A Generalized Tensor Formulation for Hyperspectral Image Super-Resolution Under General Spatial Blurring.

Area of Science:

• Remote Sensing
• Image Processing
• Computational Imaging

Background:

• Hyperspectral super-resolution commonly fuses low-resolution hyperspectral images with high-resolution multispectral images.
• Existing tensor-based methods often assume separable spatial blurring, which doesn't reflect real-world sensor limitations like anisotropic blurring.

Purpose of the Study:

• To propose a generalized tensor formulation for hyperspectral super-resolution that accommodates non-separable spatial degradation.
• To develop a practical algorithm for accurate super-resolution recovery under general spatial blurring conditions.

Main Methods:

• A generalized tensor formulation using Kronecker decomposition to model arbitrary spatial degradation matrices.
• Development of a blockwise-group-sparsity regularization-driven algorithm for super-resolution recovery.

Main Results:

• The proposed generalized tensor approach demonstrates superior performance compared to traditional matrix-based and state-of-the-art tensor-based methods.
• Significant performance gains were observed, particularly in scenarios with anisotropic spatial blurring.

Conclusions:

• The generalized tensor formulation effectively handles complex spatial blurring in hyperspectral super-resolution.
• The developed algorithm provides a robust and accurate solution for hyperspectral image fusion, outperforming existing methods, especially under anisotropic blurring conditions.