Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

Contents lists available at ScienceDirect

Commun Nonlinear Sci Numer Simulat

journal homepage: www.elsevier.com/locate/cnsns

Multiple-watermarking scheme based on improved chaotic maps

S. Behnia a,*, M. Teshnehlab b, P. Ayubi c
a
Department of Physics, IAU, Ourmia, Iran
b
Department of Elec. Eng., K.N Toosi University Technology, Tehran, Iran
c
Department of Computer, IAU, Ourmi, Iran

a r t i c l e i n f o a b s t r a c t

Article history: In this letter a new watermarking scheme for color image is proposed based on a fam-
Received 29 June 2009 ily of the pair-coupled maps. Pair-coupled maps are employed to improve the security
Accepted 29 September 2009 of watermarked image, and to encrypt the embedding position of the host image.
Available online 8 October 2009
Another map is also used to determine the pixel bit of host image for the watermark
embedding. The purpose of this algorithm is to improve the shortcoming of water-
Keywords: marking such as small key space and low security. Due to the sensitivity to the initial
Pair-coupled maps
conditions of the introduced pair-coupled maps, the security of the scheme is greatly
Pseudo random sequence
Ergodic theory
improved.
Robustness Ó 2009 Elsevier B.V. All rights reserved.
Digital signature image
Key space
security

1. Introduction

The majority of watermarking schemes proposed to date, use watermarks which are generated from pseudo ran-
dom number sequences [1–3]. Chaotic oscillators with positive Lyapunov exponents can be viewed as information
sources [3,4]. Previously, it has been demonstrated that an information signal can be encoded in the symbolic dynam-
ics of a chaotic oscillator [5,6]. Recent works have introduced the application of chaotic functions such as Markov
Maps, Bernoulli Maps, Skew Tent Map, Logistic Map [7–9]. Although theses maps have perfect dynamical properties
and can be realized simply in both hardware and software with a widespread usage [10,11] but there are fundamental
drawbacks in these chaotic cryptosystems, such as small key space, slow performance and weak security function
[12,13].
Considering a symmetric two-dimensional map which possesses invariant measure in its diagonal and anti-diagonal
invariant sub-manifolds, we have been able to propose a pair-coupled map that possesses ergodic property [14] at synchro-
nized states [15,16].
Entropy and the Lyapunov exponents were calculated, at synchronized state. We propose a secure watermarking scheme
based on the pair-coupled map. In order to enhance the security, chaos is employed to select the embedding positions for
each watermark bit and for watermark encryption. The introduced dynamical system with the level of the security proposed
can be applied to many logos in watermarking process. An advantage of these watermarks is the possibility to analyze and
control their spectral properties.

* Corresponding author. Tel.: +98 441 3458902; fax: +98 441 3460980.
E-mail address: s.behnia@iaurmia.ac.ir (S. Behnia).

1007-5704/$ - see front matter Ó 2009 Elsevier B.V. All rights reserved.
doi:10.1016/j.cnsns.2009.09.042
2470 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

The rest of the Letter is organized as follows. Section 2 describes chaotic maps and the location of embedding position of
watermark. In Section 3, the chaotic domain of the introduced model is studied via Lypaunov exponents in order to generate
the key space. The watermarking scheme based on chaotic maps is proposed in Section 4. Also, the selected example and
simulation results are discussed in Section 5. Section 6 is the conclusion. Two appendices are also provided, which contain
all algebraic calculations and proofs.

2. Coupled maps

Coupled map lattices are arrays of states whose values are continuous, usually within the unit interval, or discrete space
and time [17]. The pair-coupled map with ergodic behavior can be considered as a two-dimensional dynamical map defined
as [18]:

