Proceedings Article | 4 November 2010
KEYWORDS: Image fusion, Wavelets, Visual system, Principal component analysis, Distortion, Visualization, Multispectral imaging, Spatial resolution, Silver, Wavelet transforms
Focusing on the fusion problem of the multispectral (Ms) and panchromatic (Pan) images from the same scene, a novel
image fusion method is proposed based on nonsubsampled contourlet transform (NSCT) and human visual system
(HVS). The most traditional fusion methods are IHS, PCA and Brovey transforms, which can bring the phenomenon of
spectral distortion. Avoiding this problem, the wavelet transform is usually used in image fusion in recent years, but it
only can capture limited directional information. Compared with the wavelet and other transforms, the contourlet
transform has the characteristics of multi-scale, time-frequency localization and multi-directions. However, due to the
lack of translation invariance of the contourlet transform, this paper uses the nonsubsampled contourlet transform. The
basic procedure consists of four steps. Firstly, the NSCT is performed on Pan image and the intensity component I of Ms
image with HIS transform, which can obtain the low frequency subband and highpass directional coefficients of each
image. Then a new fusion rule is presented based on HVS: corresponding low frequency and highpass components are
divided into several blocks, and contrast variance of every block is calculated, followed by a selection of an adaptive
threshold which can be used to construct the new low frequency and highpass components. The blocks with higher
contrast variance will be chosen. Thirdly, the new intensity component Inew with high spatial resolution is obtained by
performing the inverse NSCT on the attained coefficients. Finally, the inverse IHS using Inew component is performed
and the new fused multispectral image is obtained. According to the quantitative evaluation criteria, it is shown that the
proposed method can effectively preserve spectral information, improve spatial information of the fused image, and
outperform the traditional IHS, PCA, Brovey, wavelet and contourlet methods.
