See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/370926836

Fast color image encryption algorithm based on FCSM and pre-storage Arnold
transform

Article in Multimedia Tools and Applications · May 2023

DOI: 10.1007/s11042-023-15577-6

CITATIONS READS

2 105

2 authors, including:

Xiaoqiang Zhang
China University of Mining and Technology
49 PUBLICATIONS 1,088 CITATIONS

SEE PROFILE

All content following this page was uploaded by Xiaoqiang Zhang on 19 August 2024.

The user has requested enhancement of the downloaded file.

Multimedia Tools and Applications
https://doi.org/10.1007/s11042-023-15577-6

Fast color image encryption algorithm based on FCSM

and pre‑storage Arnold transform

Xiaoqiang Zhang1 · Zhiwei Liu1

Received: 26 July 2022 / Revised: 3 October 2022 / Accepted: 21 April 2023

© The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2023

Abstract
Image encryption is an important multimedia security technology to protect the image con-
tent during network transmission. Chaotic systems are widely used in the field of image
encryption due to their pseudo-randomness. To solve the problem of poor chaotic effect of
traditional chaotic systems, this paper proposes a novel Fibonacci combined Cubic-Sine
Map (FCSM) with large key space and chaotic characteristics, which passes each of NIST
SP 800-22 tests. Besides, a fast color image encryption algorithm is designed using FCSM
and pre-storage Arnold transform. Firstly, the SHA-256 is used to generate initial values;
secondly, pseudo-random sequences are generated using FCSM, which scramble the image
in combination with the row & column shift and circular rotation; finally, the pre-storage
Arnold transform is designed to diffuse the pixel values. The experimental results and
security analyses show that the proposed algorithm has ideal encryption effect with NPCR
and UACI reaching to 99.6096% and 33.4657%, which are very close to the ideal values.
The proposed algorithm is very efficient, which takes only about 0.6 second to encrypt a
color image with the size of 512 × 512. Therefore, the proposed algorithm is secure and
efficient, which can be a candidate for the real-time encryption system.

Keyword Arnold transform · Combination chaotic system · Image encryption · Pre-storage

1 Introduction

With the development of communication technology in today’s information age, a large

amount of data and information are being shared and used all the time. Among them,
image data plays a very important role. Images are widely used in many fields, such as
medical images, military images, remote sensing images, private images in life, etc. Traf-
fic cameras installed around the world capture thousands of images every day. Many high
technologies, such as autonomous driving, telemedicine, machine vision, etc., rely on mas-
sive amounts of image data. Due to the openness of the internet, there is often a risk of

* Xiaoqiang Zhang
[email protected]
1
School of Information and Control Engineering, China University of Mining and Technology,
Xuzhou 221116, China

13
Vol.:(0123456789)
 Multimedia Tools and Applications

leakage of images in network transmission. As a result, people pay more and more atten-
tion to information security, and it is of great significance to study efficient and secure
image encryption algorithms. The current image encryption algorithms exist the following
problems.

(1) The structure of most of high-dimensional chaotic systems are complex and the itera-
tion speed is slow, which is not suitable for fast image encryption algorithms.
(2) Image encryption algorithms are difficult to balance security and efficiency, so the
performance of most fast image encryption algorithms is not satisfactory.
(3) Image algorithms based on the Arnold transform are usually inefficient for its many-
times iterations to obtain excellent scrambling effect.

To solve the above problems, this paper designs an efficient image encryption algorithm
to meet the requirements of the real-time image encryption system and protect the image
information security.

1.1 Related works

In the field of image security, chaos-based image encryption is a hot topic. Chaotic systems
have the characteristics of pseudo-randomness, initial value sensitivity, parameter sensitiv-
ity, ergodicity and unpredictability. Therefore, concepts related to chaos have been widely
used in the field of image security [21, 25, 26].
According to the different dimensions, chaotic systems can be divided into low-
dimensional chaotic systems and high-dimensional chaotic systems. The former is often
used for the lightweight image encryption [37, 54], but it has a simple structure, small
key space and weak security [36]. The latter has better chaotic space and high security
[15], but its structure is more complex and is not suitable for the real-time image encryp-
tion system [47].
To improve the efficiency of image encryption, Zhou et al. [56] proposed a novel com-
binatorial chaotic system, which can combine traditional low-dimensional chaos to form
a new chaos with a larger key space and better cryptographic properties. Jain et al. [17]
proposed a two-dimensional combined chaotic system, which combined the Logistic map
and the Sine map to obtain better chaotic properties. Wang et al. [43] proposed composite
one-dimensional nested Sine chaos, which has better initial value sensitivity than the origi-
nal Sine map. This paper combines Cubic map and Sine map to design a Fibonacci com-
bined Cubic-Sine Map (FCSM). Experimental tests shows that the FCSM has better cha-
otic properties and improves some of the defects of the original Cubic map and Sine map.
Due to the limited accuracy of the equipment, chaotic systems implemented using com-
puters may lead to dynamic degradation of the original system, resulting in short system
cycles, low linear complexity, and high correlation. Wang et al [42] proved and concluded
through theoretical analysis that it is feasible to simulate chaotic systems by computer. If
the initial point of the iteration is selected on the singular attractor, although there is a
limited precision truncation, it will not lead to the degradation of the dynamics of the cha-
otic system. Therefore, image encryption using computer simulation of chaotic system is a
secure and effective encryption scheme. At present, most of image encryption algorithms
adopt the scrambling-diffusion architecture. People combine chaos with other methods to
design a variety of encryption algorithms, including DNA encoding [24, 30, 53], S-box [1,
6], genetic algorithm [9, 28], compressed sensing [2] and other algorithms [16, 19, 49].

13
Multimedia Tools and Applications

In recent years, researchers have worked on designing the superior image encryp-
tion algorithms. In 2020, Mondal et al. [29] proposed a lightweight image encryption
scheme based on chaos and diffusion circuit. This algorithm accomplishes the disloca-
tion and diffusion of pixel values in one traversal with high encryption efficiency. How-
ever, due to the reduced computational overhead, the encryption effect of the algorithm
is poor and the histogram distribution of encrypted images is not uniform enough. Wan
et al. [41] designed a bit-level image encryption algorithm. It only permutes the higher
four-bit planes of the image to reduce the number of operations. However, the disloca-
tion effect of this algorithm is not sufficient to resist the cropping attack. Yousif et al.
[52] divided an image into two 8-bit images by replacing the binary bits 0 and 1 of the
image with 10 and 01, respectively. Then, the two images are processed separately and
finally combined into one image. This algorithm has desirable security, but the encryp-
tion efficiency is not satisfactory due to the generation of redundant data.
In 2022, Chen et al. [3] designed a double image encryption algorithm for color
and gray images. This algorithm uses the gray image as a secondary image to fill the
bits of the color image. Therefore, the two-image information can fuse and affect each
other. The algorithm has the high plaintext sensitivity, but limiting the number of bits to
10, it is difficult to meet the universal demand. Huang et al. [15] combined DNA cod-
ing with a high-dimensional hyperchaotic system, which has the remarkable features
of large key space and high noise resistance. But their algorithm has high complexity
and long encryption time. Wang et al. [47] combined quantum revolving gate and high-
dimensional chaotic system to design a high-security color image encryption algorithm.
This algorithm has weak key sensitivity and the possibility of leaking plaintext informa-
tion under wrong keys. Zhou et al. [56] proposed an encryption algorithm of bit-level
decomposition with segmented exchange. The position of the exchange is controlled by
a chaotic system, which has excellent encryption effect, but there are texture differences
in the encrypted images.
Arnold transform is a common scrambling method widely used for image encryption
[20]. The Arnold transform has some remarkable advantages. Firstly, each iteration is
globally scrambled [23]; secondly, it is highly sensitive to the iterations number, and
even a tiny change cannot restore the image [11]. However, the Arnold transform also
has its drawbacks, such as the scrambling rule only depends on the iteration time, peri-
odic, and low iteration efficiency [27, 18, 48].
In 2014, Jiang et al. [18] combined the Arnold transform and Fibonacci to expand
the period of the Arnold transform in the same condition, but the periodicity of the
Arnold transform still exists. In 2016, Singh et al. [35] applied the Arnold transform
to the frequency domain by the singular value decomposition. This algorithm has poor
dislocation effect and the neighboring pixels of the encrypted image still have high cor-
relation. In 2017, Farwa et al. [8] combined Arnold transform and S-box to somewhat
improve the effect of their respective individual displacements, but the encryption is
still undesirable and inefficient. In 2020, Mansouri et al. [27] can make one time Arnold
transform with effect result by performing operations like chunking, rotation and pixel
cleaning on the image. However, this algorithm relies on the plain image, and there are
differences in the texture of different plain images. In 2022, Tora et al. [38] enriched the
coefficient matrix of the Arnold transform by matrix operations and expanded the key
space of the Arnold transform. Inspired by the idea of trading off memory for time, a
pre-storage Arnold transform is designed, which can obviously improve the encryption
efficiency.
In summary, the pros and cons of the above algorithms are shown in Tables 1 and 2.

