Two tracks of test video Affective frame segmentation Multiple linear regression method Single highlight results Single highlight results
Video Content Highlights Extraction Based on Facial Expression of Viewers [基於觀看者的臉部表情擷取視訊醒目片段]

Video highlights are sequences of frames that impressed us most. We memorized the exciting, entertaining or interesting parts of video. For example, in football games, we consider the scene of a shot as a highlight. In previous works, most highlight frames in video were extracted using the content. However, the highlight extraction results did not completely match the emotion feedbacks of viewers due to the lack of consideration in the viewer reaction. In order to solve this problem, we present a highlight extraction system based on viewer’s affections. This work consists of several stages: video collecting, preprocessing, and highlight extraction. In our thesis, we focus on video highlight extraction. We calculate viewers’ expression intensity and normalize the distribution curve of their emotion feedbacks to extract video highlights. Our approach can couple with two modes. We can extract video highlights according to everybody’s preferences parts of videos in single viewer mode. We extended our method to multi viewer evaluation in order to decrease personal preferences toward different kinds of video.

SUMMARY (中文總結):


1.     Youtube, “Andy and Jacqueline Brave the Haunted House”, URL

2.     F. Wallhoff. 2005. The Facial Expressions and Emotions Database Homepage (FEEDTUM). URL /?waf/fgnet/feedtum.html/

3.     Youtube, “Ellen's Favorite Scares”, URL