A Two Stage Classification Approach for Handwritten Devanagari Characters

TL;DR

This paper introduces a two-stage classification method for handwritten Devanagari characters, combining structural property analysis and neural network classification, achieving 89.12% success on 50,000 samples.

Contribution

It proposes a novel two-stage approach utilizing differential distance techniques and neural networks for improved handwritten Devanagari character recognition.

Findings

Achieved 89.12% success rate on 50,000 samples

Developed a differential distance method for structural feature detection

Enhanced recognition accuracy over simple histogram methods

Abstract

The paper presents a two stage classification approach for handwritten devanagari characters The first stage is using structural properties like shirorekha, spine in character and second stage exploits some intersection features of characters which are fed to a feedforward neural network. Simple histogram based method does not work for finding shirorekha, vertical bar (Spine) in handwritten devnagari characters. So we designed a differential distance based technique to find a near straight line for shirorekha and spine. This approach has been tested for 50000 samples and we got 89.12% success