A Pan-Sharpening Based on the Non-Subsampled Contourlet Transform: Application to Worldview-2 Imagery

Two pan-sharpening methods based on the non-subsampled contourlet transform (NSCT) are proposed. NSCT is very efficient in representing the directional information and capturing intrinsic geometrical structures of the objects. It has characteristics of high resolution, shift-invariance, and high directionality. In the proposed methods, a given number of decomposition levels are used for multispectral (MS) images while a higher number of decomposition levels are used for Pan images relatively to the ratio of the Pan pixel size to the MS pixel size. This preserves both spectral and spatial qualities while decreasing computation time. Moreover, upsampling of MS images is performed after NSCT and not before. By applying upsampling after NSCT, structures and detail information of the MS images are more likely to be preserved and thus stay more distinguishable. Hence, we propose to exploit this property in pan-sharpening by fusing it with detail information provided by the Pan image at the same fine level. The proposed methods are tested on WorldView-2 datasets and compared with the standard pan-sharpening technique. Visual and quantitative results demonstrate the efficiency of the proposed methods. Both spectral and spatial qualities have been improved.