Abstract: To solve the problem of poor face recognition effect caused by serious lack of facial texture in the process of human-machine interaction system capturing large pose profile images, this paper proposes a multi-angle face recognition system for video sequences based on face synthesis. By improving the detection and alignment module in traditional face recognition system, it combined the generative adversarial networks (GAN) with two mainstream facial landmarks localization methods. The front face was generated by adding the face preprocessing process, and the system threshold was set for different side face angles to restore the lost facial features. The experimental results show that the proposed system can effectively improve the face location accuracy, and the recognition accuracy can be improved by up to 18.85%.