Research On SAR Image Registration Based On Contour

SAR(Synthetic Aperture Radar) is a kind of active microwave sensor with the capacity of all-weather, all-time and high resolution. SAR-imaging makes up for the deficiency of the optics imaging and infrared thermal imaging, and is an essential part of detection technology in the process of space-based space surveillance. SAR image registration is the basis of SAR image fusion, change detection and three dimensional reconstructions, which has important theoretical value and broad application prospects.In this paper, the improved Shape Context descriptor is constructed based on the contour features point set of SAR images, and the corresponding registration algorithm is given combing the image information with SVD(Singular Value Decomposition) method. Then, in order to capture the statistical patterns of SAR images changing with scale, the Shape Context descriptor is integrated with Steerable filter to implement the multi-scale registration of SAR images. The main research content is as follows:(1) Because the traditional Shape Context descriptor is sensitive to the speckle noise of SAR images and the matching rate is low, a SAR image registration algorithm is proposed based on the improved Shape Context and the SVD method. Firstly, the improved edge detection algorithm is used to get the contour feature point set and the Shape Context descriptor is calculated. Secondly, the pixel location information and gray cross-correlation coefficient of SAR images are given to construct the proximity matrix in the process of matching. Finally, the registration results is gained, which is decomposed from proximity matrix with SVD method. Effectiveness of the method is verified by the experiments.(2) For images with large change in geometry and gray level, contour points of the SAR image often appear in different scales. If the Shape Context descriptor is constructed directly based on SAR images, the registration results are not satisfactory. Therefore, the structure information of SAR image on different scales and directions is extracted by taking advantage of multi-scale and multiple directions of Steerable filter. Then, the multi-scale registration algorithm of SAR images is presented which combines the Steerable filter with Shape Context descriptor and adopts the coarse-to-fine matching strategy. The experimental results show that the proposed algorithm has advantages in speed and matching accuracy.