2 \section{Introduction}\label{sec:introduction}
4 Information hiding has recently become a major digital
5 technology~\cite{1411349,Xie09:BASecurity}, especially with the
6 increasing importance and widespread distribution of digital media through the Internet.
7 Spread-spectrum data-hiding techniques have been widely studied in recent years
8 under the scope of security. These techniques encompass several schemes, such as
9 Improved Spread Spectrum~(ISS), Circular Watermarking~(CW), and Natural
10 Watermarking~(NW). Some of these schemes have revealed various
11 security issues. On the contrary, it has
12 been proven in~\cite{Cayre2008} that the Natural Watermarking technique is
13 stego-secure. This stego-security is one of the security classes defined
14 in~\cite{Cayre2008}, where probabilistic models are used to
15 categorize the security of data hiding algorithms in the Watermark Only
16 Attack~(WOA) framework.
18 We have explained in our previous research works~\cite{guyeux13:bc}
19 that any algorithm can be rewritten as an iterative
20 process, leading to the possibility to study its
21 topological behavior. As a concrete example,
22 we have shown that the security level of some information hiding algorithms
23 (of the spread-spectrum kind) can be studied into a
24 novel framework based on unpredictability, as it is understood in the mathematical
25 theory of chaos~\cite{guyeux13:bc}.
26 The key idea motivating our research works is that: \emph{if artificial intelligence (AI)
27 tools seem to have difficulties to deal with chaos, then steganalyzers (software based
28 on AI that try to separate original from stego-contents) may
29 be proven defective against chaotic information hiding schemes}. Our work
30 has thus constituted in showing theoretically that such chaotic schemes can be constructed.
31 We are not looking to struggle with best available information hiding techniques and
32 we do not focus on effective and operational aspects, as
33 our questioning are more locating in a conceptual domain. Among other things, we do
34 not specify how to chose embedding coefficients, but the way to insert the hidden
35 message in a selection of these ``least significant coefficient'' in an unpredictable
36 manner. To say this another
37 way, our intention is not to realize an hidden channel that does not appear as sleazy
38 to a steganalyzer, but to construct an information hiding scheme whose behavior cannot
39 be predicted: supposing that the adversary has anything (algorithm, possible embedding
40 coefficient, etc.) but the secret key, we want to determine if he can predict which coefficients
41 will be finally used, and in which order. To do so, a new class of security has been
42 introduced in~\cite{bg10b:ip}, namely the topological security. This new class can be used to study
43 some categories of attacks that are difficult to investigate in the existing
44 security approach. It also enriches the variety of qualitative and quantitative
45 tools that evaluate how strong the security is, thus reinforcing the confidence
46 that can be added in a given scheme.
48 In addition of being stego-secure, we have proven in~\cite{gfb10:ip} that
49 Natural Watermarking (NW) technique is topologically secure. Moreover, this technique
50 possesses additional properties of unpredictability, namely, strong transitivity,
51 topological mixing, and a constant of sensitivity equal to $\frac{N}{2}$~\cite{Guyeux2012}.
52 However NW are not expansive, which is in our opinion problematic in the Constant-Message
53 Attack (CMA) and Known Message Attack (KMA) setups, when we
54 consider that the attacker has all but the embedding key~\cite{Guyeux2012}.
55 Since these initial investigations, our research works in that information hiding field have
56 thus consisted in searching more secure schemes than NW, regarding the
57 concerns presented in the first paragraph of this introduction. The objective
58 of this review paper is to list the results obtained by following such an approach.
60 This article is organized as follows. Notations and terminologies are firstly recalled in
61 the next section. Then the formerly published $\mathcal{CIW}_1$ chaotic iteration based one-bit watermarking
62 process is recalled in detail in Section~\ref{sec:ciw1}. Its steganographic version $\mathcal{CIS}_2$
63 is then explained in Section~\ref{sec:secrypt11}, while Section~\ref{di3sec} presents the
64 $\mathcal{DI}_3$ process, whose aims is to merge the two previous approaches.
65 This review article of chaotic iterations based information hiding algorithms
66 ends by a conclusion section containing intended future works.