Oracle Bone Script Intelligent Recognition: Automatic Segmentation and Recognition of Original Rubbing Single Characters

Oracle Bone Script is the earliest mature writing system discovered in China to date. It is the source of Chinese characters and the root of excellent traditional Chinese culture, playing a significant role in promoting and inheriting Chinese culture. Therefore, the automatic recognition and extraction of Oracle Bone Script characters have become an important research topic. However, due to the diversity of characters, irregular arrangement, and complex background interference, the automatic extraction of Oracle Bone Script faces many challenges. Traditional image processing techniques struggle with the complexity of Oracle Bone Script rubbings. Moreover, the scarcity of Oracle Bone Script images puts deep learning methods in a predicament of insufficient training data. This paper focuses on the research of segmentation and recognition of single characters in Oracle Bone Script rubbings. Firstly, image processing algorithms are used to normalize, expand, and denoise the Oracle Bone Script image dataset, building a preprocessing model for Oracle Bone Script images. Then, the YOLOv5 model is used as the baseline model for transfer learning to modify and debug the model, training a segmentation model for Oracle Bone Script single characters. Compared to existing models for Oracle Bone Script extraction, our model has faster training efficiency and higher extraction accuracy.