( 1
X ¼ Fðx; yÞ ¼ ½ð1  Þðf1 ðxÞÞP þ ðf2 ðyÞÞP P
UðX; YÞ ¼ 1
ð1Þ
Y ¼ Fðy; xÞ ¼ ½ð1  Þðf1 ðyÞÞP þ ðf2 ðxÞÞP P

where, in general, P is an arbitrary parameter,  the strength of the coupling, and the functions f1 ðxÞ; f 2 ðxÞ are two arbitrary
one-dimensional maps. Obviously, by choosing P ¼ 1, we get ordinary linearly coupled maps. But, in order to have a two-
dimensional dynamical system associated with the pair-coupled map with the property of possessing an invariant measure
at synchronized state [19,20], one needs to choose P as an arbitrary integer and the functions ðf1 ðxÞ; f2 ðxÞÞ as measurable
dynamical systems (see Appendix A ). At completely synchronized state x ¼ y, the pair-coupled map by considering A.1 re-
duces to:

pffiffiffi
ð; a1 ; a2 Þ tan2 ðN arctanð xÞÞ
X ¼ Fðx; xÞ ¼ a ð2Þ
1
ð; a1 ; a2 Þ ¼ ðð1  
with a ÞaP1 þ aP2 ÞP which is discussed latter (see (B.6) and Appendix B).

3. Lyapunov characteristic exponent

A chaotic system is sensitive to small changes in the initial state. This tendency to amplify small perturbations is quan-
tified by the Lyapunov exponent of the system [21]. For the one-dimensional mapping xnþ1 ¼ Uðxn ; aÞ, the Lyapunov expo-
nent can be computed from

X
n1  
dUN ðxÞ
kðx0 Þ ¼ lim ln  ; ð3Þ
n!1
k¼0
dx 

It is obvious that kðx0 Þ < 0 for an attractor, kðx0 Þ > 0 for a repeller and kðx0 Þ ¼ 0 for marginal situation. Also, the Lyapunov
number is independent of initial point x0 , provided that the motion inside the invariant manifold is ergodic, thus kðx0 Þ char-
acterizes the invariant manifold of UN as a whole [21,22].
The stability of the pair-coupled map can be assessed by computing its Lyapunov exponent spectrum. A spectrum of all
the Lyapunov exponents with respect to the synchronization solution can be evaluated in a fashion similar to that of one-
dimensional local maps [4]. At synchronized states, the Lyapunov exponents ðK Þ of the two-dimensional dynamical system
described by the map (Eq. (2)) are defined as the limn!1 1n jk ðxn Þj, where k ¼ h1 ðxÞ  h2 ðxÞ are eigenstates of the following
matrix:
  !nþ1
   @x @xnþ1 n h
Y i
 h1 ðxn Þ h2 ðxn Þ   @xnþ1 @yn  ð1  ÞaP1 aP2 1
   n  ðð1  ÞaP1 þ aP2 ÞP1 gðxk Þ
 h ðx Þ h ðx ÞÞ  ¼  @ynþ1 @ynþ1 
¼
aP2 ð1  Þap1
2 n 1 n  @x n @yn
 k¼0
xn ¼yn

which yields:

1 h 1 pffiffiffi i
Kþ ¼ lim ln jkþ ðxn Þj ¼ K ðð1  ÞaP1 þ aP2 ÞP tan2 ðN arctan xÞ ð4Þ
n
n!1
1 h 1 pffiffiffi i
K ¼ lim ln jk ðxn Þj ¼ ln jð1  ÞaP1  aP2 j  ln jð1  ÞaP1 þ aP2 j þ K ðð1  Þap1 þ aP2 ÞP tan2 ðN arctan xÞ ð5Þ
n!1 n

Result representing the general form of the Lyapunov characteristic exponent for the introduced family (Eq. (2)).

4. Watermark generation

Color is an important visual information which keeps humans fascinated since birth. The representation of color is based
on the classical three-color theory where any color can be reproduced by mixing an appropriate set of three primary colors.
 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478 2471

Color information is commonly represented in the widely used RGB (red, green, blue) Cartesian coordinate system. An RGB
color image, represented by 8 bits of R, G, and B pixels has 2563 or 16, 777,216 colors [23].
Watermark provides a natural link between chaotic dynamics and information theory in which, the recent idea of utiliz-
ing chaotic systems in encoding digital information in RGB mode is based [24]. Through considering different logo for each
color mode the watermarked image was generated according the following steps (see Fig. 2).
By considering the introduced model at synchronized state with N ¼ 2, Eq. (2) reduces to:
pffiffiffi
ð; a1 ; a2 Þ tan2 ð2 arctanð xÞÞ;
X ¼ Fðx; xÞ ¼ a ð6Þ

which is used to find the embedding position of the pixels from the watermark image to host image. The last part of the
encryption, which determines the bit of the host image pixels in watermark embedding employs the following map (see
Appendix A):

