[Paper Review] Steganography: a Class of Algorithms having Secure Properties

Jacques M. Bahi, Jean-François Couchot+1 more|arXiv (Cornell University)|Dec 7, 2011

Advanced Steganography and Watermarking Techniques10 references4 citations

TL;DR

This paper introduces a formalized class of non-blind steganographic algorithms based on Devaney’s topological chaos in finite Boolean dynamical systems. By leveraging chaotic iterations with strongly connected iteration graphs and doubly stochastic Markov matrices, the authors prove both stego-security (uniform distribution of stego states) and chaos-security (Devaney chaos), generalizing prior work on the dhCI algorithm to a broad class of modes and strategy-adapters.

ABSTRACT

Chaos-based approaches are frequently proposed in information hiding, but without obvious justification. Indeed, the reason why chaos is useful to tackle with discretion, robustness, or security, is rarely elucidated. This research work presents a new class of non-blind information hidingalgorithms based on some finite domains iterations that are Devaney's topologically chaotic. The approach is entirely formalized and reasons to take place into the mathematical theory of chaos are explained. Finally, stego-security and chaos security are consequently proven for a large class of algorithms.

Motivation & Objective

• To formalize a general class of non-blind steganographic algorithms based on chaotic iterations in finite Boolean spaces.
• To prove stego-security—i.e., uniform distribution of stego states—using Markov chain convergence to a uniform stationary distribution.
• To establish chaos-security by demonstrating Devaney chaos in the iterative process via strongly connected iteration graphs.
• To generalize the dhCI algorithm beyond the negation mode to a wide range of functions and strategy-adapters.
• To provide a theoretical foundation for secure information hiding that unifies security, discretion, and chaos-based properties.

Proposed method

• The method employs Boolean discrete-time dynamical systems (BS) with asynchronous iteration using a strategy-adapter that generates a sequence of indices for updating bits.
• A mode function $ f_n: \mathds{B}^n \to \mathds{B}^n $ is defined for each $ n $, generalizing the negation function used in dhCI.
• The strategy-adapter ensures uniform distribution (u.d.) of indices over $ \llbracket 1, l \rrbracket $, independent of the cover data.
• The iterative process is modeled as a Markov chain with a transition matrix $ M $ derived from the mode function, where $ \pi^{t+1} = \pi^t M $.
• Stego-security is proven by showing that the Markov chain converges to a uniform stationary distribution $ \pi = (1/2^l, \dots, 1/2^l) $ under the assumption of doubly stochastic $ M $.
• Chaos-security is established by proving Devaney chaos via strong connectivity of the iteration graph $ \Gamma(f) $, ensuring sensitive dependence and dense orbits.

Experimental results

Research questions

• RQ1 Can a general class of steganographic algorithms be formally defined such that both stego-security and chaos-security are provable?
• RQ2 Does the convergence of the Markov chain induced by chaotic iterations to a uniform distribution guarantee stego-security?
• RQ3 Under what conditions on the mode function and strategy-adapter is Devaney chaos achieved in the iterative process?
• RQ4 Can the dhCI algorithm’s security properties be generalized beyond the negation mode to other Boolean functions?
• RQ5 What structural properties of the iteration graph $ \Gamma(f) $ ensure both chaos-security and stego-security?

Key findings

• The method achieves stego-security because the Markov chain induced by the iterative process converges to a uniform stationary distribution $ \pi = (1/2^l, \dots, 1/2^l) $, ensuring indistinguishability of stego states.
• Stego-security is formally proven under the condition that the Markov matrix $ M $ associated with the mode function is doubly stochastic and regular.
• Chaos-security is established by proving that Devaney’s chaos conditions (sensitive dependence, dense orbits, topological transitivity) are satisfied when the iteration graph $ \Gamma(f) $ is strongly connected.
• The class of algorithms generalizes the dhCI algorithm by allowing any mode function $ f_n $ whose iteration graph $ \Gamma(f_n) $ is strongly connected and whose Markov matrix is doubly stochastic.
• The CIIS strategy-adapter is shown to produce uniformly distributed indices, satisfying the u.d. condition required for stego-security.
• The theoretical framework allows for systematic construction of secure steganographic schemes by selecting appropriate modes and strategy-adapters with the required graph-theoretic and matrix properties.