Neurocomputing 169 (2015) 150–157

Contents lists available at ScienceDirect

Neurocomputing
journal homepage: www.elsevier.com/locate/neucom

A novel image encryption algorithm based on chaos and Line map

Guomin Zhou a, Daxing Zhang b,n, Yanjian Liu b, Ying Yuan a, Qiang Liu a
a
Department of Computer and Information Technology, Zhejiang Police College, Hangzhou 310053, China
b
Institute of Graphics and Image, Hangzhou Dianzi University, Hangzhou 310018, China

art ic l e i nf o a b s t r a c t

Article history: In the age of big data, border of data security has become increasingly blurred. Our privacy security is
Received 7 May 2014 facing a new round of test. The multimedia big data especially images often carry many secrets or
Received in revised form privacy information. How to ensure safety and the authorization of accessing to confidential data
12 November 2014
becomes a hot problem of urgency in image information processing and transmission. In this paper, we
Accepted 16 November 2014
propose a novel symmetrical image encryption algorithm based on skew tent map. Utilizing a new chaos
Available online 16 May 2015
based Line map, the proposed algorithm is suitable for encryption of any size of image. In order to
Keywords: disrupt the correlations between the R, G, B components of the true color image, these three
Image encryption components are encrypted at bit level and operated at the same time. The proposed algorithm realizes
Skew tent map
fast encryption and decryption of both gray-scale image and color image. In addition, the algorithm can
Line map
be implemented parallelly because there is no complex sub-block processing operation. Results of
various analyses and numerical simulations show that the new algorithm has high security and is
suitable for practical image encryption.
& 2015 Elsevier B.V. All rights reserved.

1. Introduction map [9,10], but these maps have a common flaw that the width and
height of the original image are required to be equal, or the image
With the rapid development of computer and communication cannot be completely permuted, which greatly limits their applica-
technologies, the use of multimedia applications has increased rapidly tions. Moreover, some algorithms for color image encrypted the R, G, B
in all aspects of the society due to their easy understanding and components independently using the same method without adequate
attractive presentation characteristics. Meanwhile, the security pro- consideration of the correlations between the three components.
blems of such applications have received much attention. However, These algorithms have to encrypt the image for three times and
traditional encryption algorithms, such as DES, AES and RSA, are found theyare more vulnerable to attack. In Ref. [15], Wang et al. proposed an
unsuitable for multimedia data because these algorithms are designed image encryption algorithm that uses chaotic system to encrypt the R,
for accurate data, while digital image has some intrinsic features such G, B components of a true color image at the same time and make
as bulk data capacity and high redundancy [1]. To achieve the security these three components affect each other based on chaos, but it
requirement, a variety of methods, such as Fibonacci [2,3], Hash [4], cannot be a resistant chosen plaintext attack since the basic encryption
DNA [5], Chaos [1,6–10], Transform domain [11,12] and S-box [13], operations are processed independently and it is key invertible [16]. In
have been proposed to be applied to image encryption in the past Ref. [17], Yong et al. proposed to encrypt image using an invertible 2-D
decade. Among those existing approaches, Chaos-based scheme, with map, namely Line map. The most significant advantage is that it gets
the desirable properties of pseudo-randomness, ergodicity, high rid of the restriction on the size of the image. Besides, Line map can be
sensitivity to initial conditions and parameters [14], is recognized as formulated as equations, which make it easy to implement. But the
a great potential for multimedia encryption because such features are Line map proposed by Yong contains complex arithmetic operations
considered analogous to those of an ideal cryptosystem. Since Fridrich which lead to low efficiency of encryption and decryption. In this
proposed the permutation–diffusion architecture of chaotic image paper, we design a new Line map and apply it to shuffle the pixels in
encryption in 1998 [1], this has been widely used for chaos-based our proposed image encryption algorithm. Our Line map is a chaos
image cryptosystem [1,6–10]. Most of those algorithms adopt some based map and it can be formulated as a simple equation without
chaotic maps to destroy the high correlation among pixels of the complex arithmetic operations. In addition, we splice the R, G, B
image, such as the cat map [6,7], the baker map [1,8] and the standard components of the color image horizontally into a large grayscale
image and take bit-level permutation instead of pixel-level permuta-
tion so that the correlations between these three components can be
n
Corresponding author. Tel.: þ 86 13857199971. fully disrupted. The proposed algorithm realizes fast encryption and
E-mail address: [email protected] (D. Zhang). decryption of both gray-scale image and color image. Numerical

