See discussions, stats, and author profiles for this publication at: [Link]

net/publication/306067982

A novel sensitive image encryption algorithm based on the Zaslavsky chaotic

map

Article in Information Security Journal A Global Perspective · August 2016

DOI: 10.1080/19393555.2016.1212954

CITATIONS READS
65 2,929

2 authors:

Rafik Hamza Faiza Titouna

Tokyo International University university of algeria
23 PUBLICATIONS 695 CITATIONS 17 PUBLICATIONS 250 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Security of digital images View project

Computer Vision and Image Analysis, Understanding and Processing View project

All content following this page was uploaded by Rafik Hamza on 06 January 2018.

The user has requested enhancement of the downloaded file.

INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE
[Link]

A novel sensitive image encryption algorithm based on the Zaslavsky chaotic

map
Rafik Hamza and Faiza Titouna
Department of Computer Science, University of Batna 2, Batna, Algeria

ABSTRACT KEYWORDS
In this article, a novel sensitive encryption scheme to secure the digital images based on the Image; encryption;
Zaslavsky chaotic map is proposed. We employ the Zaslavsky chaotic map as a pseudo-random permutation-substitution
network; pseudo-random;
generator to produce the key encryption of the proposed image cryptosystem. The cipher structure security digital images;
has been chosen based on permutation-diffusion processes, where we adopt the classic permuta- Zaslavsky chaotic map
tion substitution network, which ensures both confusion and diffusion properties for the encrypted
image. Our proposed algorithm has high sensitivity in plain image and the secret key. Moreover, the
results show that the characteristics of our approach have excellent performance, with high scores
(NPRC = 99.61%, UACI = 33.47%, entropy (CipherImage) ’ 8, and correlation coefficient ’ 0).
Experimental results have been studied and analyzed in detail with various types of security
analysis. These results demonstrate that our proposed cryptosystem has highly satisfactory security
performance and can withstand various attacks compared to state-of-the-art methods.

1. Introduction expresses the idea of replacing the original image

with a nonvisual image based on some processing
1.1. Background
steps. On the other hand, the decryption restores the
Security issues with multimedia data confidentiality original image from the encrypted one. According
are growing rapidly due to the development in use to Kerckhoffs’ principle (Wang et al., 2012;
of the Internet, where information is transmitted Muhammad et al., 2015a), the security should
across public networks. Cryptography has been depend only on the secrecy of the private key such
used as a useful technique to ensure the security of that reestablishment of the original image without
this digital information, while cryptanalysis tries to the secret key should be near to impossible. These
find the weaknesses in cryptography schemes, which keys of encryption algorithms are divided into two
allow unauthorized users to access or obtain specific parts (Stinson, 2005). A secret key (in the case of a
numerical data. The security of digital images has symmetric key cryptosystem) or a public and private
become more important. Accordingly, many crypto- key pair (in the case of a asymmetric cryptosystem).
graphic methods have been described to ensure the During the asymmetric algorithm, the public key is
security of digital images information (Muhammad, used for encryption, and the private key is used for
Ahmad, Rehman, Jan, & Sajjad, 2016a; Wang, Teng, decryption, while symmetric encryption uses the
& Qin, 2012; Wang, Chen, & Wang, 2010; Zhang & same key (private key) during encryption and
Wang, 2015). Image encryption plays a vital role in decryption.
the security of digital images and is considered one
common method to protect the image information
1.2. A brief investigation
(Wang et al., 2012; Wang et al., 2010; Zhang et al.,
2014). Image encryption can be defined as the From the last decade, many encryption algorithms
process of transforming the plain image into an have been presented to improve the
incomprehensible form. This transformation performance of ciphering the information data.

CONTACT Rafik Hamza [Link]@[Link] Department of Computer Science, University of Batna 2, Constantine Road N°53. Fesdis,
Batna 05078, Algeria.
Color versions of one or more of the figures in the article can be found online at [Link]/uiss.
© 2016 Taylor & Francis
2 R. HAMZA AND F. TITOUNA

The most common encryption schemes, such as problem of the compute finite precision, which
the Data Encryption Standard (DES) and can degrade the dynamical behavior.
Advanced Encryption Standard (AES) (Pub & Furthermore, employing lower chaotic maps such
NIST FIPS, 2001), are designed with good confu- as logistic maps is insecure and cannot resist
sion and diffusion properties, but also are not attacks such as brute-force ones, and the attackers
really appropriate for digital data (Chen, Mao, & are able to beat almost all the one-dimensional
Chui, 2004; Hermassi, Belazi, Rhouma, & Belghith, chaotic maps (Short, 1994). There are some
2014; Norouzi, Mirzakuchaki, Seyedzadeh, & chaos-based algorithms that are interesting and
Mosavi, 2014a). Digital images have various inher- can provide additional information about many
ent properties such as high correlation between related studies including our work in this article.
adjacent pixels, which mean that the security of For example, (Hussain, Shah, Gondal, &
digital images still has problems. Mahmood, 2013) presented a novel image encryp-
The permutation-substitution network (PS) is a tion algorithm based on chaotic maps and GFð28 Þ
model formed for the most common symmetric key exponent transformation. The Arnold cat map was
cryptosystems over the last decades. A PS is a special also used to shuffle iteratively and transform the
type of iterated block cipher. This model has several pixels of image values into GFð28 Þ, and improve
attractive features, such as simplicity of design as well the performance in the proposed algorithm
as efficiency in hardware and software (Stinson, (Hussain et al., 2013). Khan and Shah (2015) pre-
2005). The permutation box (P-box) and the substi- sented an efficient S-box using a chaotic system to
tution box (S-box) are basically simple nonlinear synthesize an S-box showing improved crypto-
structurings for the symmetric key algorithms and graphic properties; the substitution boxes are gen-
are utilized to provide confusion and diffusion erated using the Zaslavskii chaotic map. Hussain
(Wong, Kwok, & Yuen, 2009). Indeed, these boxes and Gondal (2014) presented a modified image
are used widely in almost all references of crypto- encryption algorithm (Tang, Wang, & Fang, 2010)
graphic schemes such as DES and AES. based on coupled map lattices and using substitution
In the last decade, many chaos-based image box transformation to archive the strength encryp-
encryption schemes have been presented (Wang, tion. Zhang and Zhao (2014) showed another algo-
et al., 2010; Wang & Luan, 2013; Wang &
rithm encryption for digital images with permutation-
Yu, 2009). Some of the proposed algorithms
diffusion architecture, using a coupled map lattice
are based on one-time keys (Liu & Wang, 2010),
(CML) composed of two one-dimensional chaotic
bit-level scrambling (Wang et al., 2012), the DNA
maps to generate the random numbers PRNG.
complementary rule (Liu, Wang, & Kadir, 2012;
Moreover, Huang (2012) presented a novel encryp-
Wei, Guo, Zhang, Zhang, & Lian, 2012), and hybrid
tion algorithm for digital Images using a chaotic
and high-dimension chaotic maps (Liu & Wang,
Chebyshev function; the two-dimensional of the
2011; Wang, Liu, & Zhang, 2015). Bit-level permuta-
Chebyshev function is presented to avoid attacks on
tion is a very effective encryption method, which can
chosen known plaintext and comes with good
change the position and value of a pixel simulta-
features.
neously. Chaotic-based algorithms consist of two
steps, shuffling and diffusion, while DNA encod- However, many images encryption schemes have
ing–based methods have two phases known as been analyzed and proved insecure against hackers.
encoding and decoding phases. One commonality Basically, the hackers used many types of attacks
among these techniques is using some characteristics aiming to get sensitive information. One common
of chaotic systems, which have many excellent attack is chosen/known plain-image attack using spe-
intrinsic properties such as high sensitivity to initial cial images such as a black image (all pixels equal to
conditions and controlling parameter, being pseudo- zero) (Akhavan, Samsudin, & Akhshani, 2015). The
random, etc. Because of these advantages of chaotic attackers build their cryptanalysis on the weakness of
systems, a variety of image encryption algorithms low sensitivity in plain-image change, along with
have been presented. However, these chaos-based other reasons such as the bad ciphering scheme
algorithms have some problems, such as the regardless of the generator keys or the pseudo-key
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 3

itself. Wang et al. (2014) presented a security analysis (largest Lyapunov exponent 3.65), producing very
with totally break for the work in Huang (2012), based chaotic sequences, and can lead us to increase the
on the chosen-plain-text attack. The fractured resistance of our cryptosystem against attacks. In
algorithm presented has low sensitivity to any tiny order to get high sensitivity to the plain image, we
changing of the plain-image. Moreover, it proved the use matrix multiplication over the finite field Gf ð28 Þ,
existence of equivalent keys of the cryptosystem. where we use the matrix K with its invertible L. The
Rhouma and Belghith (2008) presented a cryptanaly- square matrix K is self-invertible, which means that
sis and an improvement of a new image encryption the equation K:L ¼ Id has a unique solution L, where
algorithm based on hyper-chaos (Gao & Chen, 2008). Id is the identity matrix. And the invertible matrix L
Furthermore, Jeng, Huang, and Chen (2015) found can be found by L ¼ K 1 .
that the modified image cryptosystem of Rhouma The rest of this article is organized as follows. The
and Belghith (2008) still exhibits a security weakness, proposed image encryption scheme is explained in
especially the problem of low security-sensitivity the next section. Then, experimental results and
to plain-image change. The main issue that allows analysis are given. Finally, we present the conclusion
cryptanalysis to break many encryption schemes is of the article, followed by future directions.
the problem of low security-sensitivity to plain-
image change (Akhavan et al., 2015). Therefore, any
adjustment in plain-image pixels should change com- 2. The proposed encryption scheme
pletely the corresponding ciphered image.
2.1. Zaslavsky chaotic map

