Combining morphological filtering, anisotropic diffusion and block-based data replication for automatically detecting and recovering unscanned gaps in remote sensing images

• Autor(a) principal: Basso, Dayara [UNESP]
• Data de Publicação: 2021
• Outros Autores: Colnago, Marilaine [UNESP], Azevedo, Samara, Silva, Erivaldo [UNESP], Pina, Pedro, Casaca, Wallace [UNESP]
• Tipo de documento: Artigo
• Idioma: eng
• Título da fonte: Repositório Institucional da UNESP
• Texto Completo: http://dx.doi.org/10.1007/s12145-021-00613-6; http://hdl.handle.net/11449/210220
• topic: Landsat 7; Image restoration; Mathematical morphology; Missing data

Filling damaged pixels in satellite images is a key task present in many Remote Sensing applications. As a representative example of image restoration issue, we can refer to the failure of the Scan Line Corrector (SLC) on board the Landsat Enhanced Thematic Mapper Plus (ETM +) sensor, in which 22% of the scanned pixels in the SLC-off images were missed, thus creating unexpected stipe-type gaps in the scenes. In order to improve the usability of ETM + SLC-off data in a straightforward manner, in this paper we propose a unified methodology that automatically segments and repairs Landsat-7 scenes occluded by stripes. The proposed framework combines Morphology-based filtering, anisotropic diffusion and block-based pixel replication as an effective, fully unsupervised restoration methodology designed to cope with different gap sizes in Landsat images. Our approach does not require having as input data any prior gap mask, side reference image or time-dependent frames of the same scene to work properly. As shown in the experimental results, the current methodology performs adequately for a variety of multispectral remote sensing images with different stripe-size thicknesses and heterogeneous segments. We attest to the accuracy and robustness of our end-to-end framework throughout a variety of qualitative and quantitative evaluations involving state-of-the-art restoration methods.