Multi-actor Emotion Recognition in Movies Using a Bimodal Approach

Approaches for emotion recognition in movie scenes using high level features, consider emotion of only a single actor. The contribution of this paper is to analyze using emotional information from multiple actors present in the scene instead of just one actor. A bimodal approach is proposed for fusing emotional cues from different actors using two different fusion methods. Emotional cues are obtained from facial expressions and dialogs. Experimental observations show that emotions of other actors do not necessarily provide helpful information about the emotion of the scene and recognition accuracy is better when emotions of only the speaker are considered.