MapReduce框架下基于正负关联规则的视频人物关系挖掘  

Mining of video character relationship based on positive and negative association rules under MapReduce frame

作　　者：朱晋 怀丽波[1] 崔荣一[1] 王齐[1] Zhu Jin;Huai Libo;Cui Rongyi;Wang Qi(Intelligent Information Processing Laboratory,Dept.of Computer Science&Technology,Yanbian University,Yanji Jilin 133002,China)

机构地区：[1]延边大学计算机科学与技术学科智能信息处理研究室,吉林延吉133002

出　　处：《计算机应用研究》2020年第8期2333-2337,共5页Application Research of Computers

基　　金：吉林省教育厅“十三五”科学技术研究项目(JJKH20180897KJ)。

摘　　要：针对目前视频人物关系挖掘中关系单一的问题,提出了一种MapReduce框架下基于关联规则的视频人物关系挖掘方法。首先对预处理过的视频图像进行人脸聚类,其次引入了关系方向、关系权重、关系影响三个关系细化的概念,对所得人脸事务数据库进行了正负关联规则挖掘,最后比较了挖掘结果关系图同客观关系图在对应图节点度数上的差异。实验表明,该方法可以挖掘出客观人物关系图主要结构,相对于其他人物关系挖掘方法完成了人物关系的细化,对视频内容分析的研究具有较强的参考价值。This paper proposed a method of video character relationship mining based on association rules under MapReduce frame aiming at the problem of single relationship in current video character relationship mining.Firstly,this method applied face clustering in the preprocessed video images.Secondly,it introduced three concepts of relational refinement,namely relational direction,relational weight and relational impact,and mined positive and negative association rules in the resulting face transaction database.Finally,this paper compared the mining result graph with the objective character relationship graph in the degree of difference among the corresponding graph nodes.Experiments show that the proposed method can extract the main structure of objective relation graph,at the same time complete the refinement of the relationship between characters compared with other video character relationship mining method.It has a strong reference value for the research of video content analysis.

关 键 词：视频人物关系 正负关联规则 人物关系细化 视频内容分析 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]