Physics Letters A 346 (2005) 153–157

www.elsevier.com/locate/pla

Chaos-based image encryption algorithm ✩

Zhi-Hong Guan a,∗ , Fangjun Huang a , Wenjie Guan b
a Department of Control Science and Engineering, Huazhong University of Science and Technology, Wuhan, Hubei 430074, PR China
b Department of Electrical and Computer Engineering, University of Waterloo, Waterloo, ON N2L 3G1, Canada

Received 26 March 2005; received in revised form 2 August 2005; accepted 2 August 2005
Available online 8 August 2005
Communicated by A.R. Bishop

Abstract
In this Letter, a new image encryption scheme is presented, in which shuffling the positions and changing the grey values of
image pixels are combined to confuse the relationship between the cipher-image and the plain-image. Firstly, the Arnold cat
map is used to shuffle the positions of the image pixels in the spatial-domain. Then the discrete output signal of the Chen’s
chaotic system is preprocessed to be suitable for the grayscale image encryption, and the shuffled image is encrypted by the
preprocessed signal pixel by pixel. The experimental results demonstrate that the key space is large enough to resist the brute-
force attack and the distribution of grey values of the encrypted image has a random-like behavior.
 2005 Elsevier B.V. All rights reserved.

Keywords: Chaos; Image encryption; Arnold map; Chen’s chaotic system

1. Introduction portant. To meet this challenge, a variety of traditional

encryption algorithms have been proposed [1–4]. Re-
With the rapid growth of multimedia production cently, along with the rapid development of theory and
systems, electronic publishing and widespread dis- application of chaos, many researchers are now fo-
semination of digital multimedia data over the Inter- cusing on the chaotic cryptography. A lot of image
net, protection of digital information against illegal encryption schemes based on chaos theory have been
copying and distribution has become extremely im- presented [5–10]. These applications have been moti-
vated by the chaotic properties such as ergodicity and
sensitive dependence on initial conditions and system
✩
This research was supported by the National Natural Science parameters, in addition to complex dynamics and de-
Foundation of China under Grant 60074009 and 60274004, the Lab terministic behaviors.
Foundation of Institute of Automation, Chinese Academy of Sci-
ences.
In [5], a systematical method was suggested for
* Corresponding author. adapting an invertible two-dimensional chaotic map
E-mail address: [email protected] (Z.-H. Guan). on a torus or on a square to create a symmetric block
0375-9601/$ – see front matter  2005 Elsevier B.V. All rights reserved.
doi:10.1016/j.physleta.2005.08.006
154 Z.-H. Guan et al. / Physics Letters A 346 (2005) 153–157

encryption scheme. The main idea is to shuffle the pixels, we adopt Arnold cat map to shuffle the pixel
positions of the pixels of the plain-image in the spatial- positions of the plain-image. Without loss of general-
domain. In [6], the authors present a new chaotic key- ity, we assume the dimension of the original grayscale
based image encryption algorithm (CKBA) to change image I is N × N . The coordinates of the pixels are
the pixel values of the plain-image. However, Li and S = {(x, y) | x, y = 0, 1, 2, . . . , N − 1}. Arnold cat
Zheng point out that CKBA is very weak to the map is described as [12]
chosen/known-plaintext attack with only one plain-


x x
image, and its security to brute-force ciphertext-only = A (mod N )
attack is overestimated by the authors [7]. In [8], y y



