IMEGE: Image-based Mathematical Expression Global Error

Abstract

Mathematical expression recognition is an active research eld that is related to document image analysis and typesetting. Several approaches have been proposed to tackle this problem, and automatic methods for performance evaluation are required. Mathematical expressions are usually represented as a coded string like LATEX or MathML for evaluation purpose. This representation has ambiguity problems given that the same expression can be coded in several ways. For that reason, the proposed approaches in the past either manually analyzed recognition results or they reported partial errors as symbol error rate. In this study, we present a novel global performance evaluation measure for mathematical expression based on image matching. In this way, using an image representation solves the representation ambiguity as well as human beings do. The proposed evaluation method is a global error measure that also provides local information about the recognition result.