Hyperspectral Image Denoising with Composite Regularization Models

Denoising is a fundamental task in hyperspectral image (HSI) processing that can improve the performance of classification, unmixing, and other subsequent applications. In an HSI, there is a large amount of local and global redundancy in its spatial domain that can be used to preserve the details and texture. In addition, the correlation of the spectral domain is another valuable property that can be utilized to obtain good results. Therefore, in this paper, we proposed a novel HSI denoising scheme that exploits composite spatial-spectral information using a nonlocal technique (NLT). First, a specific way to extract patches is employed to mine the spatial-spectral knowledge effectively. Next, a framework with composite regularization models is used to implement the denoising. A number of HSI data sets are used in our evaluation experiments and the results demonstrate that the proposed algorithm outperforms other state-of-the-art HSI denoising methods.