Abstract:

    It is difficult to segment SAR image because of the speckle noise. The traditional segmentation method of finite mixture model has been adopted extensively in SAR image segmentation. However, both the performance and the robustness are not good enough. A nonparametric Bayesian infinite mixture model, called Dirichlet Process Mixture Model (DPMM) clustering method is proposed to segment SAR image. The proposed model can simulate the intrinsic property of SAR image and the segmentation method can determine the number of clusters automatically. Experiments on both simulated data and real data are carried out. The results verify the effectiveness and robustness of the proposed method in comparison with the traditional methods.

Key words: computer application, nonparametric bayesian model, SAR image segmentation, infinite mixture model, Dirichlet process mixture model