1.3. Contribution and organization of this article In this subsection, a brief introduction of the
Zaslavsky chaotic system is presented along with
In this article, we introduce a novel image encryption its dynamical behavior. This two-dimensional
scheme based on the Zaslavsky chaotic map. In parti- (2-D) chaotic map was introduced by George
cular, the proposed algorithm has a high level of M. Zaslavsky (1978). It is a discrete-time dynami-
security and high sensitivity, which ensures the good cal system that is highly sensitive to its initial
property of confusion and eliminates the correlation values. This sensitiveness with its large extent
coefficients of the original image. It also possesses a makes this system very useful for many crypto-
large-sized keyspace, and it is very sensitive to the graphic applications as well as in other applica-
plain image and the secret key. Moreover, the pro- tions that require pseudo-randomness. The 2-D
posed algorithm has good speed of encryption/ Zaslavsky map can produce random real numbers
decryption. The original image can be recovered com- and exhibits chaotic behavior. We set this map as a
pletely if the secret keys are known exactly. These pseudo-random number generator that produces
features make our proposed algorithm secure against keys for our proposed encryption algorithm. The
many types of attack such as the brute-force attack. pseudo-random numbers can be constructed by an
Conducted experiments show that the proposed cryp- iterative procedure.
tosystem approach can achieve excellent encryption The 2-D map is defined as coupled Eqs (1):
performance, and confirms its efficiency against cryp- 8
< xnþ1 ¼ xn þ νð1 þ μyn Þ þ ενμ½cosð2πxn Þ mod 1
tographic attacks. Our work also indicates that the
ynþ1 ¼ eτ ½ yn þ ε cosð2πxn Þ
two-dimensional map of the Zaslavsky chaotic system : τ
μ ¼ 1eτ
is suitable for image encryption, due to the wide range
of the initial conditions and control parameters and (1)
high sensitivity. We avoided the problem of degrada- where xn , yn are the chaotic samples of this
tion caused by finite-precision computation with the map. x0 , and y0 are the initial values. ε, τ, and
proposed chaotic system, where we employ directly ν are the controlling parameters to oversight
the generated sequence from the coupled map of the the chaotic behavior proposed, and e is the
chaotic map. Furthermore, this map is very chaotic exponentiation.
4 R. HAMZA AND F. TITOUNA

0.9 0.1

0.8

0.7 0.05

0.6

0.5 0

0.4

0.3 -0.05

0.2

0.1 -0.1
0
0 200 400 600 800 1000 1200 1400 1600 1800 2000 0 200 400 600 800 1000 1200 1400 1600 1800 2000

(a) (b)
Dynamics of Lyapunov Exponents
0.15 4
LE1 = 3.6524

0.1 2

Lyapunov Exponents
0.05 0

0 -2

-0.05 -4

-0.1 -6 LE2 = -6.6524

-0.15 -8
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0 100 200 300 400 500 600 700 800 900 1000
i

(c) (d)

Figure 1. (a) Plot of x. (b) Plot of y. (c) Zaslavsky attractor. (d) Dynamics of Lyapunov exponents.

Figure 1c shows the Zaslavsky chaotic attractor. The chaotic map of Zaslavsky can generate ran-
The plot of data x and y of the 2-D map versus the dom numbers given the initial values ðx0 ;y0 ;ε;τ;νÞ.
index of each value are shown in Figure 1a and Therefore, these values will be set as secret keys in
Figure 1b, using the initial values in Eq. (1), where our proposed encryption algorithm.
x0 ¼ 0:14, y0 ¼ 0:15, ν ¼ 4, ε ¼ 2:3, and τ ¼ 3. The
sequence generated directly from this chaotic map has
good uniform distribution and exhibits chaotic
2.2. Generating encryption keys
behavior.
Figure 1d shows the dynamics of Lyapunov To produce appropriate keys for our proposed
exponents of the Zaslavsky chaotic map. The scheme. We iterate Eq. (2) for k times to produce
Lyapunov exponents of system 1 are found to the one-dimensional (1-D) vector Vec, which
be: λ1 ¼ 3:65 and λ2 ¼ 6:65. The Zaslavskii means that the produced vector Vec has 2  k
map is very chaotic (the largest Lyapunov expo- length, while k ¼ ceilðn=2Þ and n is length of the
nent is found to be 3.65). Moreover, the generated sequence Vec. The pseudo-random
Lyapunov spectrum of this chaotic system is number generator is defined by
computed to be around 1.55, which is positive 8
and large. In addition, the sum of all positive < Vec 2k ¼ xkþ1
Vec 2kþ1 ¼ ykþ1 (2)
Lypunov exponents of the chaotic map gives an :
estimate of the Kolmogorov-Sinai (KS) entropy k ¼ 0; 1; 2; :::
(Pesin, 1977). Therefore, the KS entropy of the The proposed image encryption image uses the
Zaslavsky map is positive and large. Hence, this pseudo-random of the Zaslavsky chaotic map by mix-
map is suitable for image encryption and can be ing and cascading its samples. That ensure a highly
very useful for cryptography applications. chaotic result for the pseudo-random number
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 5

generator. To get rid of initial-values effect, we discard ⎡

⎢
0
⎢ 0
0
1
1
0
0
2
2
0
0
3
3
0
0
4
4
0
0
5
5
0
0
6
6
0
0
7
7
0
0
8
8
0
0
9
9 0 10 0 11 0 12 0 13 0 14 0 15 0
⎤
⎥
0 10 0 11 0 12 0 13 0 14 0 15 0 16 ⎥
⎢ ⎥
⎢ ⎥
the first ten numbers of the generated sequence Vec. ⎢ 1
⎢
⎢
⎢ 0
0
2
2
0
0
3
3
0
0
4
4
0
0
5
5
0
0
6
6
0
0
7
7
0
0
8
8
0
0
9
9
0 10
0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 ⎥⎥
⎥
0 11 0 12 0 13 0 14 0 15 0 16 0 17 ⎥
⎢ ⎥
⎢ ⎥
Algorithm 1 illustrates the steps to compute P
⎢ 2
⎢
⎢
⎢ 0
⎢
0
3
3
0
0
4
4
0
0
5
5
0
0
6
6
0
0
7
7
0
0
8
8
0
0
9
9 0 10 0
0 10 0 11
11 0 12 0 13 0 14 0 15 0 16 0 17 0 ⎥
⎥
⎥
0 12 0 13 0 14 0 15 0 16 0 17 0 18 ⎥
⎥

ð:Þ
⎢ ⎥
⎢ 3 0 4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 ⎥
keys for our image encryption method. ⎢
⎢ 0
⎢
⎢
4 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12
⎥
0 13 0 14 0 15 0 16 0 17 0 18 0 19 ⎥
⎥
⎥
⎢ 4 ⎥
refers to the sum of all numbers, and bc means ⎢ 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 ⎥
⎢ ⎥
⎢ 0 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 ⎥
⎢ ⎥
⎢ ⎥
⎢ 5 0 6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 ⎥
⎢ ⎥
round to the nearest number. ⎢
⎢ 0
⎢
⎢
⎢ 6
6 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14
⎥
0 15 0 16 0 17 0 18 0 19 0 20 0 21 ⎥
⎥
⎥
⎢ 0 7 0 8 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 ⎥⎥
The chaotic map generates the matrix Rinit , and ⎢
⎢ 0
⎢
⎢
⎢ 7
7
0
0
8
8
0
0
9
9 0 10 0 11 0 12 0 13 0 14 0 15
0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 ⎥
⎥
0 16 0 17 0 18 0 19 0 20 0 21 0 22 ⎥
⎥
⎥
0 ⎢
⎢
⎥
⎥
the vectors Vinit , Vinit using the secrets keys ⎢ 0
⎢
⎢
⎢ 8
8
0
0
9
9 0 10 0 11 0 12 0 13 0 14 0 15 0 16
0 10 0 11 0 12 0 13 0 14 0 15 0 16 0
0 17 0 18 0 19 0 20 0 21 0 22 0 23 ⎥
17 0 18 0 19 0 20 0 21 0 22 0 23 0 ⎥
⎥
⎥
⎢ ⎥

