Research On The Technologies Of Face Recognition Based On Deep Learning

Biometric technology has been closely linked with people,has become an important technical means of personal identity authentication.Compared with other biometrics,face recognition is one of the most intuitive,simple and convenient authentication methods,and it has a wide application in information security area.However,the existing machine learning methods mostly use shallow network structure(traditional neural network,SVM,etc.),it is difficult for the shallow neural network to express the relation between the features accurately,and the scarcity of the training samples in the real situation makes the whole model fall into the local optimal situation.In this paper,a face recognition system based on Deep Learning is proposed,which is based on deep belief network(DBN),convolution neural network(CNN)and hybrid networkwith convolution neural network and Restricted Boltzmann machine,in order to solve the problem of face training sample,excessive training parameters and disadvantages of algorithm itself.The number of training samples is increased by segmenting and zooming the human face,at the same time,using convolution neural network model,the convolution and pooling of the image greatly reduced the training parameters.Finally,the classification of the test samples is realized by the mapping ability of the restricted Boltzmann machine.By studying the relevant papers at home and abroad,this paper has a clearer understanding of the algorithm of deep learning,introduces the model and algorithm of shallow learning,and then elaborates the key models and basic algorithm principles in deep learning.The use of deep learning to the field of face recognition,is a new challenge and development opportunities.The research contents and specific work of the thesis are as follows:(1)In the face recognition system based on deep belief network,three hidden layer deep belief network models are proposed to solve the problem that the traditional algorithm haslittle hidden layers,and the weak accuracy and generalization ability.After pretreatment of face image,the face features are extracted from three hidden layers,and feature fusion is performed in the high-level hidden layer.Finally,the softmax classifier is used to classify the face.The method enhances the extraction ability of high-level features by increasing the number of hidden layers.Finally,the experiments show that the model proposed in this paper has achieved good experimental results in the ORL face database.(2)In the face recognition system based on Convolution Neural Network.In the case of a small number of database samples,make use of convolutional neural network for displacement,deformation,scaling with high robustness,in the image preprocessing process,the proposed cut the key features of the face region to increase the number of sample training and design and optimize the network model,the size of the convolution kernel and the pooling method in the optimal network modelare determined.Finally,the classification is done by softmax.Experiments show that the algorithm proposed in this paper has achieved good experimental results in LFW face database.(3)In the face recognition system based on Convolution Neural Network and Restricted Boltzmann machine.The input feature extraction and dimension reduction are realized by convolutional neural network and the classification is performed by the Restricted Boltzmann Machine is proposed.Experiments show that the model proposed in this paper has better performance compared with deep belief network model and convolutional neural network model in ORL and LFW face database.