Statistical modeling and analysis of content identification

A number of hash-based algorithms for audio and video identification (ID) have been studied in recent literature, and some have been deployed as mobile phone applications and on file sharing sites. A fundamental question is what is the relationship between database size, hash length, and robustness, that any reliable content ID system should satisfy. This paper presents some answers under a simple statistical model for the signals of interest.

[1]  Shlomo Shamai,et al.  On information rates for mismatched decoders , 1994, IEEE Trans. Inf. Theory.

[2]  Imre Csiszár,et al.  Channel capacity for a given decoding metric , 1995, IEEE Trans. Inf. Theory.

[3]  Thomas M. Cover,et al.  Elements of Information Theory , 2005 .

[4]  Vishal Monga,et al.  Reliable Communications with Asymmetric Codebooks: An Information Theoretic Analysis of Robust Signal Hashing , 2008, ArXiv.

[5]  Pierre Moulin.  A Neyman-Pearson approach to universal erasure and list decoding , 2008, 2008 IEEE International Symposium on Information Theory.

[6]  Lizhong Zheng,et al.  Finding the best mismatched detector for channel coding and hypothesis testing , 2007, 2007 Information Theory and Applications Workshop.

[7]  Ton Kalker,et al.  A Highly Robust Audio Fingerprinting System , 2002, ISMIR.

[8]  G. David Forney,et al.  Exponential error bounds for erasure, list, and decision feedback schemes , 1968, IEEE Trans. Inf. Theory.

[9]  Shumeet Baluja,et al.  Audio Fingerprinting: Combining Computer Vision & Data Stream Processing , 2007, 2007 IEEE International Conference on Acoustics, Speech and Signal Processing - ICASSP '07.

[11]  Imre Csiszár.  Generalized cutoff rates and Renyi's information measures , 1995, IEEE Trans. Inf. Theory.