Handwritten Devanagari numeral and vowel recognition using invariant moments

Abstract

Devanagari is an alphabetic script which is used by different Indian languages such as Marathi, Hindi, Konkani and Nepali. This script consists of 13 vowels, 34 consonants and 10 numerals. Due to unconstrained shape and variation in writing style, recognizing such handwritten script is challenging task. This paper proposed a system for recognizing handwritten numerals and vowels of Devanagari Script. An Invariant Moment and Affine Moment Invariant techniques are used for extracting features from handwritten samples. 2000 samples of numerals are collected from 20 different people having variations in writing style. Also, 1250 samples of vowels are taken from 25 people. Each sample is normalized into 40 × 40 pixel size. As a classification technique, Support Vector Machine is used for handwritten numerals and Fuzzy Gaussian Membership function is applied for identifying handwritten vowels. These methods of feature extraction and classification produce more accurate results. Success rate is 99.48% and 94.56% for handwritten Devanagari numerals and vowels respectively.