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

中文关键词  均值漂移  高斯核  核密度估计  自适应带宽

英文关键词  mean shift  Gaussian kernel  kernel density estimation  adaptive bandwidth

中文摘要

      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.

英文摘要

      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.