人在回路的在线手写数学公式识别方法  

Human-in-the-Loop Based Online Handwriting Mathematical Expressions Recognition

作　　者：康文惠 黄进[1] 田丰[1] 范向民 刘杰[1] 戴国忠[1] Kang Wenhui;Huang Jin;Tian Feng;Fan Xiangmin;Liu Jie;Dai Guozhong(Beijing Key Laboratory of Human-Computer Interaction,Institute of Software,Chinese Academy of Sciences,Beijing 100190;School of Computer Science and Technology,University of Chinese Academy of Sciences,Beijing 100049)

机构地区：[1]中国科学院软件研究所人机交互北京市重点实验室,北京100190 [2]中国科学院大学计算机科学与技术学院,北京100049

出　　处：《计算机辅助设计与图形学学报》2021年第11期1773-1785,1794,共14页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家自然科学基金(61802379);中国科学院前沿基础研究重点项目(QYZDY-SSW-JSC041);中国科学院青年创新促进会项目(2020113)。

摘　　要：在线手写数学公式识别面临书写字符的不确定性、数学公式结构的复杂性,以及公式书写风格因人而异等问题,特别是在公式书写中出现偶然性错误和包含复杂结构的情况下,现有的仅依赖机器的识别算法的识别准确率较低.为了解决这一问题,提出了人在回路的手写公式识别方法,该方法主要在结构分析阶段引入了人的参与,借助人对结构中歧义笔画的修改和结构补笔操作,完善和界定结构笔画和结构内笔画信息.为了评估该方法的有效性,将其与不含用户参与信息的一个基线识别方法在结构识别率和表达式识别率方面进行了对比分析.结果表明,该方法能够有效地促进用户参与到手写识别过程,同时,针对实验收集的手写数学公式数据,引入用户参与的方法能够有效地提高手写数学公式的结构和表达式识别率,分别提高了9.26%和13.99%.Online handwritten mathematical expression recognition method faces some challenges,such as the uncertainty of writing characters,the complexity of mathematical expression structure and the various written styles of expression.Existing no-human-involved recognition algorithms have low recognition accuracy,espe-cially when accidental errors and complex structures occur in mathematical expression writing.A human-in-the-loop based online handwritten mathematical expression recognition method is proposed,which mainly intro-duces human participation in the structural analysis stage to define and perfect the structural strokes or strokes within the mathematical structure,by means of human modification for the ambiguous strokes in the structure and operation of structural stroke makeup.To evaluate the effectiveness of the proposed method,the method is com-pared with a no-human-involved recognition algorithm in rate of structural analysis and expression recognition rate.The results show that proposed method can effectively promote users to participate in the process of recogni-tion,and that incorporating user in the recognition processes can improve the accuracy of recognition of structure(9.26%)and expression(13.99%)in handwritten mathematical expression.

关 键 词：人在回路 手写数学公式 草图识别 人机交互 用户反馈 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]