http://dx.doi.org/10.1016/j.neucom.2014.11.095
0925-2312/& 2015 Elsevier B.V. All rights reserved.
 G. Zhou et al. / Neurocomputing 169 (2015) 150–157 151

simulation is performed to validate the performance of the proposed A typical trajectory xi of the dynamical system is obtained by
image encryption. The rest of the paper is organized as follows. In the iterating this map, i.e.,
next two sections, the conceptions of related major are introduced. i1
xi ¼ f ðxi  1 Þ ¼ f ðx0 Þ ð6Þ
Section 4 describes the details of the proposed image encry-
ption/decryption. Experimental results and analyses are reported in We use skew tent map to generate three chaotic sequences in our
Section 5. The last section is the conclusion. proposed algorithm.

2. Skew tent map 3. Line map principle

Chaos is a stochastic process in nonlinear dynamic systems and it Line map is a very simple transformation, and it can be
is widely used in the research of cryptography. Chaotic sequences can implemented using some software or hardware efficiently, which
be characterized by being periodic, divergent, bounded, and sensitive is very important for image encryption in practical applications. A
to initial parameters. Chaos based algorithms have shown excellent new Line map is proposed in this paper that refers to offset the
properties in aspects such as security and complexity. The commonly position of each element of a sequence in order to disrupt the
used chaotic maps include Logistic Map, Baker Map, Lorenz Map and order of the original sequence elements. In order to describe the
Super chaotic System. In this paper, the novel image encryption principle of the new Line map clearly and conveniently, assume
scheme is based on skew tent map. Skew tent map is an efficient that it contains n elements of two sequences, the original sequence
chaotic system and defined as follows: P ¼ fxi g and the disrupted sequence P ¼ fxi g, where xi and xi
( denote the values at position i of the sequences P and P , and P
x=p if x A ð0; p
f ðxÞ ¼ ð1Þ is the sequence P through one time Line map with the offset value
ð1  xÞ=ð1  pÞ if x A ðp; 1 of Kð1 o k r ⌈n=2⌉Þ. xi is calculated as follows:

Here p A ð0; 1Þ is the control parameter. The transformation is xi ¼ xi

continuous and piecewise linear, with the linear regions (0, p] and t ¼ ðði  1Þ%kÞ⌊N=Kc þ ⌈i=k⌉ þs
(p, 1]. In order to check the chaoticity of a map, the Lyapunov s ¼ minðði  1Þ%k; N%kÞ ð7Þ
exponent (LE) and the Invariant measure (IM) are usually considered
[18,19]. The LE of the map shows the divergence rate between nearby Here i ¼ 1; 2; …; N. The symbol % is the remainder operator. The
orbits. It is defined as operator ⌈ ⌉ and ⌊ c respectively stand for the supremum and
the infimum of a real number. Obviously, the map is a one-to-one
X 1  
N df ðxi Þ mapping, and xi A fx0 ; x1 ; x2 ; …; xn  1 g.
λ ¼ lim ln  ð2Þ
N-1
i¼0
dx  The proposed Line map is an invertible transformation and the
recovery scheme is just a reverse assignment using the same offset
IM determines the density of the map which further shows the
uniformity of the map and is defined as
X
N
ρðxÞ ¼ lim δ j x  f ðxi Þj ð3Þ
N-1
i¼0

Using Eq. (2), one can verify that the LE of skew tent map that is
λ ¼  plnðpÞ  ð1  pÞlnð1  pÞ 4 0 ð4Þ
Fig. 1 shows the LE curve for skew tent map with respect to the
control parameter p. For any control parameter p A ð0; 1Þ, the skew
tent map has a positive LE and thus is always chaotic. Similarly, using
Eq. (3), one can obtain the IM for the skew tent map is
ρðxÞ ¼ 1 ð5Þ
From Eq. (5), it is clear that IM is independent of initial guess. Hence,
the map uniformly covers the interval (0,1) and the system is ergodic.

