Automatic Generation of Informative Video Thumbnail

Thumbnail plays a vital role in boosting the discovery and viewership of online video. Although it can be easily obtained by simply selecting a certain image from the video sequence, most of popular videos in today's video sharing platforms come with elaborately designed custom thumbnails to better showcase the highlight within the video. Unfortunately, both the selection of salient content from thousands of frames and the creation of an eye-catching thumbnail are very time-consuming and require highly specialized skills. In this paper, we present a fully automatic approach for generating informative and eye-catching video thumbnails. The proposed pipeline first splits up a given video into shot units and then selects a set of keyframes using keyframe representativeness and quality assessment metric. After performing salient region detection, a synthetic thumbnail can be obtained by embedding as many as salient regions of keyframes into the carrier image. Experimental results demonstrate the feasibility and effectiveness of our method in informative thumbnail creation.