基于最小错误率的SAR图象分割方法研究

Minimum Error Ratio-Based Automatic SAR Image Segmentation

文章提出了一种基于最小错误率的SAR图象自动分割算法。对直方图呈现出多模分布特征的SAR图象,首先运用有限高斯混合分布对SAR图象特征空间的数据统计模型进行估计;其次基于最小错误率原理选取SAR图象自动分割阈值,在先验概率未知和估计条件下,获得目标及其阴影区域的检测结果;最后对两种图象分割结果进行了分析并与SAR图象目标检测的经典方法-恒虚警(CFAR)目标检测方法作了比较。仿真结果表明在先验概率估计下的图象自动分割具有明显的优势和较大的应用潜力。

The paper presents an algorithm of automatic SAR image segmentation based on minimum error ratio.The algorithm is divided into three steps.First,it adopts finite Gaussian mixture distribution to approximate multi-modal histogram of SAR image.Second,on the basis of the principle of minimum error ratio,thresholds are computed in both a prior probability is unknown and known respectively.Last,it compares the results of two image segmentation.Simulation results preferably indicate that,under the estimated a prior probability,an approach to image segmentation is superiority.