Plug-and-play hyperspectral image denoising tool combining low-rank and sparsity priors.
This code corresponds to the following paper:
C.F. Dantas, J.E. Cohen, and R. Gribonval. “Hyperspectral Image Denoising using Dictionary Learning.” In WHISPERS, 2019.
Related slides here.
The hyperspectral image cube is matricized (by vectorizing the two spatial dimensions at each spectral band). and later approximated in a low-rank model.
Each of the eigen-images (columns of the left SVD factor) is then used to learn a dictionary in a patch-based fashion. Finally, the learned dictionary is used to sparsely reconstruct the input image.