3D Convolutional Neural Network Based on Face Anti-spoofing

Abstract

Face anti-spoofing is very significant to the security of face recognition. Many existing literatures focus on the study of photo attack. For the video attack, however, the related research efforts are still insufficient. In this paper, instead of extracting features from a single image, features are learned from video frames. To realize face anti-spoofing, the spatiotemporal features of continuous video frames are extracted using 3D convolution neural network (CNN) from the short video frame level. Experimental results show that the two sets of face anti-spoofing public databases, Replay-Attack and CASIA, have achieved the HTER (Half Total Error Rate) of 0.04% and 10.65%, respectively, which is better than the state-of-the-art.