❓ Help Need Help: Open-Source / Free Signature Detector para sa Document Images

[fairy19]

Title: Need Help: Open-Source / Free Signature Detector para sa Document Images

Hi mga ka-PHC!

Baka may nakapag-implement na dito ng signature detection sa document images. Need ko sana ng suggestions/recommendations.

Ang goal namin ay automatic na ma-detect kung may actual handwritten signature sa specific signature fields ng isang document/POD (Proof of Delivery).

Example fields:

• Received By
• Witnessed By
• Other required signature areas

Ang challenge namin is hindi lang basta detection ng signature. May cases kasi na ang signature ay naka-overlap o tumatama mismo sa printed label/text, kaya ideally dapat ma-detect pa rin siya.

Current flow namin:
Document Image → Locate Signature Area → Detect kung Signed or Missing → Return result sa API

Target sana:

• Open-source / free
• No ρáíd API/token per request
• Puwedeng i-self-host
• Lightweight sana dahil server namin currently CPU only / walang dedicated GPU
• Preferably pre-trained na at minimal/no custom training
• Kayang mag-detect ng multiple signatures sa isang document
• Mas okay kung kaya pati signature na overlapping sa printed text/label
• Puwedeng i-integrate as REST API/service

Currently nag-eexplore kami ng object detection/VLM approaches tulad ng RF-DETR/DETR, YOLOS, MiniCPM/Qwen, OpenCV + OCR, etc., pero naghahanap pa kami ng pinaka-practical at reliable na approach for production.

Hindi namin kailangan i-verify kung kaninong signature iyon. Ang kailangan lang talaga ay:

“May valid handwritten signature ba sa required field, or wala?”

Kung may naka-experience na gumawa ng ganitong system, anong model/library/approach ang mare-recommend ninyo?

Okay lang kahit Hugging Face model, GitHub project, Python library, or combination ng OCR + OpenCV + object detection.

Maraming salamat mga ka-PHC!