Texture and geometric feature-fusion-based network for Dunhuang mural inpainting

Yutong Hou, Shiqiang Du, Huaikun Zhang, Jizhao Liu, Jinying Liu, Jing Lian

Abstract: Highlights•We proposed a dual-network framework for high-quality Dunhuang mural inpainting.•We designed PIN to recover structure and REN to refine details for better accuracy.•We presented a Mamba-based encoder to boost PIN’s feature quality and efficiency.•We proposed DMSFM to fuse texture and geometry with improved semantic flexibility.•Our method performs robustly on various mural styles and degradation scenarios.