Geometrically robust video hashing based on ST-PCT for video copy detection

Abstract

Copy videos flooding in the network infringe the video copyright and heavy the storage pressure on the video services server, which raises a huge demand for video copy detection techniques that can accurately and quickly detect the copies of video from huge video database. This paper aims to generate a video hash which has not only high discrimination but also robustness against geometrical and spatial-temporal transformations. First, we propose Spatial-Temporal Polar Cosine Transform (ST-PCT), which considers a video as a three-dimensional matrix and performs two-dimensional Polar Cosine Transforms (PCT) after performing a one-dimensional Discrete Cosine Transform (DCT) on video. This transformation can extract features of the spatial-temporal domain and has geometric invariance. Then, based on ST-PCT, we propose a geometrically robust video hashing method for video copy detection. The video features generated by ST-PCT are compressed and quantified to a compact binary hash code. Experimental results show that compared with the state-of-the-art methods, the proposed method has better robustness, higher accuracy, and faster calculation speed.