Title: Shape descriptors-based generalised scheme for handwritten character recognition

Authors: Tusar Kanti Mishra; Banshidhar Majhi; Ratnakar Dash

Addresses: Department of Computer Science and Engineering, Pattern Recognition Research Laboratory, National Institute of Technology Rourkela, India ' Department of Computer Science and Engineering, Pattern Recognition Research Laboratory, National Institute of Technology Rourkela, India ' Department of Computer Science and Engineering, Pattern Recognition Research Laboratory, National Institute of Technology Rourkela, India

Abstract: In this paper, we propose a novel scheme for recognition of handwritten numerals for a regional language Odia of the Indian continent. Additional attempts have also been made to implement this scheme for recognition of handwritten numerals of two other languages namely, Bangla and English. Thus, the proposed scheme has been generalised to three different languages. Three variants of time series description of global shapes of numerals have been wrapped up in a vector. This vector is treated as the primary features for the suggested scheme. Satisfactory overall accuracy rate of 96.25% is achieved for Odia numerals. Promising results are also obtained for recognising English and Bangla numerals.

Keywords: optical character recognition; OCR; shape descriptors; Odia; feature extraction; character recognition; handwritten characters; handwritten numerals; Bangla; English.

DOI: 10.1504/IJCVR.2016.073765

International Journal of Computational Vision and Robotics, 2016 Vol.6 No.1/2, pp.168 - 179

Received: 03 Jan 2014
Accepted: 03 Feb 2015
Published online: 18 Dec 2015 *

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