An image fusion method for misaligned panchromatic and multispectral data

摘要

Misalignment between multispectral (MS) and panchromatic (PAN) images blurs fused images and reduces the spectral quality of the latter. When a PAN pixel is misaligned with the corresponding MS pixel that is previously upsampled to match the pixel size of the PAN image, it is approximately aligned with a neighbouring MS pixel. Although the two MS pixels have different pixel vector magnitudes, their pixel vector directions are typically close to each other. Therefore, a fused MS pixel aligned with the PAN pixel can be generated by setting its pixel vector direction to be the same as the MS pixel misaligned with the PAN pixel and calculating its pixel vector magnitude regarding the PAN pixel. Using this idea, we propose an image fusion method to effectively eliminate the impact of the misalignment between PAN and MS images on fused images. An experiment on misaligned IKONOS PAN and MS images shows that this method yields fused images with sharp visualization and higher spectral quality than several current fusion approaches.