A Method of Symbol Segmentation Based on Distance Between Strokes for On-line Recognition of Handwritten Mathematical Formulas

Abstract

This paper describes a new method for on-line recognition of handwritten mathematical formulas using distances among strokes in symbol segmentation process. Segmentation methods optimizing the results of symbol recognition over whole formulas are often applied before structural analysis of handwritten mathematical formulas. Though improvements on symbol segmentation using geometrical features of strokes are reported in related previous works, they are not sufficient for recognizing handwritten mathematical formulas without restrictions such as input order. To remove the restrictions, we employ a correction value for symbol segmentation derived from a confidence value of stroke combination. This confidence value is calculated by counting up a pair of distances between strokes on a lot of actual handwritings of mathematical formulas in the learning phase. The experimental result shows effectiveness of a correction value based on a confidence value of stroke combination for symbol segmentation. The proposed method reduced 40% of segmentation errors in comparison with the previous method. More improvements will be realized to make use of detailed information of strokes and mathematical structure for symbol segmentation.