AN ACTIVE CONTOUR MODEL FOR SAR IMAGE SEGMENTATION

摘要

Based on the Chan-Vese model and Li's Scheme, a novel 2-region segmentation model is proposed to handle the problem of SAR image segmentation, and the associated level set evolution equations and numerical approximation scheme are thoroughly discussed. The proposed model includes three terms: the fidelity term of regional information, the regularizing term of boundary's length and the penalizing term of level set function. The new model has several advantages. First, it produces good segmentation results. Second, it can be easily implemented by simple explicit difference scheme and thus computationally more efficient. Finally, it is not sensitive to initial contours. The experimental results on the real MSTAR and Radarsat data show its effectiveness and practicability.