Remote sensing image fusion based on structured dictionary learning and joint sparse representation

We describe a novel image fusion method for remote sensing based on structured dictionary learning and joint sparse representation which improves both quality and execution time. First, structured dictionaries are separately trained from the panchromatic (PAN) image and multispectral (MS) image by the double-sparsity model. Then, we use the joint sparse model to obtain the innovation components of the PAN image that are absent from the MS bands. Finally, the fused image is constructed by injecting the innovation components of PAN image into the low-resolution multispectral image by injection model. Experimental results show that our method has better performance than other state-of-the-art methods in terms of the spectral information, spatial resolution and computational efficiency.