Joint denoising and unmixing for hyperspectral image

Hyperspectral image denoising and unmixing are two separate stages in traditional works. Unmixing algorithm is implemented after denoising. The performance of unmixing will be promoted if noise in hyperspectral image is removed well. But the result of unmixing can not be used to improve the result of denoising. In this paper we propose a joint denoising and unmixing algorithm for hyperspectral image. In this algorithm, denoising and unmixing processes are done in coupled way, the denoising and unmixing performance can be promoted by each other. Firstly, hyperspectral denoising and unmixing processes are represented in the common sparse representation framework. Then, the abundance is estimated by exploiting the sparsity priori of endmember on spectral library. After that, estimated abundance and endmember are used as spectral regularizer to enhance the denoising result. When noise is suppressed in the hyperspectral image, unmixing process will perform better and be more robust to noise. The obtained unmixing result will further enhance the denoising. The experiment proves that our algorithm can provide satisfying denoising and unmixing result that both of them are competitive to the state-of-the-art methods in their respective fields.