结合多特征和SVM的SAR图像分割

摘要

In order to implement multi-scale and multi-directional texture extraction,this paper proposed a texture feature extraction algorithm,which combined the nonsubsampled contourlet transform(NSCT) and gray level co-occurrence matric(GL-CM).Firstly,it translated the SAR image to be segmented via NSCT.Then,it computed the gray co-occurrence features via GLCM for the decomposed sub-bands,and selected the features extracted by correlation analysis to remove redundant features.Meanwhile,it extracted gray features to constitute a multi-feature vector with the gray co-occurrence features.Finally,making full use of advantages of resolving the small-sample statistics and generalizing ability of support vector machines (SVM),it used SVM to divide the multi-feature vector to segment the SAR image.Experimental results show that the proposed method for SAR image segmentation can improve segmentation precision,and obtain better edge preservation results.%为实现灰度共生矩阵(GLCM)多尺度、多方向的纹理特征提取,提出了一种结合非下采样轮廓变换(NSCT)和GLCM的纹理特征提取方法.先用NSCT对合成孔径雷达(SAR)图像进行多尺度、多方向分解;再对得到的子带图像使用GLCM提取灰度共生量;然后对提取的灰度共生量进行相关性分析,去除冗余特征量,并将其与灰度特征构成多特征矢量;最后,充分利用支持向量机(SVM)在小样本数据库和泛化能力方面的优势,由SVM完成多特征矢量的划分,实现SAR图像分割.实验结果表明,基于NSCT域的GLCM纹理提取方法和多特征融合用于SAR图像分割,可以提高分割准确率,获得较好的边缘保持效果.