基于距离变换与路径规划的骨架提取算法  

Extraction of skeleton based on distance transform and path planning

作　　者：张超[1] 芦勤[1] 罗述谦[1] 

机构地区：[1]首都医科大学生物医学工程学院,北京100069

出　　处：《北京生物医学工程》2012年第6期551-555,585,共6页Beijing Biomedical Engineering

摘　　要：目的骨架具有与原始物体相同的拓扑与形状信息,能够有效地描述物体,在医学图像处理中有很好的应用前景,但传统基于距离变换的骨架提取难以保证骨架的连续性,因此引入骨架候选点概念以解决连续性问题。本文提出一种基于距离变换和路径规划的骨架提取算法。方法首先利用距离变换后的约束条件(局部距离变换最大值和局部距离变换梯度的模的最小值),选择骨架候选点,利用动态规划中最短路径原理连接骨架候选点。结果本文算法产生的结果与标准骨架对比,平均相似度达94%以上。结论基于距离变换和路径规划的骨架提取算法很好地保护了骨架的拓扑性和连续性,无需过多的后处理,并引入了相似度的概念来评价骨架算法。Objective The topology and shape information of the skeleton are similar with the original object. The skeleton can effectively describe the object and has great application prospect in medical image processing. Traditional skeletonization based on distance transform cannot guarantee the connectivity of the skeleton,so we introduce candidate skeleton points （CSP） to solve the connectivity problem. In this paper, we propose a well-connected skeleton algorithm based on distance transform and path planning. Methods According to the constraints of distance transform （ local maximum value of distance transform and local minimum points of the norm of distance transform gradient）, candidate skeleton points are located, and the shortest path between CSPs can be determined using path planning theory. Results The comparison between the results of our proposed method and the standard centerline of the test images shows that the average of similarity measurement （SM） is above 94%. Conclusions The proposed algorithm based on distance transform and path planning can generate a connected skeleton with well preserved original object＇ s topology without requiring a post-processing step, and SM is defined to evaluate the accuracy of the skeleton.

关 键 词：骨架提取 距离变换 路径规划 相似度 

分 类 号：R318.04[医药卫生—生物医学工程]