LanT: finding experts for digital calligraphy character restoration

Ancient calligraphy manuscripts often suffered from degradation due to human factors and natural weathering. Image restoration techniques restore the degraded contents in calligraphy character images to help text-related research and have great significance for cultural heritage protection. Traditional manual calligraphy restoration relies heavily on professional skills and identifying experts for this process is time-consuming. Automatic digital methods do not usually require experts, which have an unsatisfactory performance for severely degraded calligraphy works restoration. In this paper, we first propose a crowdsourcing-based method to identify experts by a novel character similarity inference (CSI) algorithm to restore severely degraded calligraphy characters. Then, we develop LanT, a system to implement crowdsourcing tasks of restoration. We performed an evaluation on restoring a famous Chinese calligraphy work. The experimental results show that CSI can reliably identify experts. Additionally, we evaluate the usability of LanT and the generalization of CSI for Latin calligraphy character restoration.