Study On Remote Sensing Image Matching

Image registration is an important analysis and processing technique, which located a image area from another image area or find out the correspondence between them, where the images taken at different times, by different sensors, or by different viewpoints from the same scene. Image registration technique is used in a wide-range areas, such as navigation and location, object recognition, movement analysis, stereo vision, data fusion, change detection. It is the base of these images processing techniques. The difficulty of designing registration algorithm lies in the variety of scene styles and variety of imaging variations between images. How to design an algorithm which works with high adaptability, high precision and rapid computing efficiency is the kernel study topic. And remote image registration which belonging to image registration is a technique which best matching two or more images at space place from the same area, where the images taken at different times or taken by different sensors. Similitude to image registration, the key problem of remote image registration is how to find out the best correspondence or mathematic transformation between two images.Image matching methods cover two categories: one is based on the gray, the other is feature. Based on the gray method, this paper studies the removing mean normalized cross-correlation matching method and the feature encoding matching method.Firstly introduces the application and current development of image matching technology, the research significance and the purpose of the thesis, analyzes some traditional image matching computing and find out the advantages and disadvantages of them. It has much more work to study image matching on the gray-based method than on the feature-based methods, however the correlation computing on the gray-based method are more accurate, the feature-based method can adapt better to picture type than the gray-based method. The traditional matching computing cannot meet the need of the requirements of real-time. This paper introduces the definition, flow and the principle of the image registration. It also introduces the influencing factors and the four matching elements of the image matching: feature space, search space, searching strategy and similarity measurement, the matching methods and match-types of this paper.It analyses the primary aspects of the image registration, analyzes the image preprocess aspect of the image registration. We can not obtain the right remote sensing images for the registration because of the effects of the random noise, the change of imaging sensors and the external environment. We can not get the right results if we use these images, so we must process these images before the mage registration. This paper introduces several methods for geometric correction, and introduced the method used in this article.Then it explores the removing mean normalized cross-correlation matching method according to the gray space, similarity measurement of the normalized cross-correlation coefficient which have got de-mean, this method cannot adapt better to picture type, it has much more work to study , so its matching speed is very slowly, but the correlation computing on the method are more accurate. This thesis developed the feature encoding matching method to improve the matching speed. Firstly we binary code the gray value of the model image and the restrict area,then build the feature encoding matrix and use the matrix to match. Feature encoding matching method is stable to the changes of the holistic gray value, and it can resist the infection of the image noise by changing the image processing size.The conclusion are made to summarize the chiefly research word and major problems existed in the paper and at last have a good look at the further research.