A Fusion-Based Deep Neural Networks Approach for Face Liveness Detection

Face recognition systems, which are used for access control, security surveillance, and mobile payments, are popular in recent years among biometric systems with their ability to identify individuals quickly and accurately. However, these systems are vulnerable to spoofing attacks using a fake face input for unauthorized access. Liveness detection is a crucial component of face recognition systems that helps to improve their security against spoofing attacks. The paper first presents a comparative study of state-of-the-art liveness detection techniques and their effectiveness in detecting various types of spoofing attacks. Then a match score fusion-based liveness detection algorithm is proposed in which three deep neural networks are combined. As a result of experimental studies using NUAA and CelebA-Spoof datasets to test the performance of the proposed method in face liveness detection, 98.36% accuracy was achieved. The results show that ensemble model can significantly improve the accuracy and robustness of face liveness detection.