1 pffiffiffi
X¼ tan2 ð2 arctan xÞ: ð7Þ
a23
At the second stage, we select the key. A positive way to describe the key space [25] might be in terms of positive Lyapu-
nov exponents. Apparently, the attack complexity is determined by the size of the key space and the complexity of the ver-
ification of each key. For the key space we selected the control parameter by considering the analytically calculated
Lyapunov exponents (Eqs .(4) and (5)). In our introduced example, key space size (Eqs. (2) and (4)) consists of a coupling
parameter, three control parameters and three initial conditions. We assume the original image of the size n  n to be de-
noted as U ¼ fUðx1 ; x2 Þ; 1 6 x1 ; x2 6 ng and the binary watermark (logo) of size m  m for each mode (red, green and blue)
be denoted by / ¼ f/ði; jÞ; 1 6 i; j 6 mg, where ði; jÞ represents the pixel coordinate of the binary watermark image and
ðx1 ; x2 Þ the original host image. Watermark bits (1 bit per pixel) are embedded in the host image according to the following
process. Using the coordinate ði; jÞ position of watermark pixel as the initial condition and through setting a value for the
control parameter, the coupled map is iterated after which, the embedding position of the pixels from the watermark image
to host image can be obtained. The watermark pixels will get different embedding positions, so, the embedded watermark
pixels will spread on the host image randomly.
This process follows the iteration of the third map (Eq. (4)) to determine the bit of host image pixels. Accounting to the
chaotic domain of the third map, (see Fig. 1) the truly random sequence is generated over the interval ½0; 1Þ. Then, the inter-
val can be divided into several subintervals ð½0; x1 ; ½x1 ; x2 ; . . .Þ which correspond to different pixel bits through watermark
embedding [5,6].
As it was discussed, through dividing the sub domain which correspondes to the kth bit of host image, such as
ð½0; 0:25 ! k3 ; ½0:25; 0:5 ! k4 ; . . .Þ the watermark pixel will be embedded to the kth bit of pixel in host image.
The embedded watermark pixel is denoted as Uðx1 ; x2 Þ. If /ði; jÞ is the same as the kth bit of Uðx1 ; x2 Þ, then
/ðx1 ; x2 Þ ¼ Uðx1 ; x2 Þ, i.e., the pixel value is kept unchanged; otherwise, the kth bit of Uðx1 ; x2 Þ is substituted by /ði; jÞ. The
block diagram and the flowchart of the algorithm is shown in Figs. 2 and 3.

−1

−2
Lyapunov Exponents

−3

−4

−5

−6

−7

−8
0 0.5 1 1.5 2 2.5 3
α

Fig. 1. Lyapunov exponents: Eq. (7) vs. a.

2472 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

Fig. 2. Block diagram.

 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478 2473

Fig. 3. Flowchart.

Watermark extraction is just the inverse process of the above embedding algorithm. In the process of extraction, one
needs to know the key parameters. Since both decryption and encryption procedures have similar structure, they essentially
have the same algorithmic complexity and time consumption.
2474 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

5. Experimental results

In this section the experimental results of our proposed scheme are discussed. To demonstrate the efficiency of the pro-
posed algorithm, MATLAB simulations are performed by using 512  512 pixel RGB color ‘‘MADINEH” image and multiple
128  128 pixel binary watermark logos ‘‘MOHAMMAD”, ‘‘ALI” and ‘‘FATIMA”. Fig. 4 demonstrates the invisibility of the
watermark. Fig. 4 (a) and (b–d) show the original host image and multiple watermark logos respectively. The watermarked
image and the extracted multiple watermark logos ‘‘MOHAMMAD”, ‘‘ALI” and ‘‘FATIMA”, are respectively presented in
Fig. 4(e) and (f–h). The watermark embedding process is said to be imperceptible if the humans cannot distinguish between
the original data and watermarked data. To quantitatively evaluate the performance of the proposed scheme, the peak sig-
nal-to-noise ratio (PSNR) was adopted to measure the image quality of a watermarked image which is given by:

