Handwritten numerical recognition based on multiple algorithms

Abstract

In this paper, the authors combine two algorithms for application to the recognition of unconstrained isolated handwritten numerals. The first algorithm employs a modified quadratic discriminant function utilizing direction sensitive spatial features of the numeral image. The second algorithm utilizes features derived from the profile of the character in a structural configuration to recognize the numerals. While both algorithms yield very low error rates, the authors combine the two algorithms in different ways to study the best polling strategy and realize very low error rates (0.2% or less) and rejection rates below 4%. Character recognition Statistical pattern recognition Bayes classifier Quadratic discriminant function Structural pattern recognition Combined character recognition algorithm

RISCBOT: A WWW-Enabled Mobile Surveillance and Identification Robot

“…In [4], Sridhar et al describe a collection of topological features that can be used to classify numerals. Most of these features are properties of the outline or profile, of the numeral.…”

Ensembles of Learning Machines

“…If only labels are available or if continuous outputs are hardened, then majority voting, that is the class most represented among the base classifiers, is used [67,104,87].…”

Rank-Based Multiple Classifier Decision Combination: A Theoretical Study

“…Contributions have been made or some form of classifier combination system have been attempted, among others within the fields of machine printed word/character recognition [15], handwritten character recognition [29,16,3,23,26,18,17], speaker recognition, [11,6,10,22], face identification [1,8], text to phoneme translation [28], remote sensing [4,5], military target recognition [9] and biomedical image processing [19]. The neural networks community has also been active on this subject [27,3,21,28,2,20,13].…”