基于Mean shift和图割的极化特征图像分类

Polarized Characteristics of Image Classification Based on Mean Shift and Graph Cuts

图割法对极化SAR图像能达到很好的分类效果,但由于极化SAR数据比较庞大,直接用图割法进行分类,计算量太大,所以本文提出一种改进图割模型的分类方法.首先利用自适应的Mean shift算法结合多个极化特征把图像分成若干个同质区域,以这些同质区域的加权平均值作为超像素构建图模型,最后用图割法修正Mean shift过分割来得到最终的分类.实验证明该算法不仅在分类精度上有所提高,而且在速度上更能达到实时性的要求.

The graph cuts method could get excellent classification for polarimetric SAR. However, it would be complexed to calculating directly by graph cuts because the data of polarimetric SAR was relatively large. So an improved graph model was present in this paper. First using self-adaptive Mean shift algorithm based on multi- ple polarized characteristics the image is divided into several homogeneous region, and then an improved model is constructed with pixels which were the weighted mean of those homogeneous region. Finally the graph cuts is used to revise over-segmentation caused by Mean shift for the accurate classification. Experiment results indicate that the proposed method improves classification accuracy and meets the requirements of fast computing.