Received 9 October 2022, accepted 22 October 2022, date of publication 26 October 2022, date of current version 7 November 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3217520

A Novel Chaotic Image Encryption Algorithm

Based on Coordinate Descent and SHA-256
XIYU SUN AND ZHONG CHEN
College of Computer Science and Technology, Hengyang Normal University, Hengyang 421002, China
Hunan Provincial Key Laboratory of Intelligent Information Processing and Application, Hengyang 421002, China
Corresponding author: Zhong Chen ([email protected])
This work was supported in part by the Scientific Research Fund of Hunan Provincial Education Department under Grant 19A066, and in
part by the Science and Technology Innovation Program of Hunan Province under Grant 2016TP1020.

ABSTRACT In recent years, chaotic image encryption algorithms with key and plaintext association have
been developed, which are essentially similar to a one-time pad at a time because each encryption requires
the transmission of the key. However, some existing schemes cannot uniquely map the seed key to the
initial value of the chaotic system, which leads to the reduction of the key space of the encryption system.
In addition, some schemes use the same key to encrypt the same image, which does not conform to the
one-time pad strategy. This paper solves these problems from two aspects. On the one hand, random pixels
are inserted into a plain image and then a hash value is generated using SHA-256. Different seed keys can
be obtained even if the same image is encrypted. On the other hand, the Sequential Expansion Algorithm
(SEA) and Feedback Iterative Piece-Wise Linear Chaotic Mapping (FI-PWLCM) are proposed to realize the
one-to-one correspondence between the seed key and the encrypted key stream. SEA can quickly generate
seed key sensitive and random sequences. FI-PWLCM achieves one-to-one correspondence with the seed
key through feedback iteration with more control parameters. The mapping not only has the rapidity of
PWLCM, but also can produce more complex chaotic sequences. Besides, this paper proposes a Segmented
Coordinate Descent (SCD) method for histogram statistical optimization of images to improve the ability of
cryptosystems against statistical attacks. Experiments and security analysis show that the algorithm can resist
chosen-plaintext (chosen-ciphertext) attacks, brute force attacks, statistical attacks and so on. Compared
with most current algorithms, it achieves the best performance in the statistical properties of histogram and
entropy.

INDEX TERMS Image encryption, coordinate descent, SHA-256, associate plain image, one-time pad.

I. INTRODUCTION Therefore, ordinary image encryption generally requires less

In recent years, with the rapid development of the Internet security than text encryption. The typical data encryption
and the rise of the Internet of Things, the transmission of standard AES has high security, but this does not mean that
information has become more frequent, and the issue of infor- it is universal for encryption in any situation. A good sugges-
mation security has also received more attention. One of the tion is to design a dedicated encryption algorithm according
most effective ways to deal with the information security is to to the characteristics and application scenarios of the data.
encrypt the relevant data [1], [2]. Generally speaking, images For example, it is reasonable to use lightweight encryption
have a large amount of data and high redundancy, and their algorithms in restricted devices where security requirements
information value and accuracy are not as high as that of text are not too high. Likewise, encryption methods applicable to
(for example, an image in high-definition and not-so-high- images are worth exploring.
definition expresses almost the same amount of information). Chaos-based image encryption algorithms are considered
to be very promising due to the sensitivity of chaotic sys-
The associate editor coordinating the review of this manuscript and tems to initial values, ergodicity and pseudo-randomness [3].
approving it for publication was Wei Liu. Researchers have studied chaos-based image encryption

This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/
VOLUME 10, 2022 114597
 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

