Model Based SAR Image Segmentation
Sheng Wen, Guangqiang Li · 2006
This paper presents a SAR image segmentation approach based on Gauss-Markov random field (GMRF) model. SAR images can be considered as they are composed of different textures, and image segmentation is directly implemented by texture segmentation approaches. Because of the better capability of texture discrimination, GMRF model is employed here to classify textures and the least error estimation is used for the solution of model parameters, and the Euclidean distance approach is employed to classify different features. In order to improve the local discrimination capability, the local energy feature is introduced. From classified textures, different region boundaries can be obtained. Experiments show that this approach has distinct global feature as well as local one for image segmentation compared with traditional approaches