13
 Multimedia Tools and Applications

Table 1  Related works on chaotic image encryption

Algorithm Pros Cons Year

Ref. [29] High encryption efficiency Histogram is not uniform 2020

Ref. [41] High encryption efficiency Not sufficient to resist the cropping attack 2020
Ref. [52] High security Efficiency is not satisfactory 2022
Ref. [3] High plaintext sensitivity Not enough for general use 2022
Ref. [47] Large key space and high noise Efficiency is not satisfactory 2022
resistance
Ref. [56] High-security Weak key sensitivity 2022
Ref. [17] Desirable encryption effect Texture differences in the encrypted images 2022

1.2 Contributions

The main contributions of this paper are described as follows.

(1) A novel FCSM is proposed, which has a simple structure and a large range of param-
eters. Thus, it can be iterated efficiently with good randomness and security.
(2) A pre-stored Arnold transform is designed. Unlike the traditional usage, this design can
extend Arnold transform to the pixel diffusion stage, which allows different number of
iterations for each pixel. Therefore, it can solve the periodicity problem of the Arnold
transform and improve its reusability.
(3) An efficient color image encryption algorithm is proposed using the designed FCSM
and the pre-stored Arnold transform. The experimental results show that the proposed
algorithm has desirable encryption effect and efficiency, which can meet the require-
ments of the real-time encryption system.

1.3 Architecture and abbreviations

The rest of the paper is structured as follows: Section 2 presents the theoretical principles.
Section 3 describes the new color image encryption algorithm. Experiments are carried out
in Section 4. The performance of the proposed algorithm is evaluated in Section 5. Sec-
tion 6 provides a discussion of the proposed algorithm. The conclusions and outlooks are
drawn in Section 7.
The abbreviations and symbols used in this paper are shown in Table 3.

Table 2  Related works on Arnold transform

Algorithm Pros Cons Year

Ref. [18] Expanded period of Arnold transform Still periodic 2014

Ref. [35] Used in the frequency domain Adjacent pixels still have correlation 2016
Ref. [8] Combined with S-box Unsatisfactory results and low efficiency 2017
Ref. [26] One transform can get the desired effect Differences in cipher text 2020
Ref. [38] Expanded the key space of the Arnold Still periodic 2022
transform

13
Multimedia Tools and Applications

Table 3  Abbreviations used in Abbreviations Full spellings

this paper
FCSM Fibonacci combined Cubic-Sine Map
LE Lyapunov Exponent
NPCR Number of Pixels Change Rate
UACI Uniform Average Change Intensity

2 Theoretical principles

2.1 Cubic map

Cubic map is defined by [39]

( )
xn+1 = 𝜇xn 1 − xn 2 , (1)

where xn ∈ (−1.15,1.15), and μ ∈ (0, 3] is the control parameter.

2.2 Sine map

The Sine map is defined by [56]

( )
xn+1 = 𝜆 sin 𝜋xn , (2)

where xn ∈ (0, 1), and λ ∈ (0, 1] is the control parameter.

2.3 Proposed FCSM and its analyses

2.3.1 FCSM

The Fibonacci sequence is defined by [33]

F(n) = F(n − 1) + F(n − 2). (3)
The current item is equal to the sum of the previous two items. By combining the
Fibonacci sequence with the Cubic map, the new equation can be obtained by
( )( ( )2 )
xn = 𝜇 xn−1 + xn−2 1 − xn−1 + xn−2 . (4)

By substituting Eq. (4) into Eq. (2) and setting λ=1, FCSM can be defined by
( ( )( ( ))2 )
xn+1 = sin 𝜋𝜇 xn−1 + xn−2 1 − xn−1 + xn−2 , (5)

where xn ∈ (−1, 1), μ ∈ (0, +∞) is the control parameter, and x0, x1 are the initial values.

13
 Multimedia Tools and Applications

2.3.2 Bifurcation diagram

Bifurcation phenomenon refers to the qualitative change of the qualitative behavior of

the dynamical system with the change of the system parameters [43]. The bifurcation
diagram of a chaotic system can be used to analyze the chaotic behavior of the system
by describing the generated chaotic sequence. It means that the location of the stable
point in a system after the parameters are changed.
The bifurcation diagrams of Cubic map and Sine map are shown in Fig. 1a and b.
As shown in the figure, the chaotic effect of Cubic map occurs when μ is more than 2.3,
and the chaotic effect becomes more obvious when μ is more than 2.59. The closer μ is
to 3, the more obvious the chaotic characteristic is. Sine map has chaotic characteristics
when λ is more than 0.87. The closer λ is to 1, the better the chaotic characteristic is.
Therefore, for Cubic and Sine maps, their chaotic properties are not ideal and the space
of chaotic parameters is small. The bifurcation diagram of FCSM is shown in Fig. 1c
and d. As can be seen from the figure, FCSM exhibits chaotic characteristics when μ is
over 0.43, and the chaotic value quickly fills the entire value range. Figure 1d further
shows the chaotic behavior of FCSM in the range of 4 to 20. When μ is in a large range,
it still has stable chaotic properties. From the comparative results, it can be seen that
FCSM has better random-like performance and large parameter range than Cubic map
and Sine map.

1.5 1

0.9
1
0.8

0.7
0.5
0.6

0 0.5

0.4
-0.5
0.3

0.2
-1
0.1

-1.5 0
0 0.5 1 1.5 2 2.5 3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) Cubic map (b) Sine map

1 1

0.8 0.8

0.6 0.6

0.4 0.4

0.2 0.2

0 0

-0.2 -0.2

-0.4 -0.4

-0.6 -0.6

-0.8 -0.8

-1 -1
0 0.5 1 1.5 2 2.5 3 3.5 4 4 6 8 10 12 14 16 18 20

(c) FCSM when 0, 4 (d) FCSM when 4, 20

Fig. 1  Bifurcation diagrams of different chaotic systems

13
Multimedia Tools and Applications

2.3.3 Lyapunov Exponent

The Lyapunov Exponent (LE) represents the numerical feature of the average exponential
divergence rate of adjacent trajectories in phase space, and it is one of several numerical
features used to identify chaotic motion [50]. LE is defined by

1 ∑ | � ( )|
N
LE = lim ln |f xn |. (6)
x→∞ N | |
n=0

The LE can be used to visualize whether a system or map is chaotic or not through
images. When LE is more than 0, the system motion enters a chaotic state; when LE = 0,
the system is in a stable state; when LE is less than 0, the motion state of the system tends
to be stable and the system at this time isn’t sensitive to the initial value. Figure 2a and b
shows the LE of Cubic map and Sine map respectively. For the Cubic map, when μ is more
than 2.3, LE reaches more than 0, and it will be accompanied by obvious fluctuations. For
the Sine map, when λ is more than 0.87, LE reaches positive value. value, but the value is
small. Therefore, the chaotic properties of both maps are not ideal. Figure 2c shows the LE
of FCSM, when μ is more than 0.63, LE is stably more than 0. Figure 2d further shows that
when the parameter range is larger, the LE still has excellent performance. Therefore, the
FCSM has better chaotic properties.

