Geometric correction of historical Arabic documents

Dulla, Ali 2019, Geometric correction of historical Arabic documents , PhD thesis, University of Salford.

PDF (phd thesis)
Download (9MB) | Preview


Geometric deformations in historical documents significantly influence the success of both Optical Character Recognition (OCR) techniques and human readability. They may have been introduced at any time during the life cycle of a document, from when it was first printed to the time it was digitised by an imaging device. This Thesis focuses on the challenging domain of geometric correction of Arabic historical documents, where background research has highlighted that existing approaches for geometric correction of Latin-script historical documents are not sensitive to the characteristics of text in Arabic documents and therefore cannot be applied successfully. Text line segmentation and baseline detection algorithms have been investigated to propose a new more suitable one for warped Arabic historical document images. Advanced ideas for performing dewarping and geometric restoration on historical Arabic documents, as dictated by the specific characteristics of the problem have been implemented. In addition to developing an algorithm to detect accurate baselines of historical printed Arabic documents the research also contributes a new dataset consisting of historical Arabic documents with different degrees of warping severity. Overall, a new dewarping system, the first for Historical Arabic documents, has been developed taking into account both global and local features of the text image and the patterns of the smooth distortion between text lines. By using the results of the proposed line segmentation and baseline detection methods, it can cope with a variety of distortions, such as page curl, arbitrary warping and fold.

Item Type: Thesis (PhD)
Schools: Schools > School of Computing, Science and Engineering
Funders: Libyan government
Depositing User: Mr Ali Dulla
Date Deposited: 09 Oct 2019 08:39
Last Modified: 27 Aug 2021 21:28

Actions (login required)

Edit record (repository staff only) Edit record (repository staff only)


Downloads per month over past year