Video is a major media in the society of information under way. Unfortunately, the full use of this media is limited by the opaque character of the video which prevents content-based access. In this paepr we improve our previosu spatial temporal clues-based semantic video segmentation techniue, and propose the use of the rhythm within a video to more precisely capture temporal relations within a scene and between scenes in a video. preliminary evidence based on a 7 minutes video shows that oru spatial temporal clues-based segmentation technique coupled with the rhythm consideration fully detect the narrative structure of a video.
展开▼