Mining Semantic Structures in Movies

Video editing is the process of selecting and joining various fragments of video material (called shots) to create a final video sequence. In an editing process, there are many possible ways to make a transition from one shot to another. So, the quality of a created video depends on the editor’s skills, that is, the quality of the video created by professional editors is much higher than that of amateurs. Importantly, professional video editors carry out video editing based on their own editing patterns in order to successfully convey their intention to viewers. In this paper, we concentrate on extraction of useful editing patterns from movies by applying data mining technique. The patterns extracted from movies are called ‘semantic structure’. We propose two methods to extract two types of semantic structure about the connections between consecutive shots, and the relation between character’s appearance and what is happening to him/her. Finally, based on the extracted semantic structure, our video editing support system [2] suggests hints to amateurs how to produce a new, more attractive video.