Research And Implementation Of Mask Occlusion Recognition Algorithm Based On Deep Learning

With the continuous development of deep learning,face recognition is becoming more and more mature.Face recognition is a kind of biometric technology that uses face features to infer their identity.However,under the ravages of 2019-n Co V,people have wear mask to prevent epidemic diseases,and occlusion of mask has brought great challenges to face recognition.The existence of masks concealed more than half of the features of the human face,resulting in a huge loss of information,resulting in a significant decline in the performance of the existing face recognition algorithms.Therefore,in order to reduce the influence of mask occlusion on face recognition,this thesis proposes a face recognition algorithm combined with face completion.The main contributions of this thesis are as follows:Aiming at the loss of information in the mask occluded area,this thesis proposes a two-stage face completion algorithm based on multimodality to recover the features occluded by the mask.The algorithm uses the original image,face geometric information and face attribute information to fuse,and inputs the fusion vector into the two-stage face repair network.The first stage of the network generates rough repair results,and the second stage generates more refined final repair results.And on the mask occlusion Celeb A dataset produced in this thesis,two indicators of SSIM and PSNR are used for evaluation.Experimental results show that the proposed algorithm achieves better repair effect than other face completion algorithms.In order to make more reasonable use of the real area not covered by the mask and the repaired area covered by the mask.This thesis proposes a face recognition algorithm based on multi attention mechanism.The algorithm uses spatial attention,channel attention and global attention,so that the algorithm can focus on the real features of the uncovered area.And on the mask occlusion LFW dataset produced in this thesis,accuracy index of face recognition is used for evaluation.Experiments show show that the face recognition algorithm only using multi attention mechanism improves 4.56% compared with other better face recognition algorithms;combined with the completion algorithm,it improves 5.87% compared with other face recognition algorithms.At present,there are few mask occlusion data sets available.This thesis uses face key point detection technology to make a batch of mask occlusion face data sets on the public face data sets,which provides data basis for algorithm training.In addition,based on the face repair algorithm and multi attention face recognition algorithm proposed in this thesis,a mask occlusion face recognition system is designed.The system mainly has the functions of mask occlusion face completion,mask occlusion face recognition,face database management and system management.The test results show that the proposed algorithm still has good performance in practical application.