A novel triangulation procedure for thinning hand-written text

This paper describes a novel procedure for thinning binary text images by generating graphical representations of words within the image. A smoothed polygonal approximation of the boundaries of each word is ®rst decomposed into a set of contiguous triangles. Each triangle is then classi®ed into one of only three possible types from which a graph is generated that represents the topological features of the object. Joining graph points with straight lines generates a ®nal polygon skeleton that, by construction, is one pixel wide and fully connected. Results of applying the procedure to thinning Arabic and English handwriting are presented. Comparisons of skeleton structure and execution time with results from alternative techniques are also presented. The procedure is considerably faster than the alternatives tested when the image resolution is greater than 600 dpi and the graphical representation often needed in subsequent recognition steps is available without further processing.