快速识别密度骨架的聚类算法
发布时间：2018-11-02 16:24
【摘要】：针对如何快速寻找密度骨架、提高高维数据聚类准确性的问题,提出一种快速识别高密度骨架的聚类(ECLUB)算法。首先,在定义了对象局部密度的基础上,根据互k近邻一致性及近邻点局部密度关系,快速识别出高密度骨架;然后,对未分配的低密度点依据邻近关系进行划分,得到最终聚类。人工合成数据集及真实数据集上的实验验证了所提算法的有效性,在Olivetti Face数据集上的聚类结果显示,ECLUB算法的调整兰德系数(ARI)和归一化互信息(NMI)分别为0.877 9和0.962 2。与经典的基于密度的聚类算法(DBSCAN)、密度中心聚类算法(CFDP)以及密度骨架聚类算法(CLUB)相比,所提ECLUB算法效率更高,且对于高维数据聚类准确率更高。
[Abstract]:To solve the problem of how to find density skeleton quickly and improve the accuracy of high dimensional data clustering, a fast clustering (ECLUB) algorithm for high density skeleton recognition is proposed. Firstly, based on the definition of the local density of the object, the high density skeleton is quickly identified according to the mutual k-nearest neighbor consistency and the local density relation of the nearest neighbor. Then, the unallocated low density points are divided according to the neighborhood relationship, and the final clustering is obtained. Experiments on synthetic data sets and real data sets verify the effectiveness of the proposed algorithm. The clustering results on Olivetti Face datasets show that, The adjusted Rand coefficient (ARI) and normalized mutual information (NMI) of ECLUB algorithm are 0.877 9 and 0.962 2 respectively. Compared with the classical density-based clustering algorithm (DBSCAN), density center clustering algorithm (CFDP) and density skeleton clustering algorithm (CLUB), the proposed ECLUB algorithm is more efficient and accurate for high-dimensional data clustering.
【作者单位】： 郑州大学信息工程学院;
【基金】：河南省基础与前沿基金资助项目(152300410191)~~
【分类号】：TP311.13
[Abstract]:To solve the problem of how to find density skeleton quickly and improve the accuracy of high dimensional data clustering, a fast clustering (ECLUB) algorithm for high density skeleton recognition is proposed. Firstly, based on the definition of the local density of the object, the high density skeleton is quickly identified according to the mutual k-nearest neighbor consistency and the local density relation of the nearest neighbor. Then, the unallocated low density points are divided according to the neighborhood relationship, and the final clustering is obtained. Experiments on synthetic data sets and real data sets verify the effectiveness of the proposed algorithm. The clustering results on Olivetti Face datasets show that, The adjusted Rand coefficient (ARI) and normalized mutual information (NMI) of ECLUB algorithm are 0.877 9 and 0.962 2 respectively. Compared with the classical density-based clustering algorithm (DBSCAN), density center clustering algorithm (CFDP) and density skeleton clustering algorithm (CLUB), the proposed ECLUB algorithm is more efficient and accurate for high-dimensional data clustering.
【作者单位】： 郑州大学信息工程学院;
【基金】：河南省基础与前沿基金资助项目(152300410191)~~
【分类号】：TP311.13
【相似文献】
相关期刊论文 前10条
1 赵云飞;;基于特征的船舶快速识别[J];信息与电脑(理论版);2012年11期
2 大江东去;;借助美图看看快速识别人物照片[J];电脑迷;2010年21期
3 陶胜;;人民币纸币号码的快速识别[J];电脑编程技巧与维护;2013年03期
4 李会科;李飞;;高速公路导向系统字体快速识别研究[J];包装工程;2010年22期
5 张其可,戴连奎;基于近红外光谱的汽油牌号快速识别[J];化工自动化及仪表;2005年04期
6 孙立功;阎保定;李萍;屠文珂;;基于特征差异的彩色目标快速识别方法[J];微计算机信息;2006年32期
7 倪政林;;领域本体的快速识别研究[J];新乡学院学报(自然科学版);2012年04期
8 陈恩红,张振亚,王煦法;相接行凸约束网络的快速识别算法[J];软件学报;2002年05期
9 曹学军;一个快速识别网格化图纸的方法[J];计算机学报;1994年02期
10 王谦,刘t，
本文编号：2306328
本文链接：https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/2306328.html
