【作者】 张娜； 潘振宽； 魏伟波；

【Author】 Zhang Na,Pan Zhenkuan,Wei Weibo (College of Information Engineering,Qingdao University,Qingdao Shandong,266071)

【机构】 青岛大学信息工程学院；

【摘要】 变分水平集方法已经成为图像分割的经典方法,但该类方法的传统模型是局部最优的,且计算效率低。本文首先将分段光滑图像分割的传统的变分水平集模型——Chan-Vese模型转化为全局凸分割模型(Global Convex Segmentation,GCS),避免了水平集函数的初始化对分割结果的影响,然后采用Split-Bregman迭代方法将全局凸分割模型转化为通过简单的软阈值公式和Laplacian计算实现的两个交替优化过程,大大提高了计算效率。文末通过数值实验对所提出方法的有效性进行了验证。更多还原

【Abstract】 The classical variational level set methods for image segmentation have two demerits of local minimization and low efficiency.The famous variational level set model-Chan-Vese mode for piecewise smooth image segmentation is transformed to a global convex segmentation one to avoid the initialization of level set functions on the results of segmentation,then the Split-Bregman iteration method is used to implement the minimization process through simple and fast soft threshold formulas and Laplacian computation based on two alternative subproblems of minimization.Numerical examples validate the algorithm presented in this paper.更多还原