Super-resolution algorithm based on sub-pixels spatial Correlation for hyperspectral image classification

Msellmi, Bouthayna; Rabah, Zouhaier Ben; Farah, Imed Riadh

doi:10.1109/atsip.2016.7523154

Cited by 1 publication

(2 citation statements)

References 25 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In [61], an adaptive sub-pixel mapping algorithm is proposed. After finding the abundance maps, each pixel is divided along columns into smaller units and fractions are duplicated in the divided pixels.…”

Section: Spatial Optimization Based Methodsmentioning

confidence: 99%

“…In general, SRR methods using auxiliary data such as sub-pixel shifted LR images or HR RGB or MS image generates better results[58].However, single image SRR for HS images are taking considerable attention since sometimes supplementary data may not exist for the HS image. The output of single SRR can be either super-resolution maps (SRM)[59,60,61], a map to describe the most likely distribution of mixed pixels, or HR HS image[58,62,63].…”

mentioning

confidence: 99%

See 1 more Smart Citation

Hyperspectral imagery superresolution

Irmak

Akar²,

Yüksel

2016

2016 24th Signal Processing and Communication Application Conference (SIU)

View full text Add to dashboard Cite

pagesHyperspectral (HS) imagery consists of hundred of narrow contiguous bands extending beyond the visible spectrum. It is a three dimensional data cube with two dimensional spatial information and a spectral dimension. Despite having high spectral resolution, HS images have lower spatial resolution due to technological restrictions. This degrades the performance in HS imaging applications. Therefore, increasing the resolution is a necessity, however the problem is an ill-posed problem. In this thesis, we address this problem, namely super-resolution reconstruction (SRR) of HS images from different perspectives and propose robust solutions.First method proposes a maximum a posteriori (MAP) based SRR technique for HS images when there is only one HS image and no other source of information. The novelty of the method is converting ill-posed SRR problem in spectral domain to a quadratic optimization problem in abundance map domain. Using smoothness prior and inherent properties of abundance maps in the quadratic optimization, a unique solution is obtained. Moreover, in order to avoid over smoothing, a post processing is applied to preserve textures in the abundance maps. Finally, high resolution (HR) HS image is reconstructed using the extracted endmembers and the enhanced abundances. v Second proposed method is a fusion based SRR method. This method enhances the spatial resolution of HS image by fusing with a coinciding HR RGB or multispectral (MS) image. Again, fusion problem is converted to a quadratic optimization problem in the abundance map domain. Moreover, proposed MAP based approach is also merged into the quadratic equation. That is, this method is a superposition of MAP based and fusion based approaches and closing the gaps of the both methods. Superposition of two methods leads to more robust and efficient SRR method. Similarly, after solving quadratic problem, HR HS image reconstructed from HR abundance maps and endmember signatures.Experiments are implemented on real HS datasets and compared to state-of-the-art alternative methods using different quantitative image metrics. Spectral consistency, a critical issue for HS images, is also analysed in the experiments. Results demonstrate that proposed methods perform better than competitors based on quantitative metrics while keeping spectral consistency.

show abstract

Section: Spatial Optimization Based Methodsmentioning

confidence: 99%