MURAL RESTORING BASED ON A GENERATIVE ADVERSARIAL NETWORK WITH ENHANCED LOCAL ATTENTION MECHANISM

The mural image is rich in color, and the texture of different parts of the image is quite different. Aimed at the problems of poor color and appearance after traditional methods of restoring mural images, and poor restore effect of large-area damaged murals, a mural image restore method with enhanced local attention generative adversarial network is proposed. Redesigning on the basis of traditional generative adversarial network structure, and introducing the local attention mechanism proposed in this paper, it could better restore mural images. After digitally restoring the artificially processed mural images to be restored and the real Wutai Mountain murals, the experimental results show that the algorithm can well improve the local blur and texture loss problems caused by the deep learning method when restoring images, and has better subjective perception. Compared with other comparison algorithms, the restored murals are also better than other comparison algorithms in objective evaluation of peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) indicators.