MULTISPECTRAL AND PANCHROMATIC IMAGE FUSION VIA CONVOLUTION SPARSE CODING WITH JOINT SPARSITY

In this paper, a low spatial resolution multispectral (LR MS) and panchromatic (PAN) image fusion method based on convolution sparse coding (CSC) is proposed to model the global structures existing in source images. In the proposed method, CSC is adopted to decompose the high frequency (HF) component properly and joint sparse prior is also used to capture the correlation in the bands of MS images. By joint sparsity, the correlation is further inherited into their corresponding feature maps. Then, the spatial information in LR MS image is enhanced well after detailed fusion rule for spatial details. Finally, the fusion image is reconstructed by the fused low frequency and HF. The experimental results on real datasets from QuickBird and Geoeye-1 satellites verify that the proposed method can better preserve the spatial and spectral information in the fused images.