schemes from different perspectives [4], [5], [6], [7], [8],

[9], [10], [11], [12], [13], [14]. However, some cryptana-
lysts have demonstrated that there are obvious security prob-
lems with some of the proposed schemes [15], [16], [17],
[18]. Most of the cracked algorithms are not resistant to
plaintext-ciphertext pair attacks. Based on this, image encryp-
tion of key-associated plaintext is proposed, which is actually
an image encryption scheme similar to the one-time pad strat-
egy. The ideal one-time pad scheme is complete confidential-
ity [19]. In fact, the random key stream of arbitrary length is
usually generated by the seed key through a pseudo-random FIGURE 1. Schematic diagram of sequence extension once.
generator, so this one-time pad-like encryption scheme can-
not be completely kept secret. However, the security of this
algorithm [30], dynamic harmony search algorithm [31] are
scheme is still very high, and we will call it a one-time pad
used for image encryption. Compared with conventional
scheme for the time being. According to Katz et al. [19], the
image encryption algorithms, they improve the cryptographic
security of one-time pad can be summarized into the follow-
image entropy, but they do not achieve a sufficiently sta-
ing two principles:
ble and significant improvement. Kaur and Kumar [32] also
1) Make sure that the seed key is not repeated every time it
pointed out that the existing meta-heuristic image encryption
is encrypted. This is to prevent brute force attacks and
technology has problems such as slow calculation speed and
plaintext-ciphertext pair attacks.
difficult parameter adjustment.
2) Make sure that the keystream generated by the seed
Based on the above analysis, we propose a chaotic
key has sufficient randomness. The randomness of the
image encryption algorithm based on coordinate descent and
key stream can ensure the randomness of the output
SHA-256. This solution can effectively solve the above prob-
ciphertext to prevent statistical attacks.
lems. First, random pixels are inserted before SHA-256 acts
Some researchers use the method of correlating plaintext on the normal image to ensure that each encryption uses
to simulate the one-time pad scheme [1], [9], [10], [20], a different seed key. Second, we propose SEA, which is a
[21], [22], [23], [24], [25]. They map the seed key to the simple and fast pseudo-random sequence generator. Third,
initial value of the chaotic system, and then generate unpre- we improve PWLCM to FI-PWLCM, which has more control
dictable pseudorandom sequences for encryption by iterating parameters than PWLCM and can generate more complex
the chaotic system. However, there are some problems with chaotic sequences. Finally, we propose the SCD algorithm
this. On the one hand, the process of mapping the seed key based on the idea of coordinate descent method to enhance
to the initial value of the chaotic system is not one-to-one the resistance of cryptosystems to statistical attacks. Both
correspondence. For instance, in formula (4) in [10], a large SEA and FI-PWLCM can generate random sequences that
number of equivalent keys can be generated just by exchang- correspond one-to-one to 32-byte hashes, making the algo-
ing the positions of the hash values, so multiple hash values rithm conform to the cryptographic rule of one-time pad.
are mapped to the same initial value of the chaotic system, Compared with existing meta-heuristic image encryption
which greatly reduces the key space. In addition, for exam- algorithms, the proposed SCD algorithm can significantly
ple [1], [10], [20] encrypts the same image with the same key, optimize the histogram statistical properties of cryptographic
which does not conform to the idea of one-time pad (This images.
doesn’t mean they are not secure, but since the key for each This paper will be arranged as follows. Section II describes
encryption needs to be secretly transmitted to the decrypting the proposed image encryption scheme. Section III analyzes
party, why not take a closer approach to the one-time pad?). the experimental results and safety. Section IV concludes the
On the other hand, although chaotic systems are pseudo- research of this paper.
random, unpredictable. However, in finite precision devices,
some literatures show that low-dimensional chaotic systems II. THE PROPOSED IMAGE ENCRYPTION SCHEME
are prone to dynamic degradation [13], [26]. This makes the A. SEQUENCE EXTENSION ALGORITHM (SEA)
generated keystream predictable, which in turn leads to poor The idea of the SEA is to turn a short sequence into a long
cryptographic statistical properties of the ciphertext. Some sequence of a specified length. One feature of this long
researchers solve this problem by using higher-dimensional, sequence is pseudo-randomness, which is beneficial to image
more complex chaotic systems [1], [10], however, this greatly encryption, and another feature is the correlation with the
increases the amount of computation and reduces the speed short sequence. After the value of the short sequence changes,
of encryption. the value of the long sequence will also change. The algo-
Image encryption algorithms based on meta-heuristics rithm generates a long sequence that is required to associate
have been proposed to further improve the anti-statistical the plain image, so the short sequence uses the 32-byte hash
analysis capability of image encryption. For example, based value generated by SHA-256. The short sequences are first
on genetic algorithm [27], ant colony algorithm [28], sim- correlated and then extended round by round. The extension
ulated annealing algorithm [29], imperialist competition process of each round is shown in Figure 1, and finally a long

114598 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

Algorithm 1 SEA 2) SPEED ANALYSIS

Input: The 32-byte hash value sequence k generated by In image encryption algorithms, chaotic systems can usu-
SHA-256, and the total number L of pixels of the plain image. ally be used in pseudo-random number generators. While
Output: The output sequence K. SEA is essentially a pseudo-random integer sequence gen-
1: l ← length(k) // Get the length of k. erator, in order to verify its speed properties, we compared
2: c ← mod(k1 ⊕ k2 + k3 ⊕ k4 + . . . + k31 ⊕ k32 , 127) + it with common one-dimensional chaotic systems. On one
mod(k2 ⊕ k3 + k4 ⊕ k5 + . . . + k32 ⊕ k1 , 128); hand, we use Logistic map, Tent map, Sine map, Chebyshev,
3: ki ← mod(ki + c, 256), i = 1, 2, 3, . . . , l; PWLCM to generate chaotic sequence s of length 107 , and
4: K ← k; then use mod(s × 1010 , 256) to obtain the integer sequence.
5: while l < L do On the other hand, we generate integer sequences of length
6: l ← floor(l/2); A ← K(1 : l); B ← K(l + 1 : 2l) 107 using the SEA method. In MATLAB, the execution time
7: C(i) ← A(i) ⊕ B(i), i = 1, 2, 3, . . . , l of these two processes is calculated and the results are shown
8: K ← [K, C]; l ← length(K); in Figure 2. Obviously, SEA has a faster speed.
9: end while
10: return K ← K(1 : L);

TABLE 1. NIST SP 800-22 randomness test for SEA.

FIGURE 2. Speed comparison of SEA with some 1D chaotic

pseudo-random generators.

