Research On Matching Of Airborne SAR Images Based On SIFT Algorithm

Due to the special imaging mechanism of the SAR image, the success rate, correct rate,accuracy and matching efficiency of SAR image matching are lower, so SAR image matchingbecomes one of the technical difficulties of the SAR imaging applications.In this paper, forspecial imaging characteristics of airborne SAR images, taken airborne SAR image matchingalgorithm processes as the main line, the matching algorithms of airborne SAR image areresearched; SIFT matching algorithm, elimination mismatching points algorithm withRANSAC based on the2D homography transform and matching point prediction algorithmbased on the object-side constraints are analysed and summed up. There is an emphasisresearch on SIFT matching algorithm, and two improved SIFT matching algorithms areproposed; the first one is the SIFT matching method combined with gross error eliminationalgorithm; the second one is the SIFT matching method based on the object sideconstraints.The former can extract matching points of the stable characteristics with SIFTalgorithm, and eliminate SIFT mismatching points combined with RANSAC algorithm basedon2D homography transform to improve the accuracy of matching points. The latter is animprovement of the former, SIFT matching method based on the object-side constraints is forspecial imaging features and geometric characteristics of the airborne SAR images; thegeometric constraints of the airborne SAR images are added to the SIFT algorithm matchingto improve the accuracy, precision and efficiency of the SIFT matching points.Firstly, based on the SIFT algorithm matching process of airborne SAR image matching,SIFT matching experiment of airborne SAR images verifies the SIFT algorithm canaccurately match to the stable feature from different perspectives selected representativesregional of three different land categories of airborne SAR images. And then based on thematching algorithm process of airborne SAR image, taken representatives regional of threedifferent land categories of airborne SAR images as experiment object, matching experimentsof SIFT combined with gross error elimination algorithm and SIFT matching based on theobject-side constraints are made by programme, and there is analysis and evaluation of theaccuracy of matching points.In this paper, the contents and innovations are as follows: (1) In the paper, three different land categories of airborne SAR data are used to makeexperiments, from different angles it is shown SIFT algorithm can effectively extract thestable matching points and the correct rate is high in the presence of artificial structures,containing natural vegetation and lack of texture information for airborne SAR images, evenin the area of the lack of texture information, which can promote to achieve the matching ofoverlapping regions of the whole SAR image in practice.(2) In the paper, RANSAC algorithm based2D homography transform is used toeliminate SIFT. The mismatching points elimination experiments of airborne SAR imagesusing RANSAC algorithm based on2D homography transform verifies this method caneffectively eliminate SIFT mismatching points to improve the accuracy and precision of SIFTmatching points through the analysis of match point.(3) In the paper, imaging mechanism, imaging characteristics and conformational modelof synthetic aperture radar are analysed, and SAR image matching must take into account thecharacteristics of its own image.For special imaging characteristics of airborne SAR images,based on R-D model, and the combination of POS and DEM data are used to auxiliary likepoint positioning, geometric constraints of airborne SAR images is added to SIFT algorithmmatching; the constraints of the object side is used to predict the matching point of SARimage to be matched, and the matching search window taken prediction matching point ascentre is established, and SIFT algorithm is used to match within the scope of this constraint.(4) Compared with SIFT matching algorithm combined with gross errors elimination,SIFT matching experiment based on the object-side geometric constraints can greatlyreducing the mismatching points to improve the accuracy and precision of the match points;SIFT matching based on geometric constraints constrains the search matching range of SARimages to be matched to improve the matching efficiency. SIFT matching experiment basedon geometric constraints shows this method is a kind of effective matching algorithm forairborne SAR image.