Research On Face Recognition Improved Loss Function And Portrait Hairstyle Elimination And Transformation Network

In the field of computer vision,face recognition technology based on deep learning has been widely used in recent years.In the field of face recognition,improving the performance of recognition is an important research direction.In the improved strategy,the loss function operation can further optimize the network structure,and the portrait attribute operation can greatly expand the data set,which are the research directions to improve the performance of face recognition.In addition,due to the development of mobile Internet and short video applications,face attribute editing is also a popular direction.The main contents and innovations of this dissertation are as follows(1)The improved face recognition loss function G loss based on Center loss can maximize the distance of features extracted by deep network between different classes,and maintain the characteristics of the same class.This dissertation studies and analyzes the influence of data amplification on the performance of face recognition model in training depth network,tests the performance of the improved loss function,and trains the face recognition network based on the method of generating face data amplification of generating confrontation network.The experiment shows that the accuracy of the model is improved by0.1% in LFW test set and 0.1% in YTF test set respectively 0.17%;the accuracy of LFW test set and YTF test set improved by 0.63% and 1.65% respectively.(2)Based on the operation of facial image hairstyle attributes,this dissertation studies and analyzes the generation confrontation network,designs the network structure for the task of eliminating portrait hair and transferring hair,designs portrait hair elimination network and portrait hairstyle transformation network,and completes the corresponding loss function design.Compared with the image generated by Style-GAN,the experimental results show that the image generation effect is improved,and more facial information is kept.Through the questionnaire subjective evaluation,the network is practical.3105 pairs of hairless face data sets are collected and produced as a small data set for the study of hairless face,which can be used as open data of face recognition technology.