A Research On Sar Image Segmentation Technology Based On Geometric Active Contour

Image segmentation is important issue of Synthetic Aperture Radar (SAR) image processing, and also is one of the key technologies affected the performance of SAR images automatic interpretation. Due to the influence of coherent speckle noise, traditional segmentation algorithms have several shortcomings as follows. The accuracy of segmentation is not high; the boundary of the segmentation is incomplete and they are difficult to integrate prior knowledge provided by a high-level understanding of mechanisms, so that it can not meet the requirement of actual SAR image segmentation. Compared with the traditional image segmentation methods, the geometric active contour models based on curve evolution theory and level set have many advantages. Firstly, they can naturally change the topology of the contour curves and also provide high precise closed segmentation curve. Secondly, numerical implementation is simple. Thirdly, they can be easily extended to high-dimensional and so on. Therefore, geometric active contour models have been widely applied in the image segmentation.In this paper, in order to improve the accuracy of segmentation and the robustness of the algorithm, we make in-depth research on the SAR images segmentation based on geometric active contour models. The main tasks of this thesis focus on the following two aspects.Firstly, the level set SAR image segmentation approach based on the statistical distribution. At present, Gamma statistical model can get very good segmentation results for low resolution and uniform intensity SAR image. However, Gamma model is not suitable for SAR images which have high resolution, strong reflection or uneven coherent speckle noise, and then segmentation effect of the method is not ideal. Therefore, a level set SAR image segmentation approach based on Fisher distribution is proposed. Through theoretical analysis and simulation experiments, it is proved that the Fisher distribution is more suitable as a statistical model of the SAR images with high resolution, strong reflection, uneven coherent speckle noise and has the more accurate segmentation results. In addition, because the Fisher distribution parameter estimation is difficult, three methods are used to solve the parameter estimation of Fisher distribution, and then comparative analysis is given.Secondly, the level set for SAR image segmentation based on shape prior. As SAR images is always affected by the shadows and occlusion pollution in reality, which result in that the target can not get correct segmentation results. In order to solve the above problems, the level set for SAR image segmentation method based on shape prior is proposed. Based on energy function of the level set SAR image segmentation approach based on the statistical distribution, the shape constraint term described by the level set is added. Hence a new energy functional is obtained, which contains the shape energy term, statistical energy term and regular energy term. In the evolutionary process, the method can constrain the evolution of the curve. Through experiments on simulation images and MSTAR data, we get the expected segmentation results.