Restoration of non-structural damaged murals in Shenzhen Bao’an based on a generator–discriminator network

Abstract Shenzhen is a modern metropolis, but it hides variety of valuable cultural heritage, such as ancient murals. How to effectively preserve and repair the murals worthy discussion question. Here, we propose generation-discriminator network model based on artificial intelligence algorithms perform digital image restoration damaged In adversarial learning, this study optimizes discriminative model. First, real mural images are spliced together input discriminator network. The uses 5-layer encoder unit down-sample 1024 × 3 32 256. Then, connect layer ZeroPadding2D expand 34 256, pass Conv2D layer, 31 batch normalization, repeat above steps get 30 1 matrix. Finally, part loss emphasized in function needed improve texture detail information generated by Generator. experimental results show that compared with traditional algorithm, PSNR value algorithm proposed paper can be increased 5.86 db at most. SSIM 0.13. Judging from subjective vision. dot-like damage complex structures. may helpful for murals, also provide reference workers.