Abstract
Forensic watermarking is the application of digital watermarks for the purpose of tracing unauthorized redistribution of content. One of the most powerful types of attack on watermarks is the collusion attack, in which multiple users compare their differently watermarked versions of the same content. Collusion-resistant codes have been developed against these attacks. One of the most famous such codes is the Tardos code. It has the asymptotically optimal property that it can resist \(c\) attackers with a code of length proportional to \(c^2\). Determining error rates for the Tardos code and its various extensions and generalizations turns out to be a nontrivial problem. In recent work we developed an approach called the convolution and series expansion (CSE) method to accurately compute false positive accusation probabilities. In this paper we extend the CSE method in order to make it possible to compute a bound on the False Negative accusation probabilities.
Similar content being viewed by others
Notes
The proportionality \(m\propto c_0^2\) was already noted in the context of spread-spectrum watermarking by Kilian et al. [14]. They showed that, if the watermarks have a component-wise normal distribution, \(\varOmega (\sqrt{m/ln\;n})\) differently marked copies are required to erase any mark with non-negligible probability.
References
Amiri E., Tardos G.: High rate fingerprinting codes and the fingerprinting capacity. In: ACM-SIAM Symposium on Discrete Algorithms (SODA), pp. 336–345 (2009).
Blayer O., Tassa T.: Improved versions of Tardos’ fingerprinting scheme. Des. Codes Cryptogr. 48(1), 79–103 (2008).
Boesten D., Škorić B.: Asymptotic fingerprinting capacity for non-binary alphabets. In: Information Hiding. Lecture Notes in Computer Science, vol. 6958, pp. 1–13. Springer, Heidelberg (2011).
Boneh D., Shaw J.: Collusion-secure fingerprinting for digital data. IEEE Trans. Inf. Theory 44(5), 1897–1905 (1998).
Charpentier A., Xie F., Fontaine C., Furon T.: Expectation maximization decoding of Tardos probabilistic fingerprinting code. In: SPIE Proceedings of Media Forensics and Security, vol. 7254, p. 72540 (2009).
Furon T., Pérez-Freire L.: Worst case attacks against binary probabilistic traitor tracing codes. In: IEEE Workshop on Information Forensics and Security (WIFS). http://arxiv.org/abs/0903.3480 (2009).
Furon T., Pérez-Freire L., Guyader A., Cérou F.: Estimating the minimal length of Tardos code. In: Information Hiding. Lecture Notes in Computer Science, vol. 5806, pp. 176–190. Springer, Berlin (2009).
Furon T., Guyader A., Cérou F.: On the design and optimization of Tardos probabilistic fingerprinting codes. In: Information Hiding. Lecture Notes in Computer Science, vol. 5284, pp. 341–356. Springer, Berlin (2008).
Gradshteyn I.S., Ryzhik I.M.: Table of Integrals, Series, and Products, 5th edn. Academic Press, New York (1994).
He S., Wu M.: Joint coding and embedding techniques for multimedia fingerprinting. IEEE Trans. Inf. Forensics Secur. 1, 231–248 (2006).
Hollmann H.D.L., van Lint J.H., Linnartz J.-P., Tolhuizen L.M.G.M.: On codes with the identifiable parent property. J. Comb. Theory 82, 472–479 (1998).
Huang Y.W., Moulin P.: On fingerprinting capacity games for arbitrary alphabets and their asymptotics. In: IEEE International Symposium on Information Theory (ISIT), pp. 2571–2575 (2012).
Huang Y.W., Moulin P.: Saddle-point solution of the fingerprinting capacity game under the marking assumption. In: IEEE International Symposium on Information Theory (ISIT), pp. 2256–2260 (2009).
Kilian J., Leighton F.T., Matheson L.R., Shamoon T.G., Tarjan R.E., Zane F.: Resistance of digital watermarks to collusive attacks. In: IEEE International Symposium on Information Theory (ISIT), p. 271 (1998).