Max2I
PSNR ¼ 10  log10 MSE
ðdBÞ;
ð8Þ
1
P
M P
N P
L
 i;j;k Þ;
MSE ¼ MNL ðHi;j;k  H
i¼1 j¼1 k¼1

where Hði; j; kÞ and H0 ði; j; kÞ denote the pixel values in the location (i, j, k) of the original host image and the watermarked
image, respectively, and M  N  L is the color image size and Max2I is the maximum possible pixel value of the color image.
In this study, reliability was measured as the bit error rate (BER) of extracted watermark, which is definition as:

B
BER ¼  100; ð9Þ
MN
where B is the number of erroneously detected bits and M  N is the extracted binary watermark image dimensions. The
PSNR between watermarked image and original image is 30.11 dB, and the BER of the extracted watermark is zero. Therefore,
there is no obvious perceptual distortion between watermarked image and original one; the embedded watermark does not
degrade the quality of the original host image. To test the robustness of our proposed method, we applied several attacks to

Fig. 4. (a) Original ‘‘MADINEH” image, (b) watermark logo ‘‘FATIMA”, (c) watermarked image and (d) extracted watermark logo.
 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478 2475

Fig. 5. The watermarked ‘‘MADINEH” image under different attacks: (a) JPEG compression, (b) salt & pepper noise 10%, (c) Gaussian noise (0, 0.01), (d)
histogram equalization, (e) median filter ½3  3, (f) low-pass filter ½5  5, (g) gamma correction 0.6, (h) motion blur (45°), (i) rotation (2°), (j) cropping
(25%), (k) sharpening and (l) complement (left to right to each mode).

the watermarked image including JPEG compression, salt & pepper noise, Gaussian noise, Gaussian low-pass filter, gamma
correction, histogram equalization, sharpening, rotation, motion blur, complement, cropping and median filtering.
Fig. 5 shows an example of a watermarked image which is attacked by the methods listed in Table B.1. The corresponding
best extracted watermarks are shown in Fig. 6. The test results for ‘‘MADINEH” image are presented in Table B.1.
The key space size is the total number of different keys that can be used in the encryption. Cryptosystem is completely
sensitive to all the secret keys. The order of time complexity for watermark extraction in our proposed method is computed
as below:

hð  x30  iter  a3 Þ; ð10Þ

where  e ½0; 1; x0 e ½0; 1ðx0 e ½0; 1ÞÞ, and a e ½N1 ; 1Þ. Apparently, the key space is large enough to resist all kinds of brute-force
attacks.

6. Concluding remarks

We propose a novel watermarking scheme for image authentication based on the chaotic maps. The scheme focuses on
color images, thus fitting various network multimedia applications. Chaos is applied to design the selection approach for
watermark embedding. We have used symmetric pair-coupled maps to increase both the number of keys (control parame-
ters) and complexities involved in the algorithm. This algorithm tries to address the shortcoming of watermarking such as
small key space and low level of security. Without the correct initial condition, the watermark cannot be successfully de-
tected. Therefore, may be an effective technique for color image authentication. This watermarking scheme may have appli-
cations in law, commerce, defence, medical databases and journalism.
The aim of this work was to realize a watermarking method with a private code. So further studies must be started to
develop watermarking methods with a public key.

Appendix A. Definition of maps

We present a brief review of one-dimensional chaotic maps which are going to be used. One-parameter families of chaotic
maps of the interval [0, 1] with an invariant measure can be defined as the ratio of polynomials of degree N (see [4] for more
detail):
2476 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

Table B.1
Simulation results of PSNR and BER.

