基于多特征的SAR图像的无监督分割

针对合成孔径雷达(SAR)图像灰度变化大、纹理复杂及边界模糊等特点提出了一种基于多特征的sAR图像的无监督分割方法。首先提取了SAR图像的局部矩特征与灰度共生矩阵的统计量(对比度、相关度、嫡、同质性)特征;然后利用主元分析(PCA)的方法对这些有用的特征进行降维处理以得到含有足够类别信息的2维特征;最后使用Mcan Shift方法对具有2维特征信息的像素进行自动聚类。由于Mcan Shift聚类过程中无需提供类别数，因此，这种处理是一个无监督的自动分割过程。采用了多幅SAR图像和13rodatz纹理合成图像做分割实验，结果证明:本方法与单一利用灰度共生矩阵或矩特征的方法相比，分割的准确性明显提高。

Unsupervised SAR Image Segmentation Based on Multi-features

For synthetic aperture radar(SAR) images with the characteristics of complex texture,large brightness range and vague bridge boundary,a method of unsupervised SAR image segmentation based on multi-features was presented.First of all,fcatures of the local moments and the statistics(contrast,correlation,entropy,homogcneity) of gray level cooccurrence matrix were extracted. Secondly,the dimensional reduction operation by principal component analysis(PCA)was applied to these extracted features in order to obtain 2-dimensional features with adequate category information. Finally,pixels with 2-D feature information were automatically clustered by the Mean Shift method. As the Mean Shift clustering method needn't provide the number of cluster, this processing is an unsupervised process of automatic segmentation. Composite image with 13rodatz textures and SAR images were tested in segmenting experiments and the re- sups demonstrate the method can achieve more accurate segmentation than other two methods in which only the gray level co-occurrence matrix or moments are employed.