Hyperspectral Image Denoising Based on Subspace Low Rank Representation

Abstract

Hyperspectral images (HSIs) are often degraded by different kinds of noises. Low rank (LR)-based methods have achieved great performance in HSI denoising problem. However, the LR-based methods only consider the rank of the whole spectral space, conducting no constraints on the intrinsic structure within the LR space. In fact, the spectral vectors can be classified into different categories based on the land-covers. As a result, the spectral space can be modelled as a union of multiple LR subspaces. Regarding this structure, we introduce the framework of subspace low rank (SLR) representation into HSI denoising problem and propose a novel SLR-based denoising method for HSIs. Experiments conducted on both simulated and real data show that our method achieves great improvement over the state-of-art methods qualitatively and quantitatively.