Heterogeneous Remote Sensing Image Registration Based On Local Terrain Shape Similarity

With the diversified development of remote sensing platform,collaborative analysis of multi-source remote sensing data has been widely used.High precision automatic registration of multi-source images is a necessary step for collaborative application of multi-source remote sensing images,and the accuracy of image registration has a great impact on the accuracy of land surface change detection,classification and other results.However,due to the different imaging mechanism,there are usually significant nonlinear gray differences and geometric deformation between multi-source remote sensing images,which makes it difficult to obtain the same name point pairs between images.However,the shape feature of the same object can be clearly identified in optical and synthetic aperture radar(SAR)remote sensing images,and it has high robustness to the nonlinear gray difference between images.Therefore,this paper focuses on how to use the shape similarity between images for high-precision automatic registration of multi-source remote sensing images.The main research results are as follows(1)Aiming at the problem of using shape features of ground objects,the shape context algorithm,mathematical morphology algorithm and phase consistency algorithm are introduced and introduced in detail.Finally,the transformation model related to image registration and the evaluation standard of registration results are introduced.(2)In order to solve the problem of significant nonlinear gray difference between multi-source remote sensing images,a local terrain shape similarity(LTSs)descriptor is constructed by introducing shape context algorithm,and a registration method of heterogeneous remote sensing images based on local shape similarity is proposed.Experimental results show that,compared with NCC,HOPC and CSLTP,the proposed method can effectively improve the registration accuracy of multi-source images,and LTSs descriptor has high robustness to the nonlinear gray difference between different remote sensing images.(3)Aiming at the problem of building deformation caused by illumination difference and viewpoint change between UAV images,combined with LTSs descriptor,using phase consistency algorithm instead of Canny operator to detect edge features and remove feature points on buildings,a UAV image matching algorithm based on improved shape similarity of ground structure is proposed.Finally,RANSAC algorithm is used to remove the matching point pairs with large residuals,and then RMSE of the remaining points is calculated.If it is greater than the threshold,the feature points with large residuals are proposed again until the threshold is met.Experimental results show that,compared with surf,ECC,DLSS and SSSF,the proposed method can effectively improve the accuracy of UAV image registration.There are 26 pictures,6 tables and 87 references.