A hybrid feature extraction scheme for Off-line English numeral recognition

Abstract

This paper aims at presenting a rotation invariant feature extraction scheme to support well known result oriented recognizer HMM. Hybrid feature extraction method consists of features due to moment of inertia (FMI) and projection features. Projection features have been applied in case of digits (2 and 3) and for other numerals FMI is introduced. Any recognition system consists of two major components viz. Feature extraction method and recognizer. This paper uses Hidden Markov Model (HMM) as recognizer to recognize Off-line handwritten English numerals due to its inherent specialities and promising results in automatic speech recognition. Our data-base consists of own collected data from people of different ages and CENPARMI data. The percent recognition accuracy of self collected samples and CENPARMI samples have been found to be 91.7% and 91.2% respectively.Index Terms-Numeral recognition, HMM, features due to moment of inertia (FMI), projection features, Viter-bi algorithm, Baum-welch algorithm.

An Investigation of Novel Combined Features for a Handwritten Short Answer Assessment System

“…The comparison in this study was mainly performed with other off-line word recognition techniques found in the literature since there are not many studies sourced regarding SAAS. It can be observed that when comparing experiment results with other recognition systems [4]- [8], the proposed WRL_MDGGF and G_WRL_MDGGF, as well as the recently proposed G_GGF combined feature extraction techniques, were able to attain considerably high to comparable recognition and accuracy rates compared to the existing systems in Table IX. It could be noted however, that it was difficult to compare due to the dataset sizes and the nature of the words/characters/numerals utilised (e.g.…”