Arabic offline writer identification on a new version of AHTID/MW database

Handwriting is considered to be one of the commonly used biometric modalities to verify and identify persons in commercial, governmental and forensic applications. In order to test and compare the accuracy of a computer vision system, in general, and a biometric system in particular, standard rich databases must be publicly available. In this paper and for this purpose, we expose the different works of writer identification of Arabic handwritten text carried out on our already published database AHTID/MW. As researchers have achieved high identification rates, we propose to extend the AHTID/MW database with new Arabic native writers and raise the level of difficulty. A baseline is drawn on each text-line image, and ground truth information is provided for each text image. In addition we present our experiments on the database using a new approach based on combining a CNN for feature extraction with GMM-based emission probability estimates for classification.