结合邻域信息的Chan-Vese模型图像分割  

Chan-Vese Model Image Segmentation with Neighborhood Information

作　　者：杨名宇[1] 丁欢[1] 赵博[1] 张文生[1] 

机构地区：[1]中国科学院自动化研究所复杂系统与智能科学重点实验室,北京100190

出　　处：《计算机辅助设计与图形学学报》2011年第3期413-418,共6页Journal of Computer-Aided Design & Computer Graphics

基　　金：国家"八六三"高技术研究发展计划(2008AA01Z121;2007AA01Z338);国家自然科学基金(90924026)

摘　　要：灰度异质图像广泛存在于日常生活及医学图像中,而现有方法不能很好地解决灰度异质图像的分割问题,为此提出一种结合邻域信息的改进Chan-Vese模型.首先通过计算找出邻域内与中心点属于同一类的点,其次将这些点与中心点的距离作为它们与中心点相似程度的权值进行累加,最后通过统计整幅图像中每个局部区域内各点与中心点的相似程度,加强了该模型对区域细节的捕获能力,实现对灰度异质区域的分割.实验结果表明,与Chan-Vese模型相比,文中模型可以准确地分割包含灰度异质区域的图像;与Piecewise Smooth模型相比,2个模型分割效果几乎相同,但文中模型的速度更快.Intensity inhomogeneous images, especially medical images, widely exist in real world. Considering that the existing methods can not solve intensity inhomogeneous images segmentation properly, this paper proposes an improved Chan-Vese model with neighborhood information. Firstly, the pixels of the same kind with the central pixel were obtained. Then, the distances between these pixels and central pixel were summed as their similarity weight. Finally, the capacity of capturing region details was enforced by the statistics within pixels of each local area and the central pixel in the degree of similarity, thus it could accomplish intensity inhomogeneity segmentation. Experimental results demonstrate that the improved model can segment intensity inhomogeneous images more precise than the Chan-Vese model and much faster than the Piesewise Smooth model.

关 键 词：图像分割 灰度异质 Chan—Vese模型 水平集 邻域加权函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]