Title: Artistic mural reconstruction via a GAN network based on sorting loss

Authors: Zhiqiang Chen; Jiaqi Liu; Jianfang Cao; Cunhe Peng

Addresses: Department of Big Data and Intelligent Engineering, Shanxi Institute of Technology, Yangquan, 045000, China ' School of Computer Science and Technology, Taiyuan University of Science and Technology, Taiyuan, 030024, China ' School of Computer Science and Technology, Taiyuan University of Science and Technology, Taiyuan, 030024, China; Department of Computer Science and Technology, Xinzhou Normal University, Xinzhou, 034000, China ' School of Computer Science and Technology, Taiyuan University of Science and Technology, Taiyuan 030024, China

Abstract: To address severe mural image defects and the low-resolution and rough reconstruction details of mural reconstruction methods, which can lead to a reduction in mural artistry. This paper presents an artistic reconstruction method (RSSRGAN) for murals. This method adopts the architecture of generative adversarial networks and introduces an attention mechanism. First, channel separation is performed on the feature map obtained during preliminary feature extraction, and weight prediction is performed on the 64-dimensional features to construct the channel dependence between the mural feature maps to retain the high-frequency features lost by the murals in the LR space. Finally, mural textural features are retained to improve the artistic reconstruction effect. Compared with the four popular superresolution reconstruction baseline models, the proposed method achieves a peak signal-to-noise ratio (PSNR) increase of more than 0.58 and an increase in the structural similarity index (SSIM) of more than 0.025 on the mural dataset. Moreover, public dataset verification on the DIV2K dataset showed that the method achieved good reconstruction quality, in which the PSNR increased by more than 0.27 and the SSIM increased by more than 0.014. The RSSRGAN method has achieved significant improvements in mural image reconstruction and provides a new and effective method for artistic mural reconstruction.

Keywords: mural protection; superresolution reconstruction; generative adversarial network; twin neural network; attention mechanism.

DOI: 10.1504/IJCSM.2026.151970

International Journal of Computing Science and Mathematics, 2026 Vol.23 No.1, pp.81 - 101

Received: 30 Oct 2024
Accepted: 05 Nov 2025
Published online: 28 Feb 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article