Design And Implementation Of A Face Recognition System In Low Resolution Scene

With the development of the times and the advancement of technology,more and more commercial,civil and legal related procedures need to verify personal identity.For example,Alipay's face login function,the face unlock function of the mobile phone and the ID face comparison system used by the police for enforcement require the use of face recognition comparison technology.Such technical requirements have led to the attention of many researchers in the past few decades,and the face recognition technology has also developed rapidly.However,in some special occasions,such as community monitoring,station monitoring,street monitoring and other real scenes,the captured face image is affected by factors such as the sharpness,mounting position,angle,and light of the camera,resulting in the image resolution of the captured face is lower,not easy to identify.In the above several occasions,the collected face image contains less facial feature information,and the face feature is difficult to extract,which makes the face recognition task difficult to perform.Therefore,this thesis focuses on how to reconstruct low-resolution images into high-resolution problems,and the face recognition system in low-resolution scenes designed on this basis,the main work is as follows:1)For the needs of face recognition in low-resolution scenes,the functional modules and overall system architecture of the whole system are designed.The main functional modules include video display,face detection,face image super-resolution reconstruction,face recognition.The face information management module is mainly used for the maintenance of the face information database by the administrator,and the system interface is mainly used for the functions of face recognition and identity information output,and the business process of the face recognition system is described in detail.And through the design interface,all module functions can be quickly accepted by the user,and real-time display of the actual effects of detection,reconstruction,recognition results and operation;2)The core problem of face image recognition in low resolution scenes is studied,and the method to solve this problem,namely super resolution reconstruction,is introduced.The super-resolution reconstruction method based on generating the anti-netweork is improved,and the structure is improved according to the existing network structure improvement method,so that the network structure is lighter,the parameters are less,and the training is favorable;in the aspect of highlighting the feature information of the face,the eye is improved The special features of the mouth,nose and other details have been reconstructed,and a super-resolution reconstruction method for enhancing the facial features is proposed to enhance the detail expression of the prominent features of the face,which is more conducive to the development of face recognition tasks in low-resolution scenes.Using the data resources of the existing face database CeleA and CASIA-WebFace,the pre-processing of the downsampling is used for network training.Finally,a comparative experiment is carried out to verify its effectiveness.3)For the algorithm model of face feature extraction,based on the research of residual neural network(ResNet),the network structure ResNet-34 suitable for practical use is constructed on the basis of ResNet,and the constructed network is trained with the loss function CenterLoss.And tested on the data set LFW,and achieved a good recognition rate.Therefore,using ResNet-34 to complete the face feature vector extraction task,and after experimental testing,the feature extraction network achieved good accuracy in the face recognition test task;4)The face recognition system designed in this thesis has been tested,which can complete the face recognition task in low-resolution scenes well.It can be put into practical use and effectively improve the people in low-resolution scenes.The problem of difficult face recognition.