RICH: Coin data restoration using inpainting for cultural heritage analysis

Abstract The rapid development of artificial intelligence technology has expanded its use in the restoration of cultural heritage. Although research on the digital restoration of ancient coins is well established, most studies have focused on Western currencies—such as coins from the Roman Empire and Greece—with little prior research being conducted on ancient Korean coins. This study uses Sangpyeongtongbo coins—which, among ancient currencies, were widely circulated throughout Korea—as its subject. For this research, a new dataset is constructed using Sangpyeongtongbo specimens from museums and research institutes. The Restoration using Inpainting for Cultural Heritage (RICH) method is developed based on stable diffusion and attempted digital restoration, to successfully recover damaged Chinese inscriptions on the surface of coins. To validate the restored image, image analysis of the undamaged and restored inscriptions is conducted, with the results demonstrating similarity rates of 97.10% for entire inscriptions and 83.81% for restored areas. These results demonstrate the feasibility of using artificial intelligence for digital restoration.