3D Convolutional Neural Network Based on Face Anti-Spoofing

Spoofing face can be used to deceive face authentication systems for illegal purposes, and thus it poses a serious threat to the face authentication system. Therefore, it has become an important subject that the face authentication system must be solved in the practical stage. Many existing literatures focus on the study of photo attack. For the video attack, however, the related research efforts remain to be improved. 3D convolution neural network has the characteristics of deep learning, which can automatically learn the representation of the distributed features of image. Compared with 2D convolution, it can learn the motion features of continuous video frames. In combination with 3D convolution, in this paper, 3D convolution neural network (CNN) is proposed to learn features from video frames to implement face anti-spoofing. After training the 3D convolution neural network, the SVM classifier is trained with the spatio-temporal features extracted by the last fully connected layer of the 3D convolutional neural network to carry out the classification of the real face and the disguised face. In the experiments, two face anti-spoofing public database ReplayAttack and CASIA are used to implement multi-scale internal database test and cross database test. Compared with other methods based on texture features and 2D convolutional neural network, the experimental results have been greatly improved. Which can be applied to living face detection of video attack.