Fig. 1. Lyapunov exponent curve for skew tent map with respect to the control
parameter p. Fig. 2. The flowchart of the encryption process of the proposed algorithm.
152 G. Zhou et al. / Neurocomputing 169 (2015) 150–157

value in accordance with Eq. (7). This invertible Line map will be 4. Proposed algorithm description
applied to shuffle the pixels in our proposed algorithm. For
convenience, all rows of a image for one time linear transform The proposed algorithm is a new symmetrical image encryp-
called one row Line map (RLM) operation. Similarly, a column Line tion scheme. The whole process of encryption is provided briefly
map (CLM) operation refers to all columns of a image for one time in the flowchart in Fig. 2. Suppose that a plain-image P with the
linear transform. All rows share the same value as the offset in size of M  N, the detail of the encryption process is as follows.
every RLM operation, and all of the columns share another offset
value in every CLM operation. 4.1. Key generation

The secret key of the proposed algorithm is a combination of

seven elements that can be expressed as ðx0 ; y0 ; t 0 ; px ; py ; pt Þ, where
ðpx ; x0 Þ, ðpy ; y0 Þ, ðpt ; t 0 Þ are three pairs of the control parameter and
initial value of the skew tent chaotic system, L is the encryption
rounds. The secret key will be used in both the image encryption
and decryption.

4.2. Pre-processing

4.2.1. Transform the plain-image P to a binary image P 0

In our proposed algorithm, we encrypt the plain-image bits by
bits, so we need to transform the plain-image to a binary image.
Assume that P is a true color image sized M  N pixels. Partition P
Fig. 3. Image segmentation method. (a) Odd iteration. (b) Even iteration. into three components Rp, Gp and Bp. Then combine the Rp, Gp, Bp

Fig. 4. Experimental results. (a) Plain-image of Lena; (b) cipher-image of Lena; (c) success of decryption of Lena; (d) failure of decryption of Lena; (e) plain-image of Couple;
(f) cipher-image of Couple; (g) success of decryption of Couple; (h) failure of decryption of Couple; (i) plain-image of Couple; (j) cipher-image of Couple; (k) success of
decryption of Couple; (l) failure of decryption of Couple. (For interpretation of the references to color in this figure, the reader is referred to the web version of this paper.)
 G. Zhou et al. / Neurocomputing 169 (2015) 150–157 153

components horizontally and get a matrix P 0 with M rows and 3N CLMs, and the binary sequence So will be used in the diffusion
columns. If P is a gray-scale image, P 0 is just the plain-image P. Let process
M 0  N 0 represent the size of image P 0 . We use the method (
r i ¼ ð½xi  108 ÞmodðW=4Þ þ 2
proposed in Ref. [17] to transform P 0 to a binary image PB. The
conversion function can be described as follows: ci ¼ ð½yi  108 ÞmodðH=4Þ þ 2
i ¼ 0; 1; 2; …; 3L  1 ð10Þ
P B ði; jÞ ¼ P 0 ði%N 0 ; j%M 0 ; ⌊i=Nc þ ⌊j=N0 c þ 1Þ
0 0 (
i ¼ 1; 2; …; N ; j ¼ 1; 2; …; M ð8Þ 1; tk 4 tk þ 1
ok ¼
The corresponding inverse formula of Eq. (8), that is 0; tk r tk þ 1
0 0 0 0 0
P ði%N ; j%M ; ⌊i=N c þ ⌊j=N c þ 1Þ ¼ P B ði; jÞ k ¼ 0; 1; 2; …; L  ðW þHÞ ð11Þ
0 0
i ¼ 1; 2; …; N ; j ¼ 1; 2; …; M ð9Þ
4.3. Permutation process
After conversion by use of Eq. (8), the size of binary image PB is
2M 0  4N 0 .
First, split the binary image P 0 into two subblocks of the same
size, denoted as A, B. Let l represent the lth iteration during the
4.2.2. Generate three sequences encryption process. The size of the subblock is W  ðH=2Þ for l is
Assume that the size of the binary image B is H  W . Set the odd, otherwise is ðW=2Þ  H. Fig. 3(a) and (b) shows the segmen-
initial parameter pt and initial value t0 to iterate Eq. (6) for tation methods. Then a RLM operation is applied to sub-block A
500 þ ðH þ WÞ  L times, and discard the former 500 values to and B with the offset value r 3l  2 and r 3l  1 , respectively. Similarly, a
avoid harmful effects. Then, set ðpx ; x0 Þ and ðpy ; y0 Þ, as the initial CLM operation is applied to the sub-block A and B with the offset
conditions, perform a similar operation but only 500 þ 3L times of value c3l  2 and c3l  1 , respectively. Finally, splice the sub-blocks A,
iteration respectively. Finally, we obtain three chaotic sequences, B into a binary image of W  H size. A RLM operation and a CLM
denoted as T ¼ ft k g, X ¼ fxi g and Y ¼ fyi g. Transform the chaotic operation are applied to the binary image after splicing again with
sequences X and Y to integer sequences using Eq. (10), and the offset value r 3l and c3l , respectively.
transform the chaotic sequence T to a binary sequence using a
threshold function defined as Eq. (11). Sr ¼ fr i g, Sc ¼ fci g, So ¼ fok g 4.4. Diffusion process
represent the transformed sequences, respectively. The two inte-
ger sequences Sr and Sc will be used as offset values in RLMs and The diffusion process is completed by performing XOR opera-
tions, which is composed of the row XOR and the column XOR.
Suppose the permuted binary image is P″. The pseudo-code of XOR
Table 1
Time cost for different rounds of encryption and decryption. operations is given below.

