Research And Application Of Occluded Face Inpainting And Recognition Technology

With the rapid development of intelligent information technology,the continuous improvement of hardware computing capabilities and the continuous maturity of deep learning theory,face recognition technology has achieved unprecedented development and has been widely used in various fields of production and life.However,there are still some common problems in use.When the face is disturbed by glasses,hats,masks,etc.,certain facial features will be lost,which will seriously interfere with the accuracy of face recognition and severely limit the application range of face recognition technology.Although researchers have proposed many methods to reduce the interference caused by facial occlusion and achieved certain results,they have not completely solved the problem of facial occlusion recognition.Therefore,how to eliminate the influence of the occluded area on face recognition still has great research space and research value.Aiming at the occlusion situations that often occur in daily life such as hats,glasses,masks,etc.,an improved occlusion face recognition method is proposed.This method eliminates the negative influence of occlusion on facial recognition from the two aspects of occlusion area restoration and recognition method optimization,so as to ensure the accuracy of occlusion facial recognition.At the same time,the prototype system of occlusion face recognition was designed and developed,which realized face restoration and dynamic video face recognition,retrieval and filtering functions.The contents of this thesis are summarized as follows:(1)This paper compares and analyzes the advantages and disadvantages of GAN,WGAN,and WGAN-GP algorithms,selects WAGN-GP algorithm as the basis,adopts global discriminator and local discriminator on the algorithm framework,and introduces counter loss,reconstruction loss,content in the loss function loss,and combining the advantages of the U-Net model and the VGG16 model,an improved occlusion face repair algorithm is proposed to ensure the integrity and consistency of the repaired face image,thus laying the foundation for subsequent face recognition.(2)For the common face recognition methods based on convolutional neural networks when the occlusion area is large,there is a problem of face recognition.Firstly,combine the advantages of the SENet module and Arc Face loss function to improve the VGG16 model;Secondly,based on the improved occlusion face restoration algorithm,combined with the advantages of the improved convolutional neural network VGG16,a new occlusion face recognition algorithm model is proposed;Finally,the effectiveness of this occlusion face detection algorithm is verified through several sets of comparison experiments.With different occlusion rates,the face recognition accuracy is better.(3)Based on the proposed improved algorithm for recognition of occlusal faces,a set of prototype systems for recognition of occlusal faces was designed and developed.It not only realizes the function of face restoration,but also has video face recognition and retrieval functions,and initially achieves the purpose of effective recognition,realtime retrieval and accurate screening of faces in dynamic videos in the actual environment.By testing overall system performance,it is shown that system for the repair and recognition of the occluded face has a good effect,and it has strong practicability.The research results of this thesis not only help to solve the problem of reduced recognition accuracy caused by the loss of occluded facial features,but also realize effective face recognition,real-time retrieval and accurate screening of dynamic videos,thereby further broadening the application of face recognition technology Scene,and has a strong practicality.