Diffusion-based kernel matrix model for face liveness detection

Abstract

Face recognition and verification systems are vulnerable to video spoofing attacks. In this paper, we present a diffusion-based kernel matrix model for face liveness detection. We use the anisotropic diffusion to enhance the edges of each frame in a video, and the kernel matrix model to extract the video features which we call the diffusion kernel (DK) features. The DK features reflect the inner correlation of the face images in the video. We employ a generalized multiple kernel learning method to fuse the DK features and the deep features extracted from convolution neural networks to achieve better performance. Our experimental evaluation on two publicly available datasets shows that the proposed method outperforms the state-of-art face liveness detection methods.