Spaces:
Running
Running
File size: 8,704 Bytes
6a59fe7 fcea3b7 6a59fe7 fcea3b7 6a59fe7 ffbf0eb 6a59fe7 ffbf0eb 6a59fe7 b39683a | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 | """
Batch Investigation Mode.
Processes multiple images in a single request, returning a unified
forensic report with cross-image analysis including:
- Per-image full forensic report
- Batch-level statistics (mean AI probability, class distribution)
- Clustering: which images are most similar (hash-based)
- Duplicate detection via perceptual hash comparison
- Highest-risk images ranked by AI probability
- Cross-image provenance consistency check
Limits:
MAX_BATCH_SIZE = 10 images per request
MAX_IMAGE_SIZE = 5MB per image in batch mode
Processing is sequential to avoid OOM on GPU
"""
from backend.core.logger import setup_logger
from typing import Dict, Any, List
logger = setup_logger(__name__)
MAX_BATCH_SIZE = 10
MAX_IMAGE_BYTES = 5 * 1024 * 1024 # 5MB per image in batch
def _phash_distance(h1: str, h2: str) -> int:
"""Hamming distance between two perceptual hash hex strings."""
try:
i1 = int(h1, 16)
i2 = int(h2, 16)
xor = i1 ^ i2
return bin(xor).count("1")
except Exception:
return 64 # Max distance on error
def _find_duplicates(reports: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
"""
Identify duplicate or near-duplicate images using perceptual hash.
Threshold: Hamming distance <= 8 (out of 64 bits).
"""
pairs = []
for i in range(len(reports)):
for j in range(i + 1, len(reports)):
h1 = reports[i].get("hashes", {}).get("perceptual_hash", "")
h2 = reports[j].get("hashes", {}).get("perceptual_hash", "")
if h1 and h2:
dist = _phash_distance(h1, h2)
if dist <= 8:
pairs.append({
"image_a": reports[i]["file_info"]["filename"],
"image_b": reports[j]["file_info"]["filename"],
"phash_distance": dist,
"similarity": "identical" if dist == 0 else "near_duplicate",
})
return pairs
def _rank_by_risk(reports: List[Dict[str, Any]]) -> List[Dict[str, Any]]:
"""Return images ranked by AI probability descending."""
ranked = []
for r in reports:
ranked.append({
"filename": r["file_info"]["filename"],
"ai_probability": r["summary"]["ai_probability"],
"classification": r["summary"]["ai_classification"],
"evidence_id": r["evidence_id"],
"predicted_generator": r.get("generator_attribution", {}).get("predicted_generator", "unknown"),
"c2pa_status": r.get("c2pa_provenance", {}).get("provenance_status", "none"),
})
return sorted(ranked, key=lambda x: x["ai_probability"], reverse=True)
def _batch_statistics(reports: List[Dict[str, Any]]) -> Dict[str, Any]:
"""Compute aggregate statistics across all images in the batch."""
probs = [r["summary"]["ai_probability"] for r in reports]
classes = [r["summary"]["ai_classification"] for r in reports]
generators = [
r.get("generator_attribution", {}).get("predicted_generator", "unknown")
for r in reports
]
c2pa_statuses = [
r.get("c2pa_provenance", {}).get("provenance_status", "none")
for r in reports
]
class_counts: Dict[str, int] = {}
for c in classes:
class_counts[c] = class_counts.get(c, 0) + 1
generator_counts: Dict[str, int] = {}
for g in generators:
generator_counts[g] = generator_counts.get(g, 0) + 1
# Classification strings: "likely_ai_generated", "possibly_ai_generated",
# "likely_authentic", "possibly_authentic", "uncertain"
# Use exclusive categories to prevent double-counting.
ai_count = sum(1 for cls in classes if "ai_generated" in cls)
real_count = sum(1 for cls in classes if "authentic" in cls and "ai_generated" not in cls)
uncertain_count = len(classes) - ai_count - real_count
return {
"total_images": len(reports),
"ai_detected_count": ai_count,
"authentic_count": real_count,
"uncertain_count": uncertain_count,
"mean_ai_probability": round(sum(probs) / len(probs), 4) if probs else 0.0,
"max_ai_probability": round(max(probs), 4) if probs else 0.0,
"min_ai_probability": round(min(probs), 4) if probs else 0.0,
"classification_breakdown": class_counts,
"generator_breakdown": generator_counts,
"c2pa_has_credentials": sum(1 for s in c2pa_statuses if s != "none"),
"batch_verdict": (
"high_risk" if ai_count / max(len(reports), 1) >= 0.6 else
"mixed" if ai_count > 0 else
"likely_authentic"
),
}
def _provenance_consistency(reports: List[Dict[str, Any]]) -> Dict[str, Any]:
"""
Check if C2PA provenance is consistent across images in the batch.
A collection where some images have credentials and others do not
may indicate selective credential stripping.
"""
statuses = [
r.get("c2pa_provenance", {}).get("provenance_status", "none")
for r in reports
]
has_credentials = sum(1 for s in statuses if s != "none")
lacks_credentials = sum(1 for s in statuses if s == "none")
if has_credentials == 0:
consistency = "consistent_no_credentials"
note = "No images in batch have C2PA credentials. This is normal for most images."
elif lacks_credentials == 0:
consistency = "consistent_all_credentialed"
note = "All images have C2PA credentials. Strong provenance signal."
else:
consistency = "inconsistent"
note = (
f"{has_credentials} of {len(reports)} images have C2PA credentials. "
"Mixed provenance may indicate selective credential stripping."
)
return {
"consistency": consistency,
"images_with_credentials": has_credentials,
"images_without_credentials": lacks_credentials,
"note": note,
}
def process_batch(
images: List[Dict[str, Any]],
) -> Dict[str, Any]:
"""
Process a batch of images through the full forensic pipeline.
Args:
images: List of dicts with keys:
filename - str
data - bytes (image bytes)
Returns:
Batch forensic report with per-image results and aggregate analysis
"""
from backend.services.image_forensics import ImageForensics
if len(images) > MAX_BATCH_SIZE:
return {
"error": f"Batch too large. Max {MAX_BATCH_SIZE} images per request.",
"maximum": MAX_BATCH_SIZE,
"received": len(images),
}
results = []
errors = []
reports = []
for item in images:
filename = item.get("filename", "unknown")
data = item.get("data", b"")
if len(data) == 0:
errors.append({"filename": filename, "error": "Empty file"})
continue
if len(data) > MAX_IMAGE_BYTES:
errors.append({
"filename": filename,
"error": f"File too large for batch mode (max {MAX_IMAGE_BYTES // (1024*1024)}MB)"
})
continue
try:
forensics = ImageForensics(data, filename)
report = forensics.generate_forensic_report()
reports.append(report)
results.append({
"filename": filename,
"status": "success",
"evidence_id": report["evidence_id"],
"report": report,
})
logger.info(
f"Batch: processed {filename} "
f"(ai={report['summary']['ai_probability']:.3f})"
)
except Exception as e:
logger.warning(f"Batch: failed {filename}: {e}")
errors.append({"filename": filename, "error": str(e)})
if not reports:
return {
"status": "failed",
"processed": 0,
"errors": errors,
"error": "No images could be processed",
}
return {
"status": "complete",
"processed": len(reports),
"failed": len(errors),
"errors": errors,
"statistics": _batch_statistics(reports),
"risk_ranking": _rank_by_risk(reports),
"duplicate_pairs": _find_duplicates(reports),
"provenance_consistency": _provenance_consistency(reports),
"individual_reports": results,
}
|