Attacks PSNR (dB) BER (%)

MOHAMMAD ALI FATIMA
JPEG compression 23.15 42.53 34.5 49.7
Salt & pepper noise 10% 25.77 5 4.9 4.9
Gaussian noise (0, 0.1) 19.02 31.54 33.43 31.52
Histogram equalization 18.55 31.84 32.62 33.96
Median filtering ½3  3 25.31 45.23 45.76 49.53
Low pass filter ½5  5 25.38 21.93 19.7 15.3
Gamma correction 33.44 58.79 59.53 35.82
Motion blur 45° 21.57 47.99 47.46 50.76
Rotation 2° 23.82 28 27.17 21.37
One quarter cropped 20.16 2.4 4.8 4.5
Sharpening 20.71 21.63 26.39 27.35
Complement 17.91 100 100 100

Fig. 6. Extracted watermarks under different attacks: (a) JPEG compression, (b) salt and pepper noise 10%, (c) Gaussian noise (0, 0.01), (d) histogram
equalization, (e) median filter ½3  3, (f) low-pass filter ½5  5, (g) gamma correction 0.6, (h) motion blur (45°), (i) rotation (2°), (j) cropping (25%), (k)
sharpening and (l) complement (left to right to each mode).

a2 F
xnþ1 ðx; aÞ ¼ ; ðA:1Þ
1 þ ða2  1ÞF

where a is the control parameter that can potentially be used as the secret keys for secure communication. F substitute with
chebyshev polynomial of type one T N ðxÞ. It is shown that these maps have an interesting property, that is, for even values of
N the xnþ1 ðx; aÞ maps have only a fixed point attractor x ¼ 1 provided that their parameter belongs to interval ðN; 1Þ while, at
a P N they bifurcate to chaotic regime without having any period doubling or period-n-tupling scenario and remain chaotic
for all a 2 ð0; NÞ, but for odd values of N, these maps have only fixed point attractor at x ¼ 0 for a 2 ðN1 ; NÞ, again they bifur-
cate to a chaotic regime at a P N1 , and remain chaotic for a 2 ð0; N1 Þ, finally they bifurcate at a ¼ N to have x ¼ 1 as fixed point
attractor for all a 2 ðN1 ; 1Þ (see Fig. 1). Here in this paper we are concerned with their conjugate maps which are defined as:

ð1Þ ð1Þ 1 1 pffiffiffiffiffi

~xnþ1 ðx; aÞ ¼ h  xnþ1 ðx; aÞ  h ¼ tan2 ðNarctan xn Þ; ðA:2Þ
a2

Conjugacy means that invertible map hðxÞ ¼ 1x maps I ¼ ½0; 1 into ½0; 1Þ. In order to simplify the calculation of invariant
px ffiffiffi
measure in this paper we denote ”tan2 ðNarctan xÞ” with gðxÞ. Off course the functions given in Appendix B is not the only
choice for the functions gðxÞ that leads to the hierarchy of chaotic maps with an invariant measure. Obviously the following
choices of the functions:
pffiffiffi pffiffiffi
 1
a2
tan2 ðNarccot xÞ;  1
a2
cot2 ðNarccot xÞ,
1 1
 a
j tanðNarctanjxjÞj;  a
j tanðNarccotjxjÞj,
1 1
 a
j cotðNarctanjxjÞj;  a
j cotðNarccotjxjÞj.

lead to the hierarchy of chaotic maps of trigonometric types (with an invariant measure), where some of them are equivalent
to each others up to conjugacy [26].
 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478 2477

Appendix B. Ergodicity

Each dynamical system has specific characteristics. In fact, the differences between discrete dynamical systems arise
from these properties. Furthermore, many properties of the chaotic systems have their corresponding counterparts in
traditional cryptosystems, such as: ergodicity and confusion, sensitivity to initial conditions/control parameter and dif-
fusion [27]. We have tried to describe ergodicity from the invariant measure point of view, for pair-coupled maps.
We show the uniqueness of the invariant measure and conclude that they are ergodic. The probability measure l for
the symmetric two-dimensional map Fðx; yÞ given in Eq. (1) fulfills the following formal Fobenius–Perron (FP) integral
equation [28]:
Z Z
lðX; YÞ ¼ dx dy dðX  Fðx; yÞÞdðY  Fðy; xÞÞlðy; xÞ;

the corresponding FP equation can be written as:

X
lðX; YÞ ¼ e
lðxk ; y‘ ÞJðxk ; y‘ Þ k; ‘ ¼ 1; 2; . . . ; M ðB:1Þ
ðxk ;y‘ Þ2U1 ðX;YÞ

where xk and y‘ be roots of Eq. (1) and the jocobian J is defined as:
 0 1
 @Fðxk ;y‘ Þ @Fðxk ;y‘ Þ 
 @x @y‘ 

