Geometric Correction Of Remote Sensing Image Stitching Algorithm Analysis

With the development of remote sensing technology and the amplification of countries’space exploration, remote sensing images have become an important data source. Remotesensing image stitching, using the overlapping region of the two remote sensing images tostitch an image, is used to analyze and study the remote sensing image. Remote sensing hasbeen widely used in many fields such as military investigation, geography detection,topographic mapping and marine applications. The main task of image stitching is imageregistration. Currently the main methods are registration based on gray and registration basedon features. The registration based on gray method uses the image’s gray to measure thedegree of similarity. The method is simple, sensitive to gray, and has a narrow application.Some common characteristics of feature-based image registration method are extracted fromthe reference image and the image to be registered as the registration primitives, by solvingthe corresponding transformation model parameters to complete the registration. This methodhas a small amount of calculation, wide adaptability, and strong robustness. Since themulti-source remote sensing images are obtained by the different sensors in differentconditions, even though they are on the feature information of the same region, but there aredifferences in the electromagnetic spectrum, the imaging mechanism, the spatial resolutionand other aspects. The precision registration has become a necessary condition to obtainaccurate information from multi-source remote sensing images.In order to ensure the correction control points evenly distributed, clustering algorithm isused to cluster the points according to the coordinates of all the control points, the number ofcontrol points used by geometric correction is as the number of clusters. Control points of thecenter point of the nearest neighbor clustering are selected as the calibration points, the rest ascheckpoints. The coordinates of the calibration points are used to build quadraticprogramming model and the regression function are built by solving the quadraticprogramming. Checkpoints are used to verify the approximate geometric correction of remote sensing image errors. According to the regression function, the approximate geometriccorrection and re-sampling are done to the original input images, the image eliminatinggeometric distortion is got. With the clustering algorithm, different numbers of control pointswere selected as the control points of remote sensing image geometric correction, and the restof the control points as the testing points. The direct linear transformation, general quadraticpolynomial, the general cubic polynomials, quadratic rational function, neural networkalgorithm are used for remote sensing image geometric correction and the correction precisionis compared and analyzed. And the result stitching image is shown.