Multiscale Fourier Domain Fusion Technique for Remotely Sensed Spot Images

In this study we examine the use of Multiscale Fourier Transforms (MFT) on the effect of image fusion and segmentation. The proposed method takes SPOT-PAN and SPOT-XS images to register using MFT technique to produce high quality fused image. The algorithm works iterativly from coarsest level of decomposition to the top level. This process applied to the all sub-bands in order to give best choice of transformation. For selecting the fourier coefficients either SPOT-XS band coefficients are replaced by the SPOT-PAN coefficients considering the subband coefficients. After selecting the new fused fourier coefficients, an inverse fourier transform is done to construct the fused image. The technique is first compared respect to the spectral quality assesment with the conventional image fusion techniques such as HSV and Brovey Fusion results. To evaluate the fusion results both visual inspection and segmentation results are used for quality and the degree of similarity with the original image.