ðx0 ; y0 ; ε; τ; νÞ. Then, we sort these generated keys

⎢ ⎥
⎢ 0 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24⎥
⎢ ⎥
⎢ ⎥
⎢ 9 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 ⎥
⎢ ⎥
⎢ 0 10 0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 ⎥
⎢ ⎥
in ascending order and return the index array of ⎢
⎢ 10
⎢
⎢
0 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 ⎥
⎥
⎥
⎥
⎢ 0 ⎥
each key, Followed by computing α, K, and L. ⎢ 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 ⎥
⎢ ⎥
⎢ 11 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 ⎥
⎢ ⎥
⎢ ⎥
⎢ 0 12 0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 27 ⎥
⎢ ⎥
The generated keys from Algorithm 1 are ⎢
⎢
⎢ 12
⎢
⎢ 0
0 13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0
13 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21
21 0 22 0 23 0 24 0 25 0 26 0 27 0 ⎥
0 22 0 23 0 24 0 25 0 26 0 27 0 28 ⎥
⎥
⎥
⎥
⎢ ⎥
⎢ ⎥
0 ⎢ 13 22 0 23 0 24 0 25 0 26 0 27 0 28 0 ⎥
● The random matrix Rinit , and vectors Vinit , Vinit . ⎢ 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 ⎥
⎢ ⎥
⎢ 0 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 27 0 28 0 29 ⎥
⎢ ⎥
⎢ ⎥
⎢ 14 0 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 27 0 28 0 29 0 ⎥
● The index matrix R, and indexes vectors ⎢
⎢
⎢ 0
⎢
⎢
15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 27 0 28 0 29 0 30 ⎥
24 0 25 0 26 0 27 0 28 0 29 0 30 0 ⎥
⎥
⎥
⎥
⎣ 15 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 ⎦
0
V; V 0 . 0 16 0 17 0 18 0 19 0 20 0 21 0 22 0 23 0 24 0 25 0 26 0 27 0 28 0 29 0 30 0 0

● α is a non-null value, computed using the Figure 2. Initial matrix Kinit .

0
random sequences Vinit , Vinit and the random
matrix Rinit as shown in Algorithm 1.
● The matrixes K and L are defined over
Table 1. Primitive polynomials of degree 8.
Gf ð28 Þ. Primitive polynomials CN
These generated keys denote encryption keys, 1 x8 þ x 4 þ x 3 þ x 2 þ 1 285
2 x 8 þ x 5 þ x3 þ x þ 1 299
while the initial keys of the 2-D chaotic map are 3 x8 þ x5 þ x3 þ x2 þ 1 301
the secret keys for our proposed scheme. 4 x 8 þ x 6 þ x3 þ x2 þ 1 333
The matrix R contains indexes that sort the ele- 5 x 8 þ x 6 þ x4 þ x3 þ x2 þ x þ 1 351
6 x 8 þ x 6 þ x5 þ x þ 1 355
ments in each column of the matrix generated Rinit , 7 x8 þ x6 þ x5 þ x2 þ 1 357
which means that R is a collection of column index 8 x 8 þ x 6 þ x5 þ x3 þ 1 361
vectors obtained by sorting the generated matrix Rinit . 9 x 8 þ x 6 þ x5 þ x4 þ 1 369
10 x 8 þ x 7 þ x2 þ x þ 1 391
V and V 0 are two vectors representing the index sorts 11 x8 þ x7 þ x3 þ x2 þ 1 397
0 0
of Vinit and Vinit . The two initial vectors (Vinit , Vinit ) 12 x 8 þ x 7 þ x5 þ x3 þ 1 425
are related to the height (h) and width (w) of the plain 13 x 8 þ x 7 þ x6 þ x þ 1 451
14 x8 þ x7 þ x6 þ x3 þ x2 þ x þ 1 463
image I. 15 x 8 þ x 7 þ x6 þ x5 þ x2 þ x þ 1 487
Based on a large number of experiments to find 16 x 8 þ x 7 þ x6 þ x5 þ x4 þ x2 þ 1 501
a suitable matrix for our proposed algorithm, we
propose the ½32  32 constant symmetric matrix
denoted by Kinit . Figure 2 shows the numerical
values of the proposed matrix Kinit . We use this CN. These polynomials are irreducible. The irre-
matrix to compute and produce the encryption ducible polynomial cannot be factored by a poly-
keys K and L as shown in Algorithm 1. nomial and can generate all the elements of an
The rows of Kinit are linearly independent over extension field Gf ð28 Þ. In our work, we select the
GFð28 Þ [rank(Kinit ) = 32], which means that Kinit is irreducible polynomial x8 þ x4 þ x3 þ x2 þ 1 as
an invertible matrix. Therefore K is an invertible default polynomial. The arithmetic matrix multi-
matrix (we have already stated that αÞ0 and Kinit plication over Galois fields GFð28 Þ is applied for
is proved as an invertible matrix). each sub-block ½32  32 of the data matrix
Table 1 shows the primitive polynomials of image. The image appears distorted so that the
degree 8 with their corresponding numerical image is completely incomprehensible.
6 R. HAMZA AND F. TITOUNA

diffusion processes. The plain image I should have

one matrix with size (h  w).
Algorithm 1. Compute the encryption keys: Figure 3 shows the plain image lena of dimen-
x ; y ; ε; τ; ν; Kinit ; I sion 512  512 pixels along with its corresponding
INPUT: 0 0
encrypted and decrypted images using our pro-
0
OUTPUT: R, V, V , α, K, L posed algorithm.
The detailed encryption process includes the
[h, w] size(I)
following steps.
Apply Eq. (2), which generates Vec. Step 0. Read the 8-bit gray image I and get its
Rinit reshapeðVecð1 : 1024Þ; 32; 32Þ size h  w.
Step 1. Apply Algorithm 1 using the secret keys,
Vinit Vecð1 : hÞ: which produces the indexes ðRÞ, ðVÞ, and ðV 0 Þ,
0
alongside α, K, L.
Vinit Vecð1 : wÞ: Step 2. Shift the rows and columns of the plain-
R indexsort( Rinit ) image matrix using the sorts elements V and V 0
and then apply the following equation:
V indexsort(Vinit )
V
0 0
indexsort(Vinit ) I2 : I  α (3)
jX X X k where  is the bit-wise operation, and I2 repre-
α¼ Rinit ð:Þ þ Vinit ð:Þ þ V 0 init ð:Þ mod256
sents the obtained matrix.
Step 3. To distort the image, we permute the
If α ¼ 0; then
obtained image pixels using the indexes in matrix
jX X X k R and split I2 into ½32  32 blocks denoted by B.
α¼ Rinit ð:Þ þ Vinit ð:Þ þ V 0 init ð:Þ mod 255
We permute the rows and columns for each block
end if B of I2 using the column of matrix R as indexes.
The obtained matrix is denoted by I3 .
K α  Kinit Step 4. The following equation shows the diffu-
sion processes, where we diffuse each ½32  32
L K 1
block B within the obtained image I3 over the
finite field GFð28 Þ using the matrixes K and L.

2.3. The encryption algorithm I4 : L  B  K (4)

where I4 represents the obtained matrix.
The proposed encryption algorithm is a two-step
Step 5. Shift each row and column from the
process. The first is to produce encryption keys
obtained matrix I4 using the sorts elements V
using Algorithm 1 based on the 2-D Zaslavsky
and V 0 . Note that the vector V is defined with
map. The second step is to dismantle and distribute
0
pixels of the plain image using permutation and ðhÞ length, while V 0 has ðwÞ length, so that

(a) (b) (c)

