Muti-source Remote Sensing Image Registration Based On Feature

Image registration match the images obtained in different time, on different angle, from different sensors and another conditions. It is a critical technology in image processing and analysis, which is the first issue should be settled in image fuse, object identify, movement analysis and so on. It has been widely applied in the aspects of remote sensing image processing, photogrammetry, medical image analysis, computer graphics, computer vision and so on.The primary aim of image registration is to establish the space coherence of two images, by calculate transform parameter and image transform. That is to say, image registration should make all the points or important applied points in two images achieve matching, and the most important step is finding out the corresponding point of two images.The traditional methods of image registration obtain the control points by handcraft of operator, which can't meet real-time and automatic image process system. The automatic registration algorithms are intelligent and self-acting without handcraft, and it is the major direction, also the ultimate goal of image registration. The current automatic registration algorithms can be divided into pixel-based and feature-based. The automatic methods based on feature are the focus in this paper, and we have completed the following works:1. Introduce the concepts, principles, mathematical model and other correlative information of image registration. Then we introduce the common steps of feature-based registration method, which could be applied in many kinds of images.2. We introduce a series of extraction operators of feature points, and study the characteristic of Harris operator, SUSAN operator, SIFT operator by analyzing their performance and applicability. Then we present three automatic registration methods which are based on Harris, SUSAN and SIFT respectively, and analyze their performance by a series of experiment.3. We introduce a kind of virtual point is proposed, three forms of virtual points and their properties, then the virtual points are applied in the research of image registration and a multi-sensor image registration algorithm based on virtual points is proposed. Firstly the virtual points in two kinds of image are extracted and used as candidate control points, then the accurate pairs of control points are gained by the future matching, the automatic registration between the two images is realized with high registration accuracy. Experimental results show that, the proposed algorithm is feasible and has a fine effect.