中文摘要：

基于SAR图像边缘检测算子和变分水平集方法，依据能量最小化准则，提出了一种新颖的几何活动轮廓模型。其基本思想是，直接定义关于水平集函数的能量泛函，并将经典模型中基于梯度算子的边缘指示函数，替换为基于ROEWA算子的边缘指示函数，提高了模型对于SAR图像的边缘检测能力；同时在新模型中加入水平集函数惩罚项，确保水平集函数逼近符号距离函数。由于该项的作用，模型的数值求解可采用简单的显式差分格式迭代，并保持较快的收敛速度。针对仿真图像、Radarsat和华东电子研究所实测数据的实验结果表明，该模型具有实现简单、分割边界定位准确和收敛速度较快等优点。

英文摘要：

The geometric active contour model is a classical image segmentation model based on the curve evolution theory and the level set method, which has been successfully applied to the segmentation of medical images. Due to the existence of speckle noise, the model fails in SAR image segmentation. Moreover, there are several disadvantages with this model. First, the evolution equation isn＇ t obtained with the energy minimization method. Second, the level set function needs to be reinitialized to a signed distance function periodically during the evolution. Finally, the model is computationally inefficient. Based on SAR image edge detectors and the variational level set method, a novel geometric active contour model is proposed under the criterion of energy minimization. The basic idea is that the energy functional is defined directly on the level set function and the original edge indicator function based on gradients is replaced with a new edge indicator function based on the ROEWA operator. Thus, the ability of detecting edges and the accuracy of locating edges are greatly increased, which makes the model very appropriate for SAR image segmentation. In addition, a term penalizing the level set function is added to the energy functional in order to force the level set function to be close to a signed distance function and therefore completely eliminates the need of the costly re-initialization procedure. Thanks to the contribution of this term, the numerical calculation of the model can be implemented by a simple explicit difference scheme; at the same time the evolution speed keeps very fast. The proposed model has several advantages. For example, it can be easily implemented; it results in accurate segmentation boundaries; it converges fast and its level set function doesn ＇ t need to be reinitialized. The experimental results on the simulated image and real data show its efficiency and accuracy.