Abstract: Based on transaction sequence, an association mining method is proposed for soccer video summarization mining. The processing progress is divided into three steps which are video data preprocessing, video feature extraction and video summarization mining. In the first step, the original video flow is segmented into physical shots. During the second step, the video is transferred into shot identifier sequence through classifying the physical shots into five classes. The shot identifier sequence is segmented into transaction sequences and a transaction database is constructed for storing transaction sequences. The traditional APRIORI algorithm is applied to mining the transaction database to obtain association patterns, which are used to form video summarization in the last step. The experimental results show that the proposed method is practical because it has achieved high rates of recall and precision.

Key words: shot classification, transaction sequence, transaction database, video association mining, highlights

摘要： 提出了一种基于事务序列的关联挖掘方法实现对足球视频的摘要挖掘。处理过程分为视频数据预处理、视频属性提取和视频摘要挖掘等三个步骤。视频数据预处理阶段将原始视频流切分成物理镜头。视频属性提取阶段先将物理镜头分成五种类型，将视频转换成镜头标识序列，以事务为单位对标识序列进行切分形成事务序列，构造一种关系数据库来存储这些事物序列数据。在视频摘要挖掘阶段，采用传统的Apriori算法对事务数据库进行关联挖掘获得关联模式，以挖掘出的关联模式为依据形成视频摘要。实验结果表明，挖掘精彩事件的查全率和查准率较高，具有实用价值。

关键词: 镜头分类, 事务序列, 事务数据库, 视频关联挖掘, 精彩事件