基于窄带图割的交互式快速目标提取  

BandCut:interactive fast object segmentation using graph cuts

作　　者：王亚荣[1] 徐秋平[1] 

机构地区：[1]武警工程学院训练部,西安710086

出　　处：《计算机工程与应用》2013年第3期226-229,共4页Computer Engineering and Applications

摘　　要：针对传统算法易陷入局部极值、提取效率不高的不足,运用图割理论,提出一种将目标提取问题转化为能量最小化的组合优化问题的BandCut算法。BandCut通过人机交互获取一个将目标边界包围在内的环状窄带区域,对该区域生成距离图,构造s-t网络,进行最小代价切割获取目标。实验表明,BandCut能获取最优解,提取效率是GrabCut的5倍。Aiming at the limitations of critical point problems in local minima trap and low extracting efficiency, by means of graph cuts theory, an object extracting algorithm, BandCut, is proposed. An annular band region that encompasses object bound- ary is obtained in an interactive way. The distance map and s-t network are created successively. The object boundary is extracted via min-cut of s-t network. The experimental results show optimal and rapid extraction ability.

关 键 词：目标提取 彩色图像分割 图割 组合优化 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]