3) SENSITIVITY ANALYSIS TO SEED KEYS

The sensitivity of the key is a basic requirement for an encryp-
sequence of the specified length is obtained. Algorithm 1 is tion system. The procedure of sensitivity measurement is as
its concrete realization. follows: assume a 32-byte seed key k = {k1 , k2 , k3 , . . . , k32 },
SEA is essentially a pseudo-random integer sequence gen- generate the original sequence K 0 using SEA, then change
erator whose purpose is to replace pixels of plain images the value of ki and obtain K i respectively. The difference in
such that the replacement data has appreciable randomness. K i and K 0 can be measured by the NPCR (Number of Pixel
We performed NIST randomness testing, velocity analysis Change Rate) and UACI (Unified Averaged Changed Inten-
and sensitivity analysis on the seed key for SEA to verify its sity), which are calculated according to Eq. (16) and Eq. (17).
validity. The expected values of NPCR and UACI are 99.6 and 33.4,
respectively. The process is repeated 10,000 times for differ-
ent seed keys and the average values obtained are shown in
1) NIST RANDOMNESS TEST
Figure 3. It can be seen that the NPCR is about 95.2 and the
NIST SP 800-22 is a tool published by NIST (The National UACI is about 30.7. This indicates that SEA is sensitive to
Institute of Standards and Technology) for randomness seed keys.
testing of random number generators and pseudo-random
number generators. The tool has 15 tests and uses p-values B. SEGMENTED COORDINATE DESCENT (SCD)
and pass rates to assess whether the test passes. We took
The coordinate descent method is an optimization method
the significance level of 0.01 and generated 100 sets of
that decomposes high-dimensional optimization problems
byte streams of length 106 /8 using SEA for different 32-
into low-dimensional or even single-dimensional problems.
byte seed keys k and then tested them using NIST SP 800-
There are many variants [33], [34], [35], [36], [37], [38]. The
22. The results are shown in Table 1. It can be seen that
problem of coordinate descent can be described as
all the P-values are greater than 0.01 and their pass rates
are qualified (Note: the threshold of passable test is 96 for x̃ = arg min f (x), (1)
a total of 100 and 56 for a total of 59.). This means that x
the generated pseudo-random integer sequence is completely where f (.) is differentiabl, x ∈ Rm , and its lower bound is
random. xmin , and its upper bound is xmax . Solve Eq. (1) by an iterative

VOLUME 10, 2022 114599

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

Algorithm 2 SCD
Input: xmin , xmax , initial value x̃, accuracy Ac , and the max-
imum number maxStep of iterations.
Output: x̃ and the optimized value ỹ.
1: function [x̃, ỹ]=SCD(xmin , xmax , x̃, Ac , maxStep)
2: m ← length(x0 ); m1 ← max(m, 10);
3: λ1 ← 0.5; λ2 ← 0.9; λ3 ← 0.98;
λ3 − λ2 m1 λ3 − λ1
4: a←( ) 1 ;k ← ;
λ3 − λ1 a
5: for ns ← 1 : maxStep do
6: i ← mod(i − 1, m) + 1; λ ← λ3 − k · ans ;
7: x1 ← xmin (i), x2 ← x̃(i)
8: x3 ← x1 + (x2 − x1 )λ
9: [x1 , ∼] ← subLineSea(x1 , x2 , x3 , x̃, Ac , i);
FIGURE 3. SEA sensitivity to the seed key. 10: x2 ← xmax (i); x3 ← x2 − (x2 − x1 )λ;
11: [x̃, ỹ] ← subLineSea(x1 , x2 , x3 , x̃, Ac , i);
method, and describe the t-th solution as: 12: end for
13: end function
(t) (t) (t) (t)
x˜(t) = [x1 , x2 , . . . , xi , xi+1 , . . . , xm(t) ]. (2) 14:
15: function [x1 , y1 ] = subLineSea(x1 , x2 , x3 , x̃, Ac , i)
When f (x) is a convex continuously differentiable function, 16: x1 ← x̃; x2 ← x̃; x3 ← x̃; x4 ← x̃;
Eq. (1) can be transformed to solve the optimal value of each 17: x1 (i) ← x1 ; x2 (i) ← x2 ; x3 (i) ← x3 ;
dimension, namely: 18: y1 ← f (x1 ); y2 ← f (x2 ); y3 ← f (x3 );
(t) (t) (t)
x̃i = arg min f (x1 , x2 , . . . , xi
(t−1)
, . . . , xm(t−1) ). (3) 19: while x2 − x1 > Ac do
xi 20: if y3 ≤ y1 and y3 ≤ y2 then
21: x4 ← x3 + 0.1Ac ; x4 (i) ← x4 ; y4 ← f (x4 );
In Eqs. (1) - (3), m represents the dimensionality of the vari-
22: if y4 > y3 then
able x. The global optimum can be found by continuously and
23: x2 ← x3 ; y2 ← y3 ;
iteratively solving for the extreme values of each dimension.
This conclusion has been proved in [39]. However, the above 24: else
conclusion does not hold when f (.) is not a continuously dif- 25: x1 ← x3 ; y1 ← y3 ;
ferentiable convex function. In fact nonconvex problems are 26: end if
often solved using metaheuristic search algorithms, because 27: else
no algorithm can absolutely find the global optimal solution 28: if y1 < y2 then
to nonconvex problems. 29: x2 ← x3 ; y2 ← y3 ;
Inspired by coordinate descent, we propose a Segmented 30: else
Coordinate Descent (SCD) search algorithm. The SCD algo- 31: x1 ← x3 ; y1 ← y3 ;
rithm is shown in Algorithm 2. Note that, unlike conven- 32: end if
tional metaheuristics, the proposed SCD algorithm focuses 33: end if
more on finding a better solution in a faster way rather than 34: x3 ← x1 + (x2 − x1 )(0.2 + 0.6 × rand());
a global optimum, since finding the global optimum often 35: x3 (i) ← x3 ; y3 ← f (x3 );
requires huge computational power. A comparison with exist- 36: end while
ing heuristic-based image encryption algorithms is made in 37: if y1 > y2 then x1 ← x2 ; y1 ← y2 ; end
Section III-F, and the results show that the proposed SCD 38: end function
algorithm obtains better results in less time. We will verify
the properties of the SCD algorithm in 3 aspects: coordinate
(t)
switching, subinterval line search and instance testing. indicate the closeness to the current value x̃i . The closer the
(t)
value is to 1, the closer the split point is to the x̃i , and λ is
1) COORDINATE SWITCHING defined as:
We switch coordinates in a cyclic order to optimize the cur-
λ = λ3 − k · ai

