an actor/actress annotation system for drama videos

Abstract

Face information is quite important in videos, such as news, dramas, and movies. Especially for dramas, once all face sequences are extracted and identified, access to video fragments by means of actor/actress information becomes available. Automated extraction and identification of face sequences may enable many multimedia applications including video indexing, drama video authoring, etc. This paper describes an semi-automated actor/actress annotation system for drama videos, using a newly developed method for robust face sequence extraction and matching, and face classification. Figure 1 shows the process of the system. The system is first given the video of an episode (e.g., the first episode) of a drama series. After the system extracts face sequences from the video and has part of faces manually annotated by a person, it provides annotations to all faces of the episode. In addition to the annotated video of the given