On Extraction of Features for Handwritten Odia Numeral Recognition in Transformed Domain

Recognition of handwritten scripts has always been a challenging task before the character recognition community. The difficulty lies in the fact that different individuals have different writing styles and hence there is a lot of intra-class pattern variation. Several feature extraction techniques based on statistical, structural properties have been reported in literature. We, in this paper, propose a number of image transformation based feature extraction techniques such as, Slantlet transform based, Stockwell transform based, and Gabor-wavelet based transformed domain features for offline Odia handwritten numeral recognition. The performances of the proposed methods are evaluated on ISI Kolkata Odia numeral database with a nearest neighbor classifier and the recognition accuracies are reported.