Structural and Statistical Feature Based Multistage Recognition Approach for Handwritten Devanagari Script Recognition

Abstract

Devanagari is one of the basic Script widely used all over the India . Many Indian Languages Like Hindi, Marathi, Rajasthani etc are based on Devanagari Script. Devanagari Scripts Hindi language is third common language used all over the word. In this paper we propose statistical and structural method based feature extraction and an artificial neural network based classifier for the recognition of handwritten Devanagari characters . Optical isolated Hindi Characters are taken as an input image from the scanner. An input image is preprocessed and segmented in terms of various structural and statistical features like end points, middle bar, loop, end bar, aspect ratio etc.The feature vector is applied to Self organizing map which is one of the classifier of an artificial neural Network. Self organizing map is trained for such 500 different characters collected from 500 persons. The characters are classified into three different classes. The proposed classifier attains 93% accuracy.