Use of gradient technique for extracting features from handwritten Gurmukhi characters and numerals

Abstract

In this manuscript a recognition system for offline handwritten Gurmukhi characters and numerals using gradient information as mode of feature extraction technique is proposed. Two ways of extracting features using gradient information are explained in this paper. Both methods operate by accumulating gradient information from an image by dividing it into sub-images (blocks) and finally concatenating the obtained gradient features obtained from each block to form a vector of feature values with dimensionality 200. The efficiency of this feature vector is tested on two separate handwritten databases of Gurmukhi characters and Gurmukhi numerals containing 7000 & 2000 sample binary images respectively. Recognition rates of 97.38% for database of Gurmukhi characters and 99.65% for Gurmukhi numerals are obtained. Work has also been extended to test the effectiveness of the Gradient feature extraction technique on dataset of Gurmukhi characters and numerals combined together.