Virtual Inpainting for Dazu Rock Carvings Based on a Sample Dataset

A novel semi-automatic exemplar-based inpainting framework leverages a sample dataset to effectively restore severely damaged art images, such as the Dazu Rock Carvings, where minimal prior information is available. The approach extends the search space for candidate patches beyond the known region of the original image, addressing challenges in traditional inpainting methods. By integrating deep convolutional networks for reference image selection and melding algorithms for sample image creation, this method reduces complexity while avoiding duplication issues in art restoration. Poisson blending further enhances visual harmony between reconstructed fragments and existing regions. Applied to Buddhist face images from Dazu, the framework demonstrates its potential as a reference for both actual artificial inpainting and virtual reality presentations.

Wang, H., He, Z., Chen, D., Huang, Y., & He, Y. (2019). Virtual Inpainting for Dazu Rock Carvings Based on a Sample Dataset. Journal on Computing and Cultural Heritage, 12(3), 1–17. https://doi.org/10.1145/3303767