Research And Improvement On Airborne Synthetic Aperture Radar Registration

Rapid development of airborne SAR (synthetic aperture radar) technology and its extensive application on military and economic field demands more on SAR image processing. Image registration, as the vital step of image processing, including image mosaic, data fusion, object detection and change detection, has become an attractive and difficult research field. Airborne SAR image registration includes image registration inside the strip and between the strips. Images between the strips represent large difference of grey level distribution and large geometry deformation due to different positions and attitudes of the aircraft when obtaining data, which makes image registration between the strips a challenge. Particularly for high resolution airborne SAR image registration between strips, registration precision is still poor applying available techniques.Based on the research on existing registration methods and the imaging principle of airborne SAR system, the author found that, due to the existence of shadow in SAR images, a certain amount of unstable feature points located in or alongside shadow area are extracted when applying currently the most popular registration algorithm, scale-invariant feature transform (SIFT), to SAR images between the strips. The matching of those unstable feature points will deteriorate registration precision. Aimed to this problem, a method combining SIFT and shadow recognition techniques is proposed to improve the precision of image registration between the strips.Considering the phenomenon that shadows are generated by the side-looking imaging principle of airborne SAR system and that look angle difference of images between the strips are usually very large, the proposed method uses threshold method and a sliding window respectively to detect and discard unstable feature points located in or alongside shadow, retaining stable feature points located in object area or flat area for feature matching and increase precision of image registration between the strips.In order to verify the feasibility and effectiveness of the proposed method,20 pairs of airborne SAR images between the strips, with the resolution of 0.5m, cut out from three adjacent strips are tested 5 times by SIFT algorithm and the proposed method and the registration results are evaluated by the two criteria, root-mean-square error (RMSE) and correct matching rate (CMT). The comparison of evaluations proves that the proposed method not only increases the precision of image registration between the strips, but also betters correct matching rate. Besides, the proposed method greatly reduces time consumption due to the deceasing calculated quantity.