Article: Bayesian-optimised multiscale image inpainting for digital preservation of murals Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.44 pp.47 - 66 Abstract: Aiming at the challenges of insufficient accuracy in complex damage repair and time-consuming and inefficient parameter adjustment in traditional methods in the digital protection of murals, this paper proposes a multi-scale image restoration algorithm based on Bayesian optimisation. By constructing a multi-scale feature fusion network to capture context information and introducing Bayesian optimisation to the hyperparameters and loss weights of the automated repair model, the adaptive and efficient repair process has been achieved. Experiments on the Dunhuang Mural dataset and public damage benchmark show that, compared with mainstream deep restoration methods, this algorithm improves the structural similarity index by an average of 2.5% and the peak signal-to-noise ratio by 0.8 dB, significantly enhancing the visual fidelity and detail restoration ability of the restoration results. It provides reliable technical support for the digital archiving and virtual restoration of large-scale murals. Inderscience Publishers - linking academia, business and industry through research

Title: Bayesian-optimised multiscale image inpainting for digital preservation of murals

Authors: Jiongyi Li; Fanzhou You

Addresses: School of Art Design, Communication University of Shanxi, Taiyuan, 030619, China ' School of Animation and Digital Arts, Communication University of Shanxi, Taiyuan, 030619, China

Abstract: Aiming at the challenges of insufficient accuracy in complex damage repair and time-consuming and inefficient parameter adjustment in traditional methods in the digital protection of murals, this paper proposes a multi-scale image restoration algorithm based on Bayesian optimisation. By constructing a multi-scale feature fusion network to capture context information and introducing Bayesian optimisation to the hyperparameters and loss weights of the automated repair model, the adaptive and efficient repair process has been achieved. Experiments on the Dunhuang Mural dataset and public damage benchmark show that, compared with mainstream deep restoration methods, this algorithm improves the structural similarity index by an average of 2.5% and the peak signal-to-noise ratio by 0.8 dB, significantly enhancing the visual fidelity and detail restoration ability of the restoration results. It provides reliable technical support for the digital archiving and virtual restoration of large-scale murals.

Keywords: digital protection of murals; image restoration; Bayesian optimisation; multi-scale network.

DOI: 10.1504/IJICT.2026.153391

International Journal of Information and Communication Technology, 2026 Vol.27 No.44, pp.47 - 66

Received: 25 Dec 2025
Accepted: 23 Jan 2026
Published online: 06 May 2026 *

Title: Bayesian-optimised multiscale image inpainting for digital preservation of murals

Keep up-to-date