Eye‑D is a new kind of biometric system. Instead of your fingerprint or your whole face, it recognises you by the unique way you squeeze your eyes shut.
No more taking off gloves in winter. No more removing your niqab or scarf. Just squeeze – and you're in.
When you forcefully squeeze your eyes, your eyelid doesn't just close. It follows a personal sequence:
- How fast it accelerates and brakes (your kinetic signature)
- The 3D curve your inner eye corner draws
- The order in which tiny wrinkles appear around your eye
- Which part of your eyelid closes first (inner, middle, or outer)
Even identical twins have different squeeze patterns – because the sequence depends on your unique muscle wiring, skin elasticity, and bone structure.
Eye‑D captures these movements with a normal camera (webcam or phone), turns them into a mathematical fingerprint, and uses an adaptive AI that gets better every time you log in.
| Problem | Eye‑D Solution |
|---|---|
| ❄️ Cold climates – gloves block fingerprints, scarves block Face ID | Works with face partially covered |
| 🧕 Niqab / hijab – traditional face recognition impossible | Only needs the eye region |
| 📱 Smart glasses & AR/VR – no fingerprint sensor, no front camera | Tiny inward camera is enough |
| 🔁 Aging – fingerprints wear, faces change | Model updates with each successful login |
| 🎭 Deepfake / replay attacks | Unique dynamic sequence is hard to fake |
- You squeeze your eye in front of any camera.
- Eye‑D tracks five feature families in real time:
- Kinetic magnitude (work, power, braking)
- Main sequence (amplitude vs. peak velocity)
- 3D path of the medial canthus (inner eye corner)
- Wrinkle appearance order
- Wave‑like closure delays
- All measurements are normalised (works near or far) and encrypted.
- A Gaussian Mixture Model (GMM) decides: is this you or an impostor?
- After each successful login, the AI updates itself – so it ages with you.
- Backend: FastAPI (Python), SQLite, SQLAlchemy
- AI: scikit‑learn (GMM), incremental learning
- Computer Vision: MediaPipe, OpenCV
- Encryption: AES‑256‑CBC
- Frontend: HTML, CSS, JavaScript (GitHub‑style dark theme)
- Deployment: uvicorn, any cloud (or local)
For research and personal use only. Commercial licensing available – contact the author.
MediaPipe team for face landmarks
scikit‑learn for GMM implementation
OpenCV & FastAPI communities
cd eyed-biometric
python -m venv venv
source venv/bin/activate # or venv\Scripts\activate on Windows
pip install -r requirements.txt