2 0

1 -5

-10
0

-15
-1
LE
LE

-20
-2
-25

-3
-30

-4 -35

-5 -40
0 0.5 1 1.5 2 2.5 3 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

(a) Cubic map (b) Sine map

3 5

4
2
3
1
2
0
1
LE
LE

-1 0

-1
-2
-2
-3
-3
-4 -4

-5 -5
0 0.5 1 1.5 2 2.5 3 3.5 4 0 2 4 6 8 10 12 14 16 18 20

(c) FCSM when 0, 4 (d) FCSM when 0, 20

Fig. 2  LE of different chaotic system

13
 Multimedia Tools and Applications

2.3.4 0‑1 test

The 0-1 test is a chaos test method proposed by Georg Gottwald and Ian Melbourne
[10]. Unlike the LE, it doesn’t require phase space reconstruction. For sequences ϕ(n),
n = 1, 2, …, N and random constants r ∈ (0, π), the test value k is defined by
cov(𝜉, 𝛥)
k = corr(𝜉, 𝛥) = √ , 𝜉 = 1, 2, … , n, 𝛥 = M(1), M(2), … , M(n), (7)
var(𝜉)var(𝛥)

where

1 ∑( )( ) 1∑
q q
cov(x, y) = x(i) − x y(i) − y , x = x(i), var(x) = cov(x, x),
q i=1 q i=1

N (
1 ∑ [ ]2 [ ]2 ) 1 − cos(nr)
M(n) = lim p(i + n) − p(i) + q(i + n) − q(i) − (E𝜙)2 ,
N→∞ N
i=1
1 − cos r

and

∑
n
∑
n
1 ∑
N
p(n) = 𝜙(i) cos(ir), q(n) = 𝜙(i) sin(ir), E𝜙 = lim 𝜙(i).
N→∞ N
i=1 i=1 i=1

If k is close to 1, the sequence is chaotic; if k is close to 0, the sequence is non-

chaotic. Taking r ∈ [π/5, 4π/5] to be a random number, N =100, the images obtained by
the 0-1 test after the transformation of the three chaotic sequences with the parameters
are shown in Fig. 3. As can be seen from the figure, FCSM has a flatter curve, smaller
fluctuation and larger parameter range than the other two maps i.e., Cubic map and Sine
map. Therefore, the FCSM has excellent chaotic characteristics.

0.8

0.6

0.4

0.2

0 FCSM
Cubic Map
-0.2 Sine Map

-0.4

-0.6

-0.8

-1
0 0.5 1 1.5 2 2.5 3

Fig. 3  The 0-1 test of different chaotic system

13
Multimedia Tools and Applications

2.3.5 Sensitivity analysis

The sensitivity of initial value is an important characteristic of chaotic systems [43]. In a

qualified chaotic system, any small change of initial value will lead to completely differ-
ent output results, and make the chaotic system unpredictable. Figure 4 shows the chaotic
sequence obtained by the FCSM iterating 40 times with different initial values, where the
red curve is the iteration result of initial values x0=0.1 and x1=0.3, and the blue curve is
the iteration result of initial values x0=0.1 and x1=0.3+10−14. It can be seen that only one
of the two initial values has been slightly changed, the generated sequence trajectory is
completely disordered and difficult to predict. Therefore, the FCSM possesses a favorable
initial value sensitivity.

2.3.6 NIST test

NIST SP 800-22 is a statistical test suite of random number and pseudo-random num-
ber generators [44]. The suite contains 15 test items. Each test item generates a P-value.
When the P-value is more than 0.01, the sequence can be considered uniformly distributed.
The FCSM sequence is generated by random number and transformed into 1­ 07 bit binary
sequence stream. The test results are shown in Table 4. All the tests have passed. There-
fore, the FCSM can be well used to generate random sequences.

2.4 Arnold transform

2.4.1 2D Arnold transform

The Arnold transform, also known as the cat map, was proposed by the Russian mathema-
tician Vladimir Igovich Arnold [23]. It is a coordinate mapping method that repeatedly
folds and stretches in a limited area, and is often used to scramble images. For a matrix
with the size of N×N, the horizontal and vertical coordinates of the pixels are X = {xi} and
Y = {yi}, 2D Arnold transform is defined by

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
0 5 10 15 20 25 30 35 40

Fig. 4  FCSM with different initial values

13
 Multimedia Tools and Applications

Table 4  NIST SP800-22 test Statistical tests P-value Result

Frequency 0.9633 Success

BlockFrequency 0.1901 Success
CumulativeSums 0.9778 Success
Runs 0.7933 Success
LongestRun 0.4225 Success
Rank 0.1469 Success
FFT 0.1957 Success
NonOverlappingTemplate 0.6654 Success
OverlappingTemplate 0.8409 Success
Universal 0.4414 Success
ApproximateEntropy 0.4856 Success
RandomExcursions(x=-4) 0.1547 Success
RandomExcursions(x=4) 0.7812 Success
RandomExcursionsVariant(x=-8) 0.7722 Success
RandomExcursionsVariant(x=8) 0.7059 Success
Serial(p_value1) 0.7341 Success
Serial(p_value2) 0.3995 Success
LinearComplexity 0.3603 Success

[ ] [ ]
X� X
Y�
=A
Y
mod N, (8)

′ ′
where[ X and] Y are the horizontal and vertical coordinates of pixels after Arnold transform,
1 1
A= is the coefficient matrix, and mod (•) represents the modulo operation. The
1 2
result of the Arnold transform on the gray image is shown in Fig. 5. It can be seen from the
figure that when the iteration number is low, the scrambling effect is undesirable. After
many iterations, the image scrambling effect becomes better and better.

(a) Cat (b) 1 time iteration (c) 20 time iterations

Fig. 5  Plain image and its corresponding scrambled images by the 2D Arnold transform

13
Multimedia Tools and Applications

2.4.2 3D Arnold transform

Arnold transform can be extended to three dimensions. For a matrix with the size N×N×N,
the abscissa, ordinate and vertical coordinates of the pixels are X = {xi}, Y = {yi} and Z =
{zi}. The 3D Arnold transform is defined by

⎡ X� ⎤ ⎡X⎤
⎢ Y � ⎥ = A⎢ Y ⎥ mod N, (9)
⎢ �⎥ ⎢ ⎥
⎣Z ⎦ ⎣Z⎦

where X′, Y′ and Z′ are abscissa, ordinate and vertical coordinates of pixels after Arnold
⎡1 1 1⎤
transform, and A = ⎢ 1 2 2 ⎥ is the coefficient matrix.
⎢ ⎥
⎣1 2 3⎦

2.5 3D pre‑storage Arnold transform

Arnold transformation requires multiple iterations to obtain the desirable result, so it is

inefficient. According to mathematical induction, performing n iterations of Arnold trans-
form only requires taking the coefficient matrix to the n-th power and performing modulo
operations. Therefore, to improve the efficiency of Arnold transform, a 3D Arnold trans-
form with the pre-storage coefficient matrix is proposed to change the pixel values of color
images. The designed 3D pre-storage Arnold transform is defined by

⎡ rn+1 ⎤ ⎡ rn ⎤
⎢ gn+1 ⎥ = C{k}⎢ gn ⎥ mod 256, (10)
⎢ ⎥ ⎢ ⎥
⎣ bn+1 ⎦ ⎣ bn ⎦

where r, g and b represent the three-channel pixel value components of the color image, C
represents the cell array of Arnold transform coefficient matrices, and C{k} represents the
coefficient matrix for performing k Arnold iterations. The calculation method of the coef-
ficient set is shown in Eq. (11)
{
A, k = 1
C{k} =
C{k − 1} × A mod 256, k = 2, 3, … T
, (11)

where T is the period of the Arnold transform.

