Video Story Segmentation based on Content-independent Low-level Features

摘要

This paper proposes a content-independent video story segmentation method based on low-level audio-video features. Conventional video story segmentation methods are dependent on content-specific features, such as anchor shot extraction results for broadcast news video, hence, are difficult, to apply to video from other content domains. The proposed video story segmentation method utilizes generic audio-video features, such as color layout, motion, etc., which can be extracted from any video regardless of its content, domain. Evaluation experiments conducted with TRECVID and NII-VDB data prove that the proposed method is capable of conducting highly accurate story segmentation.