DBN - Based learning for Arabic handwritten digit recognition using DCT features

Abstract

In this paper multi-class classification system of handwritten Arabic digits using Dynamic Bayesian Network (DBN) is proposed, in which technical details are presented in terms of three stages, i.e. pre-processing, feature extraction and classification. Firstly, digits are pre-processed and normalized in size. Then, features are extracted from each normalized digit, where a set of new features for handwritten digit is proposed based on the discrete cosine transform (DCT) coefficients approach. Finally, these features are then utilized to train a DBN for classification. The proposed system has been successfully tested on Arabic handwritten digit database (ADBase) which is composed of 70,000 digits written by 700 different writers, and the results were promising and very encouraging.