Performance evaluation of classifiers applying directional features for Devnagri numeral recognition

Abstract

Handwriting recognition is a special category of pattern recognition which is matured enough for English language, but for Hindi it is in development state. Among various features directional features found to outperform than the others. So in this paper, we have evaluated the performance of various direction features and various classifiers for the handwritten Devnagri numeral recognition. The character image is preprocessed and portioned into sub-images. The standard zoning is compared against flexible zoning. An experimental comparison of gradient features and chain code histogram feature is evaluated with Bays classifier, k-nn, fuzzy k-nn. For comparison of the performance, the error rate and complexity of computation and time is used as the measure. Gradient features are found to outperform among various directional features. © (2012) Trans Tech Publications, Switzerland.