基于Mumford-Shah模型的参数估计和两阶段图像分割方法  

Parameter Estimation and Two Step Segmentation Based on Mumford-Shah Model

作　　者：李政文[1] 王卫卫[1] 水鹏朗[2] 

机构地区：[1]西安电子科技大学数学系,陕西西安710071 [2]西安电子科技大学雷达信号处理国防重点实验室,陕西西安710071

出　　处：《电子学报》2006年第12期2242-2245,共4页Acta Electronica Sinica

基　　金：国家优秀博士学位论文作者专项基金(No.200139);国家自然科学基金(No.60272058);教育部高校青年教师奖专项基金

摘　　要：Mumford-Shah两相分片常数模型是一个有效的图像分割模型,但当模型用于带有噪声的图像时,其水平集解法存在对初始解和长度参数敏感这两个问题.文中给出一种两阶段分割方法,首先利用传统的简单分割方法获得一个粗分割,再将其作为变分模型的初始解,从而实现自动选取初始解.文中还给出一个有效的自适应长度参数估计模型,该模型依据图像中噪声方差大小来确定参数.两阶段分割方法和自适应参数估计结合起来使得算法大大减弱了对参量的敏感性,而且可以正确、快速地分割.针对一些计算机生成图像和实际图像的实验结果验证了算法是有效的.Mumford and Shah＇s variational model in the 2-phase piecewise constant case is very efficient in image segmentation. However, if the original image is contaminated by some noise, the level set method for solving the model is very sensitive to the initial level set function and the parameter of the length of the evolving contour. Here we propose a two-step segmentation method, where in the first step, a coarse segmentation is obtained by using some traditional method, and in the second step, the coarse segmentation is used as an initial solution in the variational model. Moreover, we gave a model for adaptively estimating the parameter of the length of the evolving contour, where the parameter is defined as an increasing function of the variance of the noise. Combination of the two-step segmentation and the adaptive estimation model not only enables automatic evolution but also ensures fast and accurate partition. Experiment on some computer- produced images and real images shows that the algorithm proposed here is very efficient.

关 键 词：图像分割 变分模型 偏微分方程 

分 类 号：TN911.73[电子电信—通信与信息系统]