HYPERSPECTRAL IMAGE DENOISING USING MULTIPLE LINEAR REGRESSION AND BIVARIATE SHRINKAGE WITH 2-D DUAL-TREE COMPLEX WAVELET IN THE SPECTRAL DERIVATIVE DOMAIN

In this paper, a new denoising method is proposed for hyperspectral remote sensing images, and tested on both the simulated and the real-life datacubes. Predicted datacube of the hyperspectral images is calculated by multiple linear regression in the spectral domain based on the strong spectral correlation of the useful signal and the inter-band uncorrelation of the random noise terms in hyperspectral images. A two dimensional dual-tree complex wavelet transform is performed in the spectral derivative domain, where the noise level is elevated temporarily to avoid signal deformation during the wavelet denoising, and then the bivariate shrinkage is used to shrink the wavelet coefficients. Simulated experimental results demonstrate that the proposed method obtains better results than the other denoising methods proposed in the reference, improves the signal to noise ratio up to 0.5dB to 10dB. The real-life data experiment shows that the proposed method is valid and effective.