Research On Optical And SAR Images Registration And Mosaic Technology

Optical and SAR remote sensing images have essential differences in imaging mechanism,therefore,optical and SAR Remote Sensing images have great differences in geometric and physical characteristics.Subsequently,there lie great differences between the registration and mosaic methods for heterogeneous optical and SAR images and the traditional methods for optical remote sensing images.The traditional registration processing methods suitable for optical images are no longer suitable for optical and SAR.In this thesis,we conducted in-depth research on some difficulties in the registration and mosaic processing for optical and SAR images.The main research contents are listed in the followings.1.The imaging mechanism and principles of SAR images are analyzed,and the filtering method for SAR noises and SAR image geometric correction methods are studied.Firstly,three kinds of noise filtering methods,including Lee,Kuan,and Frost algorithms,are studied and a comparative experiment is carried out.Then SAR images geometric correction methods are included,including the Range-Doppler geometric imaging model,the conversion between different coordinate systems,the satellite orbit parameters interpolation method,the difficulties in SAR images geoposition and the direct and indirect positioning methods.Finally,Terra SAR-X image data are taken for the direct and indirect geometric correction experiment,which verifies the correction and accuracy of the direct and indirect correction methods.2.The edge feature extraction technologies of optical and SAR image are studied.Sobel,Canny and LSD algorithms are applied to optical image edge feature extraction and GF-1 is used in the experiments.For SAR image edge feature extraction,a combined filter of Gaussian gamma double window and Gabor algorithm is put forward,and the classical LSD algorithm is also improved.Experiments have verified that the improved LSD algorithm based on the combined filter can extract long line features consistent with those of optical images.3.A new matching algorithm based on the geometric structures are brought forward for optical and SAR image registration.Firstly,the intersections of long line images of the optical and SAR images are established.Then accurate and reliable matching points set is generated from the intersections using the angle,shape,line slope and the line length suppression as the matching rules.Subsequently,new lines are created from the matching points set and updated image matching points set is generated from the new lines intersections.Finally,the matching image points set covering the whole image is obtained.4.The principle of Delaunay Triangulation is studied,and the irregular Delaunay Triangulation network is constructed using the above matching points set.The irregular Delaunay Triangulation network is used to interpolate and correct small facets to realize the registration and mosaic of the optical and SAR images.