Morphological hand-printed character recognition by a skeleton-matching algorithm

We study the use of mathematical morphology for handprinted character recognition. Our approach uses the morphological skeleton transform as the shape descriptor. An efficient skeletonmatching algorithm, which renders the similarity between two skeletons as a distance measure, is employed. Based on this distance measure, a character is classified by a minimum distance classifier. The morphological skeleton transform contains complete shape information and is shown as a powerful descriptor for this class of shapes. We also study the pattern spectrum as a shape descriptor for hand-printed characters. However, the pattern spectrum conveys only information about the shape/size distribution of a given object, which turns out to be not very efficient for hand-printed characters. Experimental results demonstrate the efficiency of the skeletonbased approach and the inadequacy of the pattern-spectrum-based approach.