Partial Occlusion Face Recognition And Image Inpainting

With the rapid development of artificial intelligence technology,face recognition technology has been successfully applied in many fields such as intelligent security,work punch,identity and information entry and other fields.However,masks,hats,glasses and other factors will cause partial occlusion,resulting in the loss of face features,resulting in a decline in the accuracy of face recognition.For the repair of occluded face,there are still some problems such as unclear texture and unreasonable structure.These problems have aroused the attention and research of some scholars,but these problems have not been completely solved,and further research is still needed.Therefore,in order to solve the above problems,this paper studies face recognition and repair methods under partial occlusion.The main work is as follows:(1)In view of the problem that existing face recognition algorithms can not reasonably allocate different occlusion area feature attention and occlusion caused by the problem of reduced recognition accuracy.This thesis proposes a local occlusion face recognition model Mobilenet V1-Cbam-Facenet(MC-Facenet)based on the attention mechanism.The main feature extraction network of the model uses deep separable convolution,which effectively reduces the computational burden of the model.In addition,the attention mechanism is introduced into the model,and the model can generate adaptive attention to the features of different regions through training,so as to improve the ability of the model to perceive and utilize important features.In order to verify the effectiveness of the model,This thesis carried out tests on public AR and LFW data sets,and carried out several experiments under different occlusion types and different occlusion rates.The experimental results show that the recognition accuracy of MC-facenet is greatly improved compared with the original model under different types and different proportion of occlusion.(2)In view of the problems of disordered structure and unclear texture in inpainting of partially occluded face images,an improved two-stage partially occluded face image inpainting model is proposed in this thesis,which is divided into two stages: edge restoration and texture restoration.In the stage of edge inpainting,gated convolution was used to make full use of edge information,which strengthened the learning ability of the model for the association information between occluded and non-occluded regions,and improved the rationality of the structure of edge inpainting.The attention mechanism is introduced in the texture repair stage to obtain sufficient global information,which improves the model’s ability to repair texture details and solves the problem of unclear texture.In order to verify the effectiveness of the model,comparative experiments with different occlusion rates were carried out on the public data Celeb A.Experimental results show that the improved two-stage partially occluded face image inpainting model has a good effect on partially occluded face images,and the values of peak signal-to-noise ratio and structural similarity are higher than those of other models.Effectively improve the structure of confusion,texture is not clear,improve the quality of occlusion face image inpainting.