逐步求准的视频镜头聚类算法  

Stepwise Precise Clustering Algorithm for Video Shots

作　　者：廖嘉[1] 张博[1] 王国仁[1] 李淼[1] 

机构地区：[1]东北大学信息学院计算机系统研究所,辽宁沈阳110004

出　　处：《小型微型计算机系统》2008年第7期1272-1276,共5页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60273079;60573089)资助;国家重大基础研究计划"九七三"项目(2006CB303103)资助

摘　　要：提出一种有效的逐步求准的视频镜头聚类算法(SPC).该算法并不是直接处理视频数据的底层高维特征向量,而是先把这些高维特征向量通过小波变换转换到Haar小波系数空间中,然后利用多分辨率分析技术实现逐步求准聚类结果的目的.算法的每一步求准过程都属于层次聚类过程,它利用了一种巧妙的停止准则来使算法中的合并聚类过程结束.该算法解决了以往镜头聚类算法中存在的聚类中心选取问题以及需要给出相关领域经验参数问题的同时,还能够自动的进行聚类个数的估计.理论分析和大量的实验结果表明,该算法是一种非常有效的视频镜头聚类算法.This article proposes an efficient stepwise precise clustering algorithm for video shots （SPC）. Instead of working on the low-level feature vectors of video shots, the algorithm works in the Haar wavelet coefficients space by wavelet transformation and utilizes the technology of multi-resolution analysis of wavelet to achieve the goal of getting stepwise precise clustering results. Each iteration of the algorithm is a hierarchical clustering processes, and the iterative merging processes do not stop until a novel stop criterion is satisfied. This algorithm solves the problem of choosing proper cluster centers which is a dilemma of most existing shot clustering algorithms, does not need any parameters to give beforehand and estimates the number of clus- ters automatically. The theoretical analysis and experimental results witness that SPC is an efficient shot clustering algorithm.

关 键 词：HAAR小波变换 多分辨率分析 层次聚类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]