Research And Design Of Face Real-name Registration System Based On Improved CNN

With the continuous development of China's economy and society,more and more floating population pour into cities for employment,work and residence,and the vitality of cities is constantly increasing.At the same time,it also causes a burden to the management of social security.If the floating population can not be effectively monitored and managed,it will become a breeding ground for illegal crimes.Hotels and Internet cafes are areas where floating population activities are frequent,and they are also easy to become places of hiding for criminals.The implementation of the real-name registration system for places by public security departments in accordance with the law will help to strengthen social security management and combat crime.However,at present,a small number of operators of Internet cafes and hotels,driven by interests,can not conscientiously implement the registration of identity certificates,there are problems of fraudulent use,borrowing and multi-use of one card.The lack of personal identity information will lead to the loss of control in management,unable to achieve effective collision comparison with the fugitive database,and there are considerable security risks.Relying solely on the patrol of police officers will cause a serious burden on the police force.With the rapid development of deep learning,face recognition system based on convolution neural network(CNN)has been widely used in the fields of public security and financial payment because of its high accuracy,robustness and convenience.In the real-name registration scene,the most prominent need is to compare the scene face photos and identity cards in the consistency of the electronic photos,but there are widespread changes in age between the two,the quality of pixels,hair and lighting great differences.This poses a serious challenge to the accuracy of the human face verification system.Under this premise,it is of great practical significance to study and improve the CNN to gain better accuracy.In order to effectively implement the registration system of real names according to law,we need also to explore the application of face recognition in the real-names registration.In this paper,two main improvements are made on the basis of the traditional CNN classification model: 1.add channel pooling layer to enhance feature selection,2.use cosine loss instead of the softmax loss to reduce the intra-class distance and increase the inter-class distance.At the same time,this paper designs and implements a simple identification card face verification program based on the improved CNN model,which further verifies the practical ability of this method in the face real-name registration system.Firstly,this paper analyzes the research status of face recognition system at home and abroad,and points out the problems of the current real-name registration system.On this basis,it puts forward the research significance,our main work and innovation.Then the concept and algorithm of deep learning are introduced,as well as the algorithm flow and processing module of face recognition system,especially the MTCNN detector and affine alignment algorithm.Then,on the basis of CNN classification network,channel pooling and cosine loss are improved.The experimental results on open benchmark LFW and our real-name dataset show that the improved CNN model has better accuracy and robustness.Finally,a face verification program of ID card based on improved CNN is designed and implemented.The comparative experiment and trial results show that the face real name registration system based on this improved CNN can solve the problems of ID card impersonation,borrowing and multiple use of one ID card,which is helpful to implement the registration policy of the place real-name system in accordance with the law.