(t) 
rent solution step by step. The current solution x̃i will get 
λ3 − λ2 1


closer and closer to the optimal solution with the gradual

a=( )m (4)
optimization of each dimension. The SCD strategy is to use λ3 − λ1
λ λ

the current solution to partition the one-dimensional search

3 − 1
,

k =

interval into two segments and make the initial point in each a
segment search process closer and closer to the current solu- where λ3 is the upper bound of λ, i represents the current
(t)
tion x̃i . By using the dynamic proximity coefficient λ to search dimension, k and a are undetermined parameters, λ1 is

114600 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

FIGURE 5. Scheme of changing image pixels.

3) SEVERAL EXAMPLE TESTS

To visualize the results, as shown in Figure 4, we enumerate
2 unary functions and 2 binary functions to observe the search
process of SCD. It is clear that for convex problems SCD con-
verges to a minimum, while for nonconvex problems, SCD
has a certain probability of jumping out of the local optimal
solution.

C. OPTIMIZING THE STATISTICAL PROPERTIES OF CIPHER

IMAGES USING SCD
The chi-square value of the cipher image is taken as the
optimization target, since the chi-square value of the image
is easy to calculate, and the chi-square value of the image
is related to the information entropy of the image. Generally
speaking, the smaller the chi-square value of the image, the
more even the pixel distribution of the image, the greater the
information entropy of the image. Considering that the image
FIGURE 4. Several examples of applying the SCD algorithm. encryption needs to be reversible, by using the XOR operation
(a) y = x 2 , x ∈ [−2, 3.7]. to construct the scheme of changing the image pixel, as shown
(b) y = (x + 3)(x + 1.5)(x − 0.5)(x − 1.5)(x − 2)(x − 3.5), x ∈ [−3.2, 3.7]. in Figure 5. The image vector I1 is XORed with a variable X,
(c) z = 100(y − x 2r)2 + (x − 1)2 , x ∈ [−5, 5], and y ∈ [−5, 5].
x 2 +y 2 cos(2πx)+cos(2πy )
and the I 1 can be converted into the optimized image vector
(−0.2 )
(d) z = −20e 2 −e 2 + e + 20, x ∈ [−3, 3], I2 by the cyclic XOR operation. Use the SCD algorithm to
and y ∈ [−3, 3].
find an X such that the chi-square value of I2 is as small
as possible to optimize the statistical properties of the cipher
the initial value of λ, and λ2 is the value of λ when i = m. image.
The intent of setting this coefficient is to reduce the amount Let:
of calculation.
I 1 = [I 1 (1), I 1 (2), . . . , I 1 (MN )], (5)
2) SUBINTERVAL LINE SEARCH X = [X(1), X(2), . . . , X(m)]. (6)
In Algorithm 2 the function subLineSea(.) solves for the
smaller value in the interval by continually narrowing the According to Figure 5, I 2 can be expressed as
search interval. For a differentiable convex function, the func- I 2 = cycXOR(I 1 , X)
tion must be searchable to a minimum (the proof of this
conclusion is obvious, since the search process ensures that = [I 1 (1) ⊕ X(1), . . . , I 1 (m) ⊕ X(m),
the minima always lie within the search interval.). Therefore, I 1 (m + 1) ⊕ X(1), . . . , I 1 (2m) ⊕ X(m),
the SCD algorithm can solve Eq. (1). subLineSea(.)’s search I 1 (2m + 1) ⊕ X(1), . . . , I 1 (3m) ⊕ X(m),
process is randomized, so that for non-convex functions there . . . , I 1 (MN ) ⊕ X(mod(MN − 1, m) + 1)], (7)
is a certain probability that the minimum will be searched.
In addition, the design of subLineSea(.) does not use gradient where m represents the dimension of X, M represents the
information so it can be used for optimization problems with height of the image, and N represents the image’s width.
discontinuous functions. The objective function can be expressed as the chi-square

VOLUME 10, 2022 114601

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

TABLE 2. Several comparisons between FI-PWLCM and typical one-dimensional chaotic systems.

value of I 2 , witch is D. FEEDBACK ITERATIVE PIECE-WISE LINEAR CHAOTIC

X MAP (FI-PWLCM)
χ 2 = f (X) = i=0 (fr (cycXOR(I 1 , X), i) − f ) /f ,
255 ¯ 2 ¯ (8)
Piece-Wise Linear Chaotic Map (PWLCM) is defined as
where f¯ = (M × N )/256, χ 2 represents the chi-square value 
xn
of I 2 , and fr (Img, i) denotes the number of pixels at pixel level

