الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 120 |  |  |
| | | from | 120 | | |
Abstract
The field of arabic character recognition is crucial, not only for arabic speaking countries, but also for other languages wjich are written in arabic script, such as Faris, Urdu and others. Little work has been done in the area of arabic text recognition compared with those for Latin, chinese and Japanese. The main difficulty encountered when dealing with arabic text is the cursive nature of arabic writing in both printed and handwritten forms.
In this thesis an approach for off-line recognition of handwritten arabic characters is proposed. This approach consists of four main stages: preprocessing, segmentation, feature extraction, and recognition. Preprocessing includes all steps needed to transform the text image into a form suitable for the subsequent OCR stages. These steps algorithm is developed that produces skeletons of the characters without gaps or extra branches by using the shoulder to wall algorithm.
After the first stage of preprocessing a segmentation stage is required in order to isolate characters from a cursive word or sub-word. The proposed segmentation approach is based on following the base line of the thinned word or sub-word, the base line is calculated by analysis of horizontal density histogram. A fter the segmentation, the arabic cursive word is represented by a sequence of isolated characters. These characters are normalized using a nearest neighbor algorithm to a fixed of 24*24 pixels. These characters are used in learning and testing of a neural network classifier. The recognition is thus reduced to that of classifying each.
In the recognition stage, four different sets of characters have been independently considered which are isolated, beginning, middle and end. A neural network is used for each set. The neural network uses the principle component analysis PCA as a tool for feature extraction. where it compresses each character to a certain number of features. The classification is done by multilayer preceptron MLP neural network trained with back-propagation.
The proposed system was tested on a hand written database of arabic character. The system gives 94.5% correct recognition rate for the isolated characters. 95.1% the beginning characters, 92.91% for the middle characters, and 94.41% for the end characters.
All the described approaches and algorithms are fully implemented and tested in c++ programming language.