Research on Face Recognition Method Based on Deep Learning

In this paper, we mainly refers to the network structure of Alexnet, the traditional convolution layer is modified to Multi-layer Perceptron (MLP) convolution layer to enhance the face image feature extraction, adding Max-Feature-Map (MFP) excitation function segmentation of noise signal and the information signal to improve the recognition accuracy. The Center Loss loss function is added to reduce the distance between elements in the same class, which can better generalize its features and reduce the misjudgment caused by the distance between classes. The CASIA-Web data set is used for training and testing. Through 10575 tests, the recognition rate of the model is 82.3%. In this work, the face verification data set used is the LFW face database, 6000 pairs of face comparison experiments are calculated, and the average recognition rate is 84.5%, The experimental results show that the network model designed can show good recognition effect for face image prediction classification and face verification.