Research On Face Recognition Based On Deep Learning

Face recognition technology as an important research direction of biometrics has received more and more attention,and with the emergence and development of deep learning,the technology of face recognition has made great progress.Based on the research of face recognition algorithm based on deep learning,this thesis focuses on deep network,loss function and face data in face recognition.The main research content is divided into three parts.1.This thesis studies and analyzes the classical and the newest depth network,and improves the network structure for face recognition task in face recognition task,considering real-time,training difficulty and recognition performance.At the same time,by comparing and analyzing the influence of the existing feature extraction layer on the recognition performance,a reasonable feature extraction layer is selected for face recognition network.2.The advantages and disadvantages of various loss functions used in human face recognition task training are studied and analyzed.In view of the shortcoming of training difficulty and complicated operation of the existing loss function,a better monitoring signal is proposed,and the feature normalization and weighting normalization are applied to face recognition loss function,which simplifies the training process and improves the recognition rate.3.To study and analyze the impact of data enhancement on the performance of face recognition models,the thesis train face recognition networks and improve face recognition performance through three data augmentation methods: data augmentation based on image transformation,data augmentation based on 3D face model and data augmentation based on GAN face attributes.Compare and analyze the existing open database,collect and establish a face database based on Chinese face for face model training and improve face recognition capability.4.In order to verify the validity of the algorithm,a layered face recognition system is designed and constructed,which integrates the face detection,feature point detection and face feature comparing and other key modules of the face recognition task,and is used for testing in the real environment to verify the effectiveness of the algorithm.