Research On SAR And Optical Images Point Feature Registration Algorithm Based On Structure Information

SAR images and optical images are the two most commonly used data types in the field of remote sensing.Due to different imaging mechanisms,the two types of images can acquire different characteristics of ground objects and can provide good complementary information for ground observation.Therefore,the fusion processing of SAR and optical images has very important practical significance.The registration of SAR and optical images is the premise of fusion.The quality of registration results directly affects the effect of subsequent fusion processing,which has become one of the important research topics in the field of remote sensing in recent years.In this paper,several groups of measured SAR and optical image are taken as the research object.Considering the non-linear gray difference between SAR and optical images and the speckle noise of SAR image,the common structure information of SAR and optical image is deeply mined and integrated into the corresponding link of point feature registration method,to achieve the accurate registration of SAR and optical image.The main research content and innovation work of this paper are as follows:1.Aiming at the problems in the registration of SAR and optical image,this paper presents a method for SAR and optical image registration based on guided scale space and GGS-SIFT.This method is based on the SIFT algorithm framework.First,considering the influence of speckle noise on SAR images on registration accuracy,guided image filtering and Gaussian filtering are used to construct the guided scale space and Gaussian scale space of SAR and optical images,respectively.Secondly,due to the non-linear gray difference,a Gaussian gamma ratio detection operator is introduced to calculate the consistent gradient structure information of SAR and optical images,to improve the accuracy of the main direction calculation and the robustness of descriptor construction.Finally,NNDR matching method is combined with RANSAC method to filter the initial set of matching points,estimate the transformation model parameters,and achieve effective registration of SAR and optical images.The registration simulation analysis of 4 sets of measured SAR and optical images validates the effectiveness of the proposed algorithm.2.Considering the registration of SAR and optical images,there are often problems such as the number of feature points extracted is small or the distribution of feature points is uneven,and the feature matching is time-consuming,this paper proposes a fast SAR and optical image registration method based on multi-scale block phase congruency.In this method,firstly,log NL-means filtering is used to suppress the speckle noise of the input SAR image.At the same time,the CLAHE algorithm is used to reduce the influence of the nonlinear gray difference between SAR and optical image on the subsequent processing;secondly,guided image filtering is used to construct multi-scale guided space,and multi-scale phase congruency,proportional coefficient and block strategy are introduced to extract sufficient and evenly distributed stable feature points in the image;then,the phase congruency structure information is used to replace the gray-scale information of the image to construct the multi-scale block’s circular neighborhood feature descriptor,so as to enhance the distinguishability of descriptors and robustness to grayscale changes;finally,a fast feature matching algorithm combining PCA dimension reduction and KD-tree is proposed for initial matching,and FSC algorithm is further used to filter the initial matching point set,estimate the transformation model parameters,and get the final correct registration results.The registration results of several groups of measured data show that the proposed method can effectively solve many difficulties in SAR and optical image registration,and has good robustness and applicability.