 0 < xn < p
p


i in the statistical image Img. The SCD algorithm needs to xn+1 = fp (xn ) = xn − p (9)
be modified to suit the case of integer computations. Specifi-  p ≤ xn < 0.5
 0.5 − p
cally, use floor(· ) to round up lines 9, 11, 16, 18 and 45. In line


fp (1 − xn ) 0.5 < xn < 1,

31, x4 ← x4 + 1. The input parameters are set as xmin =
[0, 0, . . . , 0]1×m , xmax = [255, 255, . . . , 255]1×m , Ac = where u ∈ (0, 0.5). We conduct several analyses of com-
1, maxStep = 3m. Then the SCD algorithm is mon one-dimensional chaotic systems as shown in Table 2.
executed to obtain x̃ and ỹ. According to Eq. (7), As analyzed in Table 2, in a one-dimensional chaotic system,
I2 = cycXOR(I 1 , x̃). PWLCM has relatively fast speed and high sample entropy

114602 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

FIGURE 6. Structure of FI-PWLCM.

(The more complex the sequence, the higher the sample

entropy.). However, the system is only determined by two ini-
tial values, and cannot achieve a one-to-one mapping with a
32-byte hash value within a precision of 1016 . In addition, the
system exhibits severe dynamic degradation when u = 0.25.
The system cannot complete the iteration when x = 0.5.
Based on the above analysis, we improve PWLCM follow-
ing the structure of Figure 6 to avoid some known weaknesses
of PWLCM and ensure that the seed key k has a one-to-one
correspondence with the chaotic sequences generated by the
improved system. The PWLCM with this feedback iterative
structure is called FI-PWLCM. Its recursive formula is:
 
xn
if xn < u

 


u

 


xn − u
 

if xn < 0.5

 


 
t = 0.5 − u

n
 1 − xn (10)
 
 if 1 − x ≤ u

 

 u
 1 − xn − u otherwise

 


 

0.5 − u



n+1 = |tn + βi − 1|,

x

where n ≥ 1, x1 = β1 , i = mod(n−1, 6)+1, tn is the output FIGURE 7. Block diagram of the proposed image encryption scheme.

of the system, and βi is determined by a 32-byte hash value

k, which is calculated as: sample entropy and histogram analysis on FI-PWLCM with
 P6 8(i−1)
i=1 ki 2
a typical one-dimensional chaotic system. The experimental
β


 1 = 48 results are shown in Table 2. Although Sine map and cheby-
P12 2 8(i−7)





i=7 i 2
k shev map have higher sample entropy, they are too slow.
β2 =



48 And FI-PWLCM not only has similar speed performance
P18 2 8(i−13)



to PWLCM, but also has higher sample entropy. Therefore,
i=13 ki 2


β3 =

it can replace PWLCM for image encryption. FI-PWLCM is

48
P24 2 8(i−19) (11) iterated MN +200 times, where M and N represent the height
i=19 i 2
k
β

= and width of the image, respectively. The chaotic sequence is

4

 48
P30 2 8(i−25)

obtained by discarding the first 200 times, and then the index


i=25 ki 2

β5 =

subscript sequence P is obtained by ascending the chaotic


48

P32 2 8(i−31)


 sequence.
k i2

i=31

β6 =

.
248
E. ENCRYPTION AND DECRYPTION ALGORITHMS
u is the control parameter of the system, u ∈ (0, 0.5), to make Figure 7 is a block diagram of encryption. The key consists
the generated sequence have high enough complexity, it is of two parts. One is the 32-byte hash value of the plain image
defined as: processed by SHA-256. The second is the m-byte extended
6
X key of SCD optimization process, the length of which can
u = 0.23 + mod( βi , 1)/100. (12) be set by the user. The encryption process can be divided
i=1 into three stages. First using SEA substitution, second using
Obviously, the hash value k and β constitute a one-to-one SCD optimization, and finally using FI-PWLCM permuta-
correspondence, and βi will continue to act on FI-PWLCM, tion. The substitution process can make the plaintext infor-
so that the generated sequence and the hash value are in one- mation approximate to pseudo-random information, so that
to-one correspondence. Furthermore, we perform velocity, when some specific information (such as all 0 information) is

