Research on Face Recognition Technology Based on Deep Reinforcement Learning

Abstract

With the rise of deep neural networks, face recognition technology has developed rapidly. However, the low-quality video S2V face recognition under the conditions of poor lighting conditions and low resolution still does not achieve the expected effect due to the heterogeneous matching problem between the low-quality test video and the high-definition image of the sample library. Aiming at this problem, a low-quality video face recognition method based on super-resolution reconstruction is proposed. First, multi-dimensional features can be extracted by training video data with convolutional neural network; secondly, video features are input into the attention model, and local face features, face positions and temporal memory units are obtained according to the temporal continuity information of video data; finally, Q-learning is adopted. Iteratively calculates the output of the attention model, finds the optimal frame sequence containing the face, and uses this to calculate the video matching accuracy.The experimental results show that the method can effectively improve the accuracy of video face recognition in complex backgrounds.