Face Image Completion Based on GAN Prior

Face images are often used in social and entertainment activities to interact with information. However, during the transmission of digital images, there factors that may destroy or obscure key elements image, which hinder understanding image’s content. Therefore, study image completion human faces has become an important research branch field computer processing. Compared traditional inpainting methods, deep-learning-based methods have significantly improved results on face but case complex semantic information large missing areas, still blurred, color boundary is inconsistent does not match visual perception. To solve this problem, paper proposes a method based GAN priori guide network complete by directly using rich diverse pre-trained GAN. The model coarse-to-fine structure, where damaged corresponding masks first input coarse obtain results, then fine multi-resolution skip connections. uses from generate finally SN-PatchGAN discriminator evaluate results. experiment performed CelebA-HQ dataset. latest three qualitative quantitative experimental analysis shows our obvious improvement texture fidelity.