Semantic Face Image Inpainting based on Generative Adversarial Network

Abstract

With the popularity of Internet technology and camera equipment, people are accustomed to using images and videos to record life. Image deletion is one of the most important degradation directions when image degradation occurs. The repair process of the digital image repair method is to use the information of the missing part of the image, according to certain repair rules to repair and fill the missing part of the image, so that the repaired image is complete and natural. At present, the existing image inpainting algorithms still have some shortcomings in visual effect and algorithm efficiency. In order to solve the problems of fuzzy details and poor visual perception of the existing technology in the implementation of face image inpainting results, as well as the problem that the whole model could not be controlled due to the mode collapse caused by the use of the generative adversarial network, this paper provides a semantic inpainting method of face image based on multi-scale feature fusion. Using suppression enhancement unit to suppress useless channels, enhance useful channels, acquire long-range and multi-level dependency interaction without increasing parameters, coordinate the details of each position and the details of the far end when repairing the image, expand the receptive field, make up for the lack of information when generating the missing image edge, balance the learning ability of generating network and discriminating network to improve the inpainting effect of missing face image.