An Improved Algorithm for Facial Image Restoration Based on GAN

The paper proposed a two-stage facial image restoration method based on structure and texture network, which includes sketch restore network and texture generation network. The sketch restore network strives to restore the sketch structure in the missing area of the image, and the texture generation network generates the texture information of the missing area based on the sketch structure of the sketch restore network and the pixels of the known area. The method is not only able to successfully generate semantically reasonable and visually realistic content for missing regions, but also allows users to manipulate the structural properties of synthetic content in missing regions. Experimental results show that the proposed method not only has good performance, but also is more flexible.