According to Eq. (11), the matrix set required for Arnold transform is generated. Firstly,
the period T of 3D Arnold transform when N = 256 is calculated. To prevent exponential
overflow, the coefficient matrix Ak = mod(Ak − 1 × A, N) required for each Arnold transform
is calculated by iterative method until mod(Ak × A1, N) = E, where E is the unit matrix. The
iteration numbers k is the period T of Arnold transform. When N = 256, T = 447. There-
fore, the key set C with the size of 447 is obtained. The partial coefficients of C are shown
in Fig. 6.
The pre-storage Arnold transform has the following advantages.
(1) Traditional Arnold transform has two decryption methods. One is to continue T-
k iterations, and another is to use the inverse matrix for k iterations to recover the origi-
nal image. Both of these two decryption methods require multiple unnecessary coef-
ficient matrix calculations. Therefore, by pre-storing the coefficient matrix, unnecessary

13
13
C{1} C{2} C{3} … C{445} C{446} C{447}
1 1 1 3 5 6 14 25 31 14 242 5 5 252 1 2 255 0
1 2 2 5 9 11 25 45 56 … 242 19 247 252 6 253 255 2 255
1 2 3 6 11 14 31 56 70 5 247 5 1 253 2 0 255 1

Fig. 6  The pre-storage Arnold coefficient set

Multimedia Tools and Applications
Multimedia Tools and Applications

calculations are avoided, and only A = C{T-k} needs be selected to complete the decryption.
Therefore, the efficiency of Arnold encryption and decryption can be greatly improved.
(2) The pre-storage Arnold coefficient matrix set can be used for any 8-bit pixel value
and multiplexed. At the same time, the matrix set can be used as a key set to make each
pixel perform Arnold transformation for different iteration times, which improves the
encryption effect and randomness. The specific implementation process is shown in
Algorithm 1.
Algorithm 1:  Generation of the pre-storage Arnold coefficient matrix set

2.6 Circular rotation

The circular rotation means that the matrix is divided into layers, and each layer in the
matrix is circulated separately [55]. When a cyclic rotation operation is performed on a
layer, each element in the layer will be moved a number of distances counterclockwise.
Taking a 4×4 matrix as an example, loop 3 bits for the outer layer and 1 bit for the inner
layer, the result is shown in Fig. 7. The results of random cyclic shift on gray images are

Fig. 7  Circular rotation diagram

13
 Multimedia Tools and Applications

(a) Lake (b) Circular rotation

Fig. 8  Plain image and its corresponding scrambled images by the circular rotation

shown in Fig. 8, which shows that this method has great scrambling effect. The specific
implementation process is shown in Algorithm 2.

Algorithm 2:  Circular rotation

3 The proposed image encryption algorithm

3.1 Key generation

SHA-256 is a common message digest algorithm with high security. It is widely used in
digital signature and password encryption. Therefore, the SHA-256 is selected to gener-
ate keys by combining with the external key E = {e1, e2, ⋯e8}.

13
Multimedia Tools and Applications

The SHA-256 hash value of the plain image is K = {k1, k2, ⋯k32}, where ki is an 8-bit
decimal sequence. Four pairs of initial values of FCSM {w0, w1}, {x0, x1}, {y0, y1} and
{z0, z1} are generated by

⎧ (k1 +k2 )⊕(k3 +k4 )

⎪ w0 = −1 + 29 + mod(e1 ,1)
⎨ (k5 +k6 )⊕(k7 +k8 ) , (12)
⎪ w1 = 1 − 29 + mod(e ,1)
⎩ 2

⎧ (k9 +k10 )⊕(k11 +k12 )

⎪ x0 = −1 + 29 + mod(e3 ,1)
⎨ (k13 +k14 )⊕(k15 +k16 ) , (13)
⎪ x1 = 1 − 29 + mod(e ,1)
⎩ 4

⎧ (k17 +k18 )⊕(k19 +k20 )

⎪ y0 = −1 + 29 + mod(e5 ,1)
⎨ (k21 +k22 )⊕(k23 +k24 ) , (14)
⎪ y1 = 1 − 29 + mod(e ,1)
⎩ 6

⎧ (k25 +k26 )⊕(k27 +k28 )

⎪ z0 = −1 + 29 + mod(e7 ,1)
⎨ (k29 +k30 )⊕(k31 +k32 ) . (15)
⎪ z1 = 1 − 29 + mod(e ,1)
⎩ 8

The control parameters of the FCSM are generated by

⎧𝜇 = e1 + e 2 +
k1 ⊕k5 ⊕⋯⊕k29
⎪ 1 28
k2 ⊕k6 ⊕⋯⊕k30
⎪ 𝜇2 = e3 + e 4 +
⎨ 28
k3 ⊕k7 ⊕⋯⊕k31 . (16)
⎪ 𝜇3 = e5 + e 6 + 28
⎪ 𝜇4 = e7 + e 8 +
k4 ⊕k8 ⊕⋯⊕k32
⎩ 28

3.2 Encryption process

All the symbols and variables used in the encryption process are shown in Table 5. The
specific encryption steps are described as follows.

Step 1: Chaotic sequence generation

The plain image I is a color image with the size of m×n×3. Using the key generated
in Subsection 3.1, four chaotic sequences S1, S2, S3 and S4 can be generated by iterat-
ing the FCSM as shown in Eq. (17). Discard the first 2000 values of each sequence to
ensure better randomness. Among them, S1, S2, S3, S4 are used for row & column scram-
bling, circular rotation scrambling, determining the numbers of Arnold transform, and
generating random numbers in the diffusion phase respectively.

13
 Multimedia Tools and Applications

Table 5  Symbols in the Symbols Definitions

encryption process
S1, S2, S3, S4 Chaotic sequences
I Plain image
m, n Length and width of I
index1, index2 Index sequences
D’ Intermediate image
D” Row & column scrambled image
D Components exchanged image
D2 Circular rotation scrambled image
C Coefficient set of Arnold transform
R Encrypted image

⎧ � �
⎪ S1 (i) = FCSM �w0 , w1 , 𝜇1� , i = 1, 2, … , (m + n) × 3
⎪ S2 (i) = FCSM x0 , x1 , 𝜇2 , i = 1, 2, … , min (m, n) × 3∕2
⎨ S (i) = FCSM �y , y , 𝜇 �, i = 1, 2, … , m × n . (17)
⎪ 3 � 0 1 3�
⎪ S4 (i) = FCSM z0 , z1 , 𝜇4 , i = 1, 2, … , m × n × 3
⎩

Step 2: Chaotic sequence normalization

Perform the integer operations on S2, S3 and S4 by

⎧ S = mod �floor�S × 1014 �, 2mn� + 1

⎪ 2 � �2 � �
⎨ S3 = mod floor
� S�3 × 1014 , �447 +�1 , (18)
⎪ S4 = mod floor S4 × 1014 , 256
⎩

where floor(•) means rounding down the decimal. After that, S3 and S4 are reshaped to
obtain two-dimensional matrices by
{ ( )
S3 = reshape( S3 , m, n )
S4 = reshape S4 , m, n, 3
, (19)

where reshape(•) is the dimension transformation function.

Step 3: Row & column shift

The chaotic sequence S1 is divided into three groups with the same length of m + n,
where each group is used to scramble one layer of RGB component. In each group, the
first m values and the next n values are sorted respectively to obtain two index sequences
index1 and index2. The rows and columns of corresponding component are sorted accord-
ing to index1 and index2 respectively. After three groups of operations, the row & column
scrambled image D” can be obtained. The scrambling equation of each group is shown in
Eq. (20)
{ �
D (i, ∶) = I(index1(i), ∶)
D�� (∶, j) = D� (∶, index2(j))
, i = 1, 2, … , m;j = 1, 2, … , n, (20)

where D’ is the intermediate matrix after the row shift.

Step 4: Interlaminar displacement

13
Multimedia Tools and Applications

To hide the chroma information of plaintext, RGB components are exchanged to different
layers by Eq. (21)

⎧ D(r) = D�� (g)

⎪ ��
⎨ D(g) = D (b) . (21)
⎪ D(b) = D�� (r)
⎩

Step 5: Circular rotation

For further scrambling the image, D is scrambled by the circular rotation introduced in
Subsection 2.6. S2 is used to determine the shift length of each layer in D, and obtain the
scrambled image D2.

