Hyperspectral Image Denoising Based on Low Rank and Expected Patch Log Likelihood

Abstract

Denoising is a necessary and fundamental step in the hyperspectral image (HSI) analysis process. Since the spectral channels of HSI are highly correlated, they are characterized by a low rank structure and can be well approximated by low rank representation. Therefore, based on low rank structure and the EPLL, a 4-step algorithm is proposed to denoise the hyperspectral images with Gaussian noise. PCA is used to explore the high correlation and capture the low rank structure in spectral domain of HSI. The EPLL is used to further denoise the HSI in spatial domain. Compared with four state-of-the-art denoising algorithms, the proposed algorithm performs well in HSI denoising, especially for moderate and high noise levels.