Symmetric Axis Based Off-Line Odia Handwritten Character and Numeral Recognition

Abstract

Automation of handwritten character recognition is one of the challenging tasks in the problem domain of document analysis. However various writing style in orientation, shape and size are the key factor which affects the offline recognition system of Indian scripts. Here we have used a set of symmetry axes which are perceptually uniquely representing the handwritten Odia characters and numerals as patterns. This empirical model generates two symmetry axes such as row symmetry and column symmetry chords. In the subsequent phase we added up the mid points of both symmetric axis and along with we have reported the angular projection and distance between centre of the image and respective midpoints. Subsequently we have taken the mean values of horizontal and vertical symmetry angular projection values along with the mean of horizontal, vertical distance as the key feature values for the recognition system. We have analyzed overall recognition system with J48 Decision Tree which is considered as a classifier. All the simulation setup was build over upon standard database of NIT RKL Odia handwritten character, ISI Kolkata Odia numeral database. A 6 fold cross validation was performed in order to validate the recognition system. After all the successful simulation work we have noted down very good promising recognition accuracy from the J48 classifier such as 96.2% accuracy upon Odia numeral database and 95.6% upon Odia character database.