关键词: 视频检索, 关键帧提取, 视频分析, 知识迁移

Abstract: Extracting keyframes is of great interest in video summary, organization, browsing and indexing. Current researches mainly focus on obtaining extractions by optimizing low-level?feature diversity or representativeness of the video frames ignoring the interests of the users. And it is time consuming and labor expensive to collect a large amount of required labelled videos to model different user-interest concepts for different videos. To alleviate the labelling process, we propose to learn models for user- interest concepts on different videos by leveraging abundant Web images which cover many roughly annotated concepts and often captured in a maximally informative way. However, knowledge from the Web is noisy and diverse, brute force knowledge transfer may hurt the keyframe extraction performance. To address this problem, we propose a novel joint group weighting learning framework to leverage different but related groups of knowledges learnt from the Web images to videos. Under this framework, weights of different groups are learnt in a joint optimization problem, and each weight represents how contributive the corresponding image group is to the knowledge transferred to the video. Experimental results on several challenging video datasets demonstrate that it is effective to use grouped knowledge gained from Web images for video keyframe extraction and provides more comprehensive results.

Key words: Video retrieval, Keyframe extraction, Video analysis, Knowledge transfer