基于混合能量模型水平集的合成孔径声纳图像分割方法

A synthetic aperture sonar image segmentation method based on mixed energy model level set

本文讨论了合成孔径声纳图像分割问题。首先介绍了Chan-Vese模型水平集方法,针对该模型存在的边界定位和重复初始化等问题,提出了一种改进的水平集方法。该方法的能量模型由区域信息项、边界信息项和距离约束函数构成的内部能量项三部分混合形成,既兼顾了全局优化特性和局部定位精度,又避免了水平集函数重复初始化,提高了运算效率。实验结果表明,该模型对于噪声干扰严重、边缘模糊的合成孔径声纳图像分割效果良好。

A new level set energy model for Synthetic Aperture Sonar （SAS） image segmentation is described. In the conventional Chan-Vese （C-V） model level set, the edge is ignored, and it is time-consuming to repeat initializing level set function. In this paper, a region-edge term and a distance restriction function are used. Edge information term composed of image grads is useful for the local location. Distance restriction function can avoid seasonal initialization. The results of the experiment show that this new model can segment SAS images with noise and blurry edges and has good performance.