كاربرد تطابق شكل در بازشناسي ارقام دستنويس فارسي

Recognition of Handwritten Farsi Digits by Shape Matching

In this paper, we used a shape matching algorithm to recognize Farsi digits. For each sampled point on the contour of a shape, we obtain a descriptor showing the distribution of the other points of the contour, with respect to this point. Based on these descriptors, we find the corresponding points of the two contours and take the sum of their distances as a dissimilarity measure between two shapes. Then we define a geometric transformation that maps the sampled points of the one shape to the corresponding points of the other shape. The bending energy of this transform is taken as the second dissimilarity measure between two shapes. We optimized the parameters of the matching algorithm for the recognition of Farsi digits and used the method of minimum distance from the class prototypes for the recognition. In a test on a set of 1288 digits, we obtained a recognition rate of 89.9%. This result was obtained without any post processing.