基于主成分分析和非下采样SPT的多幅图像融合方法

Fusion of Multiple Images Based on Nonsubsampled SPT and Principal Component Analysis

为了将好的高、低频融合规则模型推广到多幅源图像的融合,提出了一种基于主成分分析和非下采样Steerable金字塔变换的多幅图像融合方法.首先,对源图像进行非下采样Steerable金字塔分解,得到低频子带、高频子带及系列带通子带系数.其次,对于低频分解系数采用基于主成分分析的加权融合;而对系列带通子带图像和高频子带图像的分解系数采用基于局部均值方差的选择与加权平均融合,得到融合图像的非下采样Steerable金字塔系数.最后,经过非下采样Steerable金字塔逆变换得到融合图像.通过对可见光与红外图像以及多聚焦图像的两组融合实验,证明了该方法不仅适合多幅图像的融合,而且还可以获得比传统的图像融合方法质量更高的融合图像.

In order to extrapolate the good models of high and low frequency fusion rules to the fusion of multipleimages,a new fusion algorithm of multiple images based on nonsubsampled steerable pyramid transform(NSSPT)and PCA is proposed. Firstly,the NSSPT is performed on the source images,thus both the low and high frequencysubband coefficients together with varieties of directional bandpass subband coefficients are obtained. Secondly,forthe low frequency subband coefficients,a low frequency fusion strategy based on principal component analysis ispresented while for the high frequency subband coefficients and varieties of directional bandpass subband coefficients,aselection based on the local mean variance with the weighted average scheme is presented,thus the nonsubsampledsteerable pyramid coefficients of the fusion image are obtained. Finally,the fused image is obtained by performingthe inverse NSSPT. Experiments are implemented with two categories of image fusion—multi-focus images,the visibleand infrared images,the result shows that the proposed algorithm is not only suitable for multiple images,but alsocan achieve better fusion performance than the traditional image fusion method.