A novel framework for web video thumbnail generation

When user uploads a video clip to the video sharing websites, a video thumbnail needs to be generated as the cover to represent the video content. In this paper, a novel video thumbnail generation framework is presented. For generating a good thumbnail, three criteria are considered: (1) the thumbnail should be distinct, in order to make user feel more pleasant; (2) the thumbnail should be easily accessed by human vision, especially for the small-scale thumbnails in the web, in order to make user understand the content of the whole video at a glance; (3) the thumbnail should reflect the main content of the video. In this paper, firstly the source video is segmented into shots. Secondly, a high quality key-frame is extracted from each shot based on image quality assessment. Thirdly, visual saliency analysis is adopted to get the accessibility value of each key frame and the visual similarity analysis is employed to obtain the representativeness value of the key-frames. Finally, a linear fusing scheme is explored to fuse these two values and find the best key-frame as the video thumbnail. By using the proposed framework, we attempt to generate accessible thumbnail which may improve the viewing experience of visitors and enhance their desire to watch the video. Experiments are conducted on the videos selected from the websites and the results show that the proposed method is effective.