Training Tangent Similarities with N-SVM for Alphanumeric Character Recognition

Hassiba Nemmour,Youcef Chibani

doi:10.5923/j.ajsp.20110101.06

Abstract

This paper proposes a fast and robust system for handwritten alphanumeric character recognition. Specifical- ly, a neural SVM (N-SVM) combination is adopted for the classification stage in order to accelerate the running time of SVM classifiers. In addition, we investigate the use of tangent similarities to deal with data variability. Experimental analy- sis is conducted on a database obtained by combining the well known USPS database with C-Cube uppercase letters where the N-SVM combination is evaluated in comparison with the One-Against-All implementation. The results indicate that the N-SVM system gives the best performance in terms of training time and error rate.

Highlights

In various applications of document analysis, alphanumeric character recognition constitutes a very important step
Hassiba Nemmour et al.: Training Tangent Similarities with neural SVM (N-SVM) for Alphanumeric Character Recognition character recognition, which is a problem of 36 classes (10 digits and 26 uppercase letters), we develop 18 SVMs designed to independently separate pairs of classes
Experiments are conducted on a dataset obtained by combining the well-known USPS handwritten digits with a set of cursive uppercase letters extracted from the C-Cube1 (Cursive Character Challenge) database

Summary

Introduction

In various applications of document analysis, alphanumeric character recognition constitutes a very important step. The fact that characters are written in various manners by different scripts and different tools conducts to high similarity between some uppercase letters and digits such as Z and 2 or O and 0. We investigate their use for solving an alphanumeric character recognition which aims to discriminate uppercase Latin letters from digits. It is obviously known that such application handles commonly large scale databases whereas the SVM training time is quadratic to the number of data. For this reason, we investigate the applicability of N-SVM combination for alphanumeric classification.

Objectives

Results

Conclusion