提出了一种基于区域分割和Contourlet变换的图像融合算法。首先,对各源图像做区域分割,并利用区域能量比和区域清晰比的概念来度量和提取区域信息;然后,对各源图像进行多尺度非子采样Contourlet分解,分解后的高频部分采用绝对值取大算子进行融合,低频部分则采用基于区域的融合规则和算子进行融合;最后进行重构得到融合图像。对红外与可见光图像进行了融合实验,并与基于像素的à trous小波变换和Contourlet变换的融合效果进行了比较。结果表明,采用本文算法的融合图像既保留了可见光图像的光谱信息,又继承了红外图像的目标信息,其熵值高于基于像素的融合方法约10％,交叉熵仅为基于像素的融合方法的1％左右。

An image fusion algorithm using region segmentation and contourlet transform is proposed. Firstly, region segmentation is done for source images, and ratio of region energy (RRE) and ratio of region sharpness (RRS) are presented to measure and extract region information. Then multiscale decomposition of the source images is carried out with the nonsubsampled contourlet transform. The high-frequency coefficients are fused with larger absolute value operator, and the low-frequency coefficients are fused with the region-based fusion rules and operators proposed. Finally the fused coefficients are reconstructed to obtain the fusion image. The fusion experiment for infrared and visible images is taken, and comparison between pixel-based à trous wavelet transform and contourlet transform is given. The results show that the fused image obtained by the presented algorithm can not only hold spectrum information of the visible image, but also inherit object information of the infrared image. The entropy of the fused image is about 10% greater than that of pixel-based fusion methods, and the cross entropy of the fused image is only 1% of that of pixel-based fusion methods.