Images Operations Rounds (i) Code for row XOR.

for i ¼ 0 to H – 1
1 4 8 12 16 P ″ ði0Þ; ¼ P ″ ði0Þ  K lr ðiÞ
Lena (512x512) Encryption 3.161 s 7.099 s 12.262 s 17.439 s 22.527 s
for j ¼ 1 to w–1
Lena (512x512) Decryption 3.113 s 7.231 s 12.669 s 18.117 s 23.456 s P ″ ði; jÞ ¼ P ″ ði; jÞ  P ″ ði; j 1Þ
Couple (256x256) Encryption 0.205 s 0.429 s 0.730 s 1.042 s 1.352 s endfor
Couple (256x256) Decryption 0.222 s 0.491 s 0.851 s 1.211 s 1.516 s
endfor

Fig. 5. Histograms of the plain-image and the cipher-image. (a) R channel of plain-image Lena; (b) G channel of plain-image Lena; (c) B channel of plain-image Lena;
(d) Plain-image Couple; (e) R channel of cipher-image Lena; (f) G channel of cipher-image Lena; (g) B channel of cipher-image Lena; (h) Cipher-image Couple.
154 G. Zhou et al. / Neurocomputing 169 (2015) 150–157

(ii) Code for column XOR. Bc into a true color image, which is the final encrypted true color
Here K lr ¼ fok þ R g, K lr ¼ fok þ C g, R ¼ ðl  1ÞðW þ HÞ, C ¼ ðl  1ÞðW þ image. The decryption process is similar to encryption, but the
HÞ þH, k ¼ 0; 1; 2; …; H  1, where l represents the lth iteration inverse process of encryption.
during the encryption process. The symbol  represents XOR
operation.
The permutation and diffusion are iterated for L times to 5. Experimental results and analysis
encrypt the plain image. We will get a final encrypted binary
image, denoted as CB. Then transform CB to a gray-scale image CG The proposed algorithm has been implemented by the use of
by using the inverse formula (Eq. (9)). Matlab7.10 in a personal computer with Intel(R) i5-2400 Duo CPU
3.10 GHz, 3.24 GB RAM, while the operating system is Windows7.
for i ¼ 0 to W – 1
P ″ ð0; iÞ ¼ P ″ ð0; iÞ  K lc ðlÞ 5.1. Speed performance
for j ¼ 1 to H–1
P ″ ðj; iÞ ¼ P ″ ðj; iÞ  P ″ ðj  1; iÞ Choose a secret key (0.4877, 0.6462, 0.3586, 0.4287, 0.8517,
endfor 0.7997, 8), we encrypt the standard color image Lena (Fig. 4(a))
endfor and the gray-scale image Couple (Fig. 4(e)) with the proposed
algorithm. The cipher-images we get are shown in Fig. 4(b) and (f).
Then we decrypt the two cipher-images with the same key, the
In the case of encryption of gray-scale image, the gray-scale recovery operations are successful and the recovered images are
image CG is the final encrypted image C. Another case of encryp- shown in Fig. 4(c) and (g). There is an another group result of
tion of true color image can be described as follows: Partition the experiment shown in Fig. 4(i)–(l). In order to evaluate the
gray-scale image CG into three matrixes Rc, Gc and Bc from left to efficiency of the algorithm, we test the encryption and decryption
right, the size of each matrix is M  N. Then combine the Rc, Gc and of images for ten times for different rounds, and the average time
cost of encryption and decryption can be found in Table 1.
Table 2
Correlation coefficients of the plain image and the cipher-image. 5.2. Key space analysis

