Addressing the damage caused to murals by natural factors and human destruction,traditional image restoration methods fail to capture the advanced semantic information of the images,while conventional deep learning-based methods often re-sult in blurred edges and poor restoration effects for large-scale damages.To more reasonably restore the original appearance of damaged murals,a context encoder-decoder is designed by leveraging the local feature extraction capability of traditional convolu-tions and the global structural understanding of the Swin Transformer.Ultimately,a two-stage generative adversarial network based on Swin Transformer is constructed.Experimental results show that,the restoration effect outperforms current mainstream algo-rithms both subjectively and objectively.
cultural relic protectionimage restorationSwin Transformer
万方数据
