Document Details
Document Type |
: |
Thesis |
Document Title |
: |
LAYOUT ANALYSIS FOR ARABIC DOCUMENTS USING DEEP LEARNING تحليل هيكل المستندات العربية باستخدام التعلم العميق |
Subject |
: |
Faculty of Computing and Information Technology |
Document Language |
: |
Arabic |
Abstract |
: |
In recent days, researchers are very interested in the field of document analysis and optical character recognition (OCR). There is a very great improvement in OCR engines with different languages, and whether for printed documents or handwrit- ten documents. We have a less concerned for processing documents with Arabic language comparing with other languages such as English, and this is due to many reasons like, the difficulty of processing Arabic language, and the limitation of the existed Arabic documents datasets. To implement any OCR, engine, the first step we need to do is analyzing the layout of the images before we send the image to the OCR. This thesis is concerning about layout analysis for Arabic documents using deep learning (DL) approach. We are using two different DL models, Faster Region- based Convolutional Neural Network RCNN and Mask (RCNN), where each model has specific design that match different type of Arabic documents. We are collecting three different types of Arabic documents datasets, early printed, printed, and historical, where each dataset has its own size, structure, and processing requirements. We are using Faster RCNN for printed and early printed Arabic documents, and Mask RCNN for historical Arabic document. Processing historical documents is more difficult because of the layout structure and the proprieties of the historical documents like, the author handwritten style, the age of the paper, the time period the document came from, the used ink and more. The accuracy result is as follow: 99.59% for early printed documents, and 99.56% for printed documents, and 51.14% for historical. Comparing our result with other existed models, we can say that our work can achieve state of the art work with an impressive result.
Key Word: Layout Analysis, Arabic Language, RoI, Faster RCNN, Mask RCN |
Supervisor |
: |
Dr..Al-Hassanin Al-Barahmatushi |
Thesis Type |
: |
Master Thesis |
Publishing Year |
: |
1445 AH
2023 AD |
Added Date |
: |
Monday, November 13, 2023 |
|
Researchers
لطيفه جلال الجفري | Al Jafri, Latifa Jalal | Researcher | Master | |
|
Back To Researches Page
|