Off-line Arabic handwriting recognition using neural network

العناوين الأخرى

التعرف الضوئي على الكتابة اليدوي العربية باستخدام الشبكات العصبونية

مقدم أطروحة جامعية

al-Sayidah, Usamah Nayil

مشرف أطروحة جامعية

Samawi, Venus W.

أعضاء اللجنة

al-Nihoud, Jihad Quball Awdah
Bani Muhammad, Sad
Rashid, Walid A. Jabbar M. Ali

الجامعة

جامعة آل البيت

الكلية

كلية الأمير الحسين بن عبد الله لتكنولوجيا المعلومات

القسم الأكاديمي

قسم علوم الحاسوب

دولة الجامعة

الأردن

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2011

الملخص الإنجليزي

This work aims to produce a system capable of segmenting handwritten Arabic documents to characters or fragments as well as recognizing those using neural networks.

It is clear that the success in the process of recognizing Arabic documents will lead to better communication between man and computer, which make the computer a more effective tool.

In order to create? system that is able to recognize Arabic documents correctly, it must be accompanied by a strong method capable of segmenting the documents properly. Our system mainly consists of four stages: scanning, preprocessing, segmentation and recognition. Scanning : is inputting the paper document in the computer by using a scanner. Preprocessing aims to improve the image through the use of the smoothing that works to remove noise and fills the gaps, then uses the Binarization for transforming the scanned image into binary image. The third stage proposes a new segmentation method.

The suggested segmentation stage consists of three steps : labeling connected parts, extracting features form labeled connected parts, segmenting the labeled image into lines and then into fragments or characters. The fourth stage is recognition.

This stage consists of two levels: the first level consists of two neural networks: one for recognizing the character and the other for fragments.

Each segmented part from the third stage is fed to the two neural networks, which are working in parallel (at the same time).

The second level is decision level which is used to determine the class of the entered character depending on the highest recognition rate in both neural networks. In this work, I have been concerned with two things.

First, I have developed an algorithm able to improve the accuracy of segmenting Arabic documents to characters or fragments.

Second, I have established a neural network to recognize the fragment or characters which were segmented.

I have tested this system on a number of handwritten documents of people.

Then, these documents were subjected to preprocessing that has in turn improved the image, segmentation and finally recognition. The developed segmentation method recorded an overall segmentation rate 74.5 %, while the recognition rate was 74.77 %.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

الموضوعات

• معالجة البيانات
• الشبكات العصبية(الحاسبات الإلكترونية)

عدد الصفحات

57

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-321609