Application Research Of Incomplete Image Restoration And Recognition Method Based On Deep Learning

With the improvement of the science and technology and the development of deep learning technology,in today's society,research on the problem of face recognition has attracted much attention and sought after.Face recognition technology with complete features has reached a very high level,but the results of face recognition studies with incomplete features are not ideal.Therefore,research on face recognition with incomplete features is a hot and difficult issue in face recognition.In this paper,we research on face recognition with incomplete features.Faces with incomplete features are divided into two cases: incomplete occlusion faces and complex illuminated faces with unclear features.The main idea of the method in this paper is: first repair incomplete feature images,complete image features,then recognize complementary feature images.The first is repair and recognition of incompletely occluded faces.First,the face is repaired through the improved WGAN-GP network,and then the generated image is recognized.Add jump connections in the generation adversarial network to preserve the underlying information of the image;and add symmetric loss and content loss to the loss function to make the generated image more realistic.Feature extraction is performed using pre-trained VGGFace network through transfer learning,and then cosine similarity is used as a classifier for classification and recognition.The results show that the method in this paper has better results.The second is to repair and recognize complex illuminated faces with unclear features.Firstly,normal light is migrated to complex light face images through the improved Cycle GAN to obtain normal light faces,and then the normal face light is recognized.Recognition is the same as occlusion face recognition.The results show the effectiveness of the method in this paper.