Hyperspectral Image Denoising via Generalized Kronecker Decomposition-Based Subspace Representation

Recently, subspace representation-based methods have shown notable success in hyperspectral image (HSI) denoising by exploiting the spectral correlation of HSIs. However, these methods cannot fully explore the spatial correlation within HSIs, resulting in unsatisfactory results. To address this limitation, we propose a novel generalized Kronecker decomposition-based subspace representation (GKDSR) method, which can simultaneously characterize the spatial and spectral correlation of HSIs. Specifically, we first decompose an HSI as the spectral subspace and coefficients. Then we impose the face-wise generalized Kronecker decomposition to the coefficients to fully characterize the spatial correlation of the HSI. Armed with the proposed GKDSR method, we build a GKDSR-based HSI denoising model, which can not only achieve a promising spectral fidelity but also preserve spatial fine details. To solve the proposed HSI denoising model, we develop an efficient alternating minimization-based algorithm. Experimental results demonstrate the significant superiority of the proposed method compared with competing methods in terms of spectral fidelity and preservation of spatial details.