Polarimetric SAR Image Segmentation Based On MRF

Synthetic Aperture Radar (SAR) is an important microwave imaging tool, which has been widely applied for many areas, because it can obtain high-resolution images under all weather and day and night. Polarimetric SAR can acquire more scattering information of targets than traditional signal-polarization SAR, which provides a better data supporting for SAR image interpretation. In the field of image processing, segmentation of polarimetric SAR images is a key procedure of image interpretation. Hence, study the methods of segmentation of polarimetric SAR is much theoretical and applicable significance for improvement of the application efficiency of polarimetric SAR systems, as well as the utilization of polarimetric SAR data.Markov Random Field (MRF) is a common model for describing the spatial correlation in images, which transforms the segmentation of SAR images to the maximum a posteriori estimation problem and completes the segmentation task by the optimization of objective function. In this thesis, considering the characteristics of polarimetric SAR data and the different forms of data, two segmentation methods of polarimetric SAR image based on MRF are proposed based on the framework of MRF. The main work is as follows:1. A new method of SAR image segmentation based on three-dimensional region-level MRF is proposed, which makes full use of the similarity of adjacent regions in single-polarized SAR image. Furthermore, the consistency of same region in different polarized SAR images is integrated. All of these can suppress the affect of speckle noise and deficiency of feature information and obtain accurate segmentation result of SAR images.2. A novel segmentation method of polarimetric SAR images based on an edge-preserving Wishart region-based MRF is presented. The proposed method uses SRAD filtering algorithm on span image to reduce speckle noise while extracting ICOV gradient information. Then the image is divided into initial regions by watershed algorithm based on the gradient image. Besides, by integrating region-based MRF and Wishart distribution of the covariance matrix, the method takes into account the degree of difference between two regions to adjust the penalty coefficient and a new way to carry out the accurate segmentation of polarimetric SAR images is supplied.Some polarimetric SAR data acquired by RADARSAR-2 and SIR-C are sufficiently tested and detailedly analyzed. The accuracy of SAR images segmentation is improved by using the proposed methods in this thesis compared with other algorithms based on MRF. In other words, the feasibility and availability of SAR interpretation of the proposed methods are validated.