Figure 3. (a) Plain image. (b) Encrypted image. (c) Decrypted image.
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 7

shifting the columns can be done using the vector using the unique indexes sorted V and V 0 . The
V 0 , and shifting the rows can be done using the obtained matrix is denoted by C2 .
vector V. The obtained matrix is denoted by I5 . Step 3. Similar to the encryption step, we apply
Step 6. Repeat the previous steps 3–5 sequen- Eq. (5 for each ½32  32 block B over the finite
tially for four rounds. The matrix results from field GFð28 Þ within the image data C2 .
these possessing steps are denoted by ðCÞ, the
C3 : K  B  L (5)
ciphered image for the plain image.
where C3 represents the obtained matrix.
Step 4. Apply an operation inverse to step 3 in
the encryption algorithm using the index column
2.4. The decryption algorithm
in R for each ½32  32 block of C3 . The obtained
Decryption involves reconstructing the original matrix is denoted by C4 .
image from the encrypted image with an inverse Step 5. Repeat the previous steps 2–4 sequen-
process of the proposed encryption algorithm. tially for four rounds. The resulting matrix is
The indexes sorted of the matrix Rinit , Vinit , and denoted by C5 .
0
Vinit are unique, which means that the original Step 6. Finally, apply Eq. (6). Then, apply the
pixels should be recovered successfully using the inverse shifting by the indexes in V and V 0 .
inverse operations of step 3 and step 5 in the D : C5  α (6)
encryption steps.
where D is the decrypting image of the ciphered
Furthermore, since K is an invertible matrix and
image C after applying the inverse shifting.
L  K ¼ Id over the finite field GFð28 Þ, L can be
Figure 4 shows the steps of pursuing encryption
computed by L ¼ K 1 , where Id is a ½32  32 iden-
and decryption for the image of girl lena
tical matrix. 512  512, where I1 ; I2 ; I3 ; I4 ; and I5 represent the
The decryption method can be performed by
obtained matrixes from the corresponding encryp-
applying the following steps.
tion step, while C1 ; C2 ; C3 ; C4 ; and C5 represent
Step 0. Read the ciphered image C and get its the obtained matrixes from the corresponding
size h  w. decryption step.
Step 1. Generate the encryption keys ðRÞ, ðVÞ,
and ðV 0 Þ, alongside α, K, L using the correct initial
2.5. Application for color images
values (the secret keys) based on Algorithm 1.
Step 2. Apply an operation inverse to step 5 in As we know, a gray-scale image is stored in a
the encryption algorithm, within the matrix C, matrix (2-D array), where each position of the
matrix represents a pixel value, whereas color

Figure 4. Illustration of encryption/decryption algorithm for gray image (Lena).

8 R. HAMZA AND F. TITOUNA

images such us RGB images are stored in three the three directions: vertical, horizontal, and
matrixes that represent the corresponding RGB diagonal. The test of the NIST suite is presented
colors. So, the proposed scheme can encrypt one to prove the randomness statistical properties of
matrix at a time, which means that the proposed the ciphered image. We also present the criterion
image encryption method can encrypt each matrix of difference between the original image with its
separately. Alternatively, the color image with corresponding encrypted image, using the num-
three ½h; w matrixes can be transformed into one ber of pixel change rate ðNPCRÞ test and the test
matrix with ½3  h; w. The obtained matrix is of unified average changing intensity ðUACIÞ.
encrypted similarly to the encryption of a gray- Also, key space analysis and key sensitivity are
level image. Therefore, our proposed scheme will presented. We show the efficiency of our pro-
be capable of encrypting color images. posed algorithm against known/chosen attacks,
even with some typical data images. Ultimately,
we are able to compare the performance achieved
3. Experimental results and discussion by our proposed image encryption, based on sev-
eral criteria, with the performance of other recent
Any proposed encryption method must give chaotic algorithms.
good performance and results compared to
existing algorithms. We run various tests using
MATLAB Version: [Link] (R2014a) under the
Windows 7 environment (64-bit) in a personal 3.1. Histogram analysis
laptop with an Intel Pentium CPU 2020 M 2.4 The histogram of the image represents the distri-
GHz and 4 GB memory. The secret keys are bution of intensity levels for its pixels. A histogram
selected as ðx0 ¼ 0:14; y0 ¼ 0:17; ε ¼ 6:66; τ ¼ shows pixel distribution values. Therefore, the his-
5:55; ν ¼ 7:77Þ for the following tests. The per- togram of the encrypted image should be uni-
formance of encryption and decryption of our formly distributed, which prevents any type of
proposed cryptosystem are analyzed and well statistical attacks.
discussed. The tests and simulation have been Figure 5 and Figure 6 show the histograms for
done using database images ([Link]/data- different images with their corresponding ciphering
base) along with standard testing images such image. The pixels in the ciphering images are uni-
as Lena, Cameraman, and Black images. formly well distributed, which is closer to equiva-
Starting with the analysis of the histogram, the lent probability of occurrence for each intensity
original image and its encrypted image, along level. The tests prove that the encrypted images
with the mathematical quantitative analysis are visually indistinguishable, even with special
using the variance, evaluates the uniformity of images, e.g., the Black image (see Figure 5). The
distributions of pixels. The correlation coeffi- proposed encryption algorithm possesses good con-
cient analysis of pixels in the cipher image is in fusion properties.

(a) (b) (c) (d)

Figure 5. (a) Plain image. (b) Histogram for the plain image. (c) Corresponding encrypted image. (d) Histogram for the encrypted
image.
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 9

(a) (b) (c) (d)

Figure 6. (a) Plain image (Lena). (b) Histogram for the plain image. (d) Corresponding encrypted image. (d) Histogram for the
encrypted image.

Furthermore, we employ the variances of his- Table 3. Percentage of variances difference of histograms com-
pared among all secret keys in the proposed algorithm.
tograms to evaluate uniformity of the ciphered
Ciphered image x0 % y0 % ε% τ% ν%
image. We computed two variances of ciphered Black 0.0286 0.5552 0.3627 0.1737 0.3712
images that are encrypted by two secret keys on Lena 0.5884 0.4774 0.4199 0.15038 0.70316
the same plain-image image. The closer of the Cameraman 0.2054 1.2151 0.1238 0.4964 0.63127

two values of variances indicates the higher

uniformity of ciphered images when the secret among all secret keys in the proposed algorithm. We
keys are varying (Zhang & Wang, 2014). The compute the percentage of variance difference
variance of histograms is presented as follows: between the secret key SC1 and the other secret keys
8 that have been used in Table 2. The simulation results
> P
255
>
< varðZÞ ¼ 2561
ðzi  υÞ2 show that the proposed scheme can resist any statis-
i¼0
(7) tical attacks.
>
> P
255
:υ ¼ 256 zi
1
i¼0
3.2. Information entropy
where Z is the vector of the histogram values and
zi is the number of pixels whose pixel value is The entropy of information expresses the uniform
equal to i. distribution of the pixel value (Shannon, 1949).
In this test, we computed two variances of histo- For a random image with 8-bit intensity levels,
grams of two ciphered images. The encrypted image the ideal entropy should be 8, which means that
was obtained from ciphering the same plain image it is a truly random image. Therefore, an ideal
with two different secret keys. These keys differ only information entropy of an encrypted image should
in one parameter with tiny changes. Table 2 lists the be equal to 8, proving that the information is
variances of histograms of the ciphered Black (zero completely random, and confirming the robust-
pixels), Lena, and Cameraman images. The first col- ness of the proposed encryption algorithm. The
umn in Table 2 was obtained using the secret key SC1, information entropy H(X) of a message source X
while the variances in the next columns were obtained can be computed as
by changing only one parameter, compared with the X
z
secret key SC1, as follows: x0 , y0 , ε, τ, and ν, respec- HðXÞ ¼  Pðxi Þ log 2 Pðxi Þ (8)
tively. Table 2 shows that the variance values are very i¼1

close to 5,450. Similarly, Table 3 shows the percentage where Z is the total number of symbols, and xi 2
of variances difference of histograms compared X and Pðxi Þ represent the probability of the sym-

Table 2. Variances of histograms compared among all secret keys in the proposed algorithm.
Ciphered image SC1 SC1(x0 ) SC1(y0 ) SC1(ε) SC1(τ) SC1(ν)
Black 5442.4712 5440.9117 5472.6901 5462.2115 5433.0136 5462.6758
Lena 5471.0227 5438.8302 5444.9006 5448.0486 5462.7949 5432.5524
Cameraman 5441.5661 5430.3845 5507.6902 5448.3061 5468.5818 5475.9174
10 R. HAMZA AND F. TITOUNA

Table 4. Entropy of the plain and cipher images of different size of the high correlation between adjacent pixels of
images. the plain image. Thus, The absolute value of the
Name Original image Encrypted image
correlation coefficient of two adjacent pixels in the
Black 0 7.9973
Lena 7.5954 7.9978 ciphered images should be as low as possible
Cameraman 7.0097 7.9973 (equal to zero), and it achieves the ideal result of
5.1.14 7.3424 7.9977
7.1.03 5.4957 7.9993
this test. First, we select randomly 2,048 pairs of
7.1.09 6.1898 7.9994 two adjacent pixels in the diagonal, vertical, and
Numbers 7.7292 7.9994 horizontal directions from the two images, the
Gray21 4.3923 7.9993
Ruler 0.5000 7.9994 original image and its corresponding ciphered
Testpat.1k 4.4077 7.9998 image. Then, we calculate the correlation coeffi-
5.3.01 7.5237 7.9998
cient of each pair by using the following formulas:

bol xi . Assuming that we have an 8-bit gray image, covðx; yÞ

corrðx; yÞ ¼ pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi (9)
the entropy of the encrypted image should ideally DðxÞ DðyÞ
be H(X) = 8. Therefore, an effective encryption
method should achieve an entropy close to 8 for where
the ciphered image.
1X N
Table 4 shows the information entropy scores for a covðx; yÞ ¼ ½xi  EðxÞ½yi  EðyÞ (10)
set of sample images from the chosen database. The N i¼1
average entropy of the cipher images, with 512  512
pixels, is computed to be 7.9993. Indeed, the entro- 1X N
DðxÞ ¼ ½xi  EðxÞ2 (11)
pies of all the images encrypted by our algorithm are N i¼1
very close to the ideal value, and prove that the
1X N
ciphered images from the proposed encryption algo- EðxÞ ¼ xi (12)
rithm are almost close to a random source. N i¼1

