A string matching based algorithm for performance evaluation of mathematical expression recognition

Abstract

In this paper, we have addressed the problem of automated performance evaluation of Mathematical Expression (ME) recognition. Automated evaluation requires that recognition output and ground truth in some editable format like [InlineMediaObject not available: see fulltext.], MathML, etc. have to be matched. But standard forms can have extraneous symbols or tags. For example, < mo > tag is added for an operator in MathML and \begin{array} is used to encoded matrices in [InlineMediaObject not available: see fulltext.]. These extraneous symbols are also involved in matching that is not intuitive. For that, we have proposed a novel structure encoded string representation that is independent of any editable format. Structure encoded strings retain the structure (spatial relationships like superscript, subscript, etc.) and do not contain any extraneous symbols. As structure encoded strings give the linear representation of MEs, Levenshtein edit distance is used as a measure for performance evaluation. Therefore, in our approach, recognition output and ground truth in [InlineMediaObject not available: see fulltext.] form are converted to their corresponding structure encoded strings and Levenshtein edit distance is computed between them. © 2014 Indian Academy of Sciences.