Step 6: Pre-storage Arnold transform

Arnold transform is performed on the RGB pixel values of D2, and the transform times of
each pixel are determined by S3. To enhance the resistance to the selective plaintext attack,
S4 is used to increase the randomness of the pixel values. Finally, the encrypted image R is
obtained. The specific operation is shown in Eq. (22).

Fig. 9  Encryption flowchart

13
 Multimedia Tools and Applications

⎡ R(i, j, r) ⎤ � �⎡ D2 (i, j, r) ⎤ ⎡ S4 (i, j, 1) ⎤

⎢ R(i, j, g) ⎥ = C S (i + j) ⎢ D2 (i, j, g) ⎥ + ⎢ S4 (i, j, 2) ⎥mod 256, (22)
⎢ ⎥ 3
⎢ ⎥ ⎢ ⎥
⎣ R(i, j, b) ⎦ ⎣ D2 (i, j, b) ⎦ ⎣ S4 (i, j, 3) ⎦

where C is the coefficient set proposed in Subsection 2.5. The complete encryption flow-
chart is shown in Fig. 9. Algorithm 3 shows the pseudo-code of the whole encryption
process.
Algorithm 3:  Encryption process

3.3 Decryption process

The decryption process is the reverse of the encryption process. The specific process is
described as follows.

Step 1: Chaotic sequence generation

Using the received secret keys {w0, w1, μ1}, {x0, x1, μ2}, {y0, y1, μ3} and {z0, z1, μ4}, four cha-
otic sequences S1, S2, S3, S4 can be generated using the same method according to Eqs. (17)
and (19).

13
Multimedia Tools and Applications

Step 2: Pre-storage Arnold transform

According to S3, S4 and the received coefficient set C, Arnold transform is used to inversely
diffuse the encrypted image R to obtain D2, as shown in Eq. (23).

⎡ D2 (i, j, r) ⎤ � �⎡ E(i, j, r) − S4 (i, j, 1) ⎤

⎢ D2 (i, j, g) ⎥ = C T − S (i + j) ⎢ E(i, j, g) − S4 (i, j, 2) ⎥ mod 256. (23)
⎢ ⎥ 3
⎢ ⎥
⎣ D2 (i, j, b) ⎦ ⎣ E(i, j, b) − S4 (i, j, 3) ⎦

Step 3: Circular rotation

To obtain the semi scrambled image D, the circular rotation is used in a clockwise direc-
tion for D2, and S2 is used to determine the shift length of each layer.

Step 4: Interlaminar displacement

R, G and B components of D are exchanged to recover the original layer position

according to Eq. (24)

⎧ D� (r) = D(b)
⎪ �
⎨ D (g) = D(r) . (24)
⎪ D� (b) = I(g)
⎩

Step 5: Row & column shift

The chaotic sequence S1 is divided into three groups with the same length of m + n.
where each group is used to scramble one layer of R, G, B components. In each group,
the first m values and the next n values are sorted respectively to obtain two sequences
index1 and index2. The rows and columns of corresponding component are sorted by the
sequences respectively. After three groups of operations, the decrypted image I can be
obtained. The scrambling equation of each group is shown in Eq. (25)
{ �
I (∶, j) = D� (;, index2(j)), j = 1, 2, … , N
I(i, ∶) = I � (index1(i), ∶), i = 1, 2, … , M
, (25)

where I’ is the intermediate variable after the column shift.

4 Experimental results

To protect the security of the transmitted information, the proposed algorithm is used to
encrypt the plaintext image. The relevant environments and configurations used in this
experiment are shown in Table 6. Figure 10 shows the images of each processing stage of
the House. It can be seen that the plaintext information has been hidden after the row &
column shift and the circular rotation. After the diffusion with Arnold transform, the final
encrypted image is an undifferentiated white noise image. When the image is transmitted
in the channel, even if the image is compromised, the attacker cannot obtain any plaintext
information. Therefore, the information security has been effectively protected.

13
 Multimedia Tools and Applications

Table 6  Experimental environments and configurations

Experimental environment MATLAB R2020a installed over the personal computer with i5-6300HQ CPU
and memory of 16 GB, and an operating system of Window 10 Enterprise.

Initial parameters {w0 = 0.800280468888800, w1 = 0.141886338627215, u1 =

27.849821886704838},
{x0 = -0.421761282626275, x1 = 0.915735525189067, u2 =
54.688151920498385},
{y0 = 0.792207329559554, y1 = -0.655740699156587, u3 =
15.761308167754828},
{z0 = 0.849129305868777, z1 = -0.933993247757551, u4 =
95.750683543429759}.

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

Fig. 10  Encryption process: a Plaintext image; b The semi-encrypted image after row column shift; c The
semi-encrypted image after circular rotation; d Cipher image

Figure 11 shows the results of encryption and decryption of Airplane, Baboon, Peppers,
and Plain black. It can be seen that the encrypted images are irregular snowflakes regard-
less of the content presented by the plain images. Therefore, the proposed algorithm has
generality. Besides, the plain images can be decrypted without distortion after transmis-
sion, which guarantees the normal access of the recipient.

5 Algorithm analyses

5.1 Key space analysis

For an excellent encryption algorithm, the key space must be at least more than 2­ 100 for
resisting the brute force attack [34]. In our algorithm, four chaotic systems are generated
for the encryption algorithm. Each chaotic system needs 2 initial values ranging from - 1
to 1. For the accuracy of the computer ­10-15, the key space is 2 × 8 × ­1015 ≈ ­2400. It is much
greater than 2­ 100. In addition, the value range of chaotic parameters is (0, +∞), so it is
impossible to break the encrypted image by the brute-force attack.

5.2 Histogram analysis

Histogram distribution analysis is a method of statistical analysis [22]. The more uni-
form the histogram distribution is, the better the encryption effect is, and the more

13
Multimedia Tools and Applications

(a) (b) (c)

(d) (e) (f)

(g) (h) (i)

(j) (k) (l)

Fig. 11  Results of encryption and decryption: a Airplane; b Encrypted image of (a); c Decrypted image of
(b); d Baboon; e Encrypted image of (d); f Decrypted image of (e); g Peppers; h Encrypted image of (g); i
Decrypted image of (h); j Plain black; k Encrypted image of (j); l Decrypted image of (k)

difficult it is to obtain any information from the encrypted image. Figure 12 shows the
plaintext and encrypted histograms of Airplane, Baboon and Peppers. It can be seen that
the histogram distribution after encryption is very stable and difficult to obtain any sta-
tistical information of the plain image.

13
 Multimedia Tools and Applications

12000 2500

10000
2000

8000
1500

6000

1000
4000

500
2000

0 0
0 50 100 150 200 250 0 50 100 150 200 250

(a) (b)
3000 2500

2500
2000

2000
1500

1500

1000
1000

500
500

0 0
0 50 100 150 200 250 0 50 100 150 200 250

(c) (d)
15000 2500

2000

10000
1500

1000
5000

500

0 0
0 50 100 150 200 250 0 50 100 150 200 250

(e) (f)

Fig. 12  Histogram analysis results: a Airplane; b Encrypted of (a); c Baboon; d Encrypted of (c); e Pep-
pers; f Encrypted of (e)

5.3 Key sensitivity analysis

Key sensitivity refers to that a small change of key leads to different encryption results
[12]. Figure 13 shows the encrypted image of Fig. 10 and the decrypted image obtained
after a slight change of one parameter. Figure 13b is the result of changing the initial value
­y0 to ­y0+10-14, Fig. 13c is the result of changing initial value ­z0 to ­z0+10-14. As shown

13
Multimedia Tools and Applications

(a) (b) (c)

2500 2500 2500

2000 2000 2000

1500 1500 1500

1000 1000 1000

500 500 500

0 0 0
0 50 100 150 200 250 0 50 100 150 200 250 0 50 100 150 200 250

(d) (e) (f)

2500 2500

2000 2000

1500 1500

1000 1000

500 500

0 0
0 50 100 150 200 250 0 50 100 150 200 250

(g) (h) (i) (j)

Fig. 13  Key sensitivity test results: a Encrypted image of house; b Decrypting (a) using incorrect ­y0; c
Decrypting (a) using incorrect ­z0; d Histogram of (a); e Histogram of (b); f Histogram of (c); g |(b) - (a)|; h
Histogram of (g); i |(c) – (a)|; j Histogram of (h)

in the figure, when the initial value is slightly transformed, the original image cannot be
restored correctly. Figure 13g and i show the differences between the decrypted image and
the encrypted image using the wrong key. It can be inferred that the images encrypted with
different keys are completely different. Therefore, it can be concluded that the proposed
algorithm has excellent key sensitivity.

