Digital content fingerprinting has emerged as a possible technique for fast, robust and privacy-preserving identification, which is a highly demanded application due to the increased interaction with humans and physical objects as well as explosive amount of multimedia data. Besides of the high attractiveness, the existing methods of content identification based on content fingerprinting still lack both deep theoretical understanding of achievable performance limits and practical methods capable to achieve these limits. Additionally, yet little is known about privacy protection technologies, which can provide zero-privacy leakage together with sufficient identification rate. In this thesis, the information-theoretic fundamentals of privacy-preserving content identification are introduced and analyzed. To cover the majority of existing practical techniques, the generalized model of content identification system is proposed which consists of decorrelation transform, dimensionality reduction mapper, privacy protection, indexed database and decoder. The state-of-the-art methods are analyzed according to the defined performance measures based on the achievable identification rate and privacy leak. More particularly, the trade-off between them is investigated with respect to each element of content identification system. The thesis provides four major contributions in part of generalized analysis of dimensionality reduction mapping, a new technique of channel decomposition according to the signs and magnitudes of signal components, a concept of channel splitting and polarization and, finally, a new method of zero-leak privacy protection.