Research On Deep Learning Based Face Lineness Detection Algorithm

In recent years, with the Internet + concept, the network-based application extends rapidly. Online identity authentication is becoming a key problem. Biometrics based authentication is focused because it could preserve the consistence of data identity and physical identity.In this thesis, we focus on face liveness detection. The primary works include as follows:Firstly, we propose a deep learning-based face liveness detection algorithm. First, we train a CNN network. Because the face liveness detection is a bi-class problem, the supervision information is not strong enough, therefore, the extracted feature is not robust and performance is not satisfying. Then we train a new network by fine tuning the known Alex-net. The experimental results confirm the efficiency of the proposed algorithms.Secondly, we propose a dynamic feature-based CNN scheme to face liveness detection. First the Pyramid LK optical flow method tracking video. Then we analyze the optical information and compute the horizontal and vertical optic flow respectively. Furthermore, we could obtain the dynamic map, which is used to fine tune the Alex-net.Thirdly, we designed a face liveness detection and face recognition system. The system contains registration and authentication stages. The registration stage contains face detection and feature extraction. The authentication stage contains face detection, face liveness detection, feature extraction and similarity calculation.