5.4 Correlation analysis

A major characteristic of digital images is the high correlation of neighboring pixels. An

excellent encryption algorithm must sufficiently reduce the correlation of adjacent pixels.
So that an attacker cannot infer any plaintext information from the encrypted image. The
definition of correlation is defined by [44].

13
 Multimedia Tools and Applications

cov(x, y)
rxy = √ , (26)
D(x)D(y)
N �
∑ �� � N �
∑ �2 1 ∑
N

where cov(x, y) = 1
N
xi − E(x) yi − E(y) , D(x) = 1
N
xi − E(x) , and E(x) = N
i=1
xi.
i=1 i=1
Table 7 shows the horizontal, vertical and diagonal direction correlation coefficients
of the images before and after encryption using the proposed algorithm and other related
algorithms. It can be seen that after encryption by the proposed algorithm, the correlation
coefficient of the image is close to 0. It indicates that there is almost no correlation between

Table 7  Correlation coefficients Algorithm Image Horizontal Vertical Diagonal Average

of different images
Proposed Airplane R 0.0061 0.0010 0.0010 0.0024
G -0.0018 0.0026 0.0028
B 0.0016 0.0007 -0.0044
Baboon R -0.0045 -0.0015 0.0013 0.0019
G -0.0038 -0.0043 0.0001
B 0.0008 0.0005 0.0007
Peppers R -0.0016 -0.0020 -0.0012 0.0017
G 0.0037 0.0020 -0.0005
B 0.0019 -0.0027 0.0000
Ref. [13] Baboon R -0.0213 0.0072 0.0011 0.0091
G 0.0126 0.0120 -0.0133
B -0.0102 0.0015 0.0025
Peppers R -0.0145 0.0149 0.0393 0.0165
G 0.0149 0.0136 0.0067
B 0.0393 -0.0019 -0.0038
House R -0.0028 0.0221 -0.0299 0.0127
G 0.0151 0.0058 0.0220
B -0.0047 -0.0095 0.0026
Ref. [47] Lena R -0.0060 -0.0049 0.0070 0.0043
G 0.0025 -0.0051 0.0020
B 0.0046 0.0019 0.0047
Baboon R 0.0052 -0.0042 0.0008 0.0030
G 0.0047 -0.0071 0.0009
B -0.0003 -0.0013 -0.0028
Peppers R -0.0011 0.0015 0.0066 0.0033
G -0.0015 0.0059 0.0028
B -0.0042 0.0037 0.0025
Ref. [40] House -0.0265 -0.0112 -0.0282 0.0022
Male -0.0007 -0.0374 -0.0052 0.0144
Baboon
Ref. [56] 0.0006 0.0046 0.0021 0.0024
Peppers
Lena 0.0018 0.0010 0.0018 0.0015
Ref. [42] Woman 0.0020 -0.0061 -0.0004 0.0028
-0.0021 0.0012 -0.0033 0.0022

13
Multimedia Tools and Applications

250 250

200 200

150 150

100 100

50 50

0 0

200 200
B B
100 G 100 G

0 R 0 R

(a) (b)

250

200

150

100

50

200
B
100 G

0 R

(c) (d)

250 250

200 200

150 150

100 100

50 50

0 0

200 200
B B
100 G 100 G

0 R 0 R

(e) (f)

Fig. 14  Adjacent pixels correlation of Airplane: a Plaintext’s horizontal correlation; ) Ciphertext’s horizon-
tal correlation; c Plaintext’s vertical correlation; d Ciphertext’s vertical correlation; e Plaintext’s diagonal
correlation; f Ciphertext’s diagonal correlation

adjacent pixels, so the plaintext information can be effectively hidden. Therefore, our algo-
rithm has better encryption effect compared with other algorithms.
Figure 14 shows the distribution of adjacent pixels on the horizontal, vertical and diago-
nal directions of the R, G, B components before and after the encryption. It can be clearly
seen that the correlated pixels in the encrypted image are evenly distributed throughout the
coordinate system, so the correlation of the plaintext image is completely broken up.

13
 Multimedia Tools and Applications

5.5 Information entropy

Shannon’s information theory is used to reflect the uncertainty of information. As we

all know, the closer the probability distribution of image pixel values is to uniform dis-
tribution, the higher the information entropy is. When all pixel values are equal, the
information entropy of the image is maximized. The calculation of information entropy
is defined by [12]

∑
255
( ) ( )
E(x) = − P xi log2 P xi , (27)
i=0

where P(xi) represents the probability of occurrence of pixel value xi. For images, the
closer E is to 8, the better the encryption effect of the encryption algorithm. Table 8 shows
the information entropy comparison between the proposed algorithm and other algorithms.
According to the table, it can be concluded that the proposed algorithm has higher infor-
mation entropy and encryption stability than other algorithms.

5.6 Differential attack

The Number of Pixels Change Rate (NPCR) is the rate of change in the number of cipher-
text pixels when the specified text changes slightly. The ideal value of NPCR is 99.6094%
[13]. The closer the NPCR is to the ideal value, the more sensitive the ciphertext is to
changes in the plaintext, and the stronger it is to resist plaintext attacks. The Uniform

Table 8  Entropy of different Algorithm Image Ciphertext

images entropy
(bit)

Proposed Airplane 7.9998

Baboon 7.9998
Peppers 7.9998
House 7.9998
Ref. [7] Lena 7.9991
Jasmin 7.9991
Baboon 7.9992
Ref. [29] Airplane 7.9988
Peppers 7.9989
Baboon 7.9989
Ref. [12] Airplane 7.9993
Baboon 7.9993
House 7.9993
Ref. [15] Airplane 7.9973
Baboon 7.9974
Peppers 7.9974
Ref. [47] Lena 7.9970
Baboon 7.9970
Peppers 7.9966

13
Multimedia Tools and Applications

Average Change Intensity (UACI) is the average intensity change rate of the corresponding
pixels in the specified text and cipher text. The ideal value of UACI is 33.4635% [13]. The
closer the UACI is to the ideal value, the stronger it is to resist differential attacks. NPCR
and UACI are calculated by Eqs. (28) and (29)

1 ∑∑
M N
NPCR = W(i, j) × 100%, (28)
M × N i=1 j=1

1 ∑∑ M N
UACI = P (i, j) − P2 (i, j) × 100%, (29)
M × N × 255 i=1 j=1 1

where P1{ , P2 are the encrypted images before and after a tiny change of plaintext,
0, P1 (i, j) = P2 (i, j)
W(i, j) = . The test results of NPCR and UACI are shown in Table 9.
1, P1 (i, j) ≠ P2 (i, j)
As shown in the table, it can be concluded that the proposed algorithm has higher ability to
resist the differential attack than other algorithms.

5.7 Cropping attacks

The clipping attack is one of the methods to test the robustness [45]. In the transmission
process, some parts of the images may be damaged. The cropping attack shows the abil-
ity of encrypted image to restore the original image after cropping. Figure 15 shows the
decryption effect of the Baboon after it has been cropped by 1/8, 1/4 and 1/2 respectively. It
can be seen that the recovered image is still recognizable when half of the encryption image
is damaged. Therefore, the proposed algorithm has desirable anti-cutting attack ability.

