Cursive Script Textline Image Transformation for Improving OCR Accuracy

Eman Eman, Syed Saqib Bukhari, Martin Jenckel, Andreas Dengel

In: 2019 International Conference on Document Analysis and Recognition Workshops (ICDARW). International Conference on Document Analysis and Recognition Workshops (ICDARW-2019) September 22-25 Sydney Australia Seiten 59-64 5 ISBN 978-1-7281-5055-0 IEEE 9/2019.


Over the years, there has been significant progress in field of OCR for cursive script, such as italic and handwritten text. However, complex cursive script is still a major challenging factor to parse OCR correctly from the image. Additionally, most of the OCR systems trained on straight scripts textline images, could not produce reasonable results for cursive script textline images. In this paper, our main focus is to transform cursive script textline images into straight scripts textline images with the intention of equally using straight script OCR systems for both straight and cursive scripts textline images. We use generative adversarial networks (GANs) models in conditional settings to transform cursive scripts into straight scripts. Our experiment shows that the transformed cursive script performs better than the original cursive script on open-source straight script OCR systems to parse character.

Weitere Links

Deutsches Forschungszentrum für Künstliche Intelligenz
German Research Center for Artificial Intelligence