Pansharpening for Incompletely Overlapping Image-Pairs via Dictionary Extension

There are currently various pansharpening methods to reconstruct a high-spatial-resolution multispectral image (HR-MSI) by fusing a low-spatial-resolution multispectral image (LR-MSI) with a high-spatial-resolution panchromatic image (HR-PAN). However, these methods can only handle situations where HR-MSI and LR-PAN cover the exactly same area, but in real practical situations, HR-PAN usually covers a larger area than the LR-MSI. As a result, these methods can only be used within the overlapping area and cannot reconstruct HR-MSI in the non-overlapping area. To solve this problem, we propose a pansharpening method for incompletely overlapping HR-PAN and LR-MSI image-pairs based on dictionary extension (termed PANDE), which can extend the dictionary learned in the overlapping area to the non-overlapping area, and reconstruct the entire HR-MSI on the entire area covered by the HR-PAN, in the framework of sparse expression. Specifically, in the overlapping area, the fusion model based on decomposition incorporating with the constraints of low-rank and sparsity is used to acquire the spectral dictionary and its associated coefficients matrix. Then, the spectral dictionary learned with the overlapping area will be extended to the non-overlapping area, under the guarantee of spectral similarity between adjacent areas, and its corresponding coefficients matrix will be obtained only using the HRMSI with a sparse constraint. Finally, the desired HR-MSI can be reconstructed by using the spectral dictionary, the coefficients matrix of the overlapping area and that of the non-overlapping area. Experimental results on different scenes show that, compared with the other related methods, our proposed PANDE achieves a better fusion effect and can solve the problem of their inability to reconstruct HR-MSI in the non-overlapping areas.