Seminar by Tatiana GELVEZ

Low rank regularizations for spectral Imaging recovery problems

Using prior information is crucial for solving ill-posed inverse problems in image processing. Specifically, a spectral image (SI) can be modeled as a three-dimensional array with two spatial and one spectral dimension. Natural scenes typically contain redundant spectral responses and self-similar spatial structures so that the so-called low-rank prior indicates that a SI lies in a lowdimensional subspace. This talk addresses the question of how to take advantage of the low-rank for solving spectral imaging recovery problems, such as SI denoising, single hyperspectral image super-resolution, spectral imaging fusion, and compressive spectral imaging reconstruction.