Handwritten English Digit Recognition: A Machine Learning Formulation

Abstract

Handwriting recognition is a challenging machine learning task. Handwritten Recognition (HR) systems have become commercially popular due to their potential applications. The challenges that arise due to wide range of variations in shape, structure ,size and individual writing style can be handled with the combination of a powerful feature extraction technique and an efficient classifier. In this paper, an attempt has been made to compare four different feature extraction cum classifier schemes for English handwritten numeral recognition in terms of computational time and accuracy of recognition. Observations show that single decision tree requires less computation time while SVM yields better accuracy.