Visual duplicate based topic linking using a robust video signature

Abstract

This paper proposes a topic linking using a robust video signature to detect visual duplicates for grouping coherent topics in a video archive. The proposed video signature, which was accepted as part of a new ISO/IEC standard “MPEG-7 Video Signature Tools”, is designed for robust and high-speed detection of visual duplicates in a large database. It represents intensity differences between various sub-regions in a frame, which provides robustness to various modifications to videos, including caption overlay and compression. We show that the proposed video signature significantly improves the detection rate of visual duplicate segments, by more than 40% under caption overlay, compared with conventional visual features. We also present our topic linking system with its visual presentation of topic groups for efficient browsing and viewing of videos.