【作者】 谭勇；

【导师】 徐佩霞； 徐守时；

【作者基本信息】 中国科学技术大学 ， 通信与信息系统， 2009， 博士

【摘要】 由于受到卫星遥感图像的分辨率限制,遥感图像中目标识别的精确度和识别程度都受到了很大的限制,继续发展高分辨率遥感技术是解决这个问题的有效方法,但也面临着越来越大的困难。本论文提出了一种基于遥感目标图像分辨率增强的目标精确识别、详细描述或分析的处理方案,通过序列遥感目标图像的超分辨率重建方法获得空间分辨率得到增强的目标图像,突破了卫星遥感图像的分辨率限制,为遥感目标的精确识别和后续处理创造了有利条件,在军事和民用领域都具有非常重要的意义。本论文对遥感目标图像空间分辨率增强技术所涉及的序列图像配准、序列图像超分辨率重建、遥感目标图像获取、序列目标图像形成等关键技术进行了深入研究。序列图像的超分辨率重建方法是遥感目标图像空间分辨率增强技术的核心。本论文将基于图像传感器阵列的超分辨率图像获取模型应用到序列图像的超分辨率重建中,提出了基于DCT变换的预处理共轭梯度迭代的序列图像超分辨率重建方法;采用分数延时滤波器,把基于传感器阵列的序列图像超分辨率重建模型转变成图像的反降晰模型,提出了基于分数延时滤波器和DCT变换的序列图像超分辨率重建新方法。这种序列图像超分辨率重建模型向图像反降晰模型的转变具有重要的意义,它为图像反降晰领域中的众多研究成果在序列图像超分辨率重建中的直接应用提供了条件。序列目标图像的高精度配准是序列目标图像超分辨率重建的基础。本论文提出了基于图像的线型特征谱线的图像旋转和缩放参数配准方法,结合改进的基于Radon变换的图像平移参数配准方法,实现了在具有平移、旋转和缩放关系的图像之间的高精度配准。论文分析了遥感图像目标形状在基于遥感目标图像分辨率增强的目标精确识别、详细描述或分析方案中的作用,研究了结合目标形状描述的基于PDE方法的遥感目标图像获取方法,提出了结合迭代实边缘特性的遥感图像复扩散增强方法和结合目标形状先验信息的图像多区域分割方法。论文研究了从遥感目标数据库中提取感兴趣目标的目标图像以形成序列目标图像的方法,提出了基于水平集方法的目标形状距离估计新方法,构造了一个基于目标形状距离的目标同一性判定结构以确保序列中的目标图像都属于同一目标。为了精确校正序列目标图像之间的图像旋转和图像缩放,论文研究了基于B样条类函数插值的高性能图像旋转和缩放方法。设计了采用基于分数延时滤波器和DCT变换的超分辨率重建方法的序列目标图像仿真实验方法,通过仿真实验取得了预期的超分辨率目标图像重建效果。论文的最后对研究内容进行了总结,并对后续的研究工作进行了展望,指出了进一步改进算法性能、开展实际应用研究的途径和方法。更多还原

【Abstract】 Due to resolution limitation of remote sensing images,the accuracy and extent of object recognition for remote sensing images is constrained greatly.The developing of high resolution remote sensing technology is an effective method for this problem,but it will be confronted with more and more difficulty.In this thesis,a new processing scheme of accurate object recognition,detailed object description or object analysis based on the resolution enhancement of remote sensing object images is proposed.The resolution enhancement of remote sensing object image based on superresolution reconstruction algorithm of image sequence is achieved which breaks through the resolution limitation of the remote sensing images and creates favorable conditions for accurate object recognition and further object processing,so this new processing scheme is of great significance both in the military and civil domain. Some key technologies of the resolution enhancement of the object images including the high accuracy registration method of object image sequence,the superresolution reconstruction method of object image sequence,the object image acquisition method and the composing method of object image sequence are studied thoroughly.The superresolution reconstruction algorithm is the core of the resolution enhancement of the object images.The superresolution image acquisition model based on image sensor array is applied to the superresolution reconstruction of image sequence and then a new algorithm of superresolution reconstruction of image sequence based on the DCT Transform and preconditioned conjugate gradient method is proposed.By using the fractional delay filter(FDF),the model of superresolution reconstruction of image sequence is turned into the model of image deblurring and a new algorithm of superresolution reconstruction of image sequence based on FDF and DCT Transform is then proposed.This model change is of great significance because a lot of algorithms of image deblurring can then be applied directly to the image superresoluiton reconstruction.The high accuracy image registration of the object image sequence is the base of the superresolution reconstruction of object images.A high accuracy image registration method for image rotation angle and image resizing factor based on the characteristic spectral line of images is proposed.Combining with the improved shifted image registration method based on the Radon Transform,the new image registration method attains the high accuracy estimation of image shift,image rotation angle and image resizing factor successfully.The function of object shape in the processing scheme of accurate object recognition and further object processing based on the object image resolution enhancement is analysed.The remote sensing object image acquisition method based on PDE method and the object shape is studied and the image enhancement algorithm with complex diffusion based on iterative real edge attributes,the multiple regions image segmentation algorithm combined with the object shape prior are proposed.The method of selecting the object of interest from the database of remote sensing object images to compose the object image sequence is studied and a new method of estimating the object shape distance based on level set is proposed.Then an identity determinant structure based on the object shape distance is designed to ensure all the object images in the sequence belong to the same object.The constitution model of object image sequence is analysed and in order to justify the existing image rotation and resizing accurately before the reconstructing operation,a new image rotation and resizing algorithm with high performance is proposed which is based on B-spline interpolation.Then the simulation method of superresolution reconstruction of object image sequence using the algorithm based on FDF and the DCT Transform is designed,and the expected superresolution result is obtained through a lot of simulation tests and performance analysis.Finally,all research contents are summarized and some important problems which need further study are discussed,such as the method of further improving the algorithm performance and the way to bringing the whole technology of superresolution reconstruction of object image sequence into effect.更多还原