Estimating Multiple Evoked Emotions from Videos

Video-sharing websites have begun to provide easy access to user-generated video content. How do we find what we want to view among the huge video database? When people search for a video, they may want to know whether the video evokes a certain emotional sensation. The evoked emotion is one of the important factors we consider when we select a video. One of the key concepts of evoked emotions from videos: the evoked emotions are different for each scene and for each viewer. Considering these differences, we obtained human-evoked emotions from 33 videos. We used these emotions to estimate the multiple emotions evoked by each scene of the videos. Using a computational model of emotion estimation based on mid-level visual features, we found that, in individual videos, the same scene evoked multiple emotions. Our results show that a video evoked different emotions from different people. A computational model might deliver probabilistic multiple-evoked emotions from video analyses.