Program segmentation for TV videos

A number of video browsing and retrieval systems use a hierarchical video structure (video, scene, group, shot, and keyframe) as the basis for information organization and access. However, this structure cannot represent and manage efficiently the videos recorded from TV channels, since TV videos are usually very long and consist of many scenes. More importantly, people tend to watch these videos by program and not by scenes or shots. To address this problem, we insert a new structure level, referred to as program, between the video and scene level to enhance the hierarchical structure representation. An automatic program segmentation algorithm is presented for constructing the video structure in this paper. Experimental results based on real-world TV videos demonstrate the effectiveness of the proposed approach.