基于相似性准则的SAR图像分割方法

SAR Image Segmentation by Likelihood Criterion

提出一种以似然差函数作为相似性衡量标准的SAR图像分割方法。首先根据SAR图像的强度分布特性 ,通过仿真 ,发现两个具有相同分布的均匀区域合并成一个区域后 ,它们的似然差函数近似与区域的大小和均值无关 ,而仅与SAR图像的视数有关。在此基础上对两个相邻区域的似然差函数进行简化 ,获得它的概率密度函数的解析形式。选取一定的虚警率 ,计算出两个相邻区域之间存在边界的似然差函数的阈值。然后根据似然差函数和区域的形状约束建立融合代价 ,使得所有可以融合的区域按照一定的顺序融合。当没有区域可以进一步融合时 ,就获得SAR图像的最终分割结果。

In this paper, a kind of SAR image segmentation method based on the criterion of likelihood difference function is proposed. At first, based on the distribution property of SAR image, we get the following conclusions by simulation: After merging of two regions with same distribution, the difference of likelihood is approximately independent of the regional size and regional mean value, and only depends on the look number. On the basis of this conclusion, we simplify the likelihood difference function, and the analytic form of the probability distribution function is obtained. Giving the false alarm probability, we can calculate the threshold which indicates whether the adjacent regions can be merged. Then we construct merging cost according to the likelihood difference function and region shape constrains, and make regions to be merged with order. When there are no regions which can be merged, the final segmentation result is achieved.