Table 9  NPCR and UACI values Algorithm Image NPCR (%) UACI (%)
of different images
Proposed Airplane 99.6029 33.4672
Peppers 99.6096 33.4755
Baboon 99.6063 33.4657
House 99.6195 33.4738
Ref. [14] Car 99.5981 33.4003
Couple 99.6098 33.4621
Ref. [4] Lena 99.6310 33.4963
Peppers 99.6074 33.4120
Ref. [47] Lena 99.6225 33.5320
Baboon 99.5422 33.4614
Peppers 99.5174 33.4682
Ref. [31] Lena 99.6148 33.4478
Baboon 99.6366 33.4586
House 99.5712 33.4577
Ref. [32] Lena 99.6951 33.3836
Baboon 99.7126 33.3964
Peppers 99.6438 33.4138

13
 Multimedia Tools and Applications

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

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

Fig. 15  Clipping attack results: a Encrypted image of baboon; b Crop 1/8 of (a); c Crop 1/4 of (a); d Crop
1/2 of (a); e Decrypted image of (a); f Decrypted image of (b); g Decrypted image of (c); h Decrypted
image of (d)

5.8 Noise attack

The Noise attack is common test method to test robustness. In the process of transmis-
sion, the image may be attacked by noise, in which the salt and pepper noise has the
strongest impact on the image effect [44]. Figure 16 simulates the decrypted image after
adding density of 0.05, 0.1 and 0.2 to the encrypted image. As shown in the figure, the
decrypted image is still recognizable when strong noise is added. Therefore, the pro-
posed algorithm can resist the noise attack.

5.9 Encryption speed analysis

The execution speed of the encryption algorithm is an important indicator to evaluate the
encryption algorithm. Encrypting a color image with the size of m × n × 3, the time com-
plexity of each part is calculated as follows:

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

Fig. 16  Noise attack results: a Decrypted peppers without noise; b Decrypted peppers with 0.05 noise den-
sity; c Decrypted peppers with 0.1 noise density; d Decrypted peppers with 0.2 noise density

13
Multimedia Tools and Applications

(1) Generation of chaotic sequences: The algorithm generates four chaotic sequences
whose lengths are (m + n) × 3, min(m, n) × 3/2, m × n and m × n × 3. Therefore, the total
time complexity is O(4 × m × n);
(2) Row & column shift: the row shift is performed m×3 times, and the column shift is
performed n×3 times. Therefore, the total time complexity is O(3 × (m + n));
(3) Circular rotation: The number of executions of the circular rotation is min(m, n) × 3/2.
Therefore, the total time complexity is O(min(m, n) × 3/2);
(4) Arnold transformation: For 3D matrix multiplication, each pixel performs 9 times
of multiplication, with a total of 9×3×M×N. Therefore, the total time complexity is
O(27 × m × n).

To sum up, the time complexity of the proposed algorithm is O(31 × m × n). The experi-
ment is done with MATLAB R2020a installed over the personal computer with i5-6300HQ
CPU and memory of 16 GB, and Window 10 Enterprise operating system. The encryption
execution speed is shown in the Table 10. It can be seen from the table that the proposed
algorithm has significant speed advantages compared with other algorithms.

6 Discussion

To improve the security and efficiency, we propose an algorithm using efficient row &
column shift and circular rotation for scrambling operations. Unlike the previous use of
Arnold transform to perform the scrambling operation, our algorithm uses Arnold trans-
form to perform the diffusion operation on the RGB component pixel values of the color
image. The proposed algorithm can execute different times of Arnold transform for each
pixel value, which greatly improves the key space of Arnold transform, solves its drawback
of short periodicity, and is very efficient.
The proposed algorithm was tested through Section Experimental results and our algo-
rithm can encrypt and decrypt images without distortion. In Section Algorithm analyses,
we performed several tests on the algorithm, including statistical test, encryption quality,
sensitivity test, robustness test, encryption speed test, etc. According to the above experi-
ments and analyses, it is clear that the proposed algorithm has excellent security and higher

Table 10  Encryption speed of Algorithm Image Size Speed (second)

different images
Proposed Airplane 512×512×3 0.6321
Baboon 512×512×3 0.6653
Peppers 512×512×3 0.6694
Ref. [5] Lena 256×256×3 1.1620
Ref. [46] Lena 512×512×3 1.4990
Ref. [51] -- 512×512×3 1.5803
Ref. [15] Baboon 512×512 5.7855
Ref. [29] Airplane 512×512 1.5044
Baboon 512×512 1.5069
Peppers 512×512 1.5813

13
 Multimedia Tools and Applications

efficiency compared with similar algorithms. Therefore, it can meet the requirement of fast
image encryption and be real-time encryption systems.

7 Conclusions and outlooks

In this paper, a new FCSM is proposed to solve the problems of poor chaotic properties
and small parameter range of Cubic map and Sine map. Experimental results show that the
proposed FCSM has excellent chaotic properties and is very suitable for generating random
sequences. In addition, to solve the problems of low iteration efficiency and periodicity of
Arnold transform, a pre-storage Arnold transform based on the idea of trading off memory
for time is proposed. Meanwhile, a fast color image encryption algorithm is designed using
the proposed chaotic system and pre-storage Arnold transform. According to the encryp-
tion quality analysis, statistical analysis and robustness analysis, the proposed algorithm
has excellent encryption effect and encryption efficiency. Therefore, it can be applied to
real-time encryption system.
The proposed algorithm can be improved from the following aspects in the future.

(1) The proposed algorithm focuses on encrypting the single-color image at present. We
can extend it to encrypt multiple color images in the future.
(2) The proposed FCSM has excellent pseudo randomness and high iterative efficiency.
We can further enhance its chaotic performance by extending it to high-dimensional
chaotic systems while maintaining its efficiency.
(3) The use of the Arnold transform for diffusion operation has excellent results
and fully incorporates the correlation between each component of color images.
Due to the idea of pre storage, we can use hardware to realize Arnold transform.
Moreover, the proposed algorithm also can be implemented in hardware to further
speed up.

Data availability Some or all of the data generated or used in this study are available from the correspond-
ing author (Xiaoqiang Zhang, [email protected]) upon reasonable request.

Declarations
Conflicts of Competing interests The authors declare that they have no known competing financial interests
or personal relationships that could have appeared to influence the work reported in this paper.

References
1. Ali TS, Ali R (2020) A new chaos based color image encryption algorithm using permutation substitu-
tion and Boolean operation. Multimed Tools Appl 79:19853–19873
2. Chen L, Yin H, Yuan L et al (2020) A novel color image encryption algorithm based on a fractional-order dis-
crete chaotic neural network and DNA sequence operations. Front Inf Technol Electron Eng 21:866–879
3. Chen Y, Xie S, Zhang J (2022) A novel double image encryption algorithm based on coupled chaotic system.
Phys Scr 97:65207

