Citation Context

...acity analysis of identification systems, more recent considerations are based on multiple HT framework with a rejection option. Possible examples for binary data statistics are presented in [20] and =-=[22]-=-. Motivated by the prior art, we extend the problem of content identification as multiple HT with rejection to a broader class of source priors including AVS’s. In this regard, the current study impli...

Citation Context

... fingerprinting where the extracted fingerprint also consists of additional information about bit reliabilities. The performance of pCFP in terms of probability of bit error rate (BER) is measured as =-=[10]-=-: Pb−pCFP = Pr {Bx = By} (7) [ ( | = Ef(|˜x|) Q ˜ )] X| = 1 π arctan ( ) σZ , (8) where Q(x) = 1 √ 2π ∫ ∞ x σZ u2 e− σX 2 du indicates Q-function. IV. ACTIVE CONTENT FINGERPRINTING In the aCFP, we wi...

Citation Context

...the identification and authentication of digital content, physical objects and people. Digital fingerprints are extracted directly from any digital data including images, video, audio, text etc. [1], =-=[2]-=-, and represent a short, robust and distinctive representation of the content. In the case of physical objects, the fingerprints are extracted from some unique unclonable features of the material in t...

Citation Context

...channels with desynchronization. For the sake of the quantitativeness of our analysis, we deviate from the identification setup analyzed in earlier Sections to a more particular construction (Fig. 2) =-=[10]-=-. In this system design it is assumed that fingerprint generation is organized in a two-stage procedure. First, a multimedia object relevant data of reduced dimensionality ˜x L (w) are obtained by pro...

Citation Context

...nce target for any such biometric identification system. At the same time, characterization of performance bounds of identification systems in the setting of optimal hypothesis testing (HT) (see [7], =-=[8]-=-) are also highly important both from practical and theoretical considerations. In this ∗Institute for Informatics and Automation Problems (IIAP), Armenian National Academy of Sciences (NAS), ashot@ie...

Citation Context

...l letters to denote vector random variables X, corresponding small letters x and small bold letters x to denote the realizations of scalar and vector random variables, respectively, i.e., x = {x(1), x=-=(2)-=-, ..., x(N)}. bx is used to denote the binary version of x. We use X ∼ p(x) to indicate that a random variable X follows pX(x). ∗ The contact author is S. Voloshynovskiy (email: svolos@unige.ch). http...

Citation Context

... 1 L( ≤ 2 L log2 |A(J) ɛ |+Rid−(I(Y ;Z)−ɛ)) , (10) where it is assumed that decoding was performed for all elements of A (J) ɛ . Therefore, if Rid satisfies Rid ≤ I(Y ; Z) − ɛ − 1 L log 2 |A (J) ɛ |, =-=(11)-=-P G e 1 → 0asL→∞and ɛ → 0. Moreover, since only typical desynchronization parameter sets are considered, L log2 |A (J) ɛ |≤ J(H(A)+ɛ) L that vanishes for L →∞, one can conclude that Rid ≤ I(Y ; Z) − ...

Citation Context

...ty issue of classical cryptographic encryption and one-way functions to small noise in input data by trading-off the security and robustness to noise. This paper is an extension of our previous works =-=[2]-=-–[4]. We have previously considered the rate-privacy-complexity trade-off for identification applications [2]. This approach is based on global privacy amplification, where all bits of stored fingerpr...

Citation Context

...d [1]. From coding perspectives, the identification problem can be considered as a form of communication with random codewords [2], which can be further implemented based on bounded distance decoding =-=[3]-=-. In contrast to the high attractiveness of binary data representations for memory storage, complexity, security and privacy, an important fraction of information is neglected once the data are binari...

Citation Context

...assical cryptographic encryption and hashing to minor noise in input data [1], [2], [3]. The binary fingerprint is typically constructed based on the dimensionality reduction followed by binarization =-=[4]-=-. Mostly, the soft information about the magnitudes of transformed components is neglected and some privacy amplification procedure is applied to binary data to avoid the recovery of the original data...