Shadow Eye: A Security System Featuring Few-Shot Face Recognition With Siamese Networks and Triplet Loss

Existing facial recognition systems struggle with scalability and data scarcity, requiring extensive retraining for new users and failing under variable lighting/angles. In this paper, we propose - <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Shadow Eye</i> - a novel few-shot learning system that achieves 94.2% accuracy on VGGFace2 (surpassing FaceNet by 4.7%) by integrating Siamese Networks with triplet loss and a pre-trained ResNet50 backbone. Unlike existing methods, our proposed Few-Shot Face Recognition (FS-FR) model eliminates retraining for new users - embedding generation from 50-100 images suffices and operates robustly in real-world conditions (pose, lighting, expressions). The <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Shadow Eye</i> system, powered by the Few-Shot Face Recognition (FS-FR) model, bridges the gap between highly accurate biometric recognition and real-world environments where only limited data is available.