Recognition of Similar Shaped Handwritten Characters Using Logistic Regression

Recognition of similar shaped characters is a difficult problem and in character recognition systems most of the errors occur in similar shaped characters. In this article we propose a generic method to differentiate between two similar shaped characters, which works well not only when the characters are rotated about its center, but also in the presence of noise. Rotation is taken care of by contour distance based approach and recognition is done based on logistic regression. We consider a training data set to estimate the parameters of the logistic model, and using these parameters we classify the test object. We have considered pairs of similar shape characters of Bengali script for testing our algorithm.