Research On The Registration Of Optical Remote Sensing Images Using Multi-Feature Combination

Visible images obtained by optical remote sensor,which is the most common type of the payloads among the earth observation systems,are pervasively applied in varieties of remote sensing products with its excellent observation characteristics such as intuition and intelligibility.The increasing means of optical remote sensing image acquisition lead to an enormous data quantity,and to implement the applications adapted to multi-source images can greatly improve the service ability of geographic information science and technology.With this realistic request,one of the key questions is how to get a stable and high-precision registration result.In this dissertation,a registration method based on multi-feature combination is proposed,which is adapted to large differences of angle,scale,illumination or resolution between multi-source remote sensing images.The main work of this dissertation is listed as follows:1.The principle of image registration technology and the research status of the subject are both introduced.The main problems existed in processing of optical remote sensing images with large differences are summarized.2.The basic theory and classic algorithms of gray-based and feature-based registration methods are studied.A multi-feature-combined registration method is proposed,and its advantages are clarified through comparative analysis.3.A new combination method of five types of image invariant features is proposed after a further study on performance differences of multi features.The analysis about each feature’s advantage is also given.4.After extraction of homonymous points,the dissertation proposed a modified principle of random sample consensus(RANSAC)based on point elimination.The RANSAC principle and uniformity analysis are applied to support the coarse registration,and the least squares optimization is added during the precise registration process.The triangulation network is built by divide-and-conquer method to realize the registration operation using small panels.Through the experiment,it is corroborated that the method proposed in this dissertation can obtain a stable and high-precision registration processing with multi-source optical remote sensing images with big differences of angle,scale,resolution or illumination.