基于PDS和ENNS的快速K-Means聚类算法
Fast K-Means Algorithm Based on PDS and ENNS
-
摘要: 在将部分失真搜索算法PDS, 等均值最近邻搜索算法ENNS集成到K-Means算法迭代过程中的基础上, 进一步利用迭代过程中已获取的历史索引信息构造优先搜索序列来减小K-Means算法的计算量, 降低时间开销.实验结果表明, 此算法提高了聚类的速度, 在利用标准测试Lena图生成不同尺寸码书的情况下, 能够将计算时间降至传统全搜索K-Means的8.6%~4.5%.Abstract: In this paper, PDS and ENNS methods are firstly plugged in the search stage of K-Means iteration.Then by using the indexes obtained from previous iterations, we propose a priority search list (PSL) in order to achieve a smaller value of initial minimum sooner to accelerate K-Means clustering.The experimental results show that the improvements are remarkable without degrading the output performance of K-Means algorithm using 4 test data sets, and the proposed method can reduce computational time to 8.6%~14.5% in the case of generating different size codebooks using the image "Lena".