In Table 5, the correlation between pixels of the

ciphered images are listed. The correlation coeffi-
3.3. Correlation of two adjacent pixels
cient of two adjacent pixels in the ciphered images
An efficient image encryption algorithm should is almost zero, and achieve the ideal value, which
eliminate the correlation of pixels. In this part of means that our approach can effectively remove
the analysis, the correlations between two adjacent correlations among the adjacent pixels. In addi-
pixels along the horizontal, vertical, and diagonal tion, Figure 7 shows the histograms for an image
directions in the original image and its corre- with its corresponding ciphered image in the three
sponding ciphering image has been analyzed. The directions, diagonal, vertical, and horizontal.
correlation through two adjacent pixels is a very Therefore, the proposed algorithm is robust
important perspective in security analysis, because against different statistical attacks.

Table 5. Correlation coefficients of two adjacent pixels in the plain and cipher images.
Original Cipher
Filename Horizontal Vertical Diagonal Horizontal Vertical Diagonal
Black 0 0 0 8.9950e-04 0.0038 0.0080
Lena 0.9201 0.9560 0.8986 0.0023 0.0049 0.0054
Cameraman 0.9335 0.9592 0.9087 6.9973e-04 5.0920e-04 0.0023
5.1.09 0.9020 0.9390 0.9037 −0.0049 −0.0015 −4.0085e-04
5.1.14 0.9466 0.8984 0.8529 0.0057 –0.0018 −9.5226e-04
7.1.03 0.9456 0.9321 0.9017 –0.0025 –0.0030 0.0055
7.1.09 0.9657 0.9304 0.9168 0.0023 8.7995e-04 0.0031
Numbers.512 0.7386 0.7159 0.6253 –0.0044 0.0012 0.0017
Gray21.512 0.9965 0.9998 0.9964 –3.5805e-04 0.0029 0.0016
Ruler.512 0.4542 0.4648 −0.0290 0.0021 0.0019 −0.0014
Testpat.1k 0.7593 0.7992 0.6978 5.5709e-04 −3.0760e-05 0.0012
3.01 0.9774 0.9813 0.9671 3.4210e-04 8.3797e-04 2.7971e-04
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 11

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure 7. Correlation analysis of two adjacent pixels in an image and its corresponding encrypted image. (a) The plain image. (b)
Distributions of two horizontally adjacent pixels in the plain image. (c) Distributions of two vertically adjacent pixels in the plain
image. (d) Distributions of two diagonally adjacent pixels in the plain image. (e) The encrypted image. (f) Distributions of two
horizontally adjacent pixels in the encrypted image. (g) Distributions of two vertically adjacent pixels in the encrypted image. (h)
Distributions of two diagonally adjacent pixels in the encrypted image.

3.4. Randomness tests First, we encrypted the 00 elaine:51200 image,

To have confirmation that the ciphered image acts which has ½512  512 pixels. The ciphered data
truly randomly, we employ the NIST suite test have been converted to binary data and passed
(Rukhin et al., 2010). This test has 15 statistics, into the suite tests of NIST. The results of the
NIST statistical tests are listed in Table 6. The
which requires a binary sequence with at least 106
cipher image passed all the NIST tests.
bits to find potential defects. The result of this test
comes with the values Pvalues , and must be bigger
than 0:01 to pass the tests. Also, we set B = 3.5. Key sensitivity
000010011 in the nonoverlapping template matching
test. Otherwise, we use the default values that come The secret keys in our proposed image encryption
with the NIST suite tests (Rukhin et al., 2010). algorithm are very sensitive to the slightest modifica-
tions. Table 7 represents the tests of NPCR and UACI
between two images, C1 , C2 . We choose to produce
Table 6. Results of the NIST SP 800–22 randomness test on 0
encrypted image. the keys R; V; V with different secret keys:
Statistical test Pvalues Results
Frequency (monobit) 0.8697 Passed ● ðxR ð0Þ; yR ð0Þ; εR ; τR ; νR Þ ¼ ð0:12; 0:13; 2:3; 5; 4Þ
Frequency (block) 0.7657 Passed
Runs test 0.6484 Passed ● ðxV ð0Þ; yV ð0Þ; εV ; τV ; νV Þ ¼ ð0:14; 0:15; 5:6; 8; 7Þ
Random excursions (x = −3) 0.2921 Passed
Random excursions variants (x = −5) 0.9032 Passed
Longest run of ones in a block 0.9523 Passed ● ðxV 0 ð0Þ; yV 0 ð0Þ; εV 0 ; τV 0 ; νV 0 Þ
Binary matrix rank 0.0776 Passed
Discrete Fourier transform (spectral) 0.6729 Passed ¼ ð0:16; 0:17; 10:11; 12; 9Þ
Cumulative sums 0.9861 Passed
Nonoverlapping template matching (B = 0.6558 Passed
000010011)
In this test, we cipher the same plain image with two
Overlapping template matching 0.1401 Passed different secret keys, SC1 and SC2 . The first secret key,
Universal 0.5945 Passed SC1 , is different from the second secret key, SC2 , by a
Linear complexity 0.6511 Passed
Serial 0.5493, 0.7315 Passed slight modification ( þ 1015 ) in one of the initial
Approximate entropy 0.2683 Passed values. The corresponding ciphering is tested using
12 R. HAMZA AND F. TITOUNA

Table 7. NPCR and UACI between cipher images with slightly SC1 , is different from the second secret key, SC2 ,
different keys +1015 .
by a slight modification ( þ 1015 ), and the second
Initial-value change NPCR UACI
xð0ÞR þ 1015 99.6323 33.5473
key, SC2 , is also different from the third secret key,
yð0ÞR þ 1015 99.6185 33.5836 SC3 , by a slight modification ( þ 1015 ). The dif-
εR þ 1015 99.5926 33.3232 ference between each pair of keys is only in one of
τR þ 1015 99.6201 33.3345
νR þ 1015 99.5987 33.6546
the initial values ðx0 ; y0 ; ε; τ; νÞ. The results
xð0ÞV þ 1015 99.6109 33.3711 show that the reconstructed image is not possible
yð0ÞV þ 1015 99.6155 33.4464 without the exact secret keys that have been
εV þ 1015 99.5804 33.3092
τV þ 1015 99.5560 33.6771 encrypted with it. Indeed, all results show that
νV þ 1015 99.6109 33.5089 our proposed algorithm is highly sensitive to its
xð0ÞV 0 þ 1015 99.6155 33.6750 secret keys for both encryption and decryption.
yð0ÞV 0 þ 1015 99.6109 33.5089
εV 0 þ 1015 99.5804 33.4856
τV 0 þ 1015 99.6033 33.3522
νV 0 þ 1015 99.6155 33.5106 3.6. Key space
A reliable ciphering algorithm should have large key
space to resist exhaustive search attacks. The space of
the NPCR and UACI tests. Table 7 shows that all initial keys in any encryption scheme should not be smaller
values are very sensitive for every single tiny change: than 2128 to make brute-force attacks infeasible
ðx0 : 1015 ; y0 : 1015 ; ε: 1015 ; τ: 1015 ; ν: 1015 Þ (Alvarez & Li, 2006). In our proposed algorithm, all
for each encryption key produced. The results show the initial values and controlling parameters are
the high sensitivity of our proposed scheme depending selected as secret keys. These keys are double-preci-
on the secret keys. sion numbers, and according to the IEEE floating-
Figure 8 shows tests of key sensitivity results in point standard (Bailey, 2005), the computational pre-
both the encryption and decryption processes. We cision of the 64-bit double-precision numbers is about
encrypt the same plain image using three different 1015 . Indeed, all initial values are very sensible for
secret keys, SC1 , SC2 , and SC3 . The first secret key, every single modification as shown in Table 7.

