Face recognition based on improved residual neural network

Abstract

For traditional face recognition algorithm based on convolution neural network in the insufficient light, side face of 45° environment, for large-scale face recognition accuracy is not high, and when the network layer increases to a certain layer, the performance of the network tends to saturation, continue to increase the network layer instead lead to performance problems. An improved neural network architecture combined with deep residual neural network and ArcFace Loss is proposed, namely, ArcFace Loss replaces the Loss function Softmax of deep residual neural network. Experimental results show that the improved residual neural network has an accuracy of 97.7% in face recognition on LFW data set, which is higher than the traditional and improved deep learning algorithm.