Research On SAR And Optical Images Registration Based On Feature Point

With the rapid development of remote sensing technology,the acquisition methods of remote sensing images are gradually diversified.Among the various remote sensing methods,the images acquired by the SAR sensor have a strong penetrating ability,which makes it easy to acquire the characteristics of the ground targets that are not easily detected.In contrast,optical image has abundant spectral information such as grayscale texture and has good visual interpretation function.Therefore,it is of great significance to integrate and complement the information in SAR and optical remote sensing images.SAR and optical image registration,as the basic part of remote sensing information fusion process,has become a research hotspot in recent years.In this paper,SAR and optical image registration technology is studied in depth,and to solve the registration difficulties two registration methods based on feature points are proposed for specific application environment.The research content and innovation work of this paper are as follows:Firstly,this paper summarizes the research progress of image registration methods at home and abroad,and analyzes the difference between SAR and optical image in radiation characteristics and geometric characteristics.To solve the problem of nonlinear radiometric difference and geometric deformation between SAR and optical images caused by this problem,this paper proposes to study the problem of SAR and optical images registration from two aspects of obtaining uniform robust feature point of the same name and improving the matching accuracy of heterologous feature vector.Secondly,due to the nonlinear radiometric difference between SAR and optical image and the speckle noise in SAR image,the number of feature points of the same name extracted by the traditional image registration method is small and the feature points are unstable.At the same time,the constructed descriptor is not reliable enough to establish a large number of effective matching relationships between SAR and optical image registration.Therefore,this paper proposes a SIFT registration method(PCUR-SIFT)based on uniform feature point extraction and consistency gradient calculation.In the stage of feature points extraction,the method combines phase consistency intensity screening and scale space grid division to obtain stable and uniform feature points from the image.In the feature description stage,the method employs the extended phase consistency method to calculate the gradient amplitude and direction of the image,and improves the correctness of the original SIFT method's main direction calculation and descriptor construction.Through multiple experiments,it is proved that this method has a good registration effect for SAR and optical images under various levels of gray difference and noise.Finally,due to the differences in sensor height and orbit,there are often large differences in perspective between SAR and optical images that need to be matched.In addition to the nonlinear radiometric difference and noise problems mentioned above,local geometric deformation will occur in the image,which will affect the correctness of the descriptor and thus reduce the success rate of matching.To solve this problem,this paper proposes a registration method based on modified affine invariant feature extraction and RANSAC matching.In the stage of feature extraction and construction,feature descriptors with global affine invariance are obtained by combining ASIFT algorithm framework and PCUR-SIFT method.In addition,this paper proposes a modified RANSAC method,which combines candidate set resampling and extreme point removal method on the original RANSAC method.This method can effectively remove the false matching of rough matching point pairs and avoid the problem that the original method is easily affected by the initial sampling.The registration experiments of multiple groups of Terra SAR-X and Google Earth images show that this method has a higher matching success rate and registration accuracy for heterogeneous image registration tasks with varied differences in perspective.