Kerckhoffs-Based Embedding Security Classes for WOA Data Hiding

Abstract

Abstract-It has recently been discovered that using pseudorandom sequences as carriers in spread-spectrum techniques for data-hiding is not at all a sufficient condition for ensuring datahiding security. Using proper and realistic a priori hypothesis on the messages distribution, it is possible to accurately estimate the secret carriers by casting this estimation problem into a Blind Source Separation problem. After reviewing relevant works on spread-spectrum security for watermarking, we further develop on this topic to introduce the concept of security classes which broaden previous notions in watermarking security and fills the gap with steganography security as defined by Cachin. We define four security classes, namely, by order of creasing security: insecurity, key-security, subspace-security and stego-security. To illustrate these views, we present two new modulations for truly secure watermarking in the Watermark-Only-Attack (WOA) framework. The first one is called Natural Watermarking and can be made either stego-secure or subspace-secure. The second is called Circular Watermarking and is key-secure. We show that Circular Watermarking has robustness comparable to that of the insecure classical spread spectrum. We shall also propose information leakage measures to highlight the security level of our new spread-spectrum modulations.Index Terms-Spread spectrum watermarking, security.

Security and robustness constraints for spread-spectrum Tardos fingerprinting

“…α ISS (i, x) and λ ISS (x) are computed to achieve host-interference rejection and error probability minimization given a target W CR and N CR: In the WOA framework, following the classification of [8], this modulation is insecure. Adversaries are able to estimate the carriers for example by ICA [16], provided that they can gather enough contents.…”

“…However, if this solution is adopted, one has also to take into account the possible security of the embedding method. The design of fingerprinting codes has been studied in various works [2,3,4,5,6] and the security analysis of watermarking schemes has also been investigated [7,8] but outside of the fingerprinting framework. In the sequel, we study the impact of the use of secure embeddings in the context of fingerprinting using spread-spectrum methods and present dedicated collusion strategies.…”

Two key estimation techniques for the broken arrows watermarking scheme

“…This shows clearly an important increases of the variance within the secret subspace, consequently the strategy that is developed in this section is to estimate the subspace spanned by the vectors {ci} by estimating the components of important variances from the observations. If such similar strategies have already been used for security analysis of watermarking systems [5,2,12], the estimation of the secret subspace in our case is challenging for different reasons:…”

“…• Contrary to the systems addressed in [5,2,12], the proposed method used a secret subspace of large dimension (30) in order to avoid basis estimation techniques such as averaging,…”

“…Since all the variances along the different cone axes are equal, one solution to estimate the direction of each axis is to look for independent directions using Independent Component Analysis (ICA). This strategy has already been used in watermarking security by former key estimation techniques and more information on the usage of ICA in this context can be found in [12].…”