الفهرس 
Chapter 1 INTRODUCTIONOnly 14 pages are availabe for public view
 |  | from | 16 |  |  |
| | | from | 16 | | |
Abstract
The character recognition system consists of four main phases: pre-processing, seg-mentation, recognition and post-processing. The segmentation is the most important phase and its accuracy greatly affects the final recognition results. In addition, the pre-processing is very important and it is usually performed to reduce the noise and en-hances the image quality. For pre-processing phase, image quality and execution time are important factors to choose the suitable noise reduction filtering type. Accordingly, in this thesis a com-parative study of six different types of noise reduction technique for Arabic handwritten images is presented. Then, the filtering type which gives good quality and less execution time for the two types of noises: salt & pepper noise and Gaussian noise is used for the proposed system. In this thesis the CMF filter gives the best result among the other filtering types. After the pre-processing, the segmentation starts. The segmentation on this thesis based on the word level. The word is segmented into characters based on a recent seg-mentation technique which is dubbed in this thesis as base technique. The base tech-nique doing the segmentation using two stages: over-segmentation and neural valida-tion. The over-segmentation find all the possible SPs and then the validation test all these SPs and keep the correct SPs only. To solve the limitation of the base technique and increase the recognition rate, three enhancements are added to the base technique. Additionally, the random forest classifier and zoning features are used at the validation stage. The proposed approaches and the base technique are implemented and tested on 1000 words from two different databases. The proposed approach achieve 88% and it is considered a promising result compared to 81% which is the result of the base tech-nique.