Abstract: This paper presents a novel strategy for image fusion by simulating visual attention mechanism. Firstly, source image is decomposed to low and high frequency components by Laplacian pyramid method. Then different frequency level is fused with different fusion rules. The fused pixels in low frequency component are chosen based on the rule of maximum saliency. And those pixels in high frequency component are chosen by the rule of weighted average based on correlation. The final fused image is reconstructed by Laplacian pyramid transform. Since pyramid transform can provide the multi-resolution of image, but cannot distinguish the importance of pixel in the image, while the saliency detection can locate the saliency region in image, but very sensitive to the noise. Combining those two algorithms may complement each other and bring benefits. The experimental results show that the performance of the proposed method is better than that of other published pyramid transform based fusion algorithms. It can be utilized to multi-focus image fusion, multiband image fusion and multispectral image fusion cases.

Key words: image fusion, pyramid transform, saliency detection

摘要： 提出了一种利用人类视觉机制进行图像融合的算法。首先对源图像进行金字塔分解；接着对低频和高频分量采用不同的融合策略，低频分量依据最大显著性准则选择融合像素，高频分量利用相关性加权准则选择融合像素。初步融合后的低频和高频分量经金字塔重建获得最终融合结果。金字塔变换可提供多分辨率的图像表示，但不区分图像区域的重要性；而视觉显著性检测可定位图像最显著区域，但对噪声敏感；两算法的结合能取长补短，获得好的融合结果。实验表明，提出的方法优于已发表的其他基于金字塔变换的图像融合算法，适用于多聚焦图像、多波段图像和多光谱图像融合。

关键词: 图像融合, 金字塔变换, 显著性检测