(a) (b) (c) (d)

(e) (f) (g) (h)

Figure 8. Key sensitivity results. (a) The encrypted image C using the secret key SC1. (b) The encrypted image C1 using the secret key
SC2. (c) The encrypted image C2 using the secret key SC3, (d) The image difference |C1 – C2|. (e) The decrypted image using the secret
key SC1. (f) The decrypted image D1 using the secret key SC2. (g) The decrypted image D3 using the secret key SC3. (h) The image
difference |D1 – D2|.
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 13

Therefore, the space key in the pseudo-random num- The mathematical representation of the tests
ber generator can compute approximately with more NPCR and UACI are presented as follows:
than ð1015 Þ5 ¼ 1075 . Indeed, if we consider that we P
Cði; jÞ
can generate the vectors V, V 0 , and ðRÞ from different NPCR ¼ (13)
NM
keys (see Table 7), this makes the space key more
than 10225 ’ 2711 for our proposed image encryption X jC1 ði; jÞ  C2 ði; jÞj
UACI ¼ (14)
algorithm. And with such large key space, it is 255  N  M
nearly impossible to estimate the selected secret where C1 and C2 are the ciphering images produced
keys, and there is no significant chance for exhaustive from two images that differ just in one pixel with 1
attacks. bit. The C size is N  M, and C is defined by

1; if C1 ði; jÞ ¼ C2 ði; jÞ
Cðl; mÞ ¼ (15)
0; otherwise
3.7. NPCR and UACI tests
Figures 9a and 9b show the scores of the NPCR and
Differential attacks are considered the most com- UACI tests for the ½256  256 Lena image. These
mon way to break any image encryption algo- tests were performed over 1,000 times to evaluate
rithm. Usually, the attacker seeks to find a plain-image sensitivity in our proposed encryption
relationship between the pixels from plain-image algorithm. These tests were carried out as follows.
pairs that can affect the results of the difference at First, we cloned the plain image as J ¼ I, where I is
the ciphering image pair. So, any change in the
the plain image and J is the cloned image. Then, we
initial image, even in 1 bit, should change the
randomly selected the position of one pixel. Finally,
ciphering image completely, so that the attacker
we applied the following equation:
cannot find any information related to differential
attacks. Jðx; yÞ ¼ Iðx; yÞ  1 (16)
The tests of NPCR (Wu, Noonan, & Agaian,
2011) and UACI (Wu et al., 2011) are proposed where J is the modified image, and ðx; yÞ are the
in this subsection. NPCR analysis shows the beha- indexes of the selected pixel.
vior of all pixels between two paired cipher images, Both images (I and J) were encrypted using the
while UACI measures the average intensity of dif- same secret keys. Finally, the paired cipher images
ferences between the two paired cipher images were tested using the NPCR and UACI over 1,000
(Wu et al., 2011). Both NPCR and UACI use two times. That should ensure approximate average
paired cipher images with one-pixel changing of results.
the plain image, which can prove the high resis- Table 8 provides the experimental results with
tance of any algorithm against differential attacks. different images in gray scale with the above test.

99.68 33.8

99.66 33.7

99.64
33.6

99.62
npcr scores

uaci scores

33.5
99.6
33.4
99.58

33.3
99.56

99.54 33.2

99.52 33.1
0 100 200 300 400 500 600 700 800 900 1000 0 100 200 300 400 500 600 700 800 900 1000
iteration iteration

(a) (b)

Figure 9. (a) NPRC test for 1,000 plain images modified in one bit with a single pixel. (b) UACI test for 1,000 plain images modified in
one bit with a single pixel.
14 R. HAMZA AND F. TITOUNA

Table 8. NPCR and UACI tests. image. This test is organized as follows. First, we gen-
Name Size UACI % NPCR % erate a ½256  256 zero-pixel image denoted by I. The
Black 256 × 256 33.4819 99.6121
image J is also ½256  256 zero pixels with one bit
Lena 256 × 256 33.5124 99.6101
Cameraman 256 × 256 33.4524 99.6068 different, where we randomly select and change the bit
5.1.13 256 × 256 33.4817 99.6080 in one pixel. CI; CJ represent the ciphered images
5.1.14 256 × 256 33.4217 99.6112
7.1.03 512 × 512 33.4668 99.6085 coresponding to the I; J images. Finally, we show
7.1.09 512 × 512 33.4981 99.6108 the mean absolute of the pair of images jI  Jj and
Numbers 512 × 512 33.4533 99.6084
Gray21 512 × 512 33.4525 99.6098
jCI  CJj. The tests confirm that the attacker cannot
ruler 512 × 512 33.4613 99.6077 find or get any useful information to build his attack.
The results come with no observation of the black
zone, and demonstrate the previous result of high
These tests were performed 1,000 times for each
security against differential attacks even with the cho-
sample image. As shown in Table 8, the values of
sen/known image.
NPCR and UACI of our method exceeding the ideal
According to Kerckhoffs’ principle, the security of a
ones, which is 99.60% for NPCR and 33.46% for
cryptosystem must not depend on keeping it secret.
UACI (Wu et al., 2011). Indeed, the score of NPCR
The security depends only on keeping the secret of the
is around 99.61%, and the score of UACI is about
keys (Lee & Chen, 2002). The hacker can access or/and
33.47% for our proposed image encryption algo-
knows exactly the design and working of the crypto-
rithm. The results of these tests indicate the high
system, and keeps the secret keys and the original
sensitivity to plain-image change: even a tiny change
image. Therefore, assuming that the attacker has
can completely change the corresponding ciphered
obtained access to the encryption algorithm, even for
image. Therefore, the proposed algorithm can pro-
a temporary time, there are four classical types of
vide high security against differential attacks.
attack:
(1) Known cipher-text attack. The attacker pos-
3.8. Known/chosen attack sesses a string of the cipher image.
(2) Known plain-text attack. The attacker pos-
The attacker should not obtain any useful information
sesses a string of the original image, and
by encrypting some special images such as the Black
obviously the corresponding ciphered image.
image (zero pixels). Figure 10 shows the resistance
(3) Chosen plain-text attack. The attacker
against chosen/known attack based on the Black
chooses an original image string and con-
structs the corresponding ciphered image
string using the encryption algorithm.
(4) Chosen cipher-text attack. The attacker
chooses a ciphered image string and constructs
the corresponding decrypted image string.

(a) (b) (c)

If a cryptosystem can resist the attack of chosen
plain text, it can resist other types of attack due to the
fact that chosen plain-text attack is the most powerful/
effective attack (Wang et al., 2012). Therefore, and
according to the previous tests, the proposed algo-
rithm can resist these classical types of attacks.

(d) (e) (f) 3.9. Comparison

Figure 10. Plain-image sensitivity results. (a) The plain image I. Our cryptosystem algorithm is divided into two
(b) The modified image J. (c) The image difference jI  Jj. (d)
The encrypted image CI. (e) The encrypted image CJ. (f) The
stages: initialization and processing operations. In
image difference jCI  CJj. the first stage, we produce keys for the encryption/
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 15

decryption processes. This initialization stage will cient values for the encrypted ½256  256 Lena
take around 0:06 s for a ½256  256 image. In the image, we use the following equation:
second, processing stage, the approximate time jCh j þ jCv j þ jCd j
can be about 0:26 s (encryption and decryption CC ¼ (17)
3
have the same speed) for a ½256  256 image.
where the correlation coefficient value is denoted
The encryption speed is highly dependent on
by CC, and the correlation in three directions
the CPU/GPU performance, RAM size, and the
(horizontal, vertical, and diagonal) are denoted
programming environment (C/C++, Java, Python,
by Ch ; Cv ; and Cd .
etc.) (Norouzi et al., 2014a). Therefore, the com-
Table 10 shows the comparison of various efficien-
parison of speed encryption is an approximate test
cies for different image encryption algorithms using
due to various system characteristics and the plat-
the standard Lena image with dimension ½256  256
form that has been used in each study.
pixels. Indeed, Table 10 confirms that our algorithm
The complexity performance of the proposed algo-
rithm is shown in Table 9. The encryption times of the has larger key space, good performance, and exceeds
algorithm are also shown in Table 9 for the same the ideal scores for an encryption algorithm.
½256  256 Lena image. The test demonstrates that
our algorithm can achieve good speed, and that it is 4. Conclusion and future directions
faster and better than the other methods presented in
This article presented a novel image encryption
Table 9. Therefore, our proposed image encryption
algorithm based on the two-dimensional Zaslavsky
algorithm is faster than state-of-the-art image encryp-
map using the permutation and diffusion structure.
tion schemes, validating its better performance.
Our proposed image encryption algorithm consists
In addition, we compared the performance of our
of two stages. In the first stage, the proposed cryp-
algorithm to several typical image encryption chaos-
tosystem employs the 2-D Zaslavsky chaotic map to
based algorithms. To compare the correlation coeffi-
produce encryption keys for our proposed algo-
rithm. These keys are applied to shuffle and diffuse
Table 9. Speed analysis between our proposed method and the pixels of the plain image in the second stage
other chaotic-cryptosystems. using permutation-diffusion processes. Simulation
Encryption Platform & system tests and security analyses were carried out carefully
Algorithm time (s) characteristics and are well discussed. These tests, including sta-
Our proposed algorithm 0.32 s MATLAB 8.3, CPU 2.4
GHz, 4 GB memory
tistical, histogram, sensibility, key space, differen-
Norouzi, Seyedzadeh, 0.4 s MATLAB 7.6, CPU 2.4 tial, quality, and speed analyses, prove the high
Mirzakuchaki, & Mosavi, GHz, 4 GB memory security level and sensitivity in the plain image
2014b
Norouzi et al., 2014b 0.41 s MATLAB 7.6, CPU 2.4 with its secret key of the proposed scheme. The
GHz, 4 GB memory proposed image encryption algorithm has good
Huang, 2012 0.55 s MATLAB 6.5, CPU 2.00
GHz, 2 GB memory
ability to resist differential attack and withstand
Xu, Li, Li, & Hua, 2016 1.32 s MATLAB 7.8, CPU 2.5 the known-plain-image, chosen-plain-image attack,
GHz, 4 GB memory and is highly capable of withstanding the various

