摘要： 合成孔径雷达(SAR)图像中存在严重的相干斑干扰，使得SAR的图像解译过程较为困难。为此，提出一种基于组件树的SAR图像分割算法。对SAR图像建立组件树，给出基于全局特征的自适应非局部判定准则，使用该准则对组件树中的相似组件进行合并，保留组件树中最重要的组件，以完成图像滤波，获取分割后的目标。实验结果表明，该算法能获得准确的分割结果，保持目标的细节信息。

关键词: 合成孔径雷达, 相干斑噪声, 组件树, 加权图, 图像分割

Abstract: The serious effect of multiplicative speckle noise makes the understanding of Synthetic Aperture Radar(SAR) images very hard. A segmentation algorithm for SAR images based on component tree is proposed. This algorithm builds component trees for SAR images. It gives a kind of adaptive non-local criterion based on global feature to merge the similar components in component trees. It keeps the most significant components to filter the images and get the targets. Experimental results show that this algorithm can get good segmentation and maintain detail feature of targets.

Key words: Synthetic Aperture Radar(SAR), speckle noise, component tree, weighted graph, image segmentation

中图分类号: 

• TP301.6