基于局部与全局信息的活动轮廓图像分割模型  

Active contour model based on local and global information for image segmentation

作　　者：李守荣[1] 周秋 周三平[2] 郝建红[1] 

机构地区：[1]华北电力大学电气与电子工程学院,北京102206 [2]西安交通大学人工智能与机器人研究所,陕西西安710049

出　　处：《系统工程与电子技术》2016年第5期1189-1194,共6页Systems Engineering and Electronics

基　　金：国家自然科学基金(61250008;61372050)资助课题

摘　　要：根据贝叶斯分类准则提出了一种改进的基于局部与全局信息的水平集图像分割模型。首先,利用图像的局部信息建立了局部能量项,引导目标附近的演化曲线停在目标边缘上;然后,利用图像的全局信息建立了全局能量项,加速远离目标边缘处演化曲线的演化;最后,提出了一种联合局部能量项和全局能量项的统一的水平集模型架构,提高了分割效率和分割灰度不均匀图像的能力。分割实验结果表明,该改进模型不但提高了对初始轮廓位置的鲁棒性,而且在分割灰度不均匀的图像时也取得了令人满意的分割结果。According to Bayesian classification criteria, an improved level set method for image segmenta- tion based on local and global information is proposed. Firstly, a local energy term based on local intensity infor- mation is defined. It can guide the evolving curve near the target settled on the boundaries. Secondly, a global energy term is built according to the global intensity information, so as to accelerate the evolution of the evol- ving curve far away from the target. Finally, a unified level set framework is proposed which combines the local energy term and global energy term together to improve the efficiency of segmentation and deal with images with intensity inhomogeneity. Experimental results show that this model is robust to the position of initial contour. In addition, it can obtain prod satisfying results in segmenting images with intensity inhomogeneity.

关 键 词：图像分割 灰度不均匀 水平集方法 活动轮廓模型 

分 类 号：TN957.52[电子电信—信号与信息处理]