Effective arabic character recognition using support vector machines

Abstract

This paper proposes an Arabic character recognition system. The system focuses on employing Support Vector Machines (SVMs) as a promising pattern recognition tool. In addition to applying SVM classification which is a novel feature in arabic character recognition systems, the problem of dots and holes is solved in a completely different way from the ones previously employed. The proposed system proceeds in several phases. The first phase involves image acquisition and character extraction, the second phase performs image binarization where a character image is converted into white with black background, while the next phase involves smoothing and noise removal. In the fourth phase a thinning algorithm is used to thin the character body. The fifth phase involves feature extraction where statistical features, such as moment invariants, and structural features, such as number and positions of dots and number of holes, are extracted. Finally, the classification phase takes place using SVMs, by applying a one-against-all technique to classify 58 Arabic character shapes. The proposed system has been tested using different sets of characters, achieving a nearly 99% recognition rate. © 2007 Springer.