Abstract
In this paper, a system is proposed for word-based recognition of handwritten Arabic scripts. Techniques are discussed in details in terms of three stages in the system, i.e. preprocessing, feature extraction and classification. Firstly, words are segmented from inputted scripts and also normalized in size. Then, DCT features are extracted for each word sample. Finally, these features are then utilized to train a neural network for classification. The proposed system has been successfully tested on database (version v2.0p1e) consisting of 32492 Arabic words handwritten by more than 1000 different writers, and the results were promising and very encouraging.