title": "Sparsity-Inducing Divergence Losses for Biometric Verification", "abstract": "Performance in face and speaker verification is largely driven by margin-penalty softmax losses such as CosFace and ArcFace. Recently introduced $\\alpha$-divergence loss functions offer a compelling alternative, particularly due to their ability to induce sparse solutions (when $\\alpha>1$). However, standard geometric margins are designed for the softmax function and do not naturally extend to this generalized probabilistic framework. In this paper we propose Q-Margin, a novel $\\alpha$-divergence loss that introduces a principled probabilistic margin. Unlike conventional methods that apply geometric penalties to the logits (unnormalized log-likelihoods), Q-Margin encodes the margin penalty directly into the reference measure (prior probabilities). This formulation naturally encourages discriminative embeddings while preserving the beneficial sparsity properties of the $\\alpha$-divergence. We demonstrate that Q-Margin achieves competitive or superior performance on the challenging IJB-B and IJB-C face verification benchmarks and similarly strong results in speaker verification on VoxCeleb. Crucially, against ArcFace and CosFace baselines trained under an identical recipe, Q-Margin consistently improves at low False Acceptance Rates (FARs), a capability critical for practical high-security applications. Finally, the extreme sparsity of the Q-Margin posteriors enables exact and memory-efficient training, offering a scalable solution for datasets with millions of identities.", "date": "2026-06-30", "categories": []}
{"id": "2606.32040", "title": "FaceMoE: Mixture of Experts for Low-Resolution Face Recognition", "abstract": "Low-resolution face recognition (LR-FR) remains a challenging task due to poor feature extraction and aggregation, as probe images often contain limited identity information resulting from extreme degradations such as blur, occlusion, and low contrast. Additionally, the domain gap between high-resolution (HR) gallery images and low-resolution (LR) probe images poses a significant challenge. A single feature encoder struggles to generalize effectively across both domains when fine-tuned on an LR dataset, and this issue is further magnified by catastrophic forgetting. To address these challenges, we propose FaceMoE, an effective adaptation of Mixture of Experts (MoE) transfomer architecture for low-resolution face-recognition . Specifically, we introduce multiple specialized feed-forward network (FFN) experts and incorporate a top-k router, which dynamically assigns tokens to appropriate experts. This design emergently promotes specialization across experts for different semantic regions of the face, which enables FaceMoE to perform resolution-aware feature extraction. Moreover, the top-k router facilitates sparse expert activation, enabling the model to preserve pretrained knowledge when finetuned on a LR dataset, while increasing model capacity without proportional computational overhead. FaceMoE is trained with a combined face recognition loss, router z-loss, and load balancing loss to ensure expert specialization and stable training. To the best of our knowledge, this is the first work leveraging MoE for LR-FR. Extensive experiments across eleven datasets, spanning HR, mixed-quality, and LR benchmarks, demonstrate that FaceMoE significantly outperforms state-of-the-art methods. Code: https://github.com/Kartik-3004/FaceMoE", "date": "2026-06-30", "categories": []}