Framelet-based random walk pan-sharpening method

A novel pan-sharpening method was proposed for the fusion of multispectral image and panchromatic image. First, the original image was transformed from spatial domain to framelet domain by intensity, hue, saturation (IHS) transform and non-subsampled framelet transform. Second, the statistical fusion model of high frequency framelet coefficients was established with the random walk method based on graph theory. On basis of the neighborhood correlation and scale correlation of high frequency framelet coefficients, the novel compatibility function for random walk was constructed. Finally, the fusion weight estimation of high frequency framelet coefficients was translated into the solution of the random walk labeled problem. Experimental results show that the proposed method is beneficial to keep the spectral information and edge contour information of the image. It can reduce the spectral distortion while improve the spatial resolution simultaneously, and it outperforms the other state-of-the-art pan-sharpening methods.