Video Highlight Shot Extraction with Time-Sync Comment

Abstract

Benefit from abundance of mobile applications, portability of large-screen mobile devices and accessibility of media resources, users nowadays much more prefer to watch videos on their mobiles no matter whether they are at home or on the way. However, constrained by available time and network flow, users may only choose to watch some hot video segments that are manually annotated by video editors. In this paper, we aim to automatically extract video highlight shot with the help of video sentimental feature of time-sync comments. First, analyzing statistical feature of real data. After, we simulate the generation process of time-sync comment after. Then, we propose a shot boundary detection method to extract highlight shot, which is proved to be more effective than traditional methods based on comment density. This experiment attests the time-sync comment is particularly suitable for sentiment-based video segment extraction for 2 reasons. 1) Text-based similarity calculation of is much faster than image-based process depending on every frame of video; 2) Time-sync comment reflects user subjective emotion therefore is useful in personalised video recommendation.