WCSE 2022
ISBN: 978-981-18-3959-7 DOI: 10.18178/wcse.2022.06.030

Mathematical Expression Character Recognition Based on Object Detection

Chunmei Xing, Jun Yue, Chunjie Zhou

Abstract—The recognition accuracy of mathematical characters is one of the key factors affecting mathematical expression recognition. The Traditional mathematical expression recognition consist of three stages which are character segmentation, character recognition and structural analysis. Such methods are cumbersome and with the error accumulation. To solve this problem, a method to use object detection technology combining the two steps of character segmentation and character recognition in expression recognition is proposed in this paper, by which the expression is simplified and the accuracy of character recognition is improved. Firstly, a dataset (PME) containing 130 character categories and a total of 2000 formula pictures is established; and then, the SE (squeeze and excitation) block is integrated into the backbone network (CSPDarkNet53) to perform feature recalibration; finally, three sizes feature maps are obtained to predict different size characters respectively. The experiments results show that mAP (mean average precision) is 98.62% on PME, which is 1.52%, 8.95% and 4.5% higher than that of benchmark model YOLOv4, contrast model UP-DETR and BiDet, which proves the effectiveness of this method.

Index Terms—mathematical character recognition, CNN, attention, object detection

Chunmei Xing, Jun Yue, Chunjie Zhou
School of Information and Electrical Engineering, Ludong University, Yantai, 264025, CHINA

[Download]


Cite:Chunmei Xing, Jun Yue, Chunjie Zhou, "Mathematical Expression Character Recognition Based on Object Detection, " Proceedings of 2022 the 12th International Workshop on Computer Science and Engineering (WCSE 2022), pp. 212-217, June 24-27, 2022.