14th March 2019 - Seminar - Non-Negative and Non-Local Tensor Dictionary Learning Based Hyperspectral Image Super-Resolution - Professor Weihong Guo (Case Western Reserve University, USA)
Venue: MATH-211, Second Floor, Department of Mathematical Sciences Building
Abstract: Hyperspectral images provide rich spectral information that could be used in industry, medicine and remote sensing etc. Hyperspectral sensors achieve high resolution along spectral direction with the price of very low spatial resolution. We provide a hypersepctral image (HIS) super resolution algorithm to increase its spatial resolution by fusing it with a multispectral image (MSI) which is usually available simultaneously during the data collection. Multispectral images have less spectral bands but higher spatial resolution compared with the hyperspectral counter part. We propose a novel non-negative tensor dictionary learning based HSI super-resolution model using non-local spatial similarity and group-block-sparsity. It aims to preserve the high dimensional data structure while achieving nonlocal regularity. The computation is done on clusters obtained by tensor cube classification. Numerical experiments demonstrate that the proposed model outperforms many state-of-the-art HSI super-resolution methods.