Research And Implementation Of Inpainting And Recognition Of The Incomplete Face Images

In recent years,the rise of convolutional neural network and generative adversarial network makes human being more imaginative in image processing,and plays an important role in image editing,face recognition,security verification and other fields.Image editing technology has been a hot topic.Traditional image editing softwares such as Photoshop has been widely used.However,due to the limitations of some traditional algorithms,it is not effective to deal with the incomplete image.Recently,a large number of image inpainting technologies based on deep learning algorithm have emerged,which greatly satisfies the limitations of traditional algorithms.However,the existing deep learning algorithms often have obvious artificial traces when dealing with the missing images,because the uncertainty of inpainting will be increased by directly repairing RGB images from three channels.Inspired by the sketch technology and the generated adversarial network in the art field,this paper considers using contour shaping first and then filling it.Therefore,a two-stage repair model is proposed to repair the incomplete face image.In this model,the existing contour is extracted by gray level and mask of the incomplete image in the first stage,and the predicted contour is obtained by the generator of the generated adversarial network;in the second stage,the predicted three-channel color image is generated by using the predicted contour map and the missing original image.The experiment shows that the method has better effect by comparing qualitative and quantitative with other deep learning methods.Because of the lack of information in the incomplete face,it is always a difficult point to recognize the incomplete face.Traditional subspace regression and robust error coding methods can not achieve satisfactory results.The missing information will also produce the difficulty of feature extraction for the deep learning algorithm based on neural network.In order to reduce the influence of missing area on recognition,this paper proposes a solution to the recognition of the incomplete face image,namely,first repair the missing part,and then use cosine similarity to replace the triple loss based on Euclidean distance in facenet,and compare and recognize face based on the similarity of weighted contrast feature vector.Finally,a simple face repair and recognition system is designed and developed according to the requirements.The function of the system is tested.The experimental results show that the system has a better accuracy.