Table 10. Comparison between our proposed method and other cryptosystems.
Image CC NPCR UACI Entropy Key space
Ideal value ’ 0 >99:60 ’ 33:46 ’ 8 > 2100
Proposed algorithm 0.0031 99.61 33.50 7.9978 2711
Akhshani, Behnia, Akhavan, Hassan, & Hassan, 2010 0.0058 0.39 0.33 7.9989 2225
Akhavan, Samsudin, & Akhshani, 2011 0.0031 39.45 33.28 7.9978 2180
Huang et al., 2013 0.0027 99.54 28.27 7.9967 2152
Wu et al., 2012 0.002 99.59 33.58 7.9972 2256
Wang, Zhang, & Bao, 2015 0.0013 99.65 33.48 7.9970 2173
Norouzi et al., 2014b 0.0001 99.61 33.45 7.9979 2512
Xu et al., 2016 0.0094 99.62 33.51 7.9974 2210
Norouzi, Seyedzadeh, Mirzakuchaki, & Mosavi, 2015 0.0078 99.61 33.45 7.9980 2256
16 R. HAMZA AND F. TITOUNA

attacks. The comparison with several recent chaos- Bailey, D. H. (2005). High-precision floating-point arithmetic
based image encryption algorithms shows that our in scientific computation. Computing in Science &
Engineering, 7(3), 54–61. doi:10.1109/MCSE.2005.52
proposed encryption image algorithm is fast with
Chen, G., Mao, Y., & Chui, C. K. (2004). A symmetric image
excellent performance, making it more appropriate encryption scheme based on 3D chaotic cat maps. Chaos,
for reliable and practical cryptographic applications. Solitons & Fractals, 21(3), 749–761. doi:10.1016/j.
In future work, we aim to investigate a probabilistic chaos.2003.12.022
approach for image encryption. Also, we intend to De Avila, S. E. F., Lopes, A. P. B., Da Luz, A., & De Albuquerque
combine the proposed image encryption scheme Araújo, A. (2011). VSUMM: A mechanism designed to pro-
duce static video summaries and a novel evaluation method.
with video summarization (Almeida, Leite, & Torres,
Pattern Recognition Letters, 32(1), 56–68. doi:10.1016/j.
2012; De Avila, Lopes, Da Luz, & De Albuquerque patrec.2010.08.004
Araújo, 2011; Mehmood, Sajjad, & Baik, 2014) and Gao, T., & Chen, Z. (2008). A new image encryption algo-
wireless capsule endoscopy (Lin, Liu, & Yuan, 2015). rithm based on hyper-chaos. Physics Letters A, 372(4),
Another future direction is to merge the proposed 394–400. doi:10.1016/[Link].2007.07.040
image encryption scheme with image and video ste- Hermassi, H., Belazi, A., Rhouma, R., & Belghith, S. M.
(2014). Security analysis of an image encryption algorithm
ganographic methods (Mstafa & Elleithy, 2015;
based on a DNA addition combining with chaotic maps.
Muhammad et al., 2016b; Muhammad, Sajjad, & Multimedia Tools and Appli- Cations, 72(3), 2211–2224.
Baik, 2016b) and watermarking algorithms (Liu, doi:10.1007/s11042-013-1533-6
Huang, et al., 2016a; Liu, Zhang, et al., 2016b) as an Huang, C. K., Liao, C. W., Hsu, S. L., & Jeng, Y. C. (2013).
extra security layer against attackers and to secure Implementation of gray image encryption with pixel shuf-
visual contents retrieval in personalized video libraries fling and gray-level encryption by single chaotic system.
Telecommunication Systems, 52(2), 563–571.
(Muhammad et al., 2015b). Alternatively, the pro-
Huang, X. (2012). Image encryption algorithm using chao-
posed encryption method can be further extended to tic Chebyshev generator. Nonlinear Dynamics, 67(4),
encryption of images captured by visual sensors in 2411–2417. doi:10.1007/s11071-011-0155-7
wireless multimedia sensor networks. Hussain, I., & Gondal, M. A. (2014). An extended image
encryption using chaotic coupled map and S-box trans-
formation. Nonlinear Dynamics, 76(2), 1355–1363.
ORCID doi:10.1007/s11071-013-1214-z
Hussain, I., Shah, T., Gondal, M. A., & Mahmood, H. (2013). A
Rafik Hamza [Link] novel image encryption algorithm based on chaotic maps and
GF (28) exponent transformation. Nonlinear Dynamics, 72
(1–2), 399–406. doi:10.1007/s11071-012-0723-5
References Jeng, F.-G., Huang, W.-L., & Chen, T.-H. (2015). Cryptanalysis
and improve-ment of two hyper-chaos-based image encryp-
Akhavan, A., Samsudin, A., & Akhshani, A. (2011). A sym- tion schemes. Signal Processing: Image Communication, 34,
metric image encryption scheme based on combination of 45–51.
nonlinear chaotic maps. Journal of the Franklin Institute, Khan, M., & Shah, T. (2015). A novel construction of substitu-
348(8), 1797–1813. doi:10.1016/[Link].2011.05.001 tion box with Za- slavskii chaotic map and symmetric group.
Akhavan, A., Samsudin, A., & Akhshani, A. (2015). Journal of Intelligent & Fuzzy Systems, 28(4), 1509–1517.
Cryptanalysis of an improvement over an image encryption Lee, W.-B., & Chen, T.-H. (2002). A public verifiable copy
method based on total shuffling. Optics Communications, protection technique for still images. Journal of Systems
350, 77–82. doi:10.1016/[Link].2015.03.079 and Software, 62(3), 195–204. doi:10.1016/S0164-1212(01)
Akhshani, A., Behnia, S., Akhavan, A., Hassan, H. A., & Hassan, 00142-X
Z. (2010). A novel scheme for image encryption based on 2D Lin, C.-C., Liu, X.-L., & Yuan, S.-M. (2015). Reversible data
piecewise chaotic maps. Optics Communications, 283(17), hiding for VQ-compressed images based on search-order
3259–3266. doi:10.1016/[Link].2010.04.056 coding and state-codebook mapping. Information Sciences,
Almeida, J., Leite, N. J., & Torres, R. D. S. (2012). Vison: Video 293, 314–326. doi:10.1016/[Link].2014.08.057
summarization for online applications. Pattern Recognition Liu, H., & Wang, X. (2010). Color image encryption based on
Letters, 33(4), 397–409. doi:10.1016/[Link].2011.08.007 one-time keys and robust chaotic maps. Computers &
Alvarez, G., & Li, S. (2006). Some basic cryptographic require- Mathematics with Applications, 59(10), 3320–3327.
ments for chaos- based cryptosystems. International Journal doi:10.1016/[Link].2010.03.017
of Bifurcation and Chaos, 16(08), 2129–2151. doi:10.1142/ Liu, H., & Wang, X. (2011). Color image encryption using
S0218127406015970 spatial bit-level permutation and high-dimension chaotic
 INFORMATION SECURITY JOURNAL: A GLOBAL PERSPECTIVE 17

