Fetching metadata from the HF Docker repository...
Feb 16 14:30 feat: add perspective correction for document images Feb 14 13:39 feat: OCR v3 — adaptive preprocessing, GB dictionary, confidence review buba19 +1: Claude Opus 4.6 Major handwriting OCR improvements:
- Background normalization for yellowed/aged documents (Gaussian subtraction)
- Adaptive denoising based on measured noise level (h=6..20)
- Histogram stretch fallback when contrast remains too low after CLAHE
- Adaptive EasyOCR parameters per image metrics (contrast, noise, ink density)
- Line-clustered reading order for legal documents with dense paragraphs
- Confidence-based review flagging (needs_review per block)
- Expanded Guinea-Bissau legal/administrative dictionary (~120 terms)
- Server-side post-correction with Levenshtein + alphanumeric confusion fix
- Added French language support (pt, fr, en)
- document_type parameter (auto/handwritten/printed) with type-specific pipelines
- image_metrics in response (brightness, contrast, noise, ink_density)
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> Feb 13 22:07 fix: remove paddleocr from requirements, keep as optional 22:04 fix: add app_port to HuggingFace Space config 21:32 Merge branch 'main' of https://huggingface.co/spaces/buba19/leigb-ocr 21:18 initial commit 21:12 feat: LeiGB OCR microservice for HuggingFace Space buba19 +1: Claude Opus 4.6 FastAPI service with EasyOCR, PaddleOCR, and Tesseract engines.
POST /ocr endpoint accepts image + lang + engine, returns text + confidence + blocks.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com> 
