Title: Statistical features based character recognition for offline handwritten Tamil document images using HMM

Authors: S. Abirami; V. Essakiammal; R. Baskaran

Addresses: Department of Information Science and Technology, Anna University, Chennai, India ' Department of Information Science and Technology, Anna University, Chennai, India ' Department of Computer Science and Engineering, Anna University, Chennai, India

Abstract: Offline handwritten recognition has been one of the active and challenging research areas in the field of pattern recognition. More research work has been done for English, Chinese, Arabic, Japanese languages and numerals recognition but limited for Indian scripts. With respect to Tamil language, handwriting recognition is still an open challenge due to richness of language and versatile shapes. In this paper, the problem of recognising offline Tamil handwritten characters has been addressed by using a symbol-modelling HMM. Here, we propose six different statistical features which are extracted from the character boundaries, to classify a character using symbol-modelling HMM. Data samples are collected from HP data sets pertaining to 60 Tamil characters for training. For testing purpose, ten different samples are collected for every character addressing four different varieties of writers from HP data sets to evaluate the recognition performance. An accuracy of 85% has been achieved through this system.

Keywords: handwriting recognition; feature extraction; HMM; hidden Markov model; OCR; computer vision; document images; statistical features; optical character recognition; handwritten documents; Tamil documents.

DOI: 10.1504/IJCVR.2015.072192

International Journal of Computational Vision and Robotics, 2015 Vol.5 No.4, pp.422 - 440

Received: 14 Mar 2014
Accepted: 29 Sep 2014

Published online: 20 Aug 2015 *

Full-text access for editors Access for subscribers Purchase this article Comment on this article