【作者】 张强；

【导师】 周海银；

【作者基本信息】 国防科学技术大学 ， 数学， 2007， 硕士

【摘要】 作为多源数据融合领域的一个重要分支,多源图像融合研究的是如何将不同类型传感器获取的图像数据,采用一定算法将其优势或互补性信息有机结合起来产生新图像的技术。其优点是不仅能够克服单幅图像在几何、光谱和空间分辨率等方面存在的局限性和差异性,进而提高图像的质量;而且有利于对感兴趣的目标进行定位、识别和变化检测等。在信息化高度发展的今天,多源图像融合技术已经成为数据融合领域中不可或缺的技术,并在许多军事和民用方面有着重要的应用。建立多源图像融合机制包含三大基本要素:合理的图像融合流程、有效的融合规则的选择以及综合的融合性能评价。本文的研究工作就是针对这三大基本要素展开的,主要工作如下:1深入讨论了多源图像融合的不同层次和模式,对基于像素级的空间域和变换域上的多源图像融合方法进行了分类研究,并将这些方法进行融合实验,在此基础上,分析了融合方法的适用性及其边界条件;2对现有融合结果评价方法加以总结,在此基础上归纳出基于信息量、基于统计特性、基于信噪比、基于梯度值等四类十二项融合结果评价指标,从而为融合性能评估提供了定量的评价准则;研究了以主观视觉分析为主,以客观定量分析为辅的多因素图像融合质量评价方法,提出了闭环评价系统的框架,为后续的研究提供了理论支撑;3通过对各种不同的图像融合方法和融合类型的分析,研究了多源图像融合流程建模方法,建立了一个以图像数据源为入口,以融合模型和融合规则为核心,以融合后图像为输出,以多因素综合判据下的图像融合评价体制为终点的闭环式图像融合流程,最后通过融合实验验证了该模型的有效性;4在运用小波变换法进行图像融合的基础上,针对局部方差融合规则和高通滤波各自的优缺点,研究了一种新的融合规则的选择方法,提出了基于局部方差和高通滤波的小波变换图像融合算法,并对全色图像和多光谱图像进行了融合实验,融合结果的性能评价表明该算法是有效的,融合后的结果图像既保留了多光谱图像的光谱特性,且同时具有较高的空间细节表现能力。更多还原

【Abstract】 As the important research branch in multi-source data fusion region, research on multi-source image fusion is how to obtain a new image with self-complementary and advantage information by combine different image data from different sensors together based on some fusion algorithm. The purpose is not only to overcome the limitation and the discrepancy of single image in geometry, spectrum and spatial resolution and then to increase image quality, but also to be good for the next application including target identification, feature extraction and change detection and so on. In this day with high-speed informationization development, multi-source image fusion technology has become an absolutely necessary technology in data fusion region, and also has paid important function in some military and civil regions.There are three basic elements for establishing a complete multi-source image fusion system, which are logical image fusion flow, effective fusion rule selection and integrated fusion performance evaluation criterion. The research work of this paper is just carried out according to these three basic elements, which are depicted as follows:1 Different levels and modes for multi-source image fusion are deeply discussed; image fusion methods based on spatial domain and transform domain of pixel level are classified and researched, and then their simulation are carried out. Based on the above analysis, the applicability and its boundary condition for each fusion method are analyzed;2 Existing evaluation criterions for image fusion are summarized, and the integrated evaluation system with four classes and twelve kinds standards including information quantity, statistic characteristics, signal-noise ratio and gradient value are presented, which can provide a quantitative evaluation criterion for image fusion; and then a new evaluation method for image fusion based on multi-factors criterions taking visual analysis as main measure and quantitative analysis as auxiliary measure is researched, and a close-loop evaluation system framework is presented, which can provide a theoretical supporting for next research;3 Based on the analysis for different image fusion method and fusion types, a modeling method on multi-source image fusion process is analyzed and researched, and moreover a closed image fusion process taking image data as input, fusion model and fusion rule as core, fused images as output and fusion appraisal system based on multifactor integrative criterion as terminal is presented, and finally, the validity of this fusion process is validated by relevant fusion experiments;4 In view of each fact of local deviation and high-pass filtering in image fusion and based on using wavelet transform to do image fusion, a new selection method for fusion rule is studied. Image fusion algorithm based on local deviation and high-pass filtering of wavelet transform is advanced. Taking the fusion between PAN image and multi-spectral image as example, the simulation results show that compared with the other single method, the new method presented is clearly better in preserving spectral and improving spatial presentation.更多还原