Research On Adaptive Registration Of Multi-Sources Remote Sensing Images

Image registration is the process of overlaying two or more images of the same scene taken at different times, from different viewpoints, and/or by different sensors. With the development of both remote sensing technology and the new sensor networks, it is improved to access to remote sensing image data. However, when we use these multi-source remote sensing images as the source of fusion, target change detection, target identification, as well as dealing with multi-source collaborative processing, we must do the multi-source image registration work first. The level of registration accuracy directly affects the quality of user's application. So this paper examined the multi-source remote sensing image registration techniques as key technologies of a collaborative system. Methods of registration require both computing power and the matching accuracy achieve good results.First of all, this dissertation researches the existing multi-source image registration techniques and introduces the principles of analysis. Take a variety of methods and comparison in consideration and at the same time notice the characteristics of remote sensing image registration, we proposed the common elements of multi-sensor image registration technology. Meanwhile we study and analyze various technical aspects of the difficulties and problems.Secondly, this dissertation focuses on key technologies involved with multi-sources remote sensing images registration and takes research on extract methods of the control point. Then various corner detection algorithms are completed and we can analyze and compare advantages and disadvantages of them. Concerning the problems about the SAR image noise can impact on the corner detection, the stable featured points can be got by the method which is applying down sampling algorithm to remove noise corner points.Then, according to the disadvantage that traditional image registration method only employs single feature points,we can propose a new method based on incorporated various features. Besides, this image registration method has the adaptive characteristic of regional extension. By the handling ways of regional extension, the suited control dot pairs can be obtained. This method can strengthen the robustness of image processing. With regard to the question that different sensors image registration can easily produce a lot of wrong matched controlling points,the strategy of random sampling iteration is proposed up. Especially regarding to the different sensors image registrations which have big recognition differences, this paper puts forward image corner detection methods which is based on regional features. The space characteristic differences of two images can be reduced by using regional featured images. Afterwards, the matching accuracy is acceptable for the subsequent processing.Finally, in this dissertation, the evaluation of image registration precision will be studied. This study can implement image change detection and meanwhile, it can inspect and verify the influences registration precision has on change detection.