Bimodal Face Recognition Based on Liveness Detection

摘要

In this paper, we proposed a three-dimensional face recognition method based on liveness detection. Firstly, one liveness detection method based on the three-dimensional structure of the face was proposed. According to the face feature point localization algorithm combined with the RANSAC fitting algorithm and the SVM model training prediction method, the depth information was used to judge the reality of the face, and the detection of the photo and video attacker behavior were solved. In this paper, the traditional loss function and the new central loss function were used as the supervised signals in the face recognition process. We used model fine-tuning to train two independent convolutional neural networks, and then we used the score fusion approach to fuse facial feature matching. Finally, we evaluated the Eurecom and Vap public test sets based on the way of actual application. The experimental results showed that compared with the traditional in liveness detection, our algorithm made use of the advantages of three-dimensional information to realize the judgment of face reality, and the algorithm had better accuracy in three-dimensional face recognition section.