Research On Face Recognition Based On Convolutional Neural Network

The face recognition algorithm recognizes face features by extracting a logo or feature from the image of the face.For example,PCA-based face recognition,LBPbased face recognition,and neural network-based face recognition use these features to search for other images having matching features.Convolutional neural networks have better effects than other algorithms in face recognition.First of all,it needs a reasonable neural network model,through which a lot of training is performed to complete face recognition and obtain false recognition rate.The research in this paper is to construct and improve a reasonable network model,and apply it to face recognition to solve the problems that affect the recognition rate caused by illumination factors,occlusion factors,expression factors,etc.Other methods for face recognition are good.Based on the classical LeNet structure convolutional neural network,this paper builds several different convolutional neural network models by deepening and expanding,and uses these models to study face recognition:(1)This paper first summarizes the basic knowledge of face recognition and convolutional neural networks,and takes a classic analysis of LeNet-5 as an example.(2)In this paper,the advantages and applications of convolutional neural networks in face recognition are introduced.Then three faces are used for face recognition on three different face databases.The maximum pooled and pooled averages are compared.The misclassification rates of the pooling method,the Re LU excitation function and the tanh excitation function on the three networks can be analyzed to solve the problems caused by illumination factors and occlusion factors in the traditional face recognition.Finally,the recognition effect is compared with the traditional face recognition method.(3)This paper constructs a deep convolutional neural network model applied in the field of face recognition.The fusion model is improved and a multi-column deep convolutional neural network is constructed.In the self-built face database and LFW face database,it is found that the improved deep convolutional neural network can better solve the problems that affect the recognition rate such as illumination factors and occlusion factors than the general convolutional neural network.