system. Optics Communications, 284(16), 3895–3903. Pub & NIST FIPS. (2001). 197: Advanced encryption stan-
doi:10.1016/[Link].2011.04.001 dard (AES). Federal Information Processing Standards
Liu, H., Wang, X., & Kadir, A. (2012). Image encryption Publication 197, 441–0311.
using DNA complementary rule and chaotic maps. Rhouma, R., & Belghith, S. (2008). Cryptanalysis of a new image
Applied Soft Computing, 12(5), 1457–1466. doi:10.1016/j. encryption algorithm based on hyper-chaos. Physics Letters
asoc.2012.01.016 A, 372(38), 5973–5978. doi:10.1016/[Link].2008.07.057
Liu, Z., Huang, J., Sun, X., & Qi, C. (2016a). A security Rukhin, A. L., Soto, J., Nechvatal, J. R., Smid, M. E., Barker,
watermark scheme used for digital speech forensics. E. B., Leigh, S., . . . Vo, S. (2010). A statistical test suite for
Multimedia Tools and Applications, 1–21. random and pseudorandom number generators for cryp-
Liu, Z., Zhang, F., Wang, J., Wang, H., & Huang, J. (2016b). tographic applications (Special publication 800-22,
Authentication and recovery algorithm for speech signal based Revision 1a). Gaithersburg, MD: National Institute of
on digital watermarking. Signal Processing, 123, 157–166. Standards and Technology.
Mehmood, I., Sajjad, M., & Baik, S. W. (2014). Mobile-cloud Shannon, C. E. (1949). Communication theory of secrecy
assisted video summarization framework for efficient manage- systems*. Bell System Technical Journal, 28(4), 656–715.
ment of remote sensing data generated by wireless capsule doi:10.1002/j.1538-7305.1949.tb00928.x
sensors. Sensors, 14(9), 17112–17145. doi:10.3390/s140917112 Short, K. M. (1994). Steps toward unmasking secure commu-
Mstafa, R. J., & Elleithy, K. M. (2015). A video steganography nications. International Journal of Bifurcation and Chaos, 4
algorithm based on Kanade-Lucas-Tomasi tracking algo- (04), 959–977. doi:10.1142/S021812749400068X
rithm and error correcting codes. Multimedia Tools and Stinson, D. R. (2005). Cryptography: Theory and practice
Applications, 1–23. (3rd ed.). Boca Raton, FL: CRC Press.
Muhammad, K., Sajjad, M., Mehmood, I., Rho, S., & Baik, S. W. Sun, F., Lü, Z., & Liu, S. (2010). A new cryptosystem based on
(2015a). A novel magic LSB substitution method (M-LSB- spatial chaotic system. Optics Communications, 283(10),
SM) using multi-level encryption and achromatic compo- 2066–2073.
nent of an image. Multimedia Tools and Applications, 1–27. Tang, Y., Wang, Z., & Fang, J.-A. (2010). Image encryption
Muhammad, K., Mehmood, I., Lee, M. Y., Ji, S. M., & Baik, S. using chaotic coupled map lattices with time-varying
W. (2015b). Ontology-based secure retrieval of semanti- delays. Communications in Nonlinear Science and
cally significant visual contents. Journal of Korean Institute Numerical Simulation, 15(9), 2456–2468. doi:10.1016/j.
of Next Generation Computing, 11(3), 87–96. cnsns.2009.09.023
Muhammad, K., Ahmad, J., Rehman, N. U., Jan, Z., & Sajjad, Wang, X.-Y., Chen, F., & Wang, T. (2010). A new compound
M. (2016a). CISSKA-LSB: Color image steganography mode of confusion and diffusion for block encryption of
using stego key- directed adaptive LSB substitution image based on chaos. Communications in Nonlinear
method. Multimedia Tools and Applications, 1–30. Science and Numerical Simulation, 15(9), 2479–2485.
Muhammad, K., Sajjad, M., & Baik, S. W. (2016b). Dual-level doi:10.1016/[Link].2009.10.001
security based cyclic18 steganographic method and its appli- Wang, X., Liu, L., & Zhang, Y. (2015). A novel chaotic block
cation for secure transmission of keyframes during wireless image encryption algorithm based on dynamic random
capsule endoscopy. Journal of Medical Systems, 40(5), 1–16. growth technique. Optics and Lasers in Engineering, 66,
Norouzi, B., Mirzakuchaki, S., Seyedzadeh, S. M., & Mosavi, M. 10–18. doi:10.1016/[Link].2014.08.005
R. (2014a). A simple, sensitive and secure image encryption Wang, X., & Luan, D. (2013). A novel image encryption algorithm
algorithm based on hyper-chaotic system with only one round using chaos and reversible cellular automata. Communications
diffusion process. Multimedia Tools and Applications, 71(3), in Nonlinear Science and Numerical Simulation, 18(11),
1469–1497. doi:10.1007/s11042-012-1292-9 3075–3085. doi:10.1016/[Link].2013.04.008
Norouzi, B., Seyedzadeh, S. M., Mirzakuchaki, S., & Wang, X., Luan, D., & Bao, X. (2014). Cryptanalysis of an
Mosavi, M. R. (2014b). A novel image encryption image encryption algorithm using Chebyshev generator.
based on hash function with only two-round diffusion Digital Signal Processing, 25, 244–247. doi:10.1016/j.
process. Multimedia Systems, 20(1), 45–64. doi:10.1007/ dsp.2013.10.020
s00530-013-0314-4 Wang, X., Teng, L., & Qin, X. (2012). A novel colour image
Norouzi, B., Seyedzadeh, S. M., Mirzakuchaki, S., & Mosavi, M. encryption algorithm based on chaos. Signal Processing, 92
R. (2015). A novel image encryption based on row-column, (4), 1101–1108. doi:10.1016/[Link].2011.10.023
masking and main diffusion processes with hyper chaos. Wang, X.-Y., & Yu, Q. (2009). A block encryption algorithm
Multimedia Tools and Applications, 74(3), 781–811. based on dynamic sequences of multiple chaotic systems.
doi:10.1007/s11042-013-1699-y Communications in Nonlinear Science and Numerical
Pesin, Y. B. (1977). Characteristic Lyapunov exponents and Simulation, 14(2), 574–581. doi:10.1016/[Link].2007.10.011
smooth ergodic the- ory. Russian Mathematical Surveys, 32 Wang, X.-Y., Zhang, Y.-Q., & Bao, X.-M. (2015). A novel
(4), 55–114. doi:10.1070/RM1977v032n04ABEH001639 chaotic image encryption scheme using DNA sequence
 18 R. HAMZA AND F. TITOUNA

operations. Optics and Lasers in Engineering, 73, 53–61. Zhang, Y.-Q., & Wang, X.-Y. (2014). A symmetric image
doi:10.1016/[Link].2015.03.022 encryption algorithm based on mixed linear–nonlinear
Wei, X., Guo, L., Zhang, Q., Zhang, J., & Lian, S. (2012). A novel coupled map lattice. Information Sciences, 273, 329–351.
color image encryption algorithm based on DNA se- quence doi:10.1016/[Link].2014.02.156
operation and hyper-chaotic system. Journal of Systems and Zhang, Y.-Q., & Wang, X.-Y. (2015). A new image encryption
Software, 85(2), 290–299. doi:10.1016/[Link].2011.08.017 algorithm based on non-adjacent coupled map lattices. Applied
Wong, K.-W., Kwok, B. S.-H., & Yuen, C.-H. (2009). An Soft Computing, 26, 10–20. doi:10.1016/[Link].2014.09.039
efficient diffusion approach for chaos-based image encryp-
tion. Chaos, Solitons & Fractals, 41(5), 2652–2663.
Biographies
doi:10.1016/[Link].2008.09.047
Wu, Y., Noonan, J. P., & Agaian, S. (2011). NPCR and UACI Rafik Hamza received the Master degree in cryptography and
randomness tests for image encryption. Cyber Journals: security from the Department of Computer Science,
Multidisciplinary Journals in Science and Technology, Journal University of Batna, Algeria, in 2014. He is currently pursuing
of Selected Areas in Telecommunications (JSAT), 31–38. his Ph.D. on the security of digital images in the department of
Wu, Y., Yang, G., Jin, H., & Noonan, J. P. (2012). Image computer science, University of Batna 2, Algeria. His research
encryption using the two-dimensional logistic chaotic interests include security the digital images, image and video
map. Journal of Electronic Imaging, 21(1), 013014–1. processing, randomness algorithms, keys cryptographic, and
doi:10.1117/[Link].21.1.013014 probabilistic approaches.
Xu, L., Li, Z., Li, J., & Hua, W. (2016). A novel bit-level image Contact information: [Link]@[Link]; rafik.
encryption algorithm based on chaotic maps. Optics and Lasers hamza@[Link]
in Engineering, 78, 17–25. doi:10.1016/[Link].2015.09.007
Zaslavsky, G. M. (1978). The simplest case of a strange Faiza Titouna received her Ph.D. in 2009 under the joint
attractor. Physics Letters A, 69(3), 145–147. doi:10.1016/ supervision from the University of Annaba and the
0375-9601(78)90195-0 University of Artois. She is currently a lecturer in the
Zhang, X., & Zhao, Z. (2014). Chaos-based image encryption Department of Computer Science, University of Batna 2,
with total shuffling and bidirectional diffusion. Nonlinear Algeria. Her research interests include probabilistic networks,
Dynamics, 75(1–2), 319–330. doi:10.1007/s11071-013- possibility theory, and IT security.
1068-4 Contact information: ftitouna@[Link]

View publication stats