Skeletonization of Arabic Characters Using a Neural Network Mapped on Maspar

In this paper, we employ a back-prop neural network for skeletonization of Arabic characters to be used in an Arabic character recognition system. The neural network is trained to learn the thinning algorithm which is a variation of the parallel algorithm proposed by Zhang and Suen. Zhang and suen’s algorithm was modified so that dots are thinned into single pixels. The neural network as well as the binary image are mapped on a massively parallel machine (Maspar) in order to speed up the skeletonization process.