【作者】 田养军；

【导师】 薛春纪；

【作者基本信息】 长安大学 ， 地球探测与信息技术， 2009， 博士

【摘要】 图像融合是以图像为研究对象的数据融合，它将同一对象两幅或多幅图像合成到一幅图像中，使得到的图像比原来任何一幅图像都更容易被人们所理解，并能反映多源图像中的信息，为遥感图像分析、图像理解和计算机视觉具有非常重要的意义。但是至今，图像融合尚无通用的融合理论和方法，现存的各种方法各有自己针对适应范围和特点。本文对前人融合算法进行大量对比试验，提出一种基于提升小波分解曲波变换的多源遥感图像融合方法，通过遥感图像融合，改善遥感图像质量，该方法具有一定的实用价值。深入研究了小波、提升小波、曲波的基本理论、本质特征以及它们内在联系；比较全面探讨小波、提升小波、曲波在遥感图像中的技术应用与发展及其局限性；根据提升小波、曲波各自的特点，提出提升小波曲波遥感图像的融合扩展方式，以及建模方法；最后，以实验对本方法与其它方法进行了分析比较研究。应该指出的是：本文方法对遥感图像处理问题具有比较广泛的适应性，它不仅可以进行图像融合，而且只需要简单地改变处理规则，就可以进行图像滤波、平滑、水印等处理。图像滤波、平滑可以作为独立一个研究区域，也可以作为遥感图像融合的预处理方法，水印是特征信息或特征图像在另一图像中的隐藏，也可以理解为图像融合，它们是对遥感图像融合方法的重要补充。本文主要研究工作包括：(1)全面综述遥感图像融合的现状和困难，指出遥感图像融合在图像理解、人工智能和视觉中应用的意义和重要性，分析了一些具体算法存在的问题。(2)全面探讨小波变换的意义和方法，研究了它产生的背景与发展现状，及其与提升小波、曲波等深刻内在关系。为遥感图像融合领域的研究拓宽了思路。(3)深入研究了基于提升小波图像融合技术，指出单纯提升小波的图像融合的局限性，并且将提升小波与其它多种算法结合，可以构成完善的提升模型。(4)研究脊波变换特点及不足，曲波变换过程及曲波变换在图像融合的方法，分析曲波变换在遥感融合图像中局限性及扩展方法。(5)分析遥感图像的特点，遥感图像面临的问题及发展方向。(6)研究提升小波分解曲波变换思想、方法及建模过程，同时对滤波、平滑、水印建模进行研究和探讨。(7)对提出提升小波分解曲波变换进行实验研究，并得出相应的结论。本文的创新点主要包括：(1)对提升小波变换、小波变换和曲波变换的特点进行深刻研究，提出了基于提升小波分解曲波变换遥感图像的融合方法。(2)比较了基于提升小波分解曲波变换遥感图像的融合方法与小波分解曲波变换遥感图像融合方法。(3)提出基于提升小波分解曲波变换的任务，它可以进行图像融合，还可以对遥感图像滤波、平滑、水印等处理，并对此进行实验性研究。存在问题：(1)小波选择问题，选择不同小波产生效果有差异；(2)遥感图像信息评价没有统一标准，一般是根据遥感图像处理方法和遥感影像特点，选择合适的方法进行评价。需要继续研究的问题：(1)基于提升小波分解曲波变换方法，使其应用进一步扩展；(2)结合其它技术进一步完善提升小波分解曲波变换方法，提高变换过程信息质量；(3)遥感影像处理后评价深入研究；(4)小波函数对图像变换特点的研究。更多还原

【Abstract】 Image fusion is data fusion to study the image.It wills synthetic a new image from sametargets in two images or multiple images.The new image is more easily understood than the originalevery image by the people,and can reflect information from the multiple original images.Imagefusion has very important significance for the remote sense image analyses,understanding imageand computer vision.But so far,there is no common theory and methods in fusion image.Extantvarious methods have aimed at the scope and the characteristic of themselves respectively.In thispaper,after several-hundred times of comparing calculation for old fusion method,a new way oflifting wavelet decomposable curvelet transformation is put forward in multiple original remotesensing images.Through remote sensing image fusion,the way can improves image quality.It hassome practical value.The fundamental theory,intrinsic characteristics and inner link for the wavelet,lifting wavelet,curvelet are studied in-depth.The technology application and development and their limitation ofthem in remote sense image with all-round investigation and discussion.Expansion way andmodeling method is put forward according to their characteristic.At last,this method and othermethod have carried out analytical comparison with experiment.It should be point out:the methodhas comprehensive range adaptability for remote sense image processing problem.It not only can usein remote sense image fusion,but also need to change the image handling regulation,you can filter,smooth and watermarks treatment image.The filtering,smoothing of the image can be used as anindependent studying area,and can also be used as pretreatment method for remote sense imagefusion.The watermark is that characteristic information or the characteristic image hides in anotherimage.It can be understand the image fused.They are important supplement methods for remotesense image fusion.This paper major studying contents include:(1) The difficulty and present situation of remote sense image fusion are summarized.It ispointed out that remote sense image fusion is very important in image understanding,artificialintelligence and computer vision.And some of concrete algorithms analyzed.(2) The significance and method of the wavelet transformation is systematically probed intowhich include backgrounds of its creation and present situation of development.The profound relationships among lifting wavelet and curvelet are revealed.This opened up our thinking indomain of remote sense image field.(3) The technologies of image fusion based on wavelet are deeply surveyed.And thelimitation of them only is apropos discussed.And it is pointed out that construct moreperfecting lifting model based on lifting wavelet combine with other various algorithms.(4) The feature and insufficient base on ridgelet transform process are discussed.The processand fusion method based on the curvelet transform are studied.The limitation and extendedmethod of the curvelet transform are analyzed in remote sense image.(5) Remote sense image characteristic are analyzed,the problem and development direction ofthe remote sense image are discussed.(6) The ideas,method,and model process of lifting wave decomposable curvelet transform areestablished.At the same time,filtering,smoothing,watermark model are studied and discussed basedon lifting wave decomposable curvelet transform(7) some experiment based on lifting wave decomposable curvelet transform are implemented,and investigated.And the corresponding conclusions are obtained.The major innovations of study included:(1) The relations and the characteristic among lifting wavelet transformation,wavelettransform,curvelet transform are summarily studied.The fusion method in the remote senseimages is put forward base on lifting wave decomposable curvelet transform.(2) The methods of lifting wave decomposable curvelet transform and the methods of waveletdecomposable curvelet transform are compared.(3) It is put forward that lifting wave decomposable curvelet transform not only can be usedmain tasks of remote sense image fusion,but also can use in image filtering,image smoothingand watermark.And it is studied by experiments.The problem in existence:(1) How to choose wavelet? The difference wavelet has the difference effect.(2) There is not uniform standard in remote sense image information evaluation.The future study tasks required(1) it is further extended that lifting wave decomposable curvelet transform is used in theremote sense image fusion (2) Algorithm of lifting wavelet decomposable curvelet transform needs further perfects withother technology,the quality of information and transformation process should be improved.(3) The evolution of remote sense images are studied deeply.(4) Wavelet functions are studied in the characteristic of the image transformation.更多还原