| Image fusion in remote sensing is a technique that offers great promise and numerous algorithms have been developed to merge a high-spectral-resolution multispectral (MS) image with a high-spatial-resolution panchromatic (PAN) image in order to synthesize an image with the high spectral and spatial resolutions. One objective of image fusion is to synthesize images as close as possible to real images of the same spectral and spatial resolutions. Although the visualization and the spectral accuracy of synthetic images have been improved significantly, the spectral distortion in synthetic images is obvious and seriously impacts on the applications of the synthetic products. Except the main challenge of spectral distortion in current image fusion, several technical challenges still exist, including the limit on the spatial resolution ratio of MS and PAN images, automatic setting of parameters, and the impact of mis-registration of input MS and PAN images with a large spatial resolution difference. To solve these challenges, a series of solutions is proposed based on results of a thorough study on current image fusion techniques; in addition, each proposed fusion method performs well in preserving the spectral characteristics of MS images and sharpening visualization.;Atmospheric effects or haze, are rarely taken into account in current fusion methods. Taking haze into account, two improvements are proposed for a generalized ratio-based PAN modulation fusion method; taking haze into account and with reference to phenology of pixels, the third method significantly improves current multiresolution analysis-based fusion methods. The third method can fuse mixed MS sub-pixels optimally in the sense of probability. Mis-registration of MS and PAN images impacts on fusion quality seriously. Utilizing a direction similarity of the pixel vectors of each MS sub-pixel and neighbors, the fourth method effectively eliminates the impact of mis-registration on fusion.;The fusion of mixed MS sub-pixels is rarely addressed in image fusion and the corresponding synthetic pixels normally remain spectrally mixed and visually blurred. The fifth method may properly fuse mixed sub-pixels and significantly sharpen related soil-vegetation boundaries in synthetic products. Employing an object-oriented classification map of a PAN image, the sixth method can effectively fuse MS and PAN images with a significant spatial resolution ratio.;Thermal-IR and reflective images are normally too weakly correlated to be fused properly. Employing a technique of multivariate analysis, the seventh method can offer a synthetic thermal-IR image with high quality. Using a non-linear transform technique combined with a technique of multivariate analysis, the eighth method further improves fusion quality. |
