Modeling content fingerprints using markov random fields

Content fingerprints are widely employed for identifying multimedia in various applications. A ¿fingerprint¿ of a video or audio is a short signature that captures unique characteristics of the signal and can be used to perform robust identification. Several fingerprinting techniques have been proposed in the literature and are often evaluated using benchmark databases. To complement these experimental evaluations, this paper develops a theoretical model for content fingerprints and evaluates the identification accuracy. Fingerprints and the noise are modeled as Markov random fields and the optimal decision rule for matching is derived. An algorithm to compute the probability of correct detection and the false alarm rate by estimating the density of states is described. Numerical results are provided for a model of a block based binary fingerprinting scheme and the influence of the fingerprint correlation and the noise on the detection accuracy is studied.