Methods Study Of Geometric Correction For High-resolution Remote Sensing Images

With the continuous deepening of the socialist construction, the demand of high spatial resolution remote sensing data in our national defense, environmental protection, land planning and so on all walks of life increased continuously. The implementation of the "high special" project, filled the blank in the field of high resolution earth observation technology in China, and promoted the development of social and economic.All researches in this paper were based on the GF-1 satellite, the first satellite of China's high resolution earth observation system, and a simple introduction about the technical characteristics of the GF-1 was made. The geometric distortion produced in the process of collecting remote sensing images are mainly caused by the exterior orientation elements, atmospheric refraction, the earth's rotation, topographic relief and other factors. The polynomial model used in the geometric correction was analyzed, the geometric transformation and resampling methods were simply introduced, and a polynomial correction experiment which selected a panchromatic image covered in Wu Han was made. In the experiment, the results of different number of control points and different orders of polynomial correction were compared. Finally, we can learn by the experiment that the correction precision can reach the sub-pixel level when selected appropriate number of control points for the second order and third order polynomial correction. Image registration is an important link in the process of geometric correction, mainly used to selecting control points. The traditional selection of control points was done by manual, time-consuming. At the same time, the reference image used in the geometric correction was difficult to obtain. In order to solve the problems, a quick matching method based on the figures of control points and the data of the first high resolution satellite was proposed. The basic flow of the method is as follows: the database building of the control points, collecting control points and input the database, image preprocessing of the uncorrected image, primitive matching of figure, quick image registration for SURF feature points, eliminate redundant control points through the RANSAC, polynomial iteration to eliminate invalid control points.The quick matching method based on the figures of control points and the data of the first high resolution satellite proposed in this paper can quickly finish the image registration not rely on large amount data of the reference image. And the time of registration was greatly reduced, and the effect was ideal, the correction results can achieve sub-pixel level.