Research And Implemention Of Face Recognition For Partial Occlusion

With the continuous enrichment of electronic products and the continuous improvement of intelligence level,people are beginning to pursue an increasingly efficient and safe lifestyle.As an emerging intelligent technology in recent years,face recognition technology has been widely applied to life.Besides,with the deepening research of face recognition,various algorithms and technologies emerge in an endless stream,the accuracy of identification is getting higher and higher.However,in real life,people's face may be obscured by wearing masks,glasses and other accessories,which lead to a lack of some representation features of people's face,resulting in a performance that isn't ideal enough of these algorithms and techniques,and the accuracy of recognition is significantly reduced.Although various algorithms have been proposed to solve the problem of occlusion face recognition,these algorithms are not ideal enough to solve the problem fundamentally.Therefore,it's very valuable to solve the occlusion problem fundamentally in the field of face recognition research.For the symmetric occlusion of masks?glasses,etc,this thesis proposed a partial occlusion face recognition algorithm based on the generated confrontation network,which fundamentally solves the interference of occlusion on recognition by repairing the occlusion area on the face.The work of this thesis is as follows:(1)Firstly,this thesis introduced the techniques of image preprocessing and face detection,which provided sufficient technical support for subsequent face recognition research.Secondly,in order to compare the local occlusion face recognition algorithm proposed in this thesis,this thesis introduced a face recognition algorithm based on VGGNet network,and proved that the algorithm is not ideal enough under conditions with large occlusion area.Then,in order to solve the influence of large area occlusion,this thesis proposed a local occlusion face repair algorithm based on Wasserstein distance to repair the occlusion area of the face,and introduced a new loss: symmetric loss,based on the symmetric idea of occlusion content,which can ensure that the content generated by the occlusion part is approximately symmetrical.By demonstrating the introduction of the new loss function,the effect of is significantly improved.(2)Secondly,this thesis designed a new partial occlusion face recognition algorithm combines the improved facial repair algorithm and the face recognition algorithm based on VGG-Net network.The basic principle of the algorithm is to repair the face occlusion part first and then recognize the repaired face.Experiments show that the recognition accuracy of the algorithm is not only higher than that of traditional algorithm and convolution network face recognition algorithm with deep learning,but also higher than that of VGG-Net based face recognition algorithm under different occlusion areas.(3)Finally,based on the partial occlusion face recognition algorithm proposed in this thesis,I designed an occlusion face recognition system.This thesis firstly designed the overall architecture of the system according to the requirements of the system.Then divided the modules and implemented functions according to the overall framework.Finally,the thesis tested the system performance according to the results of identification accuracy rate and time delay of system test evaluation index,the test results show that the system has high recognition accuracy and efficiency for occlusion face recognition.