始于凸包络线的闭主曲线学习算法研究  

Algorithm Study for Closed Principle Curves Learning Which Starts with Convex Envelope Hull

作　　者：倪劲松[1] 王宜怀[2] 

机构地区：[1]苏州大学数学科学学院,苏州215006 [2]苏州大学计算机科学与技术学院,苏州215006

出　　处：《计算机科学》2007年第4期166-170,共5页Computer Science

基　　金：国家自然科学基金(10571129)资助

摘　　要：本文提出了将凸包技术与自组织拓扑映射技术相结合的一种针对封闭曲线特征提取的主曲线学习算法,解决了一般主曲线算法无法有效模拟封闭和较为复杂分布数据集的难题。算法以数据集的凸包络线为起始步,通过分析数据集的全局和局部特征,逐步逼近数据集分布并获得封闭主曲线。算法的关键在于凹点挖掘算法的研究。实验结果表明,对于一般封闭曲线点集,该方法均能在较短的时间步内较好地逼近源数据集。该算法结构简单,复杂性在最坏情况下也不超过O(n2),同时对图像的有界连通区域外部边界特征的提取与图形识别亦将具有较高的应用价值。We combine convex technique and SOM to design an algorithm for closed principle curves, which can learn gradually to draw out ＇ center line＇ for outside boundary cloud of bounded connected domain in plane as an ＇ approximate＇ boundary. It solves the problem of extracting closed connected curves, for which the known principle curve method cannot do effectively. Since beginning with the convex envelope hull of the data set, it has higher convergence speed than the known method. The key is giving out a mining algorithm for concave points and concave Section： This method can get ＇ approximate＇ outside boundary of bounded connected domains in short time steps, if which is not too complicated. The complexity of the algorithm is not over 0 （n2） , even if it is in the worst case. This algorithm has simpler structure and adaptabilities; it is also useful for extraction of outside boundary of connected domain and recognition of geometric objects.

关 键 词：凸包 向量量化器 凹点挖掘 主曲线 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]