Abstract: A novel fusion method is proposed for multispectral and panchromatic satellite images using Principal Component Analysis（PCA） and Nonsubsampled Contourlet Transform（NSCT）. This method first performs PCA on MS, and NSCT on PAN and the first principal component（PC1） to get corresponding low-frequency and high-frequency coefficients. Then fuses the approximation coefficients using PCA again for the tradeoff between the spectral and spatial information, and fuses the subbands coefficients based on Structural Similarity Index（SSIM） and local Sobel average gradient for the spatial detail information. A fused image is formed through inverse NSCT and inverse PCA. Experimental results show that the proposed fusion method can effectively preserve spectral information while improving the spatial quality, and outperforms the general IHS-, PCA-, Wavelet-, Contourlet-based fusion methods.

Key words: image fusion, Principal Component Analysis（PCA）, Nonsubsampled Contourlet Transform（NSCT）, Sobel gradient, Structural Similarity Index（SSIM）

摘要： 研究了主分量分析（PCA）和非下采样Contourlet变换（NSCT），提出一种新的多光谱图像和全色图像的融合算法。该方法对多光谱图像进行PCA变换，对所得的第一主分量（PC1）以及全色图像进行NSCT变换。对二者的低频近似系数再次进行PCA变换以寻求多光谱信息和空间信息的平衡；对于高频细节系数，通过结构相似性指标（SSIM）和局部Sobel梯度进行融合，进一步提高空间信息量；经过逆NSCT和逆PCA变换得到融合图像。实验结果表明，提出的方法在增强融合图像空间细节表现能力的同时，尽可能地保留了多光谱图像的光谱信息，优于传统的基于IHS、PCA、小波变换和Contourlet变换的融合方法，是有效可行的。

关键词: 图像融合, 主分量分析, 非下采样Contourlet, Sobel梯度, 结构相似性指标