Facial Image Restoration Algorithm Based on Generative Adversarial Networks

Abstract

In this paper, we propose an improved adversarial generative mesh facial image restoration model that addresses problems such as protruding boundaries and blurred textures after repairing damaged areas in facial images. A multidimensional residual module and a self-attention module are used in the sublayer to improve the feature extraction capability. The generator and discriminator are alternately trained on the basis of the opponent loss function and the L1 loss function until the model becomes stable. Comparative experiments on CelebA-based datasets show that the constructed face retrieval algorithm performs better.