Hyperspectral image restoration based on spatio-spectral structure tensor regularization

We propose a regularization function for hyperspectral image restoration based on a newly-designed structure tensor. We adopt a convex optimization approach with the use of the nuclear norm of a matrix, termed as spatio-spectral structure tensor. It consists of the gradient components of a hyperspectral image cube w.r.t. the spatio-spectral domain. The proposed approach allows to penalize variations in the spectral domain as well as the spatial domain to exploit the spatio-spectral correlations. Our experiments on denoising of hyperspectral images show that the proposed regularization leads to significant improvements in restoration performance over state-of-the-art methods.