Based on Delaunay triangulation

This paper describes a novel algorithm of thinning text images for fast nicety recognition. The polygonal approximation of images' boundaries is decomposed into Delaunay triangles that represents the topological features of the object. All triangles are classified into three types that generate different local skeletons which are connected to form the whole text images skeletons. Taking full advantage of text images' whole and local information

this algorithm has the advantages of fast speed and nice effect.