J ðx‘ ; yk Þ ¼ det @ @Fðx k;y Þ
1 A ðB:2Þ
@Fðxk ;yl Þ 
 k ‘
@xk @y‘


By considering the one-dimensional maps with invariant measure lðyÞ, one can prove that, at synchronized states, the
invariant measure of pair-coupled maps may take the following form:

lðx; yÞ ¼ dðx  yÞlðyÞ; ðB:3Þ

we insert lðx; yÞ ¼ dðx  yÞlðyÞ in B.1, after using the Dirac delta function, we get:
X l ð xk Þ X dðxk  y‘ Þ
 e
k; ‘ ¼ 1; 2; . . . :; M
k
jh1 ðxk Þ þ h2 ðxk Þj ‘
jh1 ðxk Þ  h2 ðxk Þj

We have used h1 ðxÞ ¼ @Fðx;yÞ

@x
jx¼y and h2 ðxÞ ¼ @Fðx;yÞ
@y
jx¼y . For a given root xk , the last term of sum is reduced to:
X dðxk  y‘ Þ X
¼ dðf ðxk ; y‘ Þ  f ðy‘ ; xk ÞÞ ¼ dðf ðxk ; yÞ  f ðy; xk ÞÞ ¼ dðX  YÞ;
‘
jh1 ðxk Þ  h2 ðxk Þj ‘

where two last equalities result from the fact xk 2 U1 ðX; YÞ, i.e., xk is one of the roots of the map Eq. (1) for a given set of
fX; Yg and the expansion of dðFðxk ; yÞ  Fðy; xk ÞÞ(by considering y as the variable). Substituting the obtained results in B.1, we
obtain:
X lðxk Þ
lðX; YÞ ¼ dðX  YÞlðXÞ ¼ dðX  YÞ
1
jh1 ðxk Þ þ h2 ðxk Þj
xk 2U ðX;XÞ

which implies that lðXÞ should satisfy:

X X  
lðxk Þ df ðxk ; xk Þ1
lðXÞ ¼ ¼   lðxk Þ ðB:4Þ
jh1 ðxk Þ þ h2 ðxk Þj  dx 
k
xk 2U1 ðX;XÞ xk 2U1 ðX;XÞ

which is the same as PF equation of one-dimensional map X ¼ Fðx; xÞ. The required condition for the presentation of the
invariant measure of the synchronized coupled map is choosing a one-dimensional map with an invariant measure as intro-
duced in our previous work (see Appendix A). By considering (A.1), this map displays the invariant measure of the following
form [4]:
pffiffiffi
b
lðxÞ ¼ pffiffiffi pffiffiffi ; ðB:5Þ
xð1 þ bxÞ
provided that we choose the constant b as one of the positive roots of the following equation:
0 12
P½N2 
N k
C b
ð; a1 ; a2 Þ ¼ @ N1
a k¼0 2k A ðB:6Þ
P½ 2  N k
k¼0 C 2kþ1 b
2478 S. Behnia et al. / Commun Nonlinear Sci Numer Simulat 15 (2010) 2469–2478

where [ ] means the greatest integer part. Now, by substituting the invariant measure of one-dimensional maps (B.4) in the
relation (B.5), we get the following expression for the invariant measure of pair-coupled map:
pffiffiffi
b
lðx; yÞ ¼ dðx  yÞ pffiffiffi pffiffiffi : ðB:7Þ
xð1 þ bxÞ

