改进可变区域拟合模型的合成孔径声呐图像分割方法
Segmentation method for synthetic aperture sonar image using improved region-scalable fitting model
-
摘要: 为了自动且高精度地分割合成孔径声呐图像中的目标和阴影区域,提出一种核函数尺度自适应可变区域拟合(RSF)模型的分割方法。使用一种基于K-均值聚类的自动初始化方法对水平集进行初始化,减少了人为干预;提出改进的核函数尺度自适应RSF模型,其利用声呐成像中目标与阴影在沿扫测方向具有近似宽度的一般规律,自适应选择核函数尺度参数,使得对应目标和阴影的水平集函数能够同步演化,提高最终分割精度。通过对真实声呐图像的实验结果分析,验证了该方法能较为准确地实现目标和阴影区域的分割,具有一定的精确性和适应性。Abstract: A segmentation method using Region-Scalable Fitting(RSF) model with adaptive scale of kernel function was proposed in order to segment synthetic aperture sonar image automatically and precisely.An automatic initialization method based on K-means,which can reduce human intervention,was brought out to initialize level set functions.Then an improved RSF model with adaptive scale of kernel function was proposed.By utilizing the general rule of sonar imaging that the target and its shadow have approximately the same length in the along-track direction,the model can select kernel function's scale parameter automatically so that it can have two level set functions corresponding to target area and shadow area evolving synchronously,thus increasing the accuracy of segmentation.Experimental results demonstrate that the proposed method can acquire precise segmentation of target and shadow such that it has certain accuracy and adaptability.