| The issues related to writer identification are currently at the heart of numerous concerns in our modern day's society. Writer identification for Arabic text is receiving a renewed attention. We anticipate that the developed research techniques and algorithms in this thesis to help in establishing this area of research.;Writer identification of off-line Arabic handwritten text and digits is addressed by utilizing the state of the art identification and verification techniques, features, and classifiers. The presented identifiability of handwritten digits provides quantitative measurements for the uniqueness of each digit. The research work has also shown that we can design a successful writer identification system from only the basic 10 Arabic digits given that other writers write the same digits.;A successful writer identification system for handwritten Arabic text is designed and developed. Since there is no available Arabic database for this application, this research includes building a database for handwritten Arabic text by 250 writers. Several types of structural and statistical features are extracted. Connected component features as well as Overlapped gradient distribution features, gradient distribution features, windowed gradient distribution features, contour chain code distribution features, windowed contour chain code distribution features, density features, horizontal and vertical run length features, stroke features, and concavity shape features are implemented.;The accuracy results of each feature type are compared using statistical significance. The effects of increasing the number of writers on the accuracy results are presented and analyzed. Experimental results of applying these features on Arabic text and digits are presented. Feature reduction and selection techniques (e.g. PCA, LDA, MDA,…) are applied and shown to improve both computation time and accuracy results. |
