基于区域显著性的双波段图像融合方法

为了在红外与可见光图像融合中保留更多有效信息,提出一种基于区域显著性分析的融合方法.通过区域分割以及多分辨率对比度分析,获得图像的尺度不变区域显著性图（RSISM）.利用RSISM能够有效表达区域的显著性特征,合理区分不同性质的区域；根据RSISM划分显著性区域、背景区域及中间区域,对各区域制定相异融合规则,并在非降采样轮廓波变换（NSCT）变换域上融合双波段图像.实验证明,与传统方法相比,该方法能够更好地保留显著性区域的红外特征及其他区域的细节信息,同时对背景热辐射干扰不敏感,有较好融合效果,并能够拓展应用于动态图像的融合中.

Dual band image fusion method based on region saliency

In order to efficiently preserve information in infrared and visible images fusion, a region based image fusion method was proposed using region saliency analysis. Region based scale invariant salient map (RSISM) was achieved by image segmentation, pyramid decomposition and multi-scale contrast analysis. RSISM could well represent the salient information of a region and distinguish different regions. Image was separated into three parts according to RSISM, including salient area, background and other regions. Proper fusion rules were formulated for each region. Dual band images were finally fused in NSCT domain. Results showed the proposed method has a better performance compared to other fusion methods. According to experimental results, the proposed method is thus proven to be a good one for image fusion which can keep both target features and background details. Besides, with the combination of spatially salient feature and dynamic information between adjacent frames in image sequence, dynamic salient map can be achieved and used for sequential image fusion．