Research on Fusion of Hyperspectral Images Based on Nonsubsampled Contourlet Transform

常威威,郭雷,付朝阳
DOI: https://doi.org/10.3969/j.issn.1002-0640.2009.10.032
2009-01-01
Abstract:There are hundreds of bands in hyperspectral data and the bands are highly correlative.The difficulty of analysis could be reduced by using fusion technology.A novel hyperspectral image fusion algorithm based on nonsubsampled Contourlet transform is proposed.Firstly,feature images extraction is applied to hyperspectral images;then,several feature images extracted are separately decomposed into multiresolution representation by nonsubsampled Contourlet transform and the coefficients are fused according to the weights which are calculated with the energy of the corresponding neighbor region.Finally,the fused coefficients are reconstructed by reverse nonsubsampled Contourlet transform.Experimental results for real OMIS hyperspectral images show that the proposed fusion algorithm of hyperspectral images based on nonsubsampled Contourlet transform could well retain spatial and spectral feature.In addition,it outperforms than the typical fusion algorithms such as wavelet transform and Contourlet transform.
What problem does this paper attempt to address?