Abstract
This paper proposes a novel and an effective approach to classify ancient Arabic manuscripts in "Naskh" and "Reqaa" styles. This work applies SIFT and SURF algorithms to extract the features and then uses several machine learning algorithms: Gaussian Naive Bayes (GNB), Decision Tree (DT), Random Forest (RF) and K-Nearest Neighbor (KNN) classifiers. The contribution of this work is the introduction of synthetic features that enhance the classification performance. The training phase encompasses four training models for each style. For testing purposes, two famous books from the Islamic literature are used: 1) Al-kouakeb Al-dorya fi Sharh Saheeh Al-Bokhary; and 2) Alfaiet Ebn Malek: Mosl Al-tolab Le Quaed Al-earab. The experimental results show that the proposed algorithm yields a higher accuracy with SIFT than with SURF which could be attributed to the nature of the dataset.