Doc-Attentive-GAN: attentive GAN for historical document denoising

Neji, Hala ; Ben Halima, Mohamed ; Nogueras-Iso, Javier (Universidad de Zaragoza) ; Hamdani, Tarek M. ; Lacasta, Javier (Universidad de Zaragoza) ; Chabchoub, Habib ; Alimi, Adel M.

Resumen: Image denoising attempts to restore images that have been degraded. Historical document denoising is specially challenging because there is considerable background noise or variation in contrast and illumination in handwritten literature and the first times of the printing press. The main objective of this work is to propose a new method for historical document denoising based on an Attentive Generative Adversarial Network (Attentive-GAN). Our proposed model for historical document denoising is named Doc-Attentive GAN , and it employs an attention map generated by a deep network to help the generator to learn and focus on the modification between the target image and its noisy version. It has been trained and tested with different historical document collections such as well-known DIBCO datasets, Arabic Historical Documents from the Tunisian National Library, and Incunabula books. The experiments demonstrate a clear improvement in the visual quality of the images obtained by Doc-Attentive-GAN with respect to the state-of-the-art.

Idioma: Inglés

DOI: 10.1007/s11042-023-17476-2

Año: 2024

Publicado en: Multimedia Tools and Applications 83 (2024), 55509–55525
ISSN: 1380-7501

Factor impacto SCIMAGO: 0.801 - Media Technology (Q1) - Software (Q2) - Computer Networks and Communications (Q2) - Hardware and Architecture (Q2)

Tipo y forma: Artículo (PostPrint)

Área (Departamento): Área Lenguajes y Sistemas Inf. (Dpto. Informát.Ingenie.Sistms.)