This document summarizes a research paper that proposes a new method for hyperspectral image restoration using nonlocal spectral-spatial structured sparse representation. The key points are:
1) It introduces using nonlocal similarity and spectral-spatial structure of hyperspectral images in sparse representation models. Nonlocal similarity means similar image patches can be represented by shared dictionary atoms, distinguishing true signals from noise.
2) Using 3D blocks that exploit spectral and spatial correlations, rather than 2D patches, for sparse coding. This better distinguishes true signals and noise.
3) A mixed Poisson-Gaussian noise model is used to handle signal-dependent and signal-independent noise present in hyperspectral images. Variance-fitting transformation