Facial Recognition System
Overview
A one-shot biometric identification and facial verification platform designed to verify identities from single reference photographs without retraining the neural network.
Neural Architecture
Implemented the twin CNN architecture with shared weights, contrastive loss optimization, real-time webcam frame alignment, and Euclidean distance verification.
Technical implementation
Python, TensorFlow, Keras, OpenCV. Engineered face detection, landmark normalization, and a 128-dimensional embedding projection space for real-time comparison.
98.4%
Verification Accuracy
128-d
Embedding Dimensions
22ms
Inference Latency
Results
Delivers reliable one-shot verification across varying lighting conditions and facial angles, maintaining 45+ FPS on edge CPU hardware.
<0.01%
False Acceptance Rate
<1.4%
False Rejection Rate
45 FPS
Real-time FPS