Remote Sensing Image Fusion Based on Gaussian Mixture Model and Multiresolution Analysis

摘要

A novel image fusion algorithm based on region segmentation and multiresolution analysis(MRA) is proposed to make full use of advantages of different multiscale transform. Nonsubsampled contourlet transform(NSCT) processes edges better than wavelet transform does. While wavelet transform handles smooth area and singularities better than NSCT does. As an image often includes more than one feature, the proposed method is conducted on the basis of Gaussian mixture model(GMM) based region segmentation. Firstly, transform the multispectral(MS) image into intensity, hue and saturation component. Secondly, segment intensity component into dense contour and smooth regions according to GMM and NSCT. And then gain new intensity component by fusing intensity component and high resolution image with Atrous wavelet transform(ATWT) fusion in smooth areas and NSCT fusion in dense contour areas. Finally transform the new intensity together with hue component, saturation component back into RGB space and obtain the fused image. Multisource remote sensing images are tested to assess this proposed algorithm. Visual evaluation and statistics analysis are employed to evaluate the quality of fused images of different methods. The proposed improved algorithm demonstrates excellent spectrum information and high resolution. Experiment results show that the new proposed fusion algorithm incorporating with region segmentation based improved GMM and MRA outperforms those algorithms based on single multiscale transform.