融合时序特征和关键帧的视频检索方法

Video Retrieval Method Based on Spatio-temporal Feature and Key Frames

针对已有视频检索方法存在的诸多问题,提出了一种融合视频时序特征和关键帧的视频检索方法。该方法首先提取视频的时序特征,通过匹配两段视频的时序特征可直接实现画面变化率高的视频(如广告等)的检索,大大减小了视频检索的运算量。在此基础上,对于画面内容变化率较低的视频(如电视剧等),进一步将其划分为若干个子片段,每个子片段中提取出关键帧用于检索。实验结果表明,该方法检索效果好,稳健性高,应用范围广。

To solve the problems of present video retrieval algorithms, a video retrieval method base on video spatio-temporal feature curves and key frames is proposed. First, the feature curves are extracted, and then two videos＇ feature curves are compared to determine whether they have the same content. Based on the retrieval results, for a video whose content does not change frequently, the video is further divided to several clips. Key frames are selected from each of the clips and are used to compare with those of another video. Experiments show that the method is very effective, robust to various disturbances and can be used for various types of videos.