Nearest Neighbor based Digital Restoration of Damaged Ancient Chinese Paintings

Abstract

Most ancient artworks have structure damage problems, such as tears, flakes and cracks. This work gives an initial study of digital restoration of damaged ancient Chinese paintings via nearest neighboring method, which is an effective non-parameter machine learning algorithm. We first present a damage detection method to estimate the mask, and then a patch based image inpainting algorithm is performed to reconstruct the damaged paintings. Our experiments illustrate the restoration performance of the proposed method by using ancient Chinese paintings. Moreover, we also provide discussion on the future research topics about automatic damage detection and image inpainting via deep learning algorithms.