Efficient Sanskrit Word Recognition Using Segmentation and Dual Feature Extraction Techniques

Development of a Character recognition system for Devanagri is difficult because there are about 350 basic, modifier (“matra”) and compound character shapes in the script and the characters in words are topologically connected. A feature based on the combination of gradient feature and coefficients of wavelet transform is developed in this paper. In handwritten word recognition, the gradient feature represents local characteristics properly, but it is so sensitive to deformation of handwritten character. Meanwhile, wavelet transform represents the character image in multiresolution analysis and keep adequate global characteristic in different scales. In order to improve the discrimination power, we composed both local and global characteristic in a combined feature. The combination schemes are described in this paper.