🔍 Face Matcher
Upload two photos → real on-device face detection → visual similarity score 0–100. Nothing is uploaded anywhere.
⚠️ Demo-grade visual similarity — NOT biometric identification. This tool compares how alike two face regions look; it cannot prove identity. Never use it for security, KYC, access control, or legal decisions.
Loading face-detection engine…
How it works (honest version)
- Face detection: the YuNet neural network (232 KB ONNX model, bundled in this page) finds faces via ONNX Runtime WebAssembly — entirely in your browser.
- Crop: the largest face from each photo is cropped with a small margin and resized to 128×128 grayscale.
- Score: 60% grayscale-histogram correlation + 40% normalized template match, scaled 0–100. Same math as our Python original.
- If no face is found in one or both photos, the tool falls back to whole-image comparison and says so — it never pretends a face was detected.
- Works best on clear, frontal, well-lit faces. “Uncertain” is a real answer — the tool says so instead of guessing.