Dual-View Oracle Bone Script Recognition System via Temporal-Spatial Psychovisual Modulation

Abstract

Dual-view oracle bone script recognition is a useful technique for the research and application of the oracle bone script. This paper introduces a dual-view oracle bone script recognition system combining the character recognition algorithm and temporal-spatial psychovisual modulation (TPVM). There are two main parts in our system. In the oracle bone script recognition part, we have two types of recognition targets, the oracle bone inscription images and the handwriting oracle character images. The inscription images have much noise from the erosion and the scratch of tortoiseshell or the stele, which cannot be separated from the character. So the current recognition methods cannot achieve high accuracy on the oracle bone inscription recognition tasks. We adopt the SIFT feature points for the inscription recognition. The accuracy of this method is 95.6% in our experiments. Handwriting oracle character recognition is helpful to popularize the oracle bone script knowledge and ancient Chinese culture, but no research has been done on it. We build a handwriting oracle character database and use convolutional neural networks for handwriting oracle character recognition. Our methods can achieve a high recognition accuracy. In the dual-view display part, we adopt temporal and spatial psychovisual modulation for different application scenarios. Without specific glasses, people see the oracle bone inscription. When wearing specific glasses, viewers see the the Chinese character corresponding to the oracle bone inscription.