Research On Facial Recognition Method Of Micro-plastic Surgery Based On Hybrid Self-attention Network

Face because of the stability of its facial features brings convenience to identity recognition,but the facial features of the micro plastic surgery will change to a certain extent,micro plastic surgery in daily production to face recognition technology has brought new challenges.If in the authentication,to be identified in the micro plastic face image features and the database of the relevant documents on the face image features of the difference,will lead to authentication errors,to the need for authentication agencies to work has brought a lot of inconvenience.To this end,in this paper,the face recognition after micro-plastic surgery problem,mainly from three aspects of research: self-made relevant micro-plastic face image data set,selection of appropriate loss function and improvement,improve the network structure.Improving the data set,optimizing the loss function and improving the network structure are three magic weapons to break through the bottleneck of face recognition.In deep learning,the number of research samples should be large enough,the categories should be rich,and the features should be diversified.The selection of appropriate loss function can reduce the error between the expected value and will directly affect the classification results between classes or within classes.The improvement of network structure can directly improve the performance of the model in many aspects.On the basis of the research work of face recognition and cross-age face recognition and so on,aiming at the low correct recognition rate of micro facial recognition,this paper mainly does the following aspects of work and innovation:(1)Firstly,in view of the difficulties in face recognition of micro-plastic surgery,prototype micro-plastic surgery face image data set PMP and simulated micro-plastic surgery face image data set SMP were homemade.The sample subjects of PMP data set were mainly Chinese stars who had admitted micro-plastic surgery,and the pictures before and after the plastic surgery of each sample were collected.The SMP data set samples are mainly obtained by simulating the facial organs of the original face without plastic surgery to different degrees,and finally the simulated face after micro plastic surgery.(2)Aiming at the problem that faces after micro-plastic surgery(small amplitude mutation feature)lead to blurred boundaries of verification classification,an improved face recognition Loss function Arcface-F Loss was proposed based on Arcface Loss,in order to reduce the error between the expected value and the real value,so as to improve the correct recognition rate of the model for micro-plastic surgery faces.(3)Combining the advantage of convolutional neural network in extracting detailed features with the advantage of self-attention mechanism in extracting correlation degree of long-distance features,a network model with better robustness is obtained.