基于半监督聚类的微视频标注方法

摘要

Aiming at micro video semantic annotation,this paper proposed a micro video annotation method based on semi-su-pervised clustering.From the perspective of the event driven,it looked on shot event as semantic unit and used event group to annotate micro video.It proposed a novel annotation algorithm based on semi-supervised K-means clustering algorithm.In the clustering algorithm,it optimized the objective function for getting the better results with the low coupling between clusters, high polymerization in cluster,and reflected the local data distribution density in cluster.The proposed clustering method could also realize clustering on heterogeneous data,such as micro video.Experiments show that the proposed micro video an-notation algorithm has strong semantics,and the proposed clustering method also has a strong clustering accuracy.%针对微视频语义标注问题，提出一种基于半监督聚类的微视频标注方法。从事件驱动的角度，以镜头事件为单位，用事件组来标注微视频。进一步构造半监督 K-means 聚类算法，优化目标函数，使得最终的聚类结果既体现类间的低耦合及类内的高聚合，又体现类内局部的数据分布密度。该聚类算法实现了诸如微视频等多属性异构数据的聚类，提高了微视频标注效果。实验结果表明，所提微视频标注方法具有很强的语义表达能力，该聚类方法也具有很强的聚类准确度。