Text Detection and Character Recognition of Oracle Original Topographies Based on PP-YOLOE+
Xiang Li, Mingqi Zhang, Xiaoman Du · 2024
Oracle is the earliest systematic script found in China, and the research of oracle bone information processing is a hot issue nowadays. In this paper, based on the traditional digital image processing method and PP-YOLOE+ target detection method, the oracle text detection and character recognition model for the oracle original topographies is obtained through training. The innovative application of oracle bone digitization project is studied in depth from the text detection segmentation and text recognition technology of the original oracle bone topography respectively. In this paper, the models obtained after training are evaluated separately and the accuracy and practicality of the models are tested in the public dataset. Among them, the value of the mAP of the detection model is 0.550, and the value of the mAP of the recognition model is 0.923. Finally, this paper develops a real-time detection system for oracle bone characters, which detects the oracle bone characters on the original topographies of oracle bones in real time through the external camera device, and the real-time detection has an average FPS of 35.63, which provides a new technological method to support the further research of oracle bone digitization engineering.