基于线特征的多源遥感影像配准研究-摄影测量与遥感专业论文.docxVIP

ABSTRACTAbstract ABSTRACT Abstract Remote sensing has been the one and only technology so far that could provide dynamic global—wide earth·observation information．There are various image—forming patterns for remote sensing．The fusion of multi—source remote sensing images would give birth to all image of which information is more precise，more integrated and more reliable than that of a single image．Before the fusion，the multi—source remote sensing images must first be registered．With the development of image segmentation and edge detection techniques in recent decades，linear·feature based image registration has been becoming the kernel of image registration topic． Firstly,the thesis gives a brief description of the basic theories and methods of image registration，which focuses on the registration methods that take use of the linear features and their rationales，and what’s more，on comparing and analyzing their merits and drawbacks． Then when aiming to utilize straight-·line··segment features to register multi—source remote sensing images，the thesis introduces a line extraction method for multi—source remote sensing imagery named as Partition Hough Transform．Then，the thesis presents a new registration strategy named as Robust LSA-MIHT,which overcomes the drawbacks of Modified Iterated Hough Transforill(MIHT)of low convergence speed and unstable iteration procedure by integrating Huber Variable Weight Iteration Least Square Adjustment with MIHT．The modified Robust LSA—MIHT strategy is not only has the same merits of MIHT,such as strong robustness，low requirement for prior information about the image pair,and assunles no prior information of the correspondence of the primitives，but also converges more faster and stable than the MIHT strategy,which makes it more practical Experimental results using real data proved that the proposed registration method， of which strategy is to extract line features first from image pair and then to registry them with Robust LSA—MIHT strat