Kuribayashi M., Akashi N., Morii M.: On the systematic generation of Tardos’s fingerprinting codes. In: IEEE International Workshop on Multimedia Signal Processing (MMSP), pp. 748–753 (2008).
Laarhoven T., de Weger B.: Optimal symmetric Tardos traitor tracing schemes. Des. Codes Cryptogr. (2012). doi:10.1007/s10623-012-9718-y.
Laarhoven T., Doumen J., Roelse P., Škorić B., de Weger B.M.M.: Dynamic Tardos traitor tracing schemes. IEEE Trans. Inf. Theory 59, 1–13 (2013).
Meerwald P., Furon T.: Towards joint Tardos decoding: the ‘Don Quixote’ algorithm. In: Information Hiding. Lecture Notes in Computer Science, vol. 6958, pp. 28–42. Springer, Prague (2011).
Moulin P.: Universal fingerprinting: capacity and random-coding exponents. http://arxiv.org/abs/0801.3837 (2008).
Nuida K.: Short collusion-secure fingerprint codes against three pirates. In: Information Hiding. Lecture Notes in Computer Science, vol. 6387, pp. 86–102. Springer, Calgary (2010).
Nuida K., Hagiwara M., Watanabe H., Imai H.: Optimal probabilistic fingerprinting codes using optimal finite random variables related to numerical quadrature. CoRR, abs/cs/0610036 (2006).
Prudnikov A.P., Brychkov Y.A., Marichev O.I.: Integrals and Series, vol. 1. CRC Press, Boca Raton (1994).
Schaathun H.G.: On error-correcting fingerprinting codes for use with watermarking. Multimedia Syst. 13(5–6), 331–344 (2008).
Simone A., Škorić B.: Asymptotically false-positive-maximizing attack on non-binary Tardos codes. In: Information Hiding. Lecture Notes in Computer Science, vol. 6958, pp. 14–27. Springer, Berlin (2011).
Simone A., Škorić B.: Accusation probabilities in Tardos codes: beyond the Gaussian approximation. Des. Codes Cryptogr. 63(3), 379–412 (2012).
Simone A., Škorić B.: False Positive probabilities in q-ary Tardos codes: comparison of attacks. http://eprint.iacr.org/2012/522 (2012).
Somekh-Baruch A., Merhav N.: On the capacity game of private fingerprinting systems under collusion attacks. IEEE Trans. Inf. Theory 51, 884–899 (2005).
Tardos G.: Optimal probabilistic fingerprint codes. In: ACM Symposium on Theory of Computing (STOC), pp. 116–125 (2003).
Škorić B., Katzenbeisser S., Celik M.U.: Symmetric Tardos fingerprinting codes for arbitrary alphabet sizes. Des. Codes Cryptogr. 46(2), 137–166 (2008).
Škorić B., Katzenbeisser S., Schaathun H.G., Celik M.U.: Tardos fingerprinting codes in the combined digit model. In: IEEE Workshop on Information Forensics and Security (WIFS), pp. 41–45 (2009)
Škorić B., Vladimirova T.U., Celik M.U., Talstra J.C.: Tardos fingerprinting is better than we thought. IEEE Trans. Inf. Theory 54(8), 3663–3676 (2008).
Xie F., Furon T., Fontaine C.: On-off keying modulation and Tardos fingerprinting. In: ACM Workshop on Multimedia and Security (MM &Sec), pp. 101–106 (2008).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by M. Paterson.
