1 Identity document
Drop an ID image here or click to browse
No document loaded
2 Live selfie
Drop a selfie here or click to browse
No selfie loaded
3 Run verification
- Load ONNX models
- Document OCR & field extraction
- Face detection & matching
- Passive liveness
- Tamper forensics (ELA · copy-move · metadata)
How it works
VeriLens chains four independent verification signals, all computed client-side: UltraFace RFB-320 locates faces, MobileFaceNet (w600k) produces 512-d embeddings compared by cosine similarity, MiniFASNetV2 estimates the probability the selfie is a live capture rather than a photo-of-a-photo or screen replay, and a forensics suite (Error Level Analysis, copy-move block matching, JPEG metadata inspection) scores document tampering. Tesseract 5 LSTM reads the document and regex parsers extract structured fields. A weighted trust score aggregates everything into a final verdict with a downloadable JSON audit trail.