图像分割的结构张量几何活动轮廓模型  

Geometric active contours based on structure tensor for image segmentation

作　　者：李梦[1,2] 

机构地区：[1]重庆文理学院数学与财经学院,重庆402160 [2]重庆文理学院数据分析与图像处理重点实验室,重庆402160

出　　处：《计算机应用研究》2014年第12期3890-3893,共4页Application Research of Computers

基　　金：国家自然科学基金资助项目(61202349;61271452);重庆市基础与前沿研究计划一般项目(cstc2013jcyj A40058);重庆市教委科学技术研究项目(KJ131209);重庆高校创新团队建设计划资助项目(KJTD201321);重庆文理学院科研项目(R2012SC20);重庆市科委自然科学基金计划资助项目(CSTC2011jj A40029)

摘　　要：基于水平集方法和结构张量,提出几何活动轮廓模型应用于图像分割,解决水平集方法轮廓初始化和弱边缘处易于边缘泄露问题。该方法利用张量图像的散度算子构造新的外力,引导水平集函数的自适应运动,使得其可以初始为常值函数,消失其演化对初始轮廓的需要;在偏微分方程中引入张量迹信息,减少噪声对其演化的影响,避免轮廓在弱边缘处泄露。实验结果表明,该方法对噪声图像鲁棒,能提取深度凹陷目标轮廓和红外图像中的弱目标。This paper proposed a geometric active contour model based on the level method and structure tensor for image segmentation. The goal was to deal with the problems of contours initialization and the leak of weak boundaries. With the proposed model,it presented a new force via the divergence operator of the tensor-image,which guided the level set function moving up and down adaptively. So the level set function could be initialized to a constant function,which completely eliminated the need of initial contours. By incorporating the information of tensor trace into the common partial differential equation,the regularization on zero level curves could diminish the influence of image noise on level set evolution while ensuring the active contours not to pass through weak object boundaries. Experimental results show that the proposed method is robust to noising image,and can extract the concave boundaries and the weak objects in infrared image.

关 键 词：图像分割 几何活动轮廓模型 水平集方法 结构张量 偏微分方程 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]