基于全局极小解Chan-Vese模型的SAR图像分割  

SAR image segmentation based on global minimization Chan-Vese model

作　　者：东野长磊[1] 郑永果[1] 姜东焕[1] 张彬[1] 

机构地区：[1]山东科技大学信息科学与工程学院,山东青岛266590

出　　处：《计算机工程与设计》2012年第11期4255-4258,共4页Computer Engineering and Design

基　　金：国家863高技术研究发展计划重点基金项目(2009AA122003);山东自然科学基金项目(ZR2011FM024);山东省教育厅科技计划基金项目(J08LJ10)

摘　　要：活动轮廓模型是近年来最成功的分割模型之一。但由于SAR图像存在较强的斑点噪声,使用传统的Chan-Vese模型水平集分割方法会产生很多误分割。因此,需要对传统Chan-Vese模型进行改进,将非凸的Chan-Vese模型转换为凸优化问题,得到Chan-Vese模型的全局极小解。对凸优化Chan-Vese模型引入边缘检测算子,得到基于边缘和区域信息的全局极小解Chan-Vese模型。在水平集演化迭代过程中,引入一个新的迭代终止条件,可以敏感地判断演化曲线的变化幅度,根据设定条件,自动的停止迭代计算。针对合成图像和真实SAR图像进行分割实验,实验结果表明,提出的改进Chan-Vese模型能够快速、准确地提出图像中感兴趣目标,并具有较强的抗噪性。Recently, active contours model has been one of most successful model of image segmentation. But the original Chan-Vese model level set segmentation method produced a lot of false segmentation in SAR image because of strong speckle noise. Therefore, the original Chan-Vese model is improved. Firstly, the nonconvex Chan-Vese model can be reformulated as convex optimization problem, that in turn allowed us to extract a global minmizer of the model. Then edge detector operator is incorporated into convex Chan-Vese model, hybrid model with global minimization based on edge and region information is proposed. Meanwhile, a new iteration terminal condition that is sensitive to variation of evolution contour is proposed, and this can stop the curve evolution automatically under the given rule. Finally, we apply the proposed model to synthetic images and SAR images, and the results prove that the proposed model can extract rapidly and correctly the target areas from SAR images with high robustness.

关 键 词：合成孔径雷达 图像分割 全局极小解 CHAN-VESE模型 水平集方法 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]