Computer Science Capstone — Class XII B1
Facial recognition,
step by step.
A live demo of how a camera frame becomes a face embedding, how embeddings are stored and compared, and why a match is accepted or rejected. Runs entirely in your browser using face-api.js — no server, no Python.
Register
Capture photos and save 128-dimensional face embeddings to the database.
Identify
Upload one photo and find the closest saved embedding. See distance and confidence.
Verify
Compare two photos directly without using the saved database. One-to-one match.
Recognition pipeline
SSD MobileNet V1 locates face regions and outputs bounding boxes.
68-point landmark model normalises face geometry and pose.
ResNet-based descriptor network outputs a 128-value embedding vector.
Euclidean distance between embeddings decides match vs no-match.