Spectral Representation vis Data-Guided Sparsity for Hyperspectral Image Super-Resolution^†.

Area of Science:

• Remote Sensing
• Medical Imaging
• Computer Vision

Background:

• Hyperspectral imaging captures rich spectral data but often lacks high spatial resolution due to hardware constraints.
• Existing hyperspectral imaging devices struggle to achieve high spatial resolution, limiting detailed material analysis.

Purpose of the Study:

• To develop a novel hyperspectral image superresolution method.
• To generate high-resolution hyperspectral images from available low-resolution hyperspectral and high-resolution RGB images.

Main Methods:

• A non-negative sparse representation of reflectance spectra with a data-guided sparsity constraint is proposed.
• A hyperspectral dictionary is learned from low-resolution data and transformed to RGB using a camera response function.
• A sparsity map, derived from RGB image content and spectral mixing analysis, guides the sparse representation for each pixel.

Main Results:

• The method adaptively adjusts spectral representation sparsity based on local RGB image content.
• This adaptive approach yields robust spectral representations for high-resolution hyperspectral image recovery.
• Experiments on public datasets and real remote sensing images demonstrate superior performance compared to state-of-the-art methods.

Conclusions:

• The proposed method effectively enhances spatial resolution in hyperspectral images.
• It offers a promising solution for applications requiring detailed spectral and spatial information.
• The data-guided sparsity constraint improves the accuracy and robustness of hyperspectral superresolution.