VOLUME 10, 2022 114603

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

input in the plaintext, some obvious features will not appear Algorithm 4 Decryption Algorithm
after the substitution. The SCD optimization process is equiv- Input: Cipher image matrix C, and the key Key used for
alent to extracting some common features of the information decryption.
to be encrypted with the key, so that the output ciphertext Output: The decrypted image I 0 .
contains less features about the plaintext. The FI-PWLCM 1: Convert C into one-dimensional vector I 3 , and find its
permutation can make the position of the plain image differ- length l.
ent from the position of the cipher image, making it difficult 2: k ← Key(1 : 32).
for the attacker to find the corresponding relationship. The 3: The integer sequence K is obtained by Algorithm 1 of
complete encryption algorithm is described in Algorithm 3. using k and l as input.
The decryption algorithm is described in Algorithm 4. 4: The sequence P is obtained by the method in Section II-
D.
III. EXPERIMENTAL RESULTS AND PERFORMANCE 5: I 2 (P(i)) ← I 3 (i), where i = 1, 2, 3, . . . , l.
ANALYSIS 6: X ← key(33 : end) and m ← length(X).
There are some necessary clarifications that need to be done 7: I1 ← cycXOR(I 2 , X). // See Eq. (7).
before experimental analysis. First, the development environ- 8: I(i) ← I 1 (i) ⊕ K(i), where i = 1, 2, 3, . . . , l.
ment is a 64-bit Windows 10 system, 8.00GB RAM, Intel(R) 9: I0 is the final decrypted image, which is obtained by
Core (TM) i7-10510U [email protected] GHz, 2304 Mhz. Moreover, converting the vector I into an image matrix.
use MATLAB to write programs and experiments. Second,
since the key is dynamic, when the value of m is not specified,
the default m = 80. Finally, all experimental data are the
results obtained through multiple experiments. in the SCD process. The larger the m value, the more variables
will participate in the optimization, and the chi-square value
A. ENCRYPTION AND DECRYPTION RESULT of the cipher image will be reduced more easily. However,
The encryption and decryption results for different types of if m is too high, the storage space of decryption key will be
images are shown in Figure 8. This algorithm generates sim- increased, making key transmission difficult. On the other
ilar cipher images for different plain images. The naked eye hand, the higher the iteration times of SCD algorithm, the
cannot tell the difference between these cipher images and lower the chi-square value. But, too high a number of itera-
cannot detect their features related to the plain images. tions will waste a lot of time. Therefore, the value of m and the
number of iterations can be set by the user to trade off security
Algorithm 3 Encryption Algorithm and efficiency. In addition, according to Figure 9, when the
number of iterations is greater than 3m, the chi-square value
Input: Plain image I 0 , the variable part of the key has
decreases slowly, so the maximum number of iterations is rec-
length m.
ommended to be 3m. The curves of m = 80 and m = 100 are
Output:Cipher image C, and the key Key used for
very close, so m = 80 is recommended.
decryption.
1: Convert I0 into one-dimensional vector I, and find its C. SECURITY ANALYSIS
length l. According to kerckhoffs’ principle, an attacker knows every
2: Insert 10 random pixel values at the end of I0 to getN
I0 . design detail of a cryptographic algorithm except the key.
3: The 32-byte sequence k is obtained by applying the Based on this principle, attack methods can usually be divided
SHA-256 toN I0 . into the following five situations:
4: The XOR sequence K is obtained by Algorithm 1 of
1) Ciphertext-only attack: The attacker knows the encryp-
using k and l as input. tion algorithm and the ciphertext to be deciphered. This
5: I 1 (i) ← I(i) ⊕ K(i), where i = 1, 2, 3, . . . , l.
is the most difficult attack scenario.
6: According to Section II-C, x̃ and I 2 are obtained.
2) Known-plaintext attack: The attacker knows the
7: Key ← [k; x̃]. // Splice k with x̃.
encryption algorithm, some plaintext ciphertext pair.
8: The sequence P is obtained by the method in 3) Chosen-plaintext attack: The attacker knows the
Section II-D. encryption algorithm and can choose some special
9: I3 (i) ← I2 (P(i)), where i = 1, 2, 3, . . . , l. plaintext ciphertext pair.
10: C is the final encrypted image, which is obtained by 4) Chosen-ciphertext attack: The attack knows the
converting the vector I 3 into an image matrix. encryption algorithm and can choose some special
ciphertext plaintext pair.
5) Selective text attack: a combination of selective plain-
B. THE RELATIONSHIP BETWEEN CHI-SQUARE VALUE, m text attack and selective ciphertext attack.
VALUE AND ITERATION TIMES A cryptographic algorithm can be considered highly secure
Figure 9 directly reflects the relationship between chi-square if it is resistant to selective text attacks. Note that the pro-
value, m value and iteration times of cipher image. On the one posed algorithm algorithm is a one-time pad encryption
hand, the m value determines the dimension of the variables strategy. An attacker can easily obtain the key through a

114604 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

FIGURE 8. Test images. The first line represents the plain images, the second line is the corresponding cipher images, and the third is the decrypted
images.

SEA and SHA-256 makes the system complex enough, and

SCD makes the statistical properties of the generated cryp-
tographic images very random. Coupled with the one-time
pad encryption strategy used, it is difficult for an attacker to
find information related to the key from the cipher image.
Therefore, this algorithm is resistant to selective text attacks.
Based on the above analysis, the focus of this paper on the
security analysis of this algorithm is statistical analysis.

