Abstract—Extraction of feature and classification methods are important phases in recognition system. A good classifier and extraction of features that suits play a very important role in a recognition system to improving recognition rate. In this paper we propose a new system designed to recognize the ten digits of printed Arabic numerals that are the most common symbolic representation of numbers in the world today. This study has been conducted using Hu moment, number of hole and surface which are tolerate to the geometric transformations along with seven different classifiers Naive Bayesian, Multi-Layer Perceptron (MLP), Linear Discriminant Analysis (LDA), Pseudo-Inverse. Support Vector Machine (SVM), Decision Tree and K-Nearest Neighbor (KNN). Classifier combination is considered. Experimental tests demonstrated that our technique achieves good results on multi-font and multi-size printed digit dataset.

Index Terms—Classification, features extraction, printed digit recognition, cross-validation, multiple classifiers, Hu moments.

Derdour Khedidja and Mouss Hayet are with Industrial Engineering Departement, Automatic and Manufacturing Laboratory, University of Batna 2, Algeria.

Cite:Derdour Khedidja, Mouss Hayet, "Multiple Classifiers and Invariant Features Extraction for Digit Recognition," International Journal of Computer and Electrical Engineering vol. 11, no. 1, pp. 41-52, 2019.