**4. Results**

In the experiments, the test *P* and HS images are cropped from four different hyperspectral remote sensing datasets, i.e., the Salinas dataset [12], the Pavia University dataset [12], the Washington DC dataset, and Hyperion dataset [40]. Several widely used evaluation indexes are adopted to

estimate the effectiveness of the proposed hyperspectral pansharpening method. Six representative hyperspectral pansharpening methods are utilized for comparison, i.e., principal component analysis (PCA) [18], Guided filter PCA (GFPCA) [41], HySure [8], coupled nonnegative matrix factorization (CNMF) [10], MTF-GLP with High Pass Modulation (MGH) [26] and Sparse Representation [7]. The PCA method is a most representative among the CS-based methods. The MGH method is a successful MRA-based hyperspectral pansharpening method. The CNMF method is one of the matrix factorization algorithms. The Sparse Representation and HySure methods which were presented recently belong to the Bayesian category. The GFPCA method has been awarded the "Best Paper Challenge" in the 2014 IEEE data fusion contest. Therefore, in the experiments, these six methods are compared with the proposed method.
