Year: 2026
Type: article
Abstract: Northern Chinese Rock Art is an important cultural heritage resource that may record ancient life, ritual activities, hunting scenes, totemic symbols, and early belief systems. However, long-term natural erosion, weathering, pigment fading, cracking, and human interference have caused varying degrees of deterioration, creating an urgent need for non-contact digital restoration. Traditional restoration methods of recordings often produce blurred textures, discontinuous structures, or visually inconsistent results when dealing with complex damaged regions. To address these challenges, this study proposes IntraFormer, a region-aware transformer-based restoration method for Northern Chinese rock art images. The proposed model combines a five-stage transformer backbone for global contextual modelling with a region-aware module for adaptive enhancement of damaged and structurally important regions. Qualitative and quantitative experiments demonstrate that IntraFormer achieves superior restoration performance compared with representative image inpainting methods, obtaining a PSNR of 25.34, SSIM of 0.82, and LPIPS of 0.22. These results indicate that the proposed method provides an effective computational approach for the digital preservation and virtual presentation of Northern Chinese Rock Art.
Source: Rock Art Research
DOI: 10.69978/fwx8wt98
Institutions: Northwest University, Xijing University, Xizang Minzu University
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: diamond
OpenAlex ID: W7225462824