an algorithm for encoding binary images using one- 1 p x
= (mod N ) (1)
dimensional chaotic maps is presented. It can be seen q pq + 1 y
that, this method [8] is not secure enough according
where p and q are positive integers, det(A) = 1. The
to [7]. In order to improve the security of the image
map is area-preserving since the determinant of its lin-
encryption algorithm, many researchers prefer shuf-
ear transformation matrix equals (1). The (x  , y  ) is the
fling the positions and changing the grey values of
new position of the original pixel position (x, y) when
image pixels simultaneously. For instance, in [9,10]
Arnold cat map is performed once. Iterated actions of
the two-dimensional chaotic map is generalized to 3D
A on a pixel r0 ∈ S form a dynamical system
for designing a real-time secure symmetric encryp-
tion scheme. The new scheme employs the 3D map rn+1 = An r0 (mod N ) or rn+1 = Arn (mod N ),
to shuffle the positions of image pixels and uses an-
other chaotic map to confuse the relationship between where n = 0, 1, 2, . . . . The set of points {r0 , r1 , r2 , . . .}
the cipher-image and plain-image. is an orbit of the system with a period, i.e., there ex-
A new scheme is suggested in this Letter for secure ist positive integers T and n0 , such that rn+T = rn ,
image encryption. Shuffling the positions and chang- n = n0 , n0 + 1, n0 + 2, . . . . The period T depends on
ing the grey values of image pixels are performed si- the parameters p, q and the size N of the original
multaneously in our method. The rest of this Letter is image. Thus the parameters p, q and the number of
organized as follows. Section 2 describes the proposed iterations M all can be used as the secret keys. Since
method. Section 3 presents our experiments. Security there only exists a linear transformation and mod func-
analyses are given in Section 4, and finally this Letter tion, it is very efficient to shuffle the pixel positions
is concluded in Section 5. using the Arnold cat map. After several iterations, the
correlation among the adjacent pixels can be disturbed
completely. Some experiments are given in Section 3
2. The proposed cryptosystem to demonstrate the efficiency of Arnold cat map. How-
ever, the periodicity of Arnold cat map should degrade
The image encryption algorithm includes two steps: the security the encryption, because the possible at-
Firstly, the positions of the pixels of the original image tacks may iterate the Arnold cat map continuously to
are shuffled by Arnold cat map. Then the pixel values reappear the original plain-image. As a remedy, we
of the shuffled image are encrypted by Chen’s chaotic adopt Chen’s chaotic system to change the pixel values
system. next to improve the security.

2.1. Encryption by Arnold cat map 2.2. Encryption by Chen’s chaotic system

Image data have strong correlations among adja- Chen’s chaotic system is first presented by Prof.
cent pixels. Statistical analysis on large amounts of G. Chen in 1999, which is described as following [13],
images shows that averagely adjacent 8 to 16 pixels 
are correlative in horizontal, vertical, and also diago-  ẋ = a(y − x),
nal directions for both natural and computer-graphical ẏ = (c − a)x − xz + cy,

images. In order to disturb the high correlation among ż = xy − bz,
 Z.-H. Guan et al. / Physics Letters A 346 (2005) 153–157 155