Channel Plain-image Cipher-image

The secret key of the proposed algorithm is a combination of
R
seven elements that can be expressed as ðx0 ; y0 ; t 0 ; px ; py ; pt ; LÞ, here
Horizontal 0.972585 0.036800 x0 A ð0; 1, y0 A ð0; 1, t 0 A ð0; 1, px A ð0; 1Þ, py A ð0; 1Þ, pt A ð0; 1Þ and L
Vertical 0.985908 0.027494 can be any integer greater than zero. Suppose the precision of a
Diagonal 0.971038 0.016954 floating-point number is 10  12, the key space is larger than 1072,
G
which is large enough for the cryptosystem to resist exhaustive
Horizontal 0.970740 0.035227
Vertical 0.985789  0.014485 attack. If the state of all initial values and parameters are
Diagonal 0.956975 0.054777 represented by the IEEE 754 double precision floating-point
B standard [20], the computational precision of the 64-bit double
 0.001062
Horizontal 0.952494
precision numbers is about 10  15, the key space is much larger
Vertical 0.967691 0.028694
Diagonal 0.920630 0.026081
than 1090. Therefore, a sufficiently large key space is guaranteed in
the proposed algorithm for application.

Fig. 6. Correlation analysis of image Lena in R channel. (a) plain-image in horizontal; (b) plain-image in vertical; (c) plain-image in diagonal; (d) cipher-image in horizontal;
(e) cipher-image in vertical; (f) cipher-image in diagonal.
 G. Zhou et al. / Neurocomputing 169 (2015) 150–157 155

