The Face Recognition Algorithm Research Based On Neural Network

With the continuous progress and development of social science and technology,face recognition has gradually become an important research direction in the field of image classification and recognition.The implementation of traditional face recognition technology is complex and factors such as illumination,gesture,facial expression and occlusion all affect the extraction of face features.The robustness of the algorithm is poor,and the recognition accuracy is low,so it cannot meet the needs of practical application.But the algorithm based on convolutional neural network can extract deep level face features more efficiently,and it also effectively solves many problems caused by traditional technology.Convolutional neural network technology has become the main way of face recognition.Based on the research of many literatures about face recognition technology,the paper mainly carries out the following works:Firstly,the paper analyzes the idea and structure of convolutional neural network,and then introduces the calculation methods and related algorithms of each layer in the network.As the theoretical part of the following chapters,the key technologies commonly used in neural network are described,such as activation function and dropout technology.Then,based on the structure model of Le Net-5 network,the face recognition algorithm based on shallow neural network is studied.The new model is improved and optimized by the following four aspects: Firstly,the paper improves the structure and depth to extract deeper and more kinds of face features;Secondly,using the relu activation function to maximize the data features,the neural network gets faster and better results in the process of iterative operation;Thirdly,the cross entropy loss function is used to train the parameters of the network model;Finally,dropout technology layer is added to prevent the model from over fitting and optimize the generalization ability of the model to a certain degree.At last,a new network model which named CLe Net is built.CLe Net model has a deeper structure and more training parameters.The experimental results on ORL and AR face datasets show that the CLe Net network further improves the accuracy of the network model,which proves the feasibility of the improvement.Finally,the deep neural network face recognition algorithm represented by VGGNet network model is studied.It is improved by the following four aspects: Firstly,reduce one layer of full connection layer in order to reduce the calculation parameters and improve the training efficiency;Secondly,BN layer is added to prevent the gradient from disappearing and accelerate the convergence of the model;Thirdly,using the convolution layer with step size of 2 to replace the pooling layer,so that the network model can learn better nonlinear expression ability;Finally,introduce the residual block on this basis,and generate the RVGGNet network model.RVGGNet network model can not only avoid the problem of gradient disappearing,but also solve the problem of low recognition accuracy caused by the deepening of network layers.The experimental results on related datasets show that RVGGNet network model can achieve better recognition accuracy,which verifies the feasibility of the improved method.