基于全拼图的体育视频结构的无监督挖掘

摘要

This paper presents a generic approach to unsupervised structure mining for sports video analysis. Mosaic is generated for each shot as representative image of shot content. By mosaic clustering and voting process, our system may locate plays by discriminating those segments without essential content, such as breaks. The main contributions of our approach include: using mosaic to extract robust visual features for mining; comparing to existing supervised approaches to sports video structure analysis, and mining play/break without prior knowledge. The effectiveness and robustness of the proposed approach has been validated by the experiments on various sports videos. Moreover, the efficiency of mosaic for feature extraction has been proved by experimental results with contrast to key-frame.%为实现对视频更为有效的浏览和摘要,提出了一个通用的体育视频结构的无监督挖掘的方法.对每一个镜头生成一个全拼图作为表征该镜头内容的图像. 通过对全拼图的聚类和投票过程,该方法能挖掘出体育视频中的两个基本结构单元,即含有重要内容的play和不含重要内容的break. 此方法主要解决了以下问题:利用全拼图提取鲁棒的视觉特征;同现有的有监督的体育视频结构分析方法相比,该方法能在无先验知识的情况下实现对play/break的挖掘. 对几种不同体育视频进行的试验表明了该方法的有效性和鲁棒性. 另外,同基于关键帧提取特征进行对比的试验结果也表明,此方法具有更好的准确性.