Combining Structural and Statistical Approach to Online Recognition of Handwritten Mathematical Formulas

This paper introduces a novel method for online recognition of handwritten mathematical formulas. The method is based on the combination of a structural analysis with a statistical model specifying relations of individual symbols. A description of all recognition phases is given, focusing mostly on the structural analysis stage. The recognition process following a bottom-up manner is driven by a spatial grammar, which expresses mathematical formulas unambiguously. As the grammar describes the relative positions only roughly, a statistical model learned from a database of annotated formulas is employed to refine the description. Several experimental results are also reported.