Segmentation of Handwritten Mathematical Matrices Using the Area Voronoi Diagram
Anastasiia Cherneha, Dmytro Zhelezniakov, Pavlo Tytarchuk, Vasyl Tereshchenko · 2021
Robust handwritten mathematical expressions recognition systems are expected to provide support for different mathematical structures. In this paper, a mathematical matrix segmentation method based on area Voronoi diagram is proposed. The method utilizes only input strokes geometrical data and therefore can be used as the first stage of recognition system workflow. The main steps of the method describe building area Voronoi diagram, calculating weights of the resulting diagram edges, and finding edges paths that separate matrix rows and columns. The evaluation shows that the method works adequately, but has some substantial limitations. Apart from mathematical matrices, it can be used for other tabular structures segmentation, for example, tables and charts.