基于MAR-MRF的SAR图像分割方法

SAR Image Segmentation Based on Multiscale AutoRegressive and Markov Random Field Models

该文提出了一种基于多尺度自回归模型和马尔科夫随机场的SAR图像分割算法。算法引入多尺度自回归模型,建立层与层之间以及相邻层的像素点之间的数学关系,并将此模型与马尔科夫分割算法结合,实现了更为合理的多尺度分割策略。通过相邻尺度的依赖关系及同一尺度空间的马尔可夫性,使用多尺度自回归模型的预测结果来引导精细尺度图像分割,不仅使得最细尺度下的分割迭代次数减少;而且去除了最细尺度下多余的误分类斑块;同时还能够分割出清晰、平滑的目标边界,实现了较满意的SAR图像分割。

A method of SAR image segmentation based on Multiscale AutoRegressive and Markov Random Field （MAR-MRF） models is presented. MAR models is used to establish mathematic relationship among different image layers, and is combined with Markov Random Field （MRF） segment models. This method takes into account the dependence of neighbor layers Markov property of the same layer, and uses forecasting result of the MAR models to direct the fine layer segmentation. Experimental results on SAR image show that this method reduces the iterative times of segmentation and inaccuracy classify blocks, and gets clear and smooth object border.