Appendices
Appendix
Proof of Lemma 1
The proof is similar to the steps taken in Appendix D of [25]. First we split the \(q\)-dimensional integration \(\int \mathrm{d}^q {{\varvec{p}}}\; F({{\varvec{p}}}) r({{\varvec{p}}})\) as follows,
Then we write \({{\varvec{p}}}_{\setminus y}=(1-p_y){{\varvec{t}}}\). We get \(\delta (1-p_y-\sum _{\beta \in \mathcal{Q }\setminus \{y\}}p_\beta )= (1-p_y)^{-1}\delta (1-\sum _{\beta \in \mathcal{Q }\setminus \{y\}}t_\beta )\). Furthermore, \(\mathrm{d}^{q-1}{{\varvec{p}}}_{\setminus y}=(1-p_y)^{q-1}\mathrm{d}^{q-1}{{\varvec{t}}}\) and \({{\varvec{p}}}_{\setminus y}^{-1+\kappa }=(1-p_y)^{(q-1)(-1+\kappa )}{{\varvec{t}}}^{-1+\kappa }\). Combined with the fact that \(B(\kappa \mathbf{1}_q)=B(\kappa ,\kappa [q-1])B(\kappa \mathbf{1}_{q-1})\), these steps yield the end result. \(\square \)
Proof of Theorem 1
The guilty user’s symbol is denoted as \(X\). The one-segment score is either \(g_0(p_y)\) (when \(X \ne y\)) or \(g_1(p_y)\) (when \(X=y\)). Since no other values are possible, the probability distribution at given \({{\varvec{p}}}\) will consist of delta-function peaks. Each peak is multiplied by the probability that the corresponding event occurs
Notice that
and that
Next step is to compute \(\mathrm{Pr}[u=g_1(p_y)\vert {{\varvec{p}}}]\) in (62). Let be \({{\varvec{e}}}_y\) a \(q\)-ary vector entirely set to \(0\) except for the \(y\)-th element that is instead equal to 1.
The last equation is obtained as follows: \(\mathrm{Pr}[X_{ji}=y]=p_y\); \(\mathrm{Pr}[Y=y\vert X_{ji}=y, {{\varvec{p}}}]\) is equal to the sum over all the possible \({\varvec{\sigma }}\) vectors that have at least one occurrence of \(y\) (expressed with the condition \(\sigma _y>0\)). Knowing that \(X_{ji}=y\), the multinomial factor is needed to count the remaining \(c-1\) pirate symbols in \({\varvec{\sigma }}\), subtracting 1 from \(\sigma _y\) (using the \({{\varvec{e}}}_y\) vector).
In the last equation we used \( {{\varvec{p}}}^{\varvec{\sigma }}=p_y {{\varvec{p}}}^{{\varvec{\sigma }}-{{\varvec{e}}}_y}\) and \(\left( {\begin{array}{c}c-1\\ {\varvec{\sigma }}-{{\varvec{e}}}_y\end{array}}\right) =\frac{\sigma _y}{c}\left( {\begin{array}{c}c\\ {\varvec{\sigma }}\end{array}}\right) \). Then the condition \(\sigma _y>0\) becomes superfluous and (27) trivially follows. Notice that
proving that (28)=(27) and (26)=(25). Finally, from (64) combined with (63) we have
This, together with (66), completes the proof. \(\square \)
Proof of Theorem 2
The full \(\psi (u)\), without conditioning, is obtained by taking the expectation over \({{\varvec{p}}}\) of (25)+(27).
We first prove (29) starting from \(\mathbb{E }_{{\varvec{p}}}[\psi _-(u \vert {{\varvec{p}}})]\) with \(\psi _-(u \vert {{\varvec{p}}})\) as given in (25).
From Lemma 1 and \({{\varvec{p}}}_{\setminus y}^{{\varvec{\sigma }}_{\setminus y}}=\left( 1-p_y\right) ^{c-\sigma _y}\prod _{\alpha \in \mathcal{Q }\setminus \{y\}} t_\alpha ^{\sigma _\alpha }\) we have that
The second integral in (72) evaluates to \(B({\varvec{\sigma }}_{\setminus y}+\kappa \mathbf{1}_{q-1})\), having the structure shown in Def. 1. In order to evaluate the \(p_y\)-integral we have to rewrite the delta function into the form \(\delta \left( p_y-\cdots \right) \). We use the rule
for any monotonic function \(w(p)\). This gives
We substitute (74) into (72) and solve the integral
Substituting (75) into (71) we have
Now we change the summations as follows: the \(\sum _{\varvec{\sigma }}\) can be written as \(\sum _b \sum _{{\varvec{x}}}\) with \(b=\sigma _y\) and \({{\varvec{x}}}={\varvec{\sigma }}_{\setminus y}\), so \(\theta _{y|{\varvec{\sigma }}}=\Psi _b({{\varvec{x}}})\). Then the summand is a function of only \(b\) and \({{\varvec{x}}}\), which allows us to write
Now we have
where
In the last line we used Definition 3. Substituting (80) into (78) and removing \(0\) and \(c\) from the \(b\)-range, we have (29).
