Recognition of handwritten bilingual Characters-Numerals using shape context

This paper presents a methodology for recognition of handwritten Marathi and English Characters-Numerals using shape context descriptor. During pre-processing an algorithm is developed to extract the Marathi and English Characters-Numerals form grid formatted datasheets. The corresponding sample points around the boundary of a character are computed. This is followed by obtaining the centroid of the input image. Finally shape context is computed and shape context cost is used to minimize the matching distance between training images and test images. We have designed a graphical user interface for recognition of Characters-Numerals and obtained recognition accuracy in a range of 83% to 95%.