SFF-Siam: A New Oracle Bone Rejoining Method Based on Siamese Network.

Abstract

The rejoining of oracle bone rubbings is a fundamental topic in oracle bone inscriptions (OBIs) research. However, the traditional oracle bone (OB) rejoining methods are not only time-consuming and laborious but difficult to apply to large-scale OB rejoining. We proposed a simple OB rejoining model (SFF-Siam) to handle this challenge. First, the similarity feature fusion module (SFF) is designed to combine two inputs and make them relate to each other, then a backbone feature extraction network is used to evaluate the similarity between inputs, and the forward feedback network (FFN) outputs the probability that two OB fragments can be rejoined. Extensive experiments demonstrate that the SFF-Siam achieved a good effect in OB rejoining. The average accuracy of the SFF-Siam network reached 96.4 % and 90.1 % in our benchmark datasets, respectively. It provides valuable data for promoting the use of OBIs in conjunction with AI technology.