Face Anti-Spoofing Based on 3D Learnable Convolutional Operators

In face recognition systems, Face anti-spoofing (FAS) plays a crucial role in ensuring security and reliability across various applications such as identity verification and access control. Traditional methods primarily rely on two-dimensional convolutional neural networks (CNNs) for extracting discriminative features from facial images, which have achieved significant success in distinguishing real faces from spoof attacks. However, these methods may encounter challenges in capturing temporal dynamics and modeling long-range dependencies inherent in liveness detection. This paper aims to address the limitations of existing methods by leveraging the advantages of three-dimensional convolutions and vision transformers for facial liveness detection. By combining 3D convolutions that capture spatiotemporal features with vision transformers that excel at capturing global contextual information, we can effectively enhance the accuracy and robustness of liveness detection systems. Additionally, we introduce learnable gradient operators to the 3D convolution operations, enabling adaptive learning of gradient information by the model. We evaluate our proposed method on multiple publicly available datasets and compare it against current state-of-the-art approaches. The results demonstrate lower average classification error rates, showcasing the superiority of our approach.