Face Recognition Method Based on Enhanced Edge Cosine Loss Function and Residual Network

At present, the loss function in the deep convolutional neural network (CNN)can not make the face feature reach the ideal feature classification criteria, that is, maximizing the inter-class difference and minimizing the intra-class difference. Therefore, a face recognition method based on enhanced edge cosine loss function(EECL) and combined residual network(ResNet) is proposed. The RestNet used in this paper is based on the EECL. By normalizing and maximizing the cosine decision boundary, it can realize the maximization of inter-class differences and the minimization of intra-class differences more effectively, enhance the discriminant ability, make the face features more distinguishable, and thus improve the the recognition rate of the face. Through a large number of analysis and experimental verification, the proposed method has a face recognition accuracy of 99.40% on the Labeled Face in the Wild(LFW) face dataset, 72.602% and 85.420% respectively in MegaFace face recognition task and face verification task, indicating the effectiveness of the method.