Secure Face Recognition System with an Integrated Silent-Face Anti-Spoofing Module

Tóm tắt

Abstract—Traditional facial recognition systems remain highly vulnerable to spoofing, undermining their reliability for secure identity authentication. In this study, we develop a real-time facial recognition system enhanced with a Silent-Face Anti-Spoofing module to strengthen the reliability of identity authentication. The proposed framework follows a streamlined pipeline that begins with image preprocessing for normalization and contrast enhancement. Faces are then detected using the Haar Cascade algorithm, followed by a liveness verification stage powered by an ensemble model that jointly analyzes texture and depth cues. For feature representation, we integrate FaceNet and VGG-Face to generate compact 128-dimensional embeddings. The matching phase applies the Cosine Similarity metric with a threshold of 0.40 and a 0.05 margin, balancing precision and recall effectively. Designed to operate efficiently on standard CPUs without dedicated GPUs, the system consistently achieves a 97.4% F1-score and an average processing time below three seconds, demonstrating stable and practical performance in real-world deployment scenarios.

Index Terms—Anti-spoofing, Deep Learning, Face Embedding, Facial Recognition, Silent-Face Anti-Spoofing.