We can use exactly the same steps to obtain (30) from (27). The only significant difference is the delta function which in this case will be
\(\square \)
Proof of consistency check 1
Integration of (29) and (30) gives
Let be \(\lambda {:=} b+\kappa \) and \(w {:=} c-b+\kappa [q-1]\). Applying Lemma 2 we have
The result follows applying Lemma 3 followed by Lemma 4. \(\square \)
Proof of consistency check 2
Taking (29) and (30), the integral \(\int \limits _{-\infty }^\infty \!\mathrm{d}u\; u\psi (u)\) can be written as
Let \(\lambda {:=} b+\kappa -\frac{1}{2}\) and \(w {:=} c-b+\kappa [q-1]-\frac{1}{2}\). Applying Lemma 2 and the property \(\varGamma (x+1)=x\varGamma (x)\) we have
To obtain \(\tilde{\mu }\) as in (24) we use Lemma 3 to substitute \(\left( {\begin{array}{c}c\\ b\end{array}}\right) \frac{1}{B(\kappa ,\kappa [q-1])}\) with \(\frac{\mathbb{P }_1(b)}{B(\lambda +1/2,w+1/2)}\). After some simplifications, the result follows. \(\square \)
Proof of Lemma 6
The integral \(\int \limits _{-\infty }^{\infty }\mathrm{d}u\; u^2\psi (u)\) can be written as
Let \(\lambda {:=} c-b+\kappa [q-1]\) and \(w {:=} b+\kappa \). Applying Lemma 2 with (2) and the property \(\varGamma (x+1)=x\varGamma (x)\), we get
Then using (18) we have
and (33) follows after some rewriting. \(\square \)
Proof of Theorem 5
We start from Corollary 3 and write a general power series expansion,
where the \(r_t\ge 2+2\kappa \) are powers and the \(\gamma _t\in \mathbb{C }\) are coefficients of the form \(i^{\beta _t\,\mathrm{sgn}\,k}\) times a real factor. In this expression the desired relation \(\tilde{\chi }(-k)=[\tilde{\chi }(k)]^*\) evidently holds, and the properties \(\tilde{\chi }(0)=1\), \(\tilde{\chi }^{\prime }(0)=0\), \(\tilde{\chi }^{\prime \prime }(0)=-V\) are clearly present. Then we write
where the powers \(r_t^{\prime }\ge 2+2\kappa \) and coefficients \(\delta _t\propto i^{\beta _t^{\prime }\mathrm{sgn}\,k}\) are obtained (laboriously) by substituting (89) into the Taylor series for the logarithm, \(\ln (1+\varepsilon )=\varepsilon -\varepsilon ^2/2+\varepsilon ^3/3-\varepsilon ^4/4+\cdots \). It is worth noting that \(m\) disappears from the \(k^2\) term, but not from the others. Equation (56) is obtained from (90) by using the Taylor series for the \(\exp \) function,
and (again laboriously) collecting terms with equal powers of \(k\). Since we started out with powers \(r_t\ge 2+2\kappa \), we end up with powers \(\nu _t\ge 2+2\kappa \). Finally, (57) follows by applying Lemma 10 and Lemma 11 to evaluate the integrals that arise when (56) is substituted into Theorem 4. \(\square \)
Rights and permissions
About this article
Cite this article
Simone, A., Škorić, B. False Negative probabilities in Tardos codes. Des. Codes Cryptogr. 74, 159–182 (2015). https://doi.org/10.1007/s10623-013-9856-x
Received:
Revised:
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10623-013-9856-x