Research on Style Preservation and Enhancement Methods of Generative Adversarial Networks in Digital Restoration of Yangjiabu Gate God New Year Paintings

Abstract

The digital restoration of Yangjiabu Gate God New Year paintings faces problems such as color gamut drift, mainly due to the traditional "denoising completion" paradigm. This article proposes a new GAN framework: the generator embeds a learnable prior manifold, and
the collaborative attention mechanism models the local global relationship; The discriminator introduces multi-scale PatchGAN and style moment matching; The composite objective function constrains style homeomorphism to achieve consistent reproduction of "form color spirit".