摘要
提出了一种基于Mumford-Shah推广模型的水平集能量函数,引入了梯度特征,并在此基础上提出了一种新的局部水平集分割方法,提高了算法收敛速度,避免了图像中的无关边缘对分割结果的干扰.设计了窄带算法,克服了水平集方法初始化复杂的缺点.与窄带算法相结合,所提出的分割方法可以在杂波背景中得到分割的局部最优解.通过采用Otsu算子确定感兴趣目标初始位置,所提出的方法可用于具有不同灰度特征的多目标分割.实验证明了所提出的方法用于复杂背景下的目标分割以及多目标分割时的有效性和计算效率.
A general form of energy function for Mumford-Shah model, in which the gradient feature was introcuce, was deduced based on level set method. A new local level set based segmentation approach was proposed, which can improve the convergence velocity and avoid the influence of irrelevent edges. A narrow band algorithm was also designed to overcome the complex initialization of the classical level set approach. Combined with the narrow band algorithm, the proposed segmentation approach can obtain local optimal results in clutter. With Otsu operator the method to locate interested targets, the approach proposed can segment mhiple targets with different features. Experiments demonstrate that the approach is efficient in computation and is effective in target segmentation in clutter and in multiple targets segmentation.
出处
《华中科技大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2008年第3期88-91,共4页
Journal of Huazhong University of Science and Technology(Natural Science Edition)
基金
国防预研基金资助项目
“十一五”国防重点预研项目
关键词
图像分割
轮廓提取
水平集
窄带
image segmentation
contour extraction
level set
narrow band
