【作者】 强赞霞；

【导师】 彭嘉雄；

【作者基本信息】 华中科技大学 ， 模式识别与智能系统， 2005， 博士

【摘要】 近些年来,图像融合已成为图像理解和计算机视觉领域中的一项重要而有用的新技术,多源遥感图像数据融合也成为遥感领域的研究热点,其目的是将来自多信息源的图像数据加以智能化合成,产生比单一传感器数据更精确、更可靠的描述和判决,使融合图像更符合人和机器的视觉特性,更有利于诸如目标检测与识别等进一步的图像理解与分析。它在军事、民用方面有着极为广泛的应用。本文对来自不同途径的多源遥感图像的融合方法及其应用进行了研究,主要工作如下: 图像空间配准是多源遥感图像融合前非常重要的一步,其误差大小直接影响融合结果的有效性,它是进行多源遥感图像数据融合的前提与基础。在研究了信号的傅立叶变换的性质和特点的基础上,将傅立叶相位相关技术进行扩展,用于实现遥感图像自动配准。该方法的主要优点是在不需要寻找控制点和传感器参数的情况下进行图像自动配准。通过对数-极坐标变换、利用傅立叶变换的比例特性、旋转特性和相位相关技术确定图像间的比例、旋转和平移关系,实验结果表明了此方法的可行性和有效性。提出了一种基于互信息相似性判据的分层遥感图像配准方法,通过小波变换构造图像金字塔,从金字塔的最顶层开始搜索,根据互信息最大的原则确定图像间的变换参数,并作为下一层搜索的粗略位置,然后逐层细化,实现由粗到细的搜索过程。将此算法应用于遥感图像,得到了有效、精确的配准结果。在分析了基于傅立叶变换的配准方法和基于互信息准则的分层配准方法的优缺点的基础上,提出了将傅立叶变换和基于互信息相似性判据相结合的分层图像配准方法,克服了基于互信息的分层配准方法耗时长的缺点,且利用分层细化的搜索策略增加了基于傅立叶变换的的误差修正过程,提高了配准精度。分析了多源遥感图像融合的层次、模型、结构及其特点,着重研究和分析了像素级多源遥感图像融合的概念、方法,对多源遥感图像融合效果的评价方法进行了深入的研究,在已有的评价方法的基础上将它们进行了整理、分类,提出和建立了一套对图像融合效果及融合方法性能进行定性、定量评价的方法和准则。研究和分析了几种常用的多源遥感图像的方法(包括多光谱图像的融合方法):直接平均法、HIS 变换法、Borvey 法、主成分分析法和高通滤波法等,通过实验比较了这更多还原

【Abstract】 In recent years, image fusion has been an important and useful technique for image analysis and computer vision. Multi-sensor image fusion has attracted many attentions in remote sensing area. The aim of image fusion is to combine multiple source image data from various sensors intelligently and to obtain more detailed, complete description and decision than any of the individual source images. As a result of this processing, the fused image is more useful for human and machine perception or further image processing tasks such as object detection and recognition. In this dissertation, we focus on the techniques and applications of the multiple remote sensing image fusion. In this dissertation, an automated FFT-based technique is proposed for registration of remote sensing images having different spatial resolution and relative rotation and shift. The proposed method is an extension of the phase correlation technique. The main characteristic of the algorithm is to align two images automatically without requiring either control points or sensor’s parameters. Fourier scaling properties and Fourier rotational properties are used to find scale and rotational movement by Log-polar coordinate transform. The phase correlation technique determines the transnational movement. Experimental results show that satisfactory effect has been obtained by applying our method. A pyramid approach for remote sensing image registration based on mutual information is presented. The image pyramid is obtained by using the wavelet transform. Coarse-to-fine multi-resolution search approaches have been proposed to increase accuracy and efficiency. An exhaustive search algorithm is applied at the coarsest level of the image pyramid. Registration at higher levels can be performed with the result at the pervious level serving as the initial condition. Our algorithm has been applied on remote sensing images. Main advantages and drawbacks of two registration methods mentioned above were analyzed. In order to save the computarional time and improve registration accuracy, a new coarse-to-fine hierarchical strategy for image registration based on the combination of FFT with mutual information approaches. Based on the comprehensive review and summarization of previous articles and researching achievements, basic concepts, levels, models, structures, techniques and applications of the multi-source remote sensing image data fusion, especially for pixel-based image fusion, are discussed. Furthermore, assessments of the image fusion performance are studied. Several evaluation criteria are presented in this dissertation. These evaluation criteria are classified according to condition and purpose. The methods of pixel-based remote sensing image fusion are analyzed and studied (including direct average method, HIS transformation, The Brovey method, PCA method and HPF method), these methods are compared qualitatively and quantitatively according to sharpness, information content, spatial resolution and preserving the spectral characters of source multi-spectral images and so on. We mainly studied and discussed multi-resolution image fusion approaches. These approaches are classified into two types: direction and no-direction multi-resolution analysis. For no-direction multi-resolution analysis, three pyramids are presented including laplacian pyramid, contrast pyramid and ratio pyramid. The decomposition, reconstruction, and image fusion based on these pyramids were introduced. The effects of the type of pyramid, the number of pyramid decomposition level, the scheme of fusion and the size of local region to the fusion result were investigated. Some important conclusions were drawn through a great deal experiments. For direction multi-resolution analysis, sensing image fusion techniques based on wavelet transform and direction gradient pyramid transform were proposed. Their characters, decomposition, reconstruction, and image fusion based on these two techniques were detailed. Furthermore, we investigate a pixel level image fusion algorithm based on a novel multi-resolution transform called steerable pyramid, which is both aliasing free and translation invariant. The characteristics of steerable pyramid and the scheme of image fusion by using steerable pyramid were discussed. We analyzed the effect of the number of orientation band-pass filters and the decomposition level to the fusion result. In order to resolve the problem of dim target detection, a novel approach based on image fusion and mathematical morphology was proposed. First, the original images were fusedusing steerable pyramid transform technique based on effective fusion scheme. The targets can be enhanced and clearer in fused image. Second, mathematical morphology method was applied to detect the target based on the fused image. The experimental results show that the effect of our method is satisfactory. A series of experiments on image registration, fusion, and object detection are given in this dissertation. We also got some worthy conclusions and put forward some new conceptions.更多还原