Discrete Curve Evolution Based Skeleton Pruning for Character Recognition

This paper deals with the recognition of handwritten Malayalam characters using discrete features. The features are extracted from skeletonizsed images. But the presence of parasitic components in the image will degrade the performance of the pattern recognition system. So there arise needs for a pruning method to produce skeletons that are in accordance with human visual perception. The skeleton pruning by contour portioning with discrete curve evolution (DCE) showed that it never produce spurious branches. Moreover, this method doesn’t displace skeleton points. Consequently, all skeleton points are centers of maximal disks. Even in the presence of significant noise and shape variations, this approach gave same topology as that of original skeletons. As a result, we have obtained excellent results in feature extraction which in turn gave a better recognition accuracy of 90.18 percent for 33 classes.