Multiple Classifiers Combination for Arabic Manuscript Recognition: Majority Voting Scheme

  • Hayder Jasim Alhamdane, Begared Salih Hassen, Dr Adnan Hussein Ali


 In recent times, the interest of researchers has been drawn to Arabic manuscript recognition framework. Within these documents, lie a fountain of knowledge, but due to the fragile nature of the documents their availability is limited. The purpose of developing system of an ancient Arabic manuscript recognition, is to digital archive enabling. Such system consists of binarization of image, segmentation of character and recognition phases. With the use of this system, scanned and segmented characters can be recognized automatically. The constituents of segmented characters could include modifier (Matras), basic characters (vowels and consonants), in addition to the characters of diverse compound that are designed by combining two or more basic characters). A priority of the paper gives to different techniques used for the extraction and classification of features, which are also compared with the recognizing of basic characters that are separated from Arabic manuscripts. An extraction phase characteristic involves the extraction of centroid, intersection with open-end points, vertical and horizontal peak extents features. The classification is done using different techniques which include Decision tree, Neural Network, Multilayer perception, Convolutional Neural Network and Rain forest. The experiments were performed using a dataset that contained 6153 pre-segmented samples of Arabic documents. Through the integration of all the features and classifiers that were used in this study, 83.92% accuracy was achieved by simple majority voting scheme.

How to Cite
Dr Adnan Hussein Ali, H. J. A. B. S. H. (2020). Multiple Classifiers Combination for Arabic Manuscript Recognition: Majority Voting Scheme. International Journal of Advanced Science and Technology, 29(3), 398 - 404. Retrieved from