The Research Of Face Recognition Method Based On Deep Learning

As a common method of deep learning,convolution neural network performs well in image recognition classification performance.Convolution neural network combined with local receptive fields,downsampling and shared weights make it to be a certain extent reduce,Magnification and other changes in the form of a high degree of deformation.As one of the key technologies of biometric recognition,face recognition is a hot topic in the field of pattern recognition.The purpose of final research is to make computers possess the ability of identifying.In this paper,the goal is to study how to effectively extract face features,deeply studying the basic principle of convolution neural network,referring to the classic convolution neural network structure to design basis model,and combining with the basic model and common feature extraction method to form multi-layer Feature fusion model.The main research done in the following areas:(1)As one of the most popular deep learning frameworks,Caffe performs efficiently and easily constructs its own convolutional neural network model.This dissertation studies Caffe's construction process on convolutional neural networks and uses Caffe for specific people Face experimental data set to design their own basic model,the basic model of the design includes convolution layer,pooling layer and fully connected layer.(2)Combining with traditional feature extraction methods such as PCA,LDA,LPP and so on,a multi-layer features fusion face recognition network model based on convolutional neural network is constructed.The network model firstly reconstructs and merges the feature map of each layer of the basic model to get the final face features,Then,entering the SVM and KNN classifiers.Finally,comparing and analyzing the validity of the features extracted from the features fusion model and the basic model.The experimental results show that the highest recognition rate of the features fusion model on the ORL,Yale and AR data sets is 98.6%,99.1% and 99.8%,respectively.Comparing with the basic model,increased 10.4%,9.6% and 7.0% respectively,but the time efficiency of the features fusion model is lower than that of the basic model.Experiments show that the features fusion model has better performance on face feature extraction.(3)An online face detection and recognition system is designed for the face recognition core algorithm based on the features fusion model.The system can realize the function of dynamic face recognition,and the recognition rate meets the design requirements.The system tested a total of four online face detection and recognition models.Experimental results show that the fusions feature model has the highest recognition rate and the recognition rate is 13% higher than that of the basic model,which further verifies that the features fusion model can more effectively extract facial features.