1) BRUTE FORCE ATTACK ANALYSIS

A brute force attack is a method in which an attacker reaches
and cracks a password by exhausting all the keys of an
encryption algorithm. This attack method is a condition that
an encryption system must satisfy at least. The complexity
FIGURE 9. The relationship between chi-square value, m value and
of the implementation of this method is directly related to the
iteration times. size of the key space, as long as the key space is large enough,
this method is difficult to implement. Gonzalo et al. [40]
states that the key space is at least greater than 2100 . Due to
chosen-plaintext (chosen-ciphertext) attack, but the key can- the degradation of the chaotic system in the limited-precision
not be used to decipher other ciphertext information. There- computer equipment, that is, the computer has truncation
fore, any attempt to exploit a plaintext ciphertext pair to error and rounding error in the numerical calculation, which
guess the key will fail. Furthermore, differential and linear will lead to the weakening of the initial value sensitivity of
attacks would be difficult to implement because the encryp- the chaotic system. Therefore, the initial value precision of
tion key is different each time. Brute force attack can only the chaotic system should be properly selected in order to
decipher the current cipher image, because the key obtained obtain an accurate key space. In Eq. (11) in Section II-D,
by it cannot be applied to the cracking of other ciphertext it is determined that the initial value precision of FI-PWLCM
information. It is possible to use statistical attacks to estab- is 1/248 > 3.5 × 10−15 , which is within the calculation
lish the relationship between the ciphertext and the key, but range of double precision, that is, the initial value generated
this depends on the statistical properties of the cipher image by Eq. (11) is valid. The algorithm key proposed consists of
representation. It is worth noting that although this algorithm two parts, the first 32 bytes are the SHA-256 sequence, and
uses a simple chaotic system, the combination with SCD, the last m bytes are the optimized parameters using SCD.

VOLUME 10, 2022 114605

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

FIGURE 10. Encryption results and histograms for different m values.

Therefore, the key is 32 + m bytes in total, that is, the key completely different from that of the plain image, and is very
space is: 28×(32+m) = 2256+8m > 2100 . This is enough to close to the horizontal line. On the other hand, by calculating
resist brute force attacks. the chi-square value of the histogram, which can more accu-
rately reflect the horizontal characteristics of the histogram
2) STATISTICAL ATTACK ANALYSIS
of the cipher image. The calculation of the chi-square value
Statistical attack refers to the use of certain statistical laws
is shown in Eq. (8). Different images were experimented and
of plaintext and ciphertext to achieve cryptographic attacks.
compared with [33], [48], and [49]. Figure 11 is experimental
Images are a highly correlated and redundant medium, it has
result. Note: In the chi-square test, the significance level was
many statistical characteristics, and it is a common analy-
0.05, corresponding to a chi-square value of 293.2478. The
sis method to use the inherent characteristics of image to
results show that this algorithm easily passes the chi-square
perform statistical analysis on encryption system. Preishu-
test. Compared to [33], [48], and [49], the chi-square value of
ber et al. [41] and Özkaynak [42] have proposed that even if
the proposed algorithm drops by more than 70% on average.
an encryption scheme perfectly passes several statistical anal-
We are not surprised by this result, as it just proves the effec-
ysis indicators (such as NPCR, UACI, entropy, etc.), it cannot
tiveness of the SCD algorithm. SCD optimizes the replace-
be regarded as a sufficient condition for the security of the
ment process so that the pixel distribution of the replacement
cryptosystem. Even so, statistical analysis of a cryptographic
result is more uniform, so that the histogram characteristics
algorithm is still necessary, because if a cryptographic algo-
are significantly better than other encryption algorithms.
rithm cannot pass even this typical attack, then the crypto-
graphic algorithm cannot be a good algorithm. Several typical b: CORRELATION BETWEEN ADJACENT PIXELS
statistical indicators are analyzed below. The strong correlation of neighborhood pixels is one of the
typical features of available images. Based on this, image
a: HISTOGRAM AND CHI-SQUARE VALUE
encryption should destroy this feature [12]. The typical fea-
The histogram of an image can reflect important information ture of an encrypted image is that the cipher image becomes
in the original image. Attackers can use it to perform statis- very chaotic. Mathematically, it can be considered that the
tical analysis. Image encryption should change this feature. correlation between adjacent pixels of the image is highly
On the one hand, the distribution of pixel values is visu- low. Correlation is mathematically defined as:
ally analyzed by plotting the histogram of the cipher image.  PN PN
To reduce the correlation between the histogram and the plain i=1 xi i=1 yi
, E(y) =

E(x) =

image, the histogram of the cipher image should be inclined

 NP N
towards the horizontal line [43]. The experimental results are N (13)
i=1 [xi − E(x)][yi − E(y)]
shown in Figure 10. From the figure, it can be seen that the Cxy = qP

 qP ,
N N
[xi − E(x)]2 [yi − E(y)]2

histogram of the cipher image obtained by the algorithm is 
i=1 i=1

114606 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

FIGURE 11. Comparison with Ref. [29], [44], and [45] in terms of
chi-square values (images size: 512 × 512).

TABLE 3. Correlation coefficients of adjacent pixels for Lena (512 × 512).