where a, b and c are parameters. If one chooses where Abs(x) returns the absolute value of x. Floor(x)
a = 35, b = 3, c ∈ [20, 28.4], the system is chaotic rounds the elements of x to the nearest integers less
as shown by Fig. 1. The equations of Chen’s system than or equal to x. mod(x, y) returns the remainder
are quite similar to those of the Lorenz system [11], after division. The function de2bi(x) converts decimal
but topologically they are not equivalent, essentially number x to binary value. Because in our cryptosys-
due to the parameter c in front of the state vari- tem all the variables declared as type double which has
able y, which leads to abundant dynamic characters a bit-length of 64 bits. When expressed in scientific
of the new system. Therefore, the dynamical property notation, all the variables have a 15-digit precision.
of the Chen’s chaotic system is more complicated than The decimal fractions of the variables are multiplied
the Lorenz chaotic system. This feature is very use- by 1014 . Moreover, in mod(x, y) function the variable
ful in secure communications. The chaotic behavior y is chosen as 256 because the grayscale image with
of Chen’s chaotic system using fourth order Runge– 256 grey levels is used in our scheme. The shuffled
Kutta algorithm is shown in Fig. 1. The step of the image is encrypted as
Runge–Kutta is chosen as 0.001. As we can see, the
trajectory of the Chen’s chaotic system is composed C3×(i−1)+1 = B3×(i−1)+1 ⊕ Bxi ,
of a lot of discrete dots. C3×(i−1)+2 = B3×(i−1)+2 ⊕ Byi ,
In our scheme, three discrete variables of the
C3×(i−1)+3 = B3×(i−1)+3 ⊕ Bzi ,
Chen’s chaotic system are adopted to encrypt the shuf-
fled image. The encryption process consists of three where i = 1, 2, . . . represents the ith iteration of the
steps of operations. Chen’s chaotic system. The symbol ⊕ represents the
(1) The pixels of the shuffled image are arranged exclusive OR operation bit-by-bit. The Chen’s chaotic
by the order from left to right and then top to bottom system is iterated until all the elements in the set B =
and we can get a set S = {S1 , S2 , . . . , SN ×N }, in which {B1 , B2 , . . . , BN ×N } is encrypted. Then every element
each element is the decimal grey value of the pixel. in the encrypted set C = {C1 , C2 , . . . , CN ×N } is con-
Convert decimal pixel values to binary numbers and verted into decimal numbers and we can obtain the
we can get a new set B = {B1 , B2 , . . . , BN ×N }. cipher-image.
(2) Iterate the Chen’s chaotic for N0 times.
(3) The Chen’s chaotic system is iterated continu-
ously. For each iteration, we can get three values xi , yi 3. Experimental results
and zi . These decimal values are preprocessed first as
follows Some experimental results are given in this section
    to demonstrate the efficiency of our scheme. In all our
Bxi = de2bi mod Abs(xi ) − Floor Abs(xi ) experiments, the precision is 10−14 , which is easy to

× 1014 , 256 , be realized on today’s personal computer. The plain-
image with the size 124×124 is shown in Fig. 2(a) and

(a) (b)

Fig. 2. Plain-image and its histogram: (a) plain-image; (b) histogram

Fig. 1. Chaotic behavior of Chen’s system. of the plain-image.
156 Z.-H. Guan et al. / Physics Letters A 346 (2005) 153–157

(a) (b) (a) (b)

Fig. 3. Encryption by using Arnold cat map: (a) shuffled image; Fig. 5. Decrypted image and its histogram: (a) decrypted image;
(b) histogram of the shuffled image. (b) histogram of the decrypted image.

(a) (b) (a) (b)

Fig. 4. Encryption by Chen’s chaotic system: (a) cipher-image; Fig. 6. The sensitivity to the secret key x0 : (a) decrypted image
(b) histogram of the cipher-image. (x0 = −10.05800000000001, y0 = 0.368, z0 = 37.368); (b) his-
togram of the decrypted image.

the histogram of the plain-image is shown in Fig. 2(b).

Fig. 3(a) is the shuffled image and Fig. 3(b) is the the chaotic systems are deterministic, the receiver can
histogram of the shuffled image. The secret keys are reconstruct the same plain-image exactly using the se-
chosen as p = 1, q = 1 and M = 5. As can be seen cret keys.
that, Arnold cat map only shuffle the pixel positions of
the image since the histogram of the plain-image is the
same as the shuffled image. 4. Security analyses
Fig. 4(a) illustrates the cipher-image and Fig. 4(b)
is the corresponding histogram. The parameters are A good encryption scheme should be sensitive to
selected as a = 35, b = 3, c = 28. The secret keys the secret keys, and the key space should be large
to change the pixel values of the shuffled image are enough to make brute-force attacks infeasible. In our
x0 = −10.058, y0 = 0.368, z0 = 37.368. As we can encryption algorithm, the initial values of Chen’s
see, the histogram of the ciphered image is fairly uni- chaotic system are used as secret keys. If the preci-
form and is significantly different from that of the orig- sion is 10−14 , the key space size is 1042 . Moreover,
inal image. The encryption procedure complicates the the parameters p, q and M of Arnold cat map are
dependence of the statistics of the output on the statis- also used as the secret keys. The key space is large
tics of the input. enough to resist all kinds of brute-force attacks. The
Fig. 5(a) illustrates the decrypted image and experimental results also demonstrate that our scheme
Fig. 5(b) is the corresponding histogram. The parame- is very sensitive to the secret key mismatch (10−14 ).
ters are selected as a = 35, b = 3, c = 28. The secret Fig. 5 illustrates the sensitivity of our scheme to the
keys to change the pixel values of the shuffled im- secret key x0 . The cipher-image is shown in Fig. 4(a),
age are x0 = −10.058, y0 = 0.368, z0 = 37.368. Since which is decrypted using x0 = −10.05800000000001,
 Z.-H. Guan et al. / Physics Letters A 346 (2005) 153–157 157

