A Study On The Techniques Of The Multifrequency And Multipolarization SAR Image Registration

Synthetic Aperture Radar (SAR) is a kind of high resolution microwave remote sensing imaging radar used widely in modern military area to obtain important information. Images with different frequency and polarization related to the same region can be acquired using SAR. Image fusion technology can extract important information from single frequency and polarization image and obtains a fusion image with exhaustive information which is beneficial to the subsequent image interpretation. SAR images must be aligned exactly before the images will be fused. SAR image matching is a pretreatment process of SAR image fusion and a necessary step of SAR image processing. Moreover, it is difficulty for SAR image registration due to the complexity of SAR image gray distribution.The techniques of the multifrequency and multipolarization SAR image registration are studied in this paper. To homogeneous and heterogeneous(homogeneous means the same frequency or polarization, heterogeneous means the different frequency or polarization) SAR image matching, we propose an improved algorithm respectively. The related registration algorithms are implemented on the SAR image fusion platform further, giving the algorithm warehouse and algorithm evaluation index system. The main contents of this paper include:1. To improve matching precision, homogeneous SAR image registration algorithm based on sparse position hypergraph matching is proposed. Firstly, the graph model is constructed through extracting features from real SAR images. Secondly, after getting the minimum spanning tree structure which contains the main connections between nodes of graph, sparse position hypergraph is obtained by use of the position angle information of hyper-edge composed of three neighboring nodes in the minimum spanning tree. Thirdly, a inter-graph point proximity matrix is built by the position angle information and features similarity information At last, an approach of global optimal soft matching is used to achieve matching .It could be clearly indicated by the result that this algorithm is low computational complexity and robust for SAR image matching.2. A FFT-based SAR image registration algorithm via priority estimation of scale distortion is proposed for heterogeneous SAR image matching, especially for matching image of wide scale distortion. Firstly, we use the ratio between central spectrum of template images and central spectrum of sensed-images to estimate the scale changes. Secondly, on the process of phase correlation, due to multi-peaks in the IFFT of cross-power spectrum, we make use of the local search of pulse to obtain the best matching parameters. At last, sensed-image is corrected. It could be clearly indicated by the result that the algorithm are precise, robust for heterogeneous SAR images registration.3. To implement engineering application, the registration algorithm warehouse is established, including the mainstream classical algorithms and the new algorithms proposed by us. Besides, the objective evaluation index system of multifrequency and multipolarization SAR image registration algorithms is given. The relational algorithms and evaluation indexes implemented by C/C++ are embedded into SAR image fusion software system developed by our team. At last, we make a full evaluation for all algorithms in the registration algorithm warehouse.