Arabic handwriting recognition using Gabor wavelet transform and SVM

In this paper, we propose a segmentation based recognition approach for handwritten Arabic text. The approach starts by segmenting the word images into their constituent letter representatives through exploiting a set of structural features. For classification, Gabor transform-based features are extracted from each letter that passed to a SVM classifier for recognition. For training and testing, we used IESK-arDB database, which is an Arabic off-line handwritten database, that containing the most common Arabic words as well as security-related Arabic terms. The database is developed in the Institute for Electronics, Signal Processing and Communication (IESK) at Otto-von- Guericke University Magdeburg, Germany. And it is freely available at (http://www.iesk-ardb.ovgu.de/). The approach achieved an average of 70% segmentation accuracy on 600 word images. Recognition rate of 74%, on set of 5436 segmented letter images is reached, according to a Leave-one-out estimation method.