Remote Sensing Image Registration Based On Point Feature And Local Structural Information

Image registration is the process of matching two or more images of the same scene with different time,different viewpoints,and different sensors.It is one of the key techniques in image processing,and has been widely applied to many fields,such as computer vision,medical image analysis,change detection in remote sensing images and so on.With the development of image acquisition technology,the way people get images becomes divers,and the difference in gray levels between the images is also increasing.However,traditional methods can not achieve accurate registration,so there are still many problems to be solved in remote sensing image registration.Point feature-based method is currently the most widely used method for remote sensing image registration,because the point feature is not only good for the geometric transformation of the images,but also requires a low degree of coincidence between the images.In addition,it also has high computational efficiency in application.Therefore,this thesis proposes three remote sensing image registration methods based on point features:(1)A remote sensing image registration method based on phase information and local structure information retention is proposed to addresses the problem that large differences in gray levels between remote sensing image pairs can easily lead to feature matching failure.This method first maps the original image onto a phase congruency graph,and uses the classical scale invariant feature transform(SIFT)algorithm to extract and match feature points on the phase congruency graph.Then,the mismatched feature point pairs are deleted by the local structure information retention strategy,and the correct matching feature point pairs are retained.Experimental results show that this method can register images with large differences in gray levels and obtain more correct matching feature point pairs.(2)Aiming at the problem that the traditional feature point matching methods can easily lead to high error rate in the matching process,this thesis proposes a point matching method for remote sensing images based on phase congruency cross correlation and global constraint.This method first uses the SAR-SIFT algorithm to extract and match the feature points,and obtains the initial matching pairs.Then phase congruency cross correlation is used to calculate the similarity value between the two neighborhood blocks around a matching pair,and the similarity value is used to filter matching pairs.Then a global constraint method is used to increase matching pairs and obtain the final set of matching points.Experiments show that this method can obtain a large number of correct matching pairs,and can reduce the error matching rate.(3)In view of the problem that the existing feature point matching methods can’t get enough correct matching points and high matching accuracy,this thesis proposes a feature point matching method based on improved nearest neighbor distance ratio and triangular constraints.Based on the classic nearest neighbor distance ratio feature point matching algorithm,this method can eliminate the similar descriptors between feature points by judging in turn whether the second nearest neighbor,the third nearest neighbor,etc is the neighboring points of the nearest neighbor.Then the triangular constraint is used to increase the matching pairs and get the final set of matching pairs.Experiments show that this method can obtain many correct matching pairs and improve the matching accuracy.