基于图切分的交互式图像分割算法  

Interactive Image Segmentation Based on Graph Cut

作　　者：湛永松[1] 雷德斌[1] 潘春洪[2] 石民勇[1] 

机构地区：[1]中国传媒大学数字技术与艺术研发中心,北京100024 [2]中科院自动化所模式识别国家重点实验室,北京100080

出　　处：《系统仿真学报》2008年第3期799-802,共4页Journal of System Simulation

基　　金：教育部科技研究重点项目(JK2G02035)

摘　　要：提出一种基于图切分(Graph Cut)的交互式图像分割新算法。首先,使用图切分技术获取初始轮廓并采取非均匀有理B样条(NURBS)拟合边界曲线。由于NURBS具有局部可控性,其控制点可视为隐式马尔可夫模型(HMM)的隐状态,边界特征和区域同一性被视为HMM观察状态,然后采用轮廓平滑性作为约束来定义一个状态转移模型。最后,通过维特比算法递归求取最优轮廓。实验证明该法鲁棒性好,允许用户简易、高效地分割图像。A novel Graph Cut based algorithm for interactive segmentation was proposed. Firstly, the initial curve obtained by Graph Cut was fit with Non-Uniform Rational B-Spline （NURBS）. Due to the local controllability of NURBS, the control points were considered as the states of Hidden Markov Model （HMM）, and the boundary features and uniformity along the boundary were integrated as the observations. Then, a state transition model based on contour smoothness constraint was calculated. Finally, the optimal contour could be achieved efficiently by Viterbi algorithm. Experimental results show the robustness, easiness and efficiency of the method.

关 键 词：图切分 非均匀有理B样条 隐式马尔可夫模型 交互式图像分割 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]