基于瞬态系数梯度的SAR图像分割方法

SAR image segmentation using instantaneous coefficient of variation based gradient

文章针对合成孔径雷达(SAR)图像受相干斑噪声影响严重的问题,提出了一种基于瞬态系数梯度的SAR图像分割方法。该梯度计算方法是:通过各向异性相干斑降噪算法对SAR图像进行滤波,利用瞬态系数从该滤波图像中计算梯度,并将梯度图像与分水岭算法相结合进行初始分割;为解决分水岭算法导致的过分割问题,通过构建区域邻接图和区域马尔科夫场进行区域合并。与基于经典梯度的分割方法相比,该方法对边缘的定位更准确、分割速率更快。将提出的方法用于SAR图像的分割中,实验结果表明该分割方法有效、准确性好。

In view of the serious effect of speckle noise on synthetic aperture radar（SAR） image,a SAR image segmentation using instantaneous coefficient of variation（ICOV） based gradient is proposed in this paper.Speckle reducing anisotropic diffusion（SRAD） is used to filter the speckle noise in SAR image,and then ICOV detects gradient in the SAR image filtered by SRAD.ICOV-based gradient can be combined with watershed transform to produce initial image segmentation.The over-segmentation of watershed transform can be resolved by constructing region adjacency graph（RAG） and region-level Markov random field（MRF）.The proposed method is more accurate and segments faster than those methods based on classical gradient.The experiments of real SAR image segmentation prove the good performance of the proposed method.