基于相似性度量的高分辨率SAR图像无监督分割

SAR图像中斑点噪声的存在给分割造成了严重的影响.为此,基于JSEG平台,针对高分辨率SAR图像的特点,提出了一种新的相似性度量,这也是该无监督分割算法的核心.该算法主要由预处理、纹理组合和区域生长三个步骤来完成.在纹理组合中,利用新的相似性度量标准,把预处理后的SAR数据通过计算映射成一组新的数据,这组新数据可初步表征图像的分割;最后利用简单的区域生长完成分割.实验结果表明,该方法充分利用了SAR图像的特征信息,能够准确实现对SAR图像的分割,并具有很好的稳健性.

Unsupervised segmentation for high resolution SAR images based on homogeneity criterion

The segmentation of SAR(synthetic aperture radar) images is greatly complicated by the presence of coherent speckles.A novel homogeneity criterion based on JSEG(J-segmentation) was proposed according to SAR images,which is the critical part of this unsupervised segmentation algorithm.It is composed of preprocessing,texture composition where the SAR data was mapped to new data that can be taken for an initial segmentation,and region growing to finish the segmentation.The performance of the approach was verif...