On the fusion of panchromatic and multispectral images

two important aspects

up-sampling of multispectral images and the difference of channel details

are ignored. For the former

the loss details of low-resolution images are estimated by using self-similar patch at different scales to improve up-sampling. For the latter

the local weighted dynamic sparse constraint is proposed based on the structural similarity between panchromatic images and spectral images in gradient domain. The new objective function based on variational method are proposed

the fidelity term and the regularization term of whose are constructed respectively according to the former and the latter. In addition

a multi-scale iterative fusion framework is presented

where the resolution of the fused image is gradually improved through iterations. The fused results of each iteration are more accurate

so the final fused image is improved. Our algorithm is compared with Brovey and other component substitution algorithms

P+XS and other variational algorithms

MTF_GLP and other multi-resolution analysis algorithms. The experimental results show that the fusion results of this algorithm have good visual effect

and the objective evaluation index is better than the average of the optimal value of all comparison algorithms.