Research on Oracle Bone Inscriptions Recognition From Multi-Source Data Based on Deep Learning

How to carry out recognition process in a quick and effective way is important to the oracle bone inscriptions research area. In this paper, we proposed a recognition method which adopt a skeletonization process to normalize handwriting and rubbing forms of oracle bone inscriptions into the same expression. Then by using models based on CNN such as VGG19, AlexNet and ResNet34, the proposed method can simultaneously handle recognition work with mixed input. The experiment results showed that our method achieve a good performance applied in the real oracle bone inscriptions data set.