الفهرس 
AcknowledgmentOnly 14 pages are availabe for public view
 |  | from | 221 |  |  |
| | | from | 221 | | |
Abstract
Reader machine for the Blind was mentioned through the literature as an application that integrates optical character recognition and speech synthesis systems. In this work, the development of Arabic Reading Machine for the Blind is addressed. To deal with the complexity of constrained Arabic handwriting, such as overlapping lines, overlapping words, diacritic signs, and cursiveness, two novel techniques are introduced: A parsing technique and a word segmentation technique. In order to reduce the computation complexity and enhance the system performance, both techniques are based on the character contour. Various simple feature extraction techniques are employed to achieve a high recognition accuracy, while maintaining cheap computation and fast classification. To boost the classification, the system uses decision trees. Neural network, another alternative, is also tested for character classification, but offers slower recognition rate, and lower accuracy. A knowledge-based system is developed to assemble the characters, associate diacritic signs and dots to the corresponding character. The output is fed to a Text-to-Speech system. It is converted through several stages to produce phonemic strings. These strings are translated to control codes that control the different aspects of the speech synthesizer hardware.