Study On Automatic Registration Of Multiple Sourced Remote Sensing Imagery With High Spatial Resolution

Remote sensing technology is being quickly developed, and has changed every aspects deeply, such as business, strategy of national and security, of the world where we lived in. Remote sensing data is one emblematical sort of big data, besides the universality, it also has the uniqueness of vast quantity, multi-source, and heterogeneity. We can analyze the target objects in a smaller scale, consequently there will be huge number of to operate, because of the wide using of remote sensing image, particularly with high spatial resolution. So it has brought a challenge to ability of image processing.Image registration is an indispensable part in the dispose of remote sensing. The key of image registration is to seek the ground control points(GCP), and the manner of the seeking is by image visual interpretation in the common way. The workload of artificial registration is large. The people who worked on it must have much ample experience, even as this it still is low effective. On the other hand, artificial registration is hard to get enough GCPs and difficult to get satisfying accuracy on because of visual resolution. Now, there is a puzzle to deal with large image data, the core of this puzzle is to extract an automatic way to extract the GCPs to replace the original artificial registration. And the point of automatic registration is to realize the detection and generation of the GCPs between images in automation. There are several ways in the method of registration like based on the gray value, based on feature and based on frequency domain, the way feature-based is being researched and applied more and more on the reasons of little calculation, fine robustness, fast matching, and the feature with invariance, prominent, more quantity, and extensibility.This paper experiment the way of automated registration for satellite image with high-resolution from multiple source from four aspects, after the research and analysis the theories and methods of image automatic registration.(1) Research and experiment the theory and operation of this field and analyze the method based on gray value from cross-correlation and mutual information. It sets up a framework to this paper.(2) Take the image of ALOS from Japan, Worldview-2 from USA and GF-1 from China as research targets, then experiment the extraction and match of the key points, finally realizes the automatic registration on them.(3) Orthorectify the image by the Rational Polynomial Coefficients(RPC) model and the Digital Elevation Model(DEM), and eliminate the error from terrain and others, and improve the geometrical precision of the images.(4) Extract and match the key points, get the fine geometric result by taking Worldview-2 image as the base image, and ALOS, GF-1 image as the warp image, through using the MATLAB platform and SIFT algorithm.