Research On SAR Image Registration Technique And Application

Image registration is a process to match and superpose two or more images which are acquired by different sensors, or at different times and different perspectives. It has widely used in remote sensing data analysis, computer vision, image processing and other fields. Synthetic aperture radar(SAR) can conveniently obtain the high resolution images at any times and any circumstances. Thus, it is widely used in image registration.According to the possible problems of noise and geometric deformation after SAR imaging, the paper discusses the speckle noise suppression method by the SAR imaging principle, and analyses the experimental data of several common filter algorithms. The lee filter algorithm is determined by the filtering effect and the edge keeping degree. The paper analyzes the reasons of the distortion image, and determines the geometric correction model by comparison experiments.Based on the low matching precision and bad real-time of the current registration algorithm, the paper proposes a image matching algorithm based on Harris-SIFT. Using Harris operator can extract the feature points, it can instead of the process of building the scale space using SIFT operator, so the traditional SIFT operator complexity can be simplified. When searching matching points by regional block, the operation time can be shortened;the paper also uses the improved RANSAC algorithm, it can reduce the error matching points.Genetic algorithms is introduced,by using Euclidean distance to improve operator crossover rate. It can produce volatile populations and increase the number of correct matching points. So as to improve the accuracy of image registration. Simulation experiment shows the effectiveness of the proposed method in this paper.