Junk component images in multichannel (multispectral, hyperspectral) remote sensing data may have a lower quality than other components. This could be due to high noise levels or a low dynamic range of the information component in garbage channels (sub-bands). Component images like these are sometimes overlooked (not used in further processing or analysis). In the meantime, they can be subjected to pre-filtering (denoising) to improve them. To do so efficiently, we propose using so-called reference images, which are component images of reasonably good quality that are characterised by high similarity to the image being pre-filtered. We look at how to identify component photos that can be used as references, what transformations may be done on reference photos, and how to do the denoising within the context of this broad idea. It is demonstrated that component photos of the same resolution as trash photos, as well as component photos of higher resolution, can be used as references. It is possible to use not only one, but two references. After decorrelation, several filters can be used. It is possible to enhance filtered photos using various quality criteria. Denoising examples for real-world multichannel photos are provided, proving the suggested method's great efficiency.
Author (s) Details
Sergey Abramov
National Aerospace University, Kharkov, Ukraine.
Mikhail Uss
National Aerospace University, Kharkov, Ukraine.
Vladimir Lukin
National Aerospace University, Kharkov, Ukraine.
Benoit Vozel
University of Rennes 1, Lannion, France.
Kacem Chehdi
University of Rennes 1, Lannion, France.
Karen Egiazarian
Tampere University, Tampere, Finland.
View Book :- https://stm.bookpi.org/NUPSR-V5/article/view/1396
No comments:
Post a Comment