Theoretical Modeling and Analysis of Content Fingerprinting

Multimedia identification via content fingerprints is used in many applications, such as content filtering on usergenerated content websites, automatic multimedia identification and tagging. A compact “fingerprint” is computed for each multimedia signal that captures robust and unique properties of the perceptual content, which is used for identifying the multimedia. Several different multimedia fingerprinting schemes have been proposed in the literature and have been evaluated through experiments. To complement these experimental evaluations and provide guidelines for choosing system parameters and designing better schemes, this paper develops models for content fingerprinting and provides an analysis of the identification performance under these models. Firstly, fingerprinting schemes that generate independent and equally likely fingerprint bits are examined and bounds are obtained on the identification accuracy. Guidelines for choosing the fingerprint length to attain a desired accuracy are derived, and it is shown that identification with a false alarm requirement is similar to joint source-channel coding. A Markov Random Field based model for fingerprints with correlated components is proposed and a statistical physics inspired approach for computing the probability of detection is described. The analysis shows that the commonly used Hamming distance detection criterion is susceptible to correlations among fingerprint bits, whereas the optimal log-likelihood ratio decision rule yields 5 − 20% improvement in the accuracy over a wide range of correlations. Simulation results demonstrate the validity of the theoretical predictions.