Evaluation of Missing Image Restoration with a Binary Character Image Diffusion Model

Ancient Japanese literature is a Japanese cultural heritage with high historical value, and digital preservation increases opportunities for secondary use and prevents loss of cultural heritage due to natural or human causes. However, they are subject to various types of deterioration over time and the preservation environment. In particular, loss of text is seen as a challenge from the perspective of ensuring data quality for digital preservation. In this study, masks of various proportions are applied to publicly provided binary ancient text images to simulate missing characters. In this study, a diffusion model is used to embed the character image in the masked region to facilitate its generation, and the accuracy of the recovery is compared with GAN.