分析不同来源视频中关键帧的相似度是新闻视频分析与组织中一项重要的支持技术,当前工作直接利用图像匹配分析方法进行研究,而没有充分结合新闻视频本身的特点和特定的需求,因此存在种种局限.针对这些不足,本文提出了一种快速有效的层次化过滤方法来识别新闻视频关键帧.该方法首先对局部关键点的获取和精减技术进行了研究,在第一层过滤中应用基于熵的方法选择相似关键帧候选集,在第二层过滤分析中利用相似关键帧的对称性得到可信的识别结果.实验显示,这种方法显著的提高了匹配分析的速度,并且具有较好的识别效率.
Detecting near-duplicate keyframes in news video across different sources is an important underlying technology for information analysis and organizing in news video. Currently works are directly using the techniques of image matching. Specialty and requirement of news video analysis are not considered. This paper presents a hierarchical approach for fast and effective identifying near-duplicate keyframes. At first, it detects and prunes local keypoints using DOG function. Then, the first filter builds a entropy measure to identify near-duplicate keyframes candidates. Finally, the credible results are filtered based on symmetry. Experiment results show that this method remarkably improves the matching speed and has high efficiency.