基于可变形状参数Gamma混合模型的区域化模糊聚类SAR图像分割

为了解决传统模糊聚类算法无法准确刻画SAR图像强度分布特征以及抗噪性差等问题,提出一种基于可变形状参数Gamma混合模型(GaMM)的区域化模糊聚类SAR图像分割方法.首先,利用Voronoi划分技术将SAR图像完备地划分为若干个Voronoi多边形;然后,假设SAR图像强度服从可变形状参数的GaMM,以GaMM的负对数函数刻画多边形与聚类间的非相似性关系,并结合具有邻域多边形空间约束作用的规则化项定义区域化模糊聚类目标函数;在模型参数求解的过程中,对于无法直接通过导数求解的形状参数及生成点集,设计以目标函数最小化为准则的移动更新操作以逐步逼近最优解.通过对真实及模拟SAR图像分割结果进行定性定量分析,有效地验证了所提出算法对SAR图像强度分布拟合的准确性及分割的抗噪性.

SAR image segmentation algorithm of regionalized fuzzy clustering based on the Gamma mixture model with variable shape parameter

For the problem of that the traditional fuzzy clustering algorithm cannot precisely describe the distribution characteristics of synthetic aperture radar(SAR) intensity image and overcome the inherently existed speckle noises, the SAR image segmentation algorithm of regionalized fuzzy clustering based on the Gamma mixture model(GaMM) with variable shape parameter is proposed. Firstly, the image domain is completely divided into several Voronoi polygons by Voronoi tessellation. Assuming that the pixel intensities follow the GaMM with variable shape parameter, the non- similarity measure between the intensities of pixels in Voronoi polygons and clusters is described by the negative logarithmic function of the GaMM. Then, the regionalized fuzzy objective function is defined by combining the GaMM and regularization term with spatial constraint between neighbor Voronoi polygons. In the parameter estimation procedure, the moving-updating operations are designed to solve the implicit parameters according to the criterion of minimizing the objective function. The qualitative and quantitative analyses for the segmentation results of real and simulated SAR images effectively prove the fitting ability of the regionalized GaMM with variable shape parameters to SAR data and the noise-tolerant ability of the proposed algorithm.