References

[1] Tsekeridou S, Solachidis V, Nikolaidis N, Nikolaidis A, Tefas A, Pitas I. Statistical analysis of a watermarking system based on Bernoulli chaotic
sequences. Signal Process 2001;81:1273–93.
[2] Gonza leza CM, Larrondoa HA, Rosso OA. Intensive statistical complexity measure of pseudorandom number generators. Physica A 2005;354:133–8.
[3] Falcioni M, Palatella L, Pigolotti S. Properties making a chaotic system a good pseudo random number generator. Phys Rev E 2005;72:016210–20.
[4] Jafarizadeh MA, Behnia S, Khorram S, Nagshara H. Hierarchy of chaotic maps with an invariant measure. J Stat Phys 2001;516:1013–28.
[5] Behnia S, Akhshani A, Ahadpour S, Mahmodi H, Akhavan A. A fast chaotic encryption scheme based on piecewise nonlinear chaotic maps. Phys Lett A
2007;366:391–6.
[6] Behnia S, Akhshani A, Ahadpour S, Mahmodi H, Akhavan A. Cryptography based on chaotic random maps with position dependent weighting
probabilities. Chaos Solitons Fract 2008;40(1):362–9.
[7] Nikolaidis S, Pitas I. Comparison of different chaotic maps with application to image watermarking. In: Proceedings of IEEE international symposium on
circuits and systems, Geneva; 2002. p. 509–12.
[8] Mooney A, Keating JG, Heffernan DM. A detailed study of the generation of optically detectable watermarks using the logistic map. Chaos Solitons Fract
2006;30:1088–97.
[9] Tong X, Cui M. Image encryption scheme basedon 3D-baker with dynamical compound chaotic sequence ciphergenerator. Signal Process
2009;89:480–91.
[10] Baranousky A, Daems D. Design of one-dimensional chaotic maps with prescribed statistical properties. Int J Bifurcation Chaos 1995;5:1585–98.
[11] Li S, Mou X, Cai Y, Ji Z, Zhang J. On the security of a chaotic encryption scheme: problems with computerized chaos in finite computing precision.
Comput Phys Commun 2003;153:52–8.
[12] Mazloom S, Eftekhari-Moghadam AM. Color image encryption based on coupled nonlinear chaotic map. doi:10.1016/j.chaos.2009.03.084.
[13] Alvarez G, Montoya F, Romera M, Pastor G. Cryptanalysis of a discrete chaotic cryptosystem using external key. Phys Lett A 2003;319:334339.
[14] Medio A. Chaotic dynamics; theory and applicatios to economics. Cambridge: Cambridge University Press; 1999.
[15] Balmforth NJ, Jacobson A, Provenzale A. Synchronized family dynamics in globally coupled maps. Chaos 1999;9(3):738–54.
[16] Rivera M, Martinez Mekler G, Parmananda P. Synchronization phenomena for a pair of locally coupled chaotic electrochemical oscillators: a survey.
Chaos 2006;16:037105–8.
[17] Kaneko K. Supertransients spatiotemporal intermittency and stability of fully developed spatiotemporal chaos. Phys Lett A 1990;149(2,3):105–12.
[18] Jafarizadeh MA, Behnia S, Faizi E, Ahadpour S. Global synchronization and anti-synchronization in n-coupled map lattices. Int J Theor Phys
2008;47:1005–15.
[19] Pecora LM, Carroll TL. Synchronization in chaotic systems. Phys Rev Lett 1990;64(8):821–5.
[20] Shabunin A, Astakhov V. Quantitative analysis of chaotic synchronization by means of coherence. Phys Rev E 2005;72:016210–8.
[21] Ott E. Chaos in dynamical system. Canada: Cambridge university pess; 1993.
[22] Dorfman JR. An introduction to chaos in nonequilibrium statistical mechanics. Cambridge: Cambridge University Press; 1999.
[23] Acharyand T, Ray AK. Image preessing principle application. Willy; 2005.
[24] Peng Z, Liu W. Color image authentication based on spatiotemporal chaos and SVD. Chaos Solitons Fract 2008;36:946–52.
[25] Tsekeridou S, Solachidis V, Nikolaidis N, Nikolaidis A, Tefas A, Pitas I. Theoretic investigation of the use of watermark signals derived from Bernoulli
chaotic. In: Proceedings of IEEE international conference on acoustics, speech and signal processing; 2001. p. 1989–97.
[26] Jafarizadeh MA, Foroutan M, Ahadpour S. Hierarchy of rational order families of chaotic maps with an invariant measure. Pramana
2006;67(6):1073–86.
[27] Behnia S, Akhshani A, Mahmodi H, Akhavan A. Chaotic cryptoghraphic scheme based on composition maps. Int J Bifurcation Chaos 2008;18(1):251–61.
[28] Cornfeld IP, Fomin SV, Sinai Ya G. Ergodic theory. Berlin: Springer-Verlag; 1982.