A comparative study of classifiers on recognition of offline handwritten Odia numerals

Extensive work has been done on recognition of many Indian languages. But character recognition work done on Odia character is very less. There are many fields like banking, postal system, form processing etc. which require effective digit recognition system for faster processing. In this paper a sincere attempt has been made to do a comparative study using different types of classifiers for handwritten Odia numerals. For feature extraction gradient and curvature based approaches are used. After the generation of feature vector Principal Component Analysis (PCA) is applied to reduce the size of feature vector. The reduced features are passed to a number of classifiers such as SVM (Support Vector Machine), Artificial Neural Network (ANN), Decision tree (C5.0) and Discriminant Analysis (DA). A comparative study is carried out among the classifiers. From the experimental result it is observed that SVM based classifier achieved 90.5% accuracy with curvature feature and 95.5 % accuracy with gradient feature during validation.