Research On Fuzzy Face Recognition Based On Image Restoration And Generative Adversarial Network

The application of image restoration technology and generative adversarial network to solve face recognition in fuzzy situations is a subject worth studying.This thesis introduces the background knowledge in the field of face recognition and the current status of research in the field of face recognition,and at the same time summarizes the research status of face recognition in the case of blur.The main work of this thesis is as follows: First,this thesis designs a face recognition system,which is divided into three modules: face detection,face database establishment and face recognition.The face detection module uses the Haarcascade algorithm,the face database combines a public database and a self-built database,and the face recognition module integrates classic face recognition algorithms and modern deep learning models,where the classic face recognition algorithm is based on the PCA algorithm,deep learning The model uses the Face Net model.The face recognition system can choose different modules according to different application scenarios,and can complete face recognition on the face data set or real-time video stream,which has certain practical value.Secondly,this thesis studies the recognition of blurred facial images based on image restoration technology and generative adversarial network technology for the problem of reduced recognition rate of facial images found in system development due to motion blur.First introduced the principles of image blur and restoration,summarized the classic image restoration techniques,and focused on the Wiener filter and the least squares filter,and compared the restoration effects of the two filters through experiments.According to the experimental results It is concluded that the least squares filtering method is superior to Wiener filtering.Then introduced the principle of generative adversarial networks,and finally proposed a SRGAN method based on least squares filtering,specifically introduced the network structure and model training process,using this method to restore the blurred face image and Wiener Filtering,least squares filtering and SRGAN method are used for comparative experiments.At the same time,these four methods are put into the face recognition module based on Face Net for testing.The recognition rate after the method proposed in this paper can be increased from 80.37% when blurred 97.02%,which is close to the original recognition rate of 97.75%,which proves that the SRGAN method based on least squares filtering proposed in this paper has certain effectiveness,and also proves that the research on fuzzy face recognition based on image restoration and generative adversarial network in this thesis has certain practical value.