An improved algorithm for symbol segmentation of mathematical formula images
Haiyan Wang, Yu Wang, Liying Lu, Jianhang Liu, Shibao Li, Yang Zhang · 2016
Due to the complex species, the changing dimensions and 2D nesting structure of mathematical formula symbols, the accuracy of symbol segmentation still cannot meet the actual needs. Projection is only suitable for simple mathematical formula without subscript and hierarchy. This paper presents a kind of improved algorithm based on the connected domain for symbol segmentation of mathematical formula images. At first, the image of mathematical formula is pre-processed, including gray scale processing, adaptive median filter, image thinning and other operations. Then the segmentation of mathematical formula image is studied and a segmentation method based on improved connected domain presented. This method makes full use of the inherent connected features of the symbols to obtain connected domain, then combine them according to the structural characteristics of symbols, finally achieves a precise character segmentation effect on segmenting the mathematical formula symbols with complex 2D nesting structure.