A New Dunhuang Murals Inpainting Method with Adaptive Selection Module

The Dunhuang murals represent the world's largest and best-preserved collection of grotto art. However, prolonged exposure to environmental factors may lead to fading, deformation, and damage, and the preservation and restoration of these murals is a critical task. Existing restoration methods often fail to achieve optimal results due to inadequate consideration of the distinct properties of structural and textural elements in the murals. This paper proposes a novel inpainting method for Dunhuang murals, featuring an adaptive selection module designed to utilize both structural and textural information effectively. Our approach involves dividing the restoration area into texture and structure regions and employing a restoration strategy that prioritizes texture restoration followed by structural restoration. For each region, we introduce a patch texture similarity function and a patch structure complexity function. These functions derive a priority function, enabling the adaptive selection of blocks. Additionally, we have developed new quality evaluation functions that facilitate the adaptive selection of parameters, further enhancing the adaptive selection module. The experimental results demonstrate that our method outperforms several leading algorithms, offering superior restoration quality for the complex structural and textural information characteristic of Dunhuang murals.