一种有效的半监督视频镜头聚类算法

视频镜头聚类是基于内容的视频分析和检索领域中的一个重要问题.提出了一种对视频镜头的半监督聚类算法(SSCA),该算法首先在初始化时对已知的成对实例约束集进行聚类,利用在初始化时生成的簇来指导高维空间中其他视频镜头数据的聚类.由于高维空间中不同的维度存在着不同的相关性,所以为每一个簇引入权重向量.之后提出了一种基于最大距离的聚类中心分割策略,来解决聚类中心的选取问题.最后,考虑到对于聚类个数的选择往往对最终的结果有很大的影响,算法中采用贝叶斯信息准则来评估给定范围的聚类个数.实验结果表明,提出的算法有效地提高了聚类算法的准确性并减少了算法的响应时间.

An Efficient Semi-Supervised Clustering Algorithm for Video Shots

Clustering of video shots is a very important issue in video analysis and application of content-based retrieval.In this paper,an efficient semi-supervised clustering algorithm for video shots (SSCA)is proposed.It firstly takes advantage of the given pairwise instance-level constraint sets during initialization to guide the clustering of other video shots and associate each cluster of different dimensions with a weighting vector.Then,a cluster centroid splitting strategy is proposed based on maximum distance to resolve the problem of selecting cluster centroid.In addition,the number of clustering used to be considered usually has an important influence on the result of the clustering.The proposed algorithm uses the Bayesian information criterion(BIC)to judge the number of clustering for the given range.The experimental results show that the proposed method improves the accuracy of the clustering algorithm efficiently and reduces the response time.