Research On Multi-source Remote Sensing Image Fusion

The development of remote sensing technique provides people with vast amountsof final data. Due to some factors such as the limitations of satellite sensors, they areunable to supply detailed information of the targets. Multi-source remote sensingimage fusion technique offers an effective way to make full use of multi-sourceremote sensing images.The main work of the thesis includes:A new fusion strategy of Curvelet transform is proposed. The low-frequencycoefficients of the Curvelet transform decomposition coefficients contain most of thespectral information. In this part, a strategy based on local window standard deviationis adopted, which can retain all the spectral information of t he multi-spectral imagewhile introducing some spatial detail information of panchromatic image. Spatialdetail information is mainly contained in high-frequency coefficients. The sharper ofthe image parts, the higher of the energy they contain. In this paper a strategy of localenergy based on the structural similarity is adopted, which can effectively aggregatehigh-energy parts, and thus enhance the spatial resolution of the image.A new image fusion algorithm based on two-dimension PCA and Curvelettransform is proposed. Two-dimensional PCA has good performance in preservingspectral information as well as enhancing spatial resolution; Curvelet transformexcels in describing geometric features such as image edge and texture features andupgrading image’s spatial resolution. To improve final fusion’s spectral and spatialresolution, a new algorithm possessing the superiorities of2DPCA and Curvelettransform is proposed in this paper.Experiments are carried out to validate the overall performance of the algorithms,and the final results are validated from both subjective and objective aspects. Byusing QuickBird data, experiments are carried out with the algorithms mentioned inthis paper respectively. The results of the experiments show that: the improve dCurvelet algorithm and the new algorithm have better visual effect and a much moreclear contour of the target, and the spectral features are more close to ground truth.Meanwhile, compared to multispectral image, spatial resolution has been greatlypromoted.