双窗口特征的SAR图像丛林区域MRF分割算法

MRF segmentation algorithm for jungle areas in SAR image based on double-window textural feature

针对固定窗口灰度共生矩阵纹理特征对合成孔径雷达(SAR)图像丛林区域分割存在的局限性,讨论了丛林区域纹理特征值的聚类特性,分析计算窗口大小对分割的影响。基于马尔科夫随机场(MRF)分割方法对SAR图像噪声抑制能力,提出一种基于小窗口纹理特征分割作为初始标记计算初始吉布斯分布,大窗口纹理特征作为样本估计高斯分布的MRF分割方法。该方法经实验验证,能够改善分割噪声和边缘模糊的问题,很好地对SAR丛林区域进行分割。

For the limitation of fixed window gray level co-occurrence matrix texture features for jungle region segmentationin SAR image,the clustering characteristics of jungle region texture feature value are discussed and the influence of calculationwindow size on the segmentation is analyzed in this paper. Based on the ability of MRF segmentation method to inhibit SARimage noise,a MRF segmentation method,which uses a small window texture segmentation as the initial marking results tocavculate the initial Gibbs distribution,and a large window texture matrix as the sample to estimate the Gauss distribution,ispresented in this paper. The method was verified by the experiment. The result indicates that the method can improve the theability of segmentation noise suppression,solve the problem of edge ambiguity,and segment the jungle areas in SAR image.