Isolation of Dots for Arabic OCR using Voronoi Diagrams

Abstract

In Arabic script, most characters share the same primary body and are differentiated solely by the location and number of dots, e.g. ج , ح and خ. At a certain stage, such as segmentation or feature extraction, it is important to temporary eliminate the dots. In this paper, a method based on area-Voronoi Diagram to separate Arabic dots from the main Arabic word body is proposed. Area-Voronoi diagram has the capability of representing the neighbourhood of connected components as polygons. The inner and outer contours of all word's components are traced first to choose samples using an estimated writing stroke thickness. The point-Voronoi diagram is constructed from those samples, and then the area-Voronoi diagram is created based on the point-Voronoi diagram. Area-Voronoi diagram will draw line segments between the connected components. The complete set of those lines will be able to separate the dots from the main body. The method is perfect in separating those components. The time consumed is also optimized by setting the sampling interval to be equivalent to the estimated thickness value. This choice proved to be perfect because the area-VD constructed does not differ from the one constructed using all boundary points. Hence, gaining time and preserving the structure from any distortion. The method was tested on a variety of printed and handwritten Arabic documents and showed promising results.