5.3. Statistical analysis Table 2 summarizes the results corresponding to our proposed
algorithm by testing the true color image Lena.
According to Shannons theory [21], a secure cryptographic The result indicates that the correlation coefficients of the plain
scheme should be able to cover up the statistical properties of image are always nearly equals 1, while that of the ciphered image
the origin data. Usually, the histogram and the correlations among are greatly reduced to close to 0. The chaotic encryption algorithm
adjacent pixels are two common statistical features of an image. satisfies zero correlation, and the encryption effect is rather good.
Therefore, we analyze the statistical properties from the following To show this feature graphically, the correlation distributions of R
two aspects. channel of the plain-image and cipher-image are plotted in Fig. 6.
It is clear that the dots are located along the diagonal (Fig. 6(a)–(c))
5.3.1. Histograms of encrypted images indicating high correlation of adjacent plain-image pixels in the
The histogram of an image illustrates how pixels in an image three directions, while those of the cipher-image are scattered
are distributed by graphing the number of pixels at intensity level over the entire plane (Fig. 6(d)–(f)).
and it leads to the leakage of image content. In Fig. 5, (a)–(c) are
the histograms of plain-image Lena of R, G, and B channels, (d) is
the histogram of plain-image Couple, and (e)–(h) are histograms 5.4. Information entropy
corresponding to the encrypted image respectively. It is clear that
the histograms of encrypted image are fairly uniform and sig- Entropy is one of the most important features of randomness.
nificantly different from that of the original image and it does not Information entropy is a mathematical theory for data commu-
provide any useful clue for an attacker to launch any statistical nication and storage that was introduced in 1949 by Shannon [21].
attacks on the proposed image encryption procedure. Let m be the information source, and entropy H(m) of the source m
is defined as follows:
5.3.2. Correlation of adjacent pixels
2X
N
1
The adjacent pixels in plain image are usually highly correlated, HðmÞ ¼  Pðmi Þ  log 2 ðPðmi ÞÞ ð14Þ
which is a weakness to statistical attack. An image encryption i¼0
should decrease the correlation of two adjacent pixels in the
ciphered image. To test the correlation between two vertically Here Pðmi Þ represents the probability of symbol mi. Assume that there
adjacent pixels, two horizontally adjacent pixels, and two diag- are 28 states of the information source and they appear with the same
onally adjacent pixels, respectively, in a ciphered image, the probability. According to the formula (16), we can get the ideal
following procedure was carried out. We randomly selected HðmÞ ¼ 8, which shows that the information is completely random.
2000 pairs of two adjacent (in horizontal, vertical, and diagonal The information entropy of the cipher-image should be close to 8 after
direction) pixels from plain-image, ciphered image, and calculated encryption. The closer to 8 that the value obtained for the entropy is,
their correlation coefficients by using the following two formulas: the smaller the possibility of the method being anticipated and hence
pffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffi the more secure the method. Table 3 shows the values obtained for
r xy ¼ covðx; yÞ=ð DðxÞ DðyÞÞ ð12Þ the entropy of the image Lena for different rounds of encryption by
use of the proposed algorithm. Experiments show that the informa-
1 XN
tion entropy of the cipher-image is much larger than that of the plain-
covðx; yÞ ¼ ðx  EðxÞÞðyi  EðyÞÞ ð13Þ
N i¼1 i image, and they are very close to 8.
Here D(x), E(x) respectively stand for the variance and expectation
of x:
5.5. Sensitivity analysis
1XN
1XN
DðxÞ ¼ ðx  EðxÞÞ2 ; EðxÞ ¼ x:
Ni¼1 i Ni¼1 i An ideal cryptosystem should be of high sensitivity, including key
sensitivity and plain-text sensitivity. Key sensitivity means that a tiny
change of the secret key should produce a completely different
Table 3
encrypted image or the decryption will fail even via using a slight
Entropy of image Lena for different rounds of encryption.
change key. Plain-text sensitive means that a tiny disturbance of the
Channel Plain-image Rounds plain-image will lead to dramatic changes in cipher-image. In order to
measure the difference between the resulting cipher-images, we use
1 2 4 6 8 two common measurements number of pixels change rate (NPCR) and
R 7.260888 7.999338 7.999347 7.999313 7.999367 7.999298
unified average changing intensity (UACI) defined by
G 7.598154 7.999272 7.999274 7.999297 7.999381 7.999322 P
i;j Dði; jÞ
B 6.977446 7.999367 7.999251 7.999263 7.999305 7.999312 NPCR ¼  100% ð15Þ
W H

Table 4
Difference when encryption key is changed slightly.

Item Channel Condition

1 2 4 5 6 7 8

R 99.6113 99.5800 99.6239 99.6147 9.5953 99.6174 99.5941

NPCR (%) G 99.5941 99.6025 99.6212 99.6166 99.6056 99.6223 99.6281
B 99.6120 99.5850 99.6265 99.6136 99.5850 99.6208 99.6190

R 33.3430 33.4773 33.5347 33.3505 33.4220 33.3973 33.4113

UACI (%) G 33.4488 33.3048 33.5361 33.4631 33.4960 33.5102 33.5345
B 33.4747 33.3317 33.4730 33.4821 33.4532 33.5067 33.4157
156 G. Zhou et al. / Neurocomputing 169 (2015) 150–157

Table 5
NPCR and UACI performance.

Item Channel Rounds

1 2 4 6 8

R 83.0078 66.7419 98.6710 99.5983 99.5949

NPCR (%) G 83.0078 66.4672 98.4398 99.6006 99.5845
B 83.0078 66.1926 98.4978 99.5888 99.6056

R 0.3255 15.7948 31.6366 33.4279 33.4187

UACI (%) G 0.3788 15.9258 31.5039 33.4791 33.4232
B 0.6513 15.9321 31.4032 33.4810 33.4768

Fig. 7. Plain-text sensitive analysis. (a) NPCR versus ciphering rounds. (b) UACI versus ciphering rounds.

