摘要
均值漂移谱聚类(MSSC)算法为模式识别聚类任务提供了一种较新的方案.然而由于其内嵌均值漂移过程的时间复杂度与样本容量呈平方关系,其在大数据集环境的实用性受到大大削弱.利用快速压缩集密度估计器(FRSDE)替代Parren窗密度估计式(PW)并融合基于图的松弛聚类(GRC)方法,提出了快速均值漂移谱聚类(FMSSC)算法.相比原MSSC,该算法的总体渐进时间复杂度与样本容量呈线性关系,并具有自适应性和便捷性.
Mean shift spectral clustering(MSSC) provides an alternative for clustering tasks. However, due to the time complexity of its embedded mean shift is quadratic scaling in the sample size, the usefulness of MSSC is weakened greatly on large data sets. In this paper, the fast mean shift spectral clustering(FMSSC) algorithm is proposed by replacing parren window estimator(PW) with the fast reduced set density estimator(FRSDE) and combining with the graph-based relaxed clustering(GRC) technique. Compared with MSSC, the asymptotic time complexity of the proposed algorithm is linear with the data size, and the proposed method is straightforward and adaptable.
出处
《控制与决策》
EI
CSCD
北大核心
2010年第9期1307-1312,共6页
Control and Decision
基金
国家自然科学基金项目(60773206
60903100
90820002)
江苏省自然科学基金项目(BK2009067)
关键词
密度估计
均值漂移
谱聚类
时间复杂度
图像分割
Density estimator
Mean shift
Spectral clustering
Time complexity
Image segmentation