Research on Symbol Recognition for Mathematical Expressions

Abstract

The recognition of mathematical symbols is an important step in the mathematical expression recognition system. It consists of two stages: symbol segmentation and symbol recognition. Aiming at the fact that the commercial OCR systems cannot recognize expression symbols correctly because of the differences between mathematical symbols and ordinary characters, we do lots of research work on the recognition of printed expressions. An approach that contains feedback mechanism is proposed for symbol segmentation and recognition. Firstly, projective features and connected components labeling method are used to segment symbols in expressions. Then the peripheral features and directional line element features are extracted from symbols. Finally, a coarse-to-fine classification strategy is employed to recognize symbols with these features. The experiments show that these methods can obtain satisfactory recognition accuracy with a high speed.