Off-line Unconstrained Handwritten Numeral Character Recognition with Multiple Hidden Markov Models

A. Namane (France, Algeria), M. Arezki, A. Guessoum (Algeria), E.H. Soubari, P. Meyrueis, and M. Bruynooghe


Hidden Markov Model (HMM), handwritten numeral, character recognition, slant correction, multiple classifier


This paper presents a multiple classifier for offline unconstrained numeral characters recognition based on a multiple hidden Markov model (HMM). A new feature extraction method based on contour background transition is presented. A horizontal slant correction method using the lower and upper centroid of the handwritten numeral is proposed. Classification is made using individual classifiers (IC) based on HMM models as wall as combining the four IC based on HMM models for each class using two decision strategies; equal combination weight (ECW) and unequal combination weight (UCW). The combined methods yield best results comparing to individual classifiers. Experimental results are conducted with number of training set, various state number and iteration number are presented.

Important Links:

Go Back