UQLIPS: A Real-time Near-duplicate Video Clip Detection System

Near-duplicate video clip (NDVC) detection is an important problem with a wide range of applications such as TV broadcast monitoring, video copyright enforcement, content-based video clustering and annotation, etc. For a large database with tens of thousands of video clips, each with thousands of frames, can NDVC search be performed in real-time? In addition to considering inter-frame similarity (i.e., spatial information), what is the impact of frame sequence similarity (i.e., temporal information) on search speed and accuracy? UQLIPS is a prototype system for online NDVC detection. The core of UQLIPS comprises two novel complementary schemes for detecting NDVCs. Bounded Coordinate System (BCS), a compact representation model ignoring temporal information, globally summarizes each video to a single vector which captures the dominating content and content changing trends of each clip. The other proposal, named FRAme Symbolization (FRAS), maps each clip to a sequence of symbols, and takes temporal order and sequence context information into consideration. Using a large collection of TV commercials, UQLIPS clearly demonstrates that it is feasible to perform real-time NDVC detection with high accuracy.