Title: A novel architecture of CNN based on SVM classifier for recognising Arabic handwritten script
Authors: Mohamed Elleuch; Najiba Tagougui; Monji Kherallah
Addresses: National School of Computer Science (ENSI), University of Manouba, Manouba, 2010, Tunisia ' Faculty of Computer Science and Information Technology, Albaha University, P.O. Box 1988, KSA ' Faculty of Sciences, University of Sfax, Sfax, 3000, Tunisia
Abstract: Convolutional neural network (CNN), as a deep learning algorithms being developed for years, have been successfully applied in various domains of computer vision and pattern recognition. Recently, support vector machine (SVM) classifier has received more attention for script recognition. In this paper, we investigated a new model based on the integration of two classifiers which are CNN and SVM methods for offline Arabic handwriting recognition. The proposed system modified the CNN trainable classifier by the SVM classifier. CNN-based SVM aims at extracting an automatically a high representation of the data via multi-layers in a deep hierarchical structure and performing classification. The training and the test sets were taken from the HACDB database. Experimental results gave 97.35% and 93.41% character recognition accuracy using 24 and 66 classes, respectively. This is statistically significantly important in comparison with character recognition accuracies obtained from state-of-the-art Arabic OCR.
Keywords: Arabic handwriting recognition; SVM; support vector machines; CNN; convolutional neural networks; deep hierarchical structure; HACDB; Arabic script; handwritten script; script recognition; character recognition.
International Journal of Intelligent Systems Technologies and Applications, 2016 Vol.15 No.4, pp.323 - 340
Received: 08 Nov 2015
Accepted: 19 Mar 2016
Published online: 02 Nov 2016 *