Research And Implementation Of Inpainting And Recognition Of Partial Information Missed Face Images

With the huge improvement of informatization and rapid development and applica-tion of computer science and technology such as big data,artificial intelligence and com-puter vision,The development and application of face recognition using face image pro-cessing and face feature extraction has received increasing attention and use from all walks of life.In recent years,in the field of face recognition,many new models and algorithms are emerging every year,constantly refreshing the records of face recognition.However,in real life,human faces are inevitably occluded by wearing items such as masks and sun-glasses,or are affected by objective conditions such as light,angle,and different cameras,and human face images are artificially smeared and added with mosaics.This makes partial information missing from face images,which adversely affects the recognition capabilities of existing face recognition models and algorithms.In order to eliminate or reduce these adverse effects of the situation mentioned above,researchers have proposed and demon-strated many methods including some traditional methods such as subspace regression and robust error coding,which are not only easily affected by the designer's ability,but can only reduce the adverse effects of occlusion on face recognition to a limited extent,and cannot achieve satisfactory results.Therefore,the research on how to effectively perform face recognition under partial facial information loss has huge development space.In this paper,the related methods of deep learning and neural network are adapted and applied to face image inpainting and recognition with partial information loss,and good results have been achieved.A inpainting and recognition algorithm for face images with partial information loss is proposed using face image inpainting as an entry point.The complete face image after repairing missing information is used for face recogni-tion to reduce its interference on recognition.This thesis firstly introduces the basic theo-ries and principles of deep convolutional neural networks and generative adversarial net-works.Based on yolo object detection algorithm,a face and face occlusion detection model is implemented and trained,and achieved good results.The VGG16 face recognition model is implemented and trained,and experiments verified that the performance and recogni-tion accuracy of the model were significantly reduced under partial facial information loss.Secondly,this thesis introduces the principle of Wasserstein generative adversarial network and partial convolution.Based on these principles,a face image inpainting model is designed and implemented.A new facial feature loss is added under the premise of per-ceptual loss,valid pixels loss,and total variation loss.it is verified through experiments that after adding the new loss function,it can effectively suppress ”face changing”.Finally,based on the improved face image inpainting model and VGG16 face recognition network model,a partial facial information missing face image recognition algorithm combining face inpainting and face recognition is proposed.The experiment proved that the algo-rithm is better than convolutional network based Face recognition model and traditional algorithms,having a higher recognition accuracy.