A robust and fast video fingerprinting based on 3D-DCT and LSH

A video fingerprinting that extracts fingerprints from videos and finds other similar ones from a reference database, is used for media indexing, content retrieval and copyright management. In this paper, we attempt to extract spatial and temporal information of a video as its fingerprint by Three-dimension Discrete Cosine Transform (3D-DCT) and apply Locality-Sensitive Hashing (LSH) to search the similar ones in the video fingerprint database. It combines the robustness of coefficients of 3D-DCT with the retrieval efficiency of LSH. This system can yield a high precision rate of 98.25% and a recall rate of 88.78%. Experimental results illustrate that our system outperforms existing methods in copy detection based on fingerprint.