基于最大后验概率的SAR图像目标分割算法

文中使用最大后验概率(MAP)分类方法实现合成孔径雷达(SAR)图像目标分割,并与基于偏微分方程(PDE)的各向异性扩散(AD)过程结合起来,使MAP分类准则得到更好的分割结果。AD过程是作用在后验概率上的空域滤波器,具有高效、精确和简洁的优点,并对图像数据的分布特性具有很强的适应性。这种方法需要先将图像从灰度域转化到后验概率域,因此需要对像素灰度分布进行条件概率分布建模,并进行参数估计。文中巧妙的使用有限混合高斯分布模型来逼近条件概率分布,并用期望最大化(EM)方法用来实现参数估计。在引入这种新奇的混合高斯分布模型后,基于MAP-AD的分割算法对地面SAR图像获得了很好的分割结果并对图像灰度分布具有很强的鲁棒性。

The maximum-a-posterior based SAR image target segmentation algorithm

The paper combines the maximum-a-posterior(MAP) classify method and an anisotropic diffusion(AD) smoothing process to realize the SAR image target segmentation.The AD smoothing process is based on partial differentialequation(PDE)’s accurate numerical implementations,and can be considered as spatial filtering of posterior probabilities and takes the benefits of high efficiency,accuracy and simplicity,and it is robust to data distribution properties.This method needs a transformation from image gradation domain to posterior probability domain,which needs an adaptive conditional distribution modeling of pixel intensive with accurate and efficient parameter estimation.It ingeniously used the mixed Gaussian distribution model to approximate actual pixel intensity and expectation-maximize(EM) method to implement the estimation.The MAP-AD based segmentation method gains qualified results for the ground SAR image,and becomes robust to image gradation distribution property after introducing the novel mixed Gaussian distribution model.