一种基于区域动态轮廓的交互式目标分割算法  

An Interactive Object Segmentation Algorithm for a Region-Based Active Contour

作　　者：宋纯贺[1] 赵海[1] 景巍[1] 朱宏博[1] 

机构地区：[1]东北大学信息科学与工程学院,辽宁沈阳110819

出　　处：《东北大学学报（自然科学版）》2011年第4期501-504,共4页Journal of Northeastern University(Natural Science)

基　　金：国家自然科学基金资助项目(60973022);教育部重大项目培育基金资助项目(708026)

摘　　要：研究了以用户与系统之间的交互信息为先验知识的目标分割算法,提出一种基于区域动态轮廓的交互式目标分割算法.采用基于区域动态轮廓的CV模型及形状先验引导进化思想,并引入了基于滤波后图像梯度和Laplace的分段自适应加权算法.为了克服由于对先验差值区域加权而产生的目标轮廓萎缩问题,对所构建的进化模型引入了面积激励项.实验结果表明:算法无需基于精确先验知识训练而获得的先验知识模型,用户仅需要选择待分割目标的大致区域,算法即可对该区域进行分割,在选用的灰度图像级上,完成分割所需要的迭代次数仅为基本CV模型的26.4%.Research was conducted on object segmentation of the interactive information comprising prior knowledge between user and system.An interactive object segmentation algorithm was proposed based on an active contour without an edge（CV） and the prior shape.Due to imprecision in the prior shape user inputs,a weighted pix algorithm based on gradient and Laplace was introduced.To overcome problems with object contour shrink,a new area actuator was introduced.The proposed algorithm did not need a prior knowledge model obtained by training based on accurate prior knowledge.The user only needed to choose an approximate region of object segmentation,and the algorithm was able to segment the region.Given the selected gray scale levels,the iterative frequency was only about 26.4% of the basic CV model.

关 键 词：图像处理 交互式分割 区域动态轮廓 自适应加权 

分 类 号：TG335.58[金属学及工艺—金属压力加工]