Block-Gaussian-Mixture Priors for Hyperspectral Images
Figueiredo, M. A. T.
Block-Gaussian-Mixture Priors for Hyperspectral Images, Proc SIAM Conf. on Imaging Science - SIAM-IS, Conference Online, Vol. , pp. - , March, 2022.
Digital Object Identifier:
Download Full text PDF ( 75 KBs)
We present a denoiser for hyperspectral data cubes that is based on Gaussian mixture models, exploiting the corresponding minimum mean squared error (MMSE) denoising of data “blocks" (3D “cubicles"). The rationale of the proposed approach is that it takes advantage of, simultaneously, the spatial and inter-band (spectral) dependencies. The denoiser is used in a plug-and-play fashion in ADMM (alternating direction method of multipliers) framework, with convergence guarantees. Experiments suggest that the proposed method outperforms other state-of-the-art methods for this type of data.