Research On Face Verification Based On Convolutional Neural Network

Face verification is a key problem in face recognition research is also a difficult point.Due to the development of Internet technology in recent years,how to quickly authenticate to ensure that personal information security has become a hot topic.As the face verification problem is the more important bio-verification method in authentication,face verification has become a new research hotspot.Face verification is a two-type validation problem that is given a face picture and it is known as the identity of the face picture to determine whether the two pictures are the same person.The main research work of this paper is based on convolution neural network.Two kinds of face verification models based on convolutional neural network are proposed under the condition of restrictive and unrestricted conditions.The main research work is as follows :1 In the Yale B face database and the AR face database,a hybrid model of the convolutional neural network model of the face validation method is proposed.Compared to the traditional face verification method,this method performs the segmentation operation on the face verification operation.And uses the PCA dimensionality reduction and SVM verification classification operation.Compared with the traditional method,the accuracy of face verification is improved in the restrictive experimental environment2 In the case of face images under unrestricted conditions,the hybrid convolutional neural network is optimized and improved due to the limitations of the mixed convolution neural network,and a face verification model of a three-channel parallel convolution neural network structure is proposed3 The three-channel parallel convolutional neural network model of two different connection modes is studied on the LFW face database.The first is a traditional way of using a full connection.The second is to use the local connection.The use of two connected networks is to compare the way in which the connection is made in different connection modes can improve the accuracy.In order to further improve the accuracy of the whole model,the model has been trained twice.The first use of the SGD optimization function for the first training,and save the best results of the model.The second use of the Adadelta optimization function in the first training model on the second training in order to improve the accuracy of the entire model...