13
Multimedia Tools and Applications

4. Duan C, Zhou J, Gong L et al (2022) New color image encryption scheme based on multi-parameter
fractional discrete Tchebyshev moments and nonlinear fractal permutation method. Opt Lasers Eng
150:106881–106894
5. Fang J, Jiang M, Yin N et al (2022) An image block encryption algorithm based on hyperchaotic system and
DNA encoding. Multimed Tools Appl 81:17245–17262
6. Farah MA, Farah A, Farah T (2020) An image encryption scheme based on a new hybrid chaotic map and
optimized substitution box. Nonlinear Dyn 99:3041–3064
7. Farah MB, Guesmi R, Kachouri A et al (2020) A novel chaos based optical image encryption using fractional
Fourier transform and DNA sequence operation. Opt Laser Technol 121:105777–105785
8. Farwa S, Muhammad N, Shah T et al (2017) A novel image encryption based on algebraic S-box and Arnold
transform. 3D Res 8:1-14
9. Ghazvini M, Mirzadi M, Parvar N (2020) A modified method for image encryption based on chaotic map and
genetic algorithm. Multimed Tools Appl 79:26927–26950
10. Gottwald G A, Melbourne I (2016) The 0-1 test for chaos: A review. Chaos detection and predictability
915:221-247
11. Guleria V, Mishra DC (2020) A new multi-layer RGB image encryption algorithm based on Diffie-Hell-
man cryptography associated with FrDCT and arnold transform. Multimed Tools Appl 79:33119–33160
12. Hosny KM, Kamal ST, Darwish MM (2022) Novel encryption for color images using fractional-order
hyperchaotic system. J Ambient Intell Humaniz Comput 13:973–988
13. Hosny KM, Kamal ST, Darwish MM (2022) A color image encryption technique using block scrambling
and chaos. Multimed Tools Appl 81:505–525
14. Hu C, Xie X, Zhou N (2022) Colour image encryption scheme based on the real-valued discrete Gabor
transform. J Mod Opt 69:511–522
15. Huang Z, Zhou N (2022) Image encryption scheme based on discrete cosine Stockwell transform and
DNA-level modulus diffusion. Opt Laser Technol 149:107879
16. Huang H, He Y, Yang S et al (2020) Chaotic image encryption based on bidimensional empirical mode
decomposition and double random phase encoding. Multimed Tools Appl 79:28065–28078
17. Jain K, Aji A, Krishnan P (2021) Medical image encryption scheme using multiple chaotic maps. Pattern
Recogn Lett 152:356–364
18. Jiang N, Wu W, Wang L (2014) The quantum realization of Arnold and Fibonacci image scrambling.
Quant Info Proc 13:1223–1236
19. Joshi AB, Kumar D, Mishra DC et al (2020) Colour-image encryption based on 2D discrete wavelet trans-
form and 3D logistic chaotic map. J Mod Opt 67:933–949
20. Kaur M, Kumar V, Li L (2019) Color image encryption approach based on memetic differential evolution.
Neural Comput Appl 31:7975–7987
21. Li H, Deng L, Gu Z (2020) An image encryption scheme based on precision limited chaotic system. Mul-
timed Tools Appl 79:19387–19410
22. Li S, Zhao L, Yang N (2021) Medical image encryption based on 2D Zigzag confusion and dynamic diffu-
sion. Secur Commun Netw 2021:6624809–6624829
23. Liu Q, Liu L (2020) Color image encryption algorithm based on DNA coding and double chaos system.
IEEE Access 8:83596–83610
24. Liu Y, Zhang J (2020) A multidimensional chaotic image encryption algorithm based on DNA coding.
Multimed Tools Appl 79:21579–21601
25. Malik DS, Shah T (2020) Color multiple image encryption scheme based on 3D-chaotic maps. Math
Comput Simul 178:646–666
26. Mansouri A, Wang X (2020) A novel one-dimensional sine powered chaotic map and its application in a
new image encryption scheme. Inf Sci 520:46–62
27. Mansouri A, Wang X (2021) Image encryption using shuffled Arnold map and multiple values manipula-
tions. Vis Comput 37:189–200
28. Mondal B, Mandal T (2020) A secure image encryption scheme based on genetic operations and a new
hybrid pseudo random number generator. Multimed Tools Appl 79:17497–17520
29. Mondal B, Singh JP (2022) A lightweight image encryption scheme based on chaos and diffusion circuit.
Multimed Tools Appl 81:34547–34571
30. Niu Y, Zhou Z, Zhang X (2020) An image encryption approach based on chaotic maps and genetic opera-
tions. Multimed Tools Appl 79:25613–25633
31. Patel S, Thanikaiselvan V, Pelusi D et al (2021) Colour image encryption based on customized
neural network and DNA encoding. Neural Comput & Applic 33:14533–14550
32. Pourjabbar Kari A, Habibizad Navin A, Bidgoli AM et al (2021) A novel multi-image cryptosystem based
on weighted plain images and using combined chaotic maps. Multimedia Systems 27:907–925
33. Prasad B (2021) Dual complex Fibonacci p-numbers. Chaos, Solitons Fractals 145:109922

13
 Multimedia Tools and Applications

34. Sheela SJ, Sanjay A, Suresh KV et al (2022) Image encryption based on 5D hyperchaotic system using
hybrid random matrix transform. Multidim Syst Sign Process 33:579–595
35. Singh P, Yadav AK, Singh K (2017) Phase image encryption in the fractional Hartley domain using
Arnold transform and singular value decomposition. Opt Lasers Eng 91:187–195
36. Talhaoui MZ, Wang X (2021) A new fractional one dimensional chaotic map and its application in high-
speed image encryption. Inf Sci 550:13–26
37. Talhaoui MZ, Wang X, Midoun MA (2021) Fast image encryption algorithm with high security level
using the Bülban chaotic map. J Real-Time Image Proc 18:85–98
38. Tora H, Gokcay E, Turan M et al (2022) A generalized Arnold’s Cat Map transformation for image scram-
bling. Multimed Tools Appl 81:31349–31362
39. Vallabhaneni M, Maddula B, Jayan S et al (2022) Chaotic Hooke-Jeeves Algorithm using Cubic map with
MATLAB code. 2020 IEEE international conference for innovation in technology (INOCON), pp 1-4
40. Vidhya R, Brindha M (2022) A novel approach for chaotic image encryption based on block level permu-
tation and bit-wise substitution. Multimed Tools Appl 81:3735–3772
41. Wan Y, Wang S, Du B (2022) A bit plane image encryption algorithm based on compound chaos. Mul-
timed Tools Appl 6:1–19
42. Wang X, Liu P (2021) A new full chaos coupled mapping lattice and its application in privacy image
encryption. IEEE Trans Circ Syst I Regul Pap 69:1291–1301
43. Wang X, Zhang M (2021) An image encryption algorithm based on new chaos and diffusion values of a
truth table. Inf Sci 579:128–149
44. Wang X, Guan N, Yang J (2021) Image encryption algorithm with random scrambling based on one-
dimensional logistic self-embedding chaotic map. Chaos, Solitons Fractals 150:111117–111137
45. Wang X, Liu C, Jiang D (2021) A novel triple-image encryption and hiding algorithm based on chaos,
compressive sensing and 3D DCT. Inf Sci 574:505–527
46. Wang X, Su Y, Liu L et al (2021) Color image encryption algorithm based on Fisher-Yates scrambling and
DNA subsequence operation. Vis Comput 10:1–16
47. Wang X, Su Y, Luo C et al (2022) Color image encryption algorithm based on hyperchaotic system and
improved quantum revolving gate. Multimed Tools Appl 81:13845–13865
48. Wu J, Liu Z, Wang J et al (2021) A compact image encryption system based on Arnold transformation.
Multimed Tools Appl 80:2647–2661
49. Xu C, Sun J, Wang C (2020) A novel image encryption algorithm based on bit-plane matrix rotation and
hyper chaotic systems. Multimed Tools Appl 79:5573–5593
50. Yang F, An X (2022) A new discrete chaotic map application in image encryption algorithm. Phys Scr
97:35202
51. Yang S, Tong X, Wang Z et al (2022) Efficient color image encryption algorithm based on 2D coupled
chaos and multi-objective optimized S-box. Phys Scr 97:45204–45231
52. Yousif SF, Abboud AJ, Alhumaima RS (2022) A new image encryption based on bit replacing, chaos and
DNA coding techniques. Multimed Tools Appl 81:27453–27493
53. Zefreh EZ (2020) An image encryption scheme based on a hybrid model of DNA computing, chaotic sys-
tems and hash functions. Multimed Tools Appl 79:24993–25022
54. Zhang Y (2020) The fast image encryption algorithm based on lifting scheme and chaos. Inf Sci
520:177–194
55. Zhang X, Yang X (2022) Color image encryption algorithm based on 3D spiral transform and radial diffu-
sion. Phys Scr 97:95210
56. Zhou W, Wang X, Wang M et al (2022) A new combination chaotic system and its application in a new
Bit-level image encryption scheme. Opt Lasers Eng 149:106782–106797

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and
institutional affiliations.

Springer Nature or its licensor (e.g. a society or other partner) holds exclusive rights to this article under
a publishing agreement with the author(s) or other rightsholder(s); author self-archiving of the accepted
manuscript version of this article is solely governed by the terms of such publishing agreement and applicable
law.

13
View publication stats