Scholarlink

Mathematical Optical Character Recognition (OCR) extracts math equation text from images and scanned documents so that it can be edited, formatted, indexed, searched, or translated. Recognition of mathematical symbols plays a significant role in the pipeline of math OCR. Compared to the natural language text, recognizing math symbols is more challenging due to their complex structures. Deep architectures have the potential to give the best recognition performance, however, they are data hungry. In this paper, we explore all the existing handcrafted features of handwritten and printed text recognition and propose a novel two-stage feature selection algorithm for selecting the best subset of features to recognize math symbols. To the best of our knowledge, feature selection is not applied to math symbol recognition. Our method performs remarkably well on limited datasets, achieving results comparable to deep learning, which typically requires large amounts of data for mathematical symbol recognition.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Efficient Hybrid Feature Selection Approach for Offline Handwritten Mathematical Symbol Recognition

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ridhi Aggarwal,
  • Shilpa Pandey,
  • Anil Kumar Tiwari,
  • Gaurav Harit

摘要

Mathematical Optical Character Recognition (OCR) extracts math equation text from images and scanned documents so that it can be edited, formatted, indexed, searched, or translated. Recognition of mathematical symbols plays a significant role in the pipeline of math OCR. Compared to the natural language text, recognizing math symbols is more challenging due to their complex structures. Deep architectures have the potential to give the best recognition performance, however, they are data hungry. In this paper, we explore all the existing handcrafted features of handwritten and printed text recognition and propose a novel two-stage feature selection algorithm for selecting the best subset of features to recognize math symbols. To the best of our knowledge, feature selection is not applied to math symbol recognition. Our method performs remarkably well on limited datasets, achieving results comparable to deep learning, which typically requires large amounts of data for mathematical symbol recognition.