Research On Guided Filter Based Pansharpening

Pansharpening has been one of the important research branches in remote sensing image fusion.It focuses on using panchromatic(PAN)image to improve the spatial resolution of multispectral(MS)images as well as preserving their spectral signature.With the fast development of sensors,the remote sensing satellites,e.g.,IKONOS,Geo Eye-1,World View-2,can provide higher spatial resolution PAN image and higher spectral resolution MS images at almost the same time.Some existing pansharpening methods when carried on these new kind of high-resolution spaceborne images often suffer from spectral and/or spatial information distortions more or less,which usually lead to unsatisfied pansharpened products and have weak universality.The guided filter is a new proposed edge-preserving filter having advanced features,such as edge-preserving smoothing,structure transferring,etc.Recently,it has been introduced into the image fusion field.In order to solve the problem of information distortions,this thesis makes a deep research on pansharpening method based on guided filtering.The main research work is as follows:1.Several existing pansharpening methods are studied.Their disadvantages and limitations are analyzed through the fusion experiments.The experimental results on IKONOS and World View-2 data show that the existing methods are not suitable for fusing the high-resolution MS and PAN images.2.The principle of the guided filter is studied.The experiments are carried out to verify the edge-preserving smoothing and structure transferring characteristics.The edge-preserving characteristic can be used to obtain the smoothed result keeping the edges of the input image.The structure transferring characteristic effectively transfers the guidance image's structures into the output image.The experimental results demonstrate that the above advantages are helpful to designing pansharpening method.3.A detail enhancement pansharpening method based on guided filter is proposed.First,the method takes advantage of the structure transferring characteristic to eliminate the edges with blocking-effects effectively,which can reduce the spatial distortions caused by interpolation.Thanks to the edge-preserving smoothing capability of guided filtering,the sufficient details from the PAN image are extracted which help to improve the spatial resolution of fusion results.Moreover,the proposed method does not consider the relative spectral response between MS and PAN bands and can be implemented in a band-by-band manner,which has strong universality for different satellite images.The experiments are carried out on IKONOS,Geo Eye-1,and World View-2 data respectively.The experimental results show that the proposed method can enhance the spatial details while keeping the spectral information well.4.A guided filtering and hyperspherical color transformation(HCT)integrated pansharpening method is presented.First,it takes advantage of HCT to obtain the intensity component and spectral components(angles).Then,the details from the intensity component and PAN image are extracted by a detail extraction model based on guided filter.Finally,all the fused bands can be obtained by the inverse HCT.The experimental results show that the proposed method can improve the spatial resolution of the fused image as well as maintain original spectral characteristics.