基于局部相似性测度的SAR图像多层分割

针对谱聚类算法计算复杂度高，不适用于合成孔径雷达图像分割的问题，利用谱聚类算法与权核k均值之间的等价性，提出一种基于局部相似性测度的SAR图像多层分割算法．首先提取图像中每个像素的小波纹理特征，利用每个像素点的纹理特征计算各自的局部尺度参数，进而构造像素点之间的邻接关系，然后利用最近邻规则对此邻接关系进行逐层合并，进行基础聚类和逐层细化实现像素点聚类，最终得到图像的分割结果．对人工纹理图像和SAR图像的分割结果表明了新算法避免了传统谱聚类算法对尺度参数的敏感性，获得了更优的分割性能．

SAR image multilevel segmentation based on local similarity measure

Aiming at the high computational complexity of spectral clustering algorithms and their unsuitablity to synthetic aperture radar image segmentation, a novel SAR image multilevel segmentation based on local similarity measure is proposed which utilizes the equivalence of spectral clustering and weighted kernel k-means. First, the wavelet texture features of every pixel are extracted from the image and the sparse adjacent matrix among pixels is constructed by computing the local scale parameter of each pixel, and then multilevel merging, basic clustering and multilevel refining operations are used to cluster the image pixels, and finally the image segmentation result is obtained. Experimental results on artificial texture images and SAR images show that the proposed method can avoid the sensitivity of traditional spectral clustering to the scale parameters and obtain a better segmentation performance.