where N represents the number of pairs of adjacent pixel

points selected, and x and y represent the gray values of
a pair of adjacent points. Experiments are conducted by
randomly selecting 5000 pairs of pixels in the horizontal,
vertical and diagonal directions. The results are shown in FIGURE 12. Correlation of Lena’s neighboring pixels. (a), (c) and (e) are
Figure 12 and Table 3. As can be seen from Table 3, the the correlations of Lena plain images in horizontal, vertical and diagonal
directions, respectively. (b), (d) and (f) are the correlations of Lena cipher
correlation of the proposed encryption algorithm in three images in horizontal, vertical and diagonal directions, respectively.
directions is lower than 10−3 . Compared with the algorithm
of Hua et al. [44], the proposed algorithm is more stable and
has lower average values in three directions. This result seems where S represents the gray value, N represents the gray
to have a probability factor, but it is actually an inevitable level of the image, and P(si ) represents the probability of
result. Assuming that an image with uniform histogram and the i-th gray level in the entire image. For an ideal gray
an image with non-uniform histogram are arranged by the image, the gray value of each level has the same probability,
same shuffling method, the pixel values with high frequency at this time, H (S) = 8. Figure 13 shows the changes in
of the image with non-uniform histogram are more likely to the chi-square value and information entropy of the cipher
be assigned to adjacent positions, which leads to increased image for Lena (256 × 256) as the value of m changes. It can
correlation. Therefore, the image with uniform histogram be seen from Figure 13 (a) that the chi-square value of the
will have smaller correlation between adjacent pixels after cipher image decreases as m increases. This is reflected in
shuffling. The proposed algorithm SCD process optimizes the changes in the histograms in Figure 13 (c), (d) and (e),
the histogram characteristics, and FI-PWLCM can produce where it can be seen that the histograms become more and
highly complex chaotic sequences, which is very suitable more horizontal. This is also reflected in the information
for pixel shuffling. Therefore, the algorithm can obtain quite entropy as shown in Figure 13 (b), the entropy value gradually
good correlation between adjacent pixels. increases with the increase of m, and finally approaches the
limit value of 8. One can see from Table 4 that proposed
c: INFORMATION ENTROPY method get higher entropy value. Figure 14 is a compari-
Information entropy can reflect whether the distribution of son of the information entropy of various algorithms [20],
pixels in the image is uniform [46]. To make the encrypted [25], [43], [45], [47], [48], [49], [50], [51], [52], [53], [54],
image more random, its information entropy should be larger. [55], [56]. The experimental results show that, compared
The information entropy of an image is defined as: with most current image encryption algorithms, the algo-
X rithm has the best performance in terms of information
N −1
H (S) = − i=0 P(si ) log2 P(si ), (14) entropy.

VOLUME 10, 2022 114607

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

TABLE 4. Entropy of cipher image information of different images TABLE 5. NIST SP 800-22 randomness test for cipher images.
(images size: 512 × 512).

FIGURE 13. The relationship between chi-square value and information

entropy to m. (a) The relationship between the chi-square value and m.
(b) The relationship between information entropy and m. (c) The
histograms of the cipher images when m = 0. (d) The histograms of the FIGURE 15. Key sensitive analysis diagram. (a) Cipher-image. (b) Use the
cipher images when m = 80. (e) The histograms of the cipher images correct key to decrypt. (c) Decrypt using a key that has been changed by
when m = 500. one bit.

and the length of each group is 106 . The NIST test results are
shown in Table 5. The significance level of the test is 0.01,
and when the number of samples is 123, the threshold for
passing the test is 118. When the number of samples is 85,
the threshold for passing the test is 81. The results in Table 5
show that the generated cipher images have all passed the
NIST detection. Therefore, the proposed method can resist
statistical attacks very well.

3) SENSITIVITY ANALYSIS
a: KEY SENSITIVITY ANALYSIS
The cipher image should change drastically when image
encrypted with a slightly changed key. Since the key is gen-
erated after the image is encrypted, this does not need to be
FIGURE 14. Comparison of information entropy for Lena (256 × 256). considered. Furthermore, decrypting with a slightly changed
key, the cipher image cannot be decrypted correctly [13].
d: NIST RANDOMNESS DETECTION FOR CIPHER IMAGES Based on this, by randomly changing a key for testing. The
A good enough cryptographic system should be able to experimental results are shown in Figure 15. By repeating this
encrypt any image into a messy one. This is reflected in experiment, and the experimental results obtained are similar
the independence of the plain image and the cipher image to Figure 15, indicating that the algorithm has excellent key
in many statistical characteristics. It is very important to sensitivity.
ensure the randomness of the ciphertext when the plaintext
is unchangeable. NIST has released a set of test packages b: PLAINTEXT SENSITIVITY ANALYSIS
NIST SP800-22 for randomness detection of sequences. The The sensitivity of an encryption algorithm to plaintext is a
test package includes randomness tests for 15 statistical prop- necessary condition for its ability to resist chosen-plaintext
erties, and the P-value and pass rate measure whether the attack, and this sensitivity is used in many literatures to char-
sequence to be tested passes these tests. We use the proposed acterize the ability to resist differential attacks. Note that
algorithm to encrypt a total of 200 images in the training set this is a necessary condition but not a sufficient one, that is,
of BSD300 [57] and concatenate the generated cipher images a good encryption algorithm should have sensitivity to plain-
into a bit stream. The bitstream is divided into 123 groups, text. This is usually characterized using NPCR and UACI,

114608 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

which are calculated as: TABLE 6. Quantitative results of noise and shear attacks.
(
0 if C1 (i, j) = C2 (i, j)
D(i, j) = (15)
1 if C1 (i, j) 6 = C2 (i, j),
1 XM XN
NPCR = D(i, j) × 100, (16)
MN i=1 i=1
PM PN
i=1 i=1 |C1 (i, j) − C2 (i, j)|
UACI = × 100, (17)
255MN
where M and N represent the number of rows and columns
of the image respectively, and C1 , C2 represent two cipher
images, and their corresponding plain images have only a
one-pixel difference. This paper randomly selects a pixