基于自适应带宽的快速动态高斯核均值漂移算法  

Fast dynamic Gaussian mean-shift algorithm based on adaptive bandwidth

作　　者：周芳芳[1] 樊晓平[1] 叶榛[2] 

机构地区：[1]中南大学信息科学与工程学院,湖南长沙410075 [2]清华大学智能技术与系统国家重点实验室,北京100084

出　　处：《控制理论与应用》2008年第4期608-612,共5页Control Theory & Applications

基　　金：国家自然科学基金(69975003)．

摘　　要：由核密度估计推导获得的高斯核均值漂移算法因收敛速度慢在应用中效率不高.本文提出基于自适应带宽的动态更新改进方法.首先采用空间离散方法对数据集化简,然后引入动态更新机制,每次迭代后将数据集更新到均值点,并将聚集在一起的数据点用一个收敛点表示,同时根据数据集直径的变化,自适应地计算各向异性的带宽参数.实验表明,该方法提高了算法的收敛速度,降低了计算复杂度.The Gaussian kernel mean-shift algorithm which is deduced from kernel density estimation has not been widely employed in applications because of its low convergence rate. We propose a dynamic mean-shift algorithm based on adaptive bandwidth. The number of data sets is reduced by adaptive space discretization; the convergence rate is improved by dynamically updating the data set, and the efficiency is promoted by replacing the overlapping points with a special point in the iterations. The anisotropic bandwidth is updated according to the diameter of the data set. Experiments validate the improvement of the convergence rate of Gaussian mean-shift with lower complexity in computation.

关 键 词：均值漂移 高斯核 核密度估计 自适应带宽 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]