A noise-adjusted iterative randomized singular value decomposition method for hyperspectral image denoising

In this paper, a new denoising algorithm is proposed for hyperspectral image data cubes. With the strong correlations of the image bands, the low-rank structure of the hyperspectral image is explored by lexicographically ordering the 3-D data cube into 2-D matrix. Based on this property, the traditional principal component analysis (PCA) denoising model is established. For hyperspectral images (HSIs), the noise intensity in different bands is different. Therefore, a noise-adjusted iterative randomized singular value decomposition (NAIRSVD) algorithm is proposed to solve this PCA model. Combined with adaptive noise estimation and upper bound rank estimation, the proposed NAIRSVD algorithm is free from manual parameter determination. Several experiments were conducted to illustrate the performance of the proposed algorithm.