SAR Image Segmentation Using GHM-Based Dirichlet Process Mixture Models

摘要

This paper proposes a robust SAR image segmentation scheme for SAR images with speckle noise. Our method can simulate the intrinsic property of SAR image by the proposed infinite mixture model-Dirichlet process mixture model and determine the cluster number automatically. The Gaussian-Hermite moment is applied to extract features to improve the robust of segmentation and reduce the influence of speckle noise. The effectiveness of proposed method is demonstrated via experiments with the simulated data and real data.