Selecting Representative Thumbnail Image and Video Clip from a Video via Bullet Screen

Finding representative thumbnails or key video clips in a long video usually requires laborious manual editing and compilation. Although there have been deep-learning-based methods that aim to understand videos and pictures, these approaches rely on a large number of computing resources and training data, and the results may still be unsatisfactory. This paper tackles the task of thumbnail selection and highlighting video selection from a different perspective - we leverage on the bullet screen, an emerging new feature on the online video streaming sites that are popular in East Asia, to select thumbnails and video clips. We compared the proposed method with a thumbnail and video clip selecting tool developed by KKStream, a leading streaming service provider in East Asia. We recruited 100 individuals to conduct subjective tests. The experimental results show that most participants are satisfied with the thumbnails and video clips selected by our method, suggesting that the bullet screen could be a valuable resource for video understanding.