Pan-sharpening of remote sensing images via graph regularized sparse coding

Sharpening of a multispectral (MS) image with a panchromatic (Pan) image is an important way to synthesize a high resolution MS (HRMS) image that cannot be directly acquired by satellite sensor. In this paper, we propose a novel pan-sharpening scheme via graph regularized sparse coding (GRSC). Compared to the existing pan-sharpening methods based on sparse coding, our work has the following novelties. First, the image patches extracted from all the MS bands are clustered into several subsets with the K-mean algorithm. The image patches in each subset share the same or very similar structures. Second, we take into account the sparse and local manifold structures of the image patches in each subset, and use the GRSC algorithm to precisely describe these characteristics. Experimental results with QuickBird and IKONOS datasets demonstrate that the proposed method can acquire high quality HRMS images that preserve better spatial and spectral information.