2 3
6. Conclusions
1 4X j C 1 ði; jÞ  C 2 ði; jÞj 5
UACI ¼  100% ð16Þ
W  H i;j 255
The symmetrical image encryption algorithm proposed in this
paper get rid of the limitation of the width be equal to the height
Here C1 and C2 are two cipher-images with th size of ðWHÞ. C 1 ði; jÞ
of the image using a new designed Line map. The algorithm is
denotes the gray-scale values of the pixels at position (i,j) of the
suitable for encryption of both gray-scale image and color image.
cipher-image C1. The difference array Dði; jÞ is determined by the
In addition, the algorithm can be implemented parallelly to
following rule: If C 1 ði; jÞ aC 2 ði; jÞ, then Dði; jÞ ¼ 1, otherwise it is 0.
improve speed performance. Security analysis shows great robust-
ness against related attacks such as chosen plaintext attack due to
5.5.1. Key sensitivity large key space, uniform pixels distribution and high input
In order to evaluate the key sensitivity of the proposed algorithm, sensitivity. More rounds of encryption are strongly recommended
the following key sensitivity tests have been performed: px, py, pt, x0, when there are much higher requirements for security.
y0, t0, a Δ is added, which equals to 10  12, and L is changed from 8 to
9. We test the color image Lena under these seven different conditions,
and the results are listed in Table 4. It is clear from Table 1 that a tiny Acknowledgements
difference in the key causes substantial changes in the corresponding
cipher-images. In addition, we also try to decrypt the cipher-images The National Natural Science Foundation of China (Grant no.
(Fig. 4(b) and (f)) by use of a tiny change key under the condition (1). 61272391), Open Fundation of Jiangsu Engineering Center of
The decrypted images of (d) and (h) in Fig. 4 obviously shows a fail Network Monitoring, Nanjing University of Information Science
decryption. Therefore, the proposed algorithm possesses high key and Technology (KJR1403) supports this research.
sensitivity.
References

5.5.2. Plain-text sensitivity [1] Jiri Fridrich, Symmetric ciphers based on two-dimensional chaotic maps, Int. j.
In our simulations, two plain images are used. One is the original Bifurc. Chaos 8 (06) (1998) 1259–1284.
plain image Lena, and the other is obtained by changing the image [2] Yicong Zhou, Sos Agaian, Valencia M Joyner, Karen Panetta, Two fibonacci p-
code based image scrambling algorithms, in: Electronic Imaging 2008, Inter-
Lena of upper-left pixel value from 10000001 to 10000000 in B national Society for Optics and Photonics, San Jose, CA, 2008, p. 681215.
channel. These two test images are encrypted by using the same key [3] Yicong Zhou, Karen Panetta, Sos Agaian, C.L. Chen, Image encryption using p-
for a few rounds to generate the corresponding cipher images C1 and fibonacci transform and decomposition, Opt. Commun. 285 (5) (2012)
594–608.
C2. The test results are listed in Table 5. It shows that the NPCR and
[4] Abbas Cheddad, Joan Condell, Kevin Curran, Paul McKevitt, A hash-based
UACI performance of the proposed scheme can reach 99.6% and image encryption algorithm, Opt. Commun. 283 (6) (2010) 879–893.
33.4% in the sixth round, respectively. It is clear that the proposed [5] Xingyuan Wang Hongjun Liu, et al., Image encryption using dna complemen-
scheme has a good ability to antidifferential attack. To show these tary rule and chaotic maps, Appl. Soft Comput. 12 (5) (2012) 1457–1466.
[6] Guanrong Chen, Yaobin Mao, K. Charles Chui, A symmetric image encryption
growing trends graphically, the NPCR and UACI data are plotted in scheme based on 3d chaotic cat maps, Chaos Solitons Fractals 21 (3) (2004)
Fig. 7(a) and (b), respectively. 749–761.
 G. Zhou et al. / Neurocomputing 169 (2015) 150–157 157

