Research On The Scene Matching Algorithms Based On SAR Images

Image matching is a fundamental task in image processing, and the effect of matching directly works on the following processing steps. Scene matching is a branch of image matching. Existing scene matching techniques broadly fall into two categories: the intensity-based and feature-based methods. Synthetic Aperture Radar (SAR) can not only provide sufficiently high resolution images as visible-band ones, but also work all-weather, day-or-night without being influenced by weather, so it is more and more widely used in many fields. This dissertation studies on the scene matching algorithms based on SAR images.The technology of intensity-based scene matching is addressed and a layered-searching mean-residual normalized production correlation (NNPROD) algorithm based on wavelet analysis is researched in this dissertation. The traditional NNPROD algorithm is time consuming, so a layered-searching algorithm based on wavelet analysis is presented to reduce the computing time. Besides, enhanced Lee filter is designed for speckle suppression in SAR images to reduce its affection for matching performance. Experimental results show that comparing with NNPROD, the method based on wavelet analysis can speed the matching process obviously with its precision and do well for image matching of the same sensor.Then a normalized production correlation (NPROD)-improved partial Hausdorff distance (PHD) compound scene matching algorithm is put forward in this paper. In this method, the edge comes as the common feature, and we use wavelet transform to decompose the images, match them first by simple NPROD algorithm coarsely, then by improved partial PHD of strong anti-interference ability and fault tolerance precisely. When calculating the normalized correlation coefficients, we use logic operation instead of multiplication operation to make the matching algorithm faster. A great deal of simulated experiments have been made to demonstrate the validity of the computation. The results show that the matching algorithm is of high probability, small error, good real-time and better robustness.