A feature fusion based optical character recognition of Bangla characters using support vector machine

Abstract

This paper introduces a feature fusion based optical Bangla character recognition approach using Support Vector Machine (SVM) including the feature extraction method named Zoning and Gabor Filter. Optical character recognition (OCR) is mostly dependent on the features extracted from characters which highly influences the classification and recognition accuracy. This paper emphasizes on the feature fusion of two different feature vectors obtained by Zoning and Gabor filter of suitable frequencies and orientations along with finding the recognition rate by passing the features into a well-developed classifier, Support Vector Machine (SVM) for classification. Here, a comparison has also been made among the recognition accuracy by individual features and by feature fusion which reveals that feature fusion based method performs better (92.99%) than a single feature extraction method (68.15% for Zoning, 89.73% for Gabor filter) during classification.