Proposition to distinguish machine-printed from handwritten Arabic and Latin words

Abstract

In this work, we gathered some contributions to identify script and its nature. We successfully employed many features to distinguish between handwritten and machine-printed Arabic and Latin scripts at word level. Some of them are previously used in the literature, and the others are here proposed. The new proposed structural features are intrinsic to Arabic and Latin scripts. The performance of all extracted features is studied towards this paper. We also compared the performance of three classifiers: Bayes (AODEsr), k-Nearest Neighbor (k-NN) and Decision Tree (J48), used to identify the script at word level. These classifiers have been chosen enough different to test the feature contributions. We carried experiments using standard databases. Obtained results demonstrate used feature capability to capture differences between scripts. Using a set of 58 selected features and a Bayes-based classifier, we achieved an average identification rate equals to 98.72%, which considered a very satisfactory rate compared to some related works.