Randomized challenge-response active liveness detection for the web: a 23-challenge library spanning facial actions and hand gestures, scored client-side against MediaPipe landmark geometry and re-verified server-side (trace matching, palm orientation, finger-extension ratios), so that photographs, replayed videos, and masks fail an unpredictable live challenge.
Reference-implementation extraction is planned. The production implementation currently ships inside the FIVUCSAS platform:
- Client (TypeScript):
web-app/src/features/biometric-puzzles/— challenge library, MediaPipe FaceLandmarker / HandLandmarker scoring, capture UI. - Server (Python):
biometric-processor— authoritative re-scoring (active_gesture_liveness_manager.py), confidence and duration floors, anti-replay checks. - Prototype lineage:
practice-and-test/GestureAnalysis/.
This repository will host the standalone library (client scorer + server verifier with thin adapters) together with its evaluation harness.
MIT