基于非采样Contourlet变换的高光谱图像融合算法

高光谱图像数据具有大量波段,波段之间的相关性较高,可以利用数据融合技术来降低其分析难度.提出了一种新颖的基于非采样Contourlet变换的高光谱图像融合算法.首先对高光谱数据进行特征图像提取;再将提取出的多幅特征图像分别进行非采样Contourlet变换,并按区域能量进行加权融合;最后对融合系数进行非采样Contourlet重构.实际的OMIS高光谱遥感图像融合的实验结果表明,所提基于非采样Contourlet变换的加权融合算法能够很好地保持图像的空间特性和光谱特性,且效果明显好于典型的小波变换和Contourlet变换的融合方法.

Research on Fusion of Hyperspectral Images based on Nonsubsampled Contourlet Transform

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.