Video retrieval by means of robust fingerprinting

Uniquely identifying visual content remains a challenging issue for a large variety of nowadays applications, as video browsing, database search and multimedia security, for instance. In this respect, our study brought to light a simple yet efficient fingerprinting technique allowing short video sequences to be tracked. Three corpora, all of them containing 3780 video excerpts, with different excerpts lengths (20 seconds, 40 seconds and 60 seconds) were considered in the experiments. The quantitative results established that the average probability errors for both missed detection and false alarm are lower than 0.0007. These good practical results derive from the very fine mathematical properties of stationarity governing the DWT coefficients representing the fingerprint.