A Median Regularized Level Set for Hierarchical Segmentation of SAR Images

Abstract

An efficient strategy of image processing algorithms to deal with the speckle noise is to incorporate data knowledge and models into them. In this letter, we introduce a hierarchical level set algorithm, which is fast and precise for multiregion segmentation of synthetic aperture radar (SAR) images. Our algorithm performs curve regularization with a nonparametric median filter instead of using the curvature formulation, and hence it reduces the computation time. The proposed algorithm also replaces the front propagation derivatives by morphological operations, and finally, the arithmetic-geometric distance measures the contrast between regions and controls the hierarchical segmentation. We conducted experiments on synthetic and real SAR images modeled by the $\mathcal {G}_{I}^{0}$ distribution. The performance evaluation of the proposed algorithm and two related methods comprises the computation time and measures based on segmentation accuracy and stochastic distance. Overall, our segmentation algorithm performed faster and more precise on both synthetic and real SAR images.