Handwritten Arabic writer identfication

Abstract

Automatic Writer Identification is an essential research area to forensic analysis. It is considered under pattern recognition domain problem. Writer identification mainly consists of three typical phases: pre-processing, feature extraction, and classification. Another phase, the Discretization phase, has been added to the framework to improve identification performance. The Arabic handwriting has less attention in writer identification research area, with about 10 researches found in the literature. This research intended to study and evaluate the effects of discretization process on writer identification performance, for off-line text-independent Arabic handwriting. It is tested on the IFN/ENIT Arabic DB with 100 writers and the Regional Arabic DB containing samples of Arabic words written by 30 writers from 6 Arabian countries and counties that can write Arabic. The results disclose an achievement of 99.8% accuracy of identification by using 3920 training data and 980 testing data.