论文部分内容阅读
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。
A SAR image segmentation method based on the likelihood function as a measure of similarity is proposed. First, according to the intensity distribution of SAR images, we find that when two homogeneous regions with the same distribution are merged into one region, the likelihood function of them approximates the size and mean of the region, but only with the apparent number of SAR image related. Based on this, the likelihood function of two adjacent regions is simplified and the analytic form of its probability density function is obtained. A certain false alarm rate is selected to calculate the threshold of the likelihood difference function with the boundary between two adjacent regions. Then, the fusion cost is established according to the likelihood difference function and the shape constraint of the region, so that all regions that can be fused are merged in a certain order. When no region can be further fused, the final segmentation result of the SAR image is obtained.