WLD-TOP Based Algorithm against Face Spoofing Attacks

Face liveness detection is more and more important in face recognition systems, which are vulnerable to spoof attacks made by non-real faces. Recent work has revealed that some algorithms based on image descriptors are applied to face liveness detection against face spoofing attacks, such as LBP and LBP-TOP. However, these image descriptors are not robust to spoofing attacks. In this paper, we propose a robust and powerful local descriptor, called WLD-TOP. It combines temporal and spatial information into a single descriptor with a multiresolution strategy. Extensive experiments on CASIA and our new SYSU-MFSD database demonstrate that the descriptor can achieve a better liveness detection performance in both intra and cross-databases compared to the state-of-the-art techniques based on descriptors.