[7] Chong Fu, Bin-bin Lin, Yu Sheng Miao, Xiao Liu, Jun-Jie Chen, A novel chaos- Daxing Zhang received the BS degree from Dalian
based bit-level permutation scheme for digital image encryption, Opt. Com- University of Technology, China, in 1993, MS degree
mun. 284 (23) (2011) 5415–5423. from Hangzhou Dianzi University, China, in 1998, the
[8] Yaobin Mao, Guanrong Chen, Shiguo Lian, A novel fast image encryption Ph.D. degree from Zhejiang University, China, in 2011.
scheme based on 3d chaotic baker maps, Int. J. Bifurc. Chaos 14 (10) (2004) He is currently an associate professor in the College of
3613–3624. Computer School, Hangzhou Dian University, China. His
[9] Shiguo Lian, Jinsheng Sun, Zhiquan Wang, A block cipher based on a suitable research interests include multimedia security, digital
use of the chaotic standard map, Chaos Solitons Fractals 26 (1) (2005) watermarking, cryptographic algorithm.
117–129.
[10] Kwok-Wo Wong, Bernie Sin-Hung Kwok, Wing-Shing Law, A fast image
encryption scheme based on chaotic standard map, Phys. Lett. A 372 (15)
(2008) 2645–2652.
[11] Jun Lang, Image encryption based on the reality-preserving multiple-para-
meter fractional fourier transform, Opt. Commun. 285 (10) (2012) 2584–2590.
[12] Nanrun Zhou, Yixian Wang, Lihua Gong, Novel optical image encryption
scheme based on fractional Mellin transform, Opt. Commun. 284 (13) (2011)
3234–3242.
Yanjian Liu is a Master Degree Candidate in Hangzhou Dianzi University, China.
[13] Iqtadar Hussain, Tariq Shah, Muhammad Asif Gondal, An efficient image
encryption algorithm based on s8 s-box transformation and nca map, Opt.
Commun. 285 (24) (2012) 4887–4890.
[14] Zbigniew Kotulski, Janusz Szczepański, Discrete chaotic cryptography, Ann.
Phys. 509 (5) (1997) 381–394. Ying Yuan received the BS degree from Zhejiang
[15] Xingyuan Wang, Lin Teng, Xue Qin, A novel color image encryption algorithm University, China, in 2008, the Ph.D. degree from
based on chaos, Signal Process. 92 (4) (2012) 1101–1108. Zhejiang University, China, in 2013. She is currently a
[16] Chengqing Li, LeoYu Zhang, Rong Ou, Kwok-Wo Wong, Shi Shu, Breaking a lecturer in Zhejiang Police College, China. Her research
novel color image encryption algorithm based on chaos, Nonlinear Dyn. 70 (4) interests include multi-label image annotation, cross-
(2012) 2383–2388. media retrieval, and machine learning.
[17] Yong Feng, Xinghuo Yu, A novel symmetric image encryption approach based
on an invertible two-dimensional map, in: 35th Annual Conference of IEEE
Industrial Electronics, 2009. IECON'09, IEEE, Porto, Portugal, 2009, pp. 1973–
1978.
[18] Gonzalo Alvarez, Shujun Li, Some basic cryptographic requirements for chaos-
based cryptosystems, Int. J. Bifurc. Chaos 16 (08) (2006) 2129–2151.
[19] G. Chen, Y. Mao, C.K. Chui, A symmetric image encryption scheme based on 3D
chaotic cat maps, Chaos, Solitons & Fractals 21 (2004) 749–761.
[20] William Kahan, for Binary Floating-point Arithmetic, Lecture Notes on the
Status of IEEE Standard, vol. 754, 94720-1776, 1996, p. 11.
[21] Claude E Shannon, Communication theory of secrecy systems*, Bell Syst. Tech. Qiang Liu received the BS degree from Harbin Engi-
J. 28 (4) (1949) 656–715. neering University, China, in 2004, the MS degree from
Harbin Engineering University, China, in 2007, and Ph.
D. degree from Southeast University, China, in 2014. He
is currently a lecturer in the Department of Computer
and Information Technology, Zhejiang Police College,
Guomin Zhou received the BS degree from Xidian China. His research interests include police information
University, China, in 1993, the MS degree from Hang- technology, inertial navigation and integrated
zhou Dianzi University, China, in 1996. He is currently a navigation.
deputy director in the Key Laboratory of Public Security
Information Application Based on Big-data Architec-
ture, Ministry of Public Security, People's Republic of
China. His research interests include big-data based
policing information application, image encryption and
information security.