5. Conclusions

In this Letter, a new image encryption scheme is

presented. Shuffling the positions and changing the
grey values of image pixels are combined simultane-
ously to ensure the security of our method. The pro-
posed algorithm has three merits: (1) the algorithm has
a large enough key space to resist all kinds of brute-
(a) (b) force attacks; (2) the cipher-image has a good sta-
tistical property; (3) the encryption algorithm is very
Fig. 7. The sensitivity to the secret key y0 : (a) decrypted image
(x0 = −10.058, y0 = 0.36800000000001, z0 = 37.368); (b) his-
sensitive to the secret keys.
togram of the decrypted image.

References

[1] S.S. Maniccam, N.G. Bourbakis, Pattern Recognition 37

(2004) 725.
[2] C.-C. Chang, M.-S. Hwang, T.-S. Chen, J. Syst. Software 58
(2001) 83.
[3] N. Bourbakis, C. Alexopoulos, Pattern Recognition 25 (6)
(1992) 567.
[4] H. Cheng, X.B. Li, IEEE Trans. Signal Process. 48 (8) (2000)
2439.
[5] J. Fridrich, Int. J. Bifur. Chaos 8 (6) (1998) 1259.
(a) (b)
[6] J.-C. Yen, J.-I. Guo, A New Chaotic Key-Based Design for
Image Encryption and Decryption, in: Proceedings IEEE In-
Fig. 8. The sensitivity to the secret key z0 : (a) decrypted image ternational Conference on Circuits and Systems, vol. 4, 2000,
(x0 = −10.058, y0 = 0.368, z0 = 37.36800000000001); (b) his- pp. 49–52.
togram of the decrypted image. [7] S. Li, X. Zheng, Cryptanalysis of a Chaotic Image Encryption
Method, Scottsdale, AZ, USA, 2002, in: Proceedings IEEE In-
ternational Symposium on Circuits and Systems, vol. 2, 2002,
pp. 708–711.
y0 = 0.368, z0 = 37.368, p = 1, q = 1 and M = 5. As [8] F. Beldhouche, U. Qidwai, Binary Image Encoding Using 1D
can be seen that, even the secret key x0 is changed a Chaotic Maps, in: IEEE Annual Technical Conference, 11
little (10−14 ), the decrypted image is absolutely dif- April 2003, pp. 39–43.
ferent from the plain-image. Similar results for other [9] Y. Mao, G. Chen, Chaos-Based Image Encryption, Handbook
secret keys y0 and z0 all can be obtained, which are of Computational Geometry for Pattern Recognition, Com-
puter Vision, Neurocomputing and Robotics, Springer-Verlag,
shown in Figs. 6–8. As we can see, the decrypted im- Berlin, 2003.
age with wrong keys has a histogram with random [10] G. Chen, Y. Mao, C.K. Chui, Chaos Solitons Fractals 21 (2004)
behavior. The sensitivity to initial conditions which is 749.
the main characterization of chaos guarantees the se- [11] E.N. Lorenz, J. Atmos. Sci. 20 (1963) 130.
curity of our scheme. Undoubtedly, the secret keys are [12] http://online.redwoods.cc.ca.us/instruct/darnold/laproj/Fall97/
Gabe/catmap.pdf.
secure enough even a chosen plaintext/ciphertext at- [13] G. Chen, T. Ueta, Int. J. Bifur. Chaos 9 (7) (1999) 1465.
tack is adopted.