A novel triangulation procedure for thinning cursive text

Abstract

Thinning or skeletonization can contribute tremendously to efficient feature extraction and classification. This paper describes a novel thinning procedure which decomposes a polygonal approximation of the inner and outer boundaries of a word image into a set of triangles. These triangles are classified into three types and replaced either by three lines at branches or by single lines to form a completely connected skeleton which is one pixel wide. The procedure also automatically generates a graph representation of the skeleton for use by subsequent analysis steps. The procedure has been tested on a variety of Arabic and English handwritten words and found to produce very few spurious features. The speed of this method does not depend strongly on the resolution of the digital image unlike commonly used thinning methods which iteratively remove layers from the boundary of words. In addition, it can be used for thinning any line-like patterns such as fingerprints, chromosomes and line drawings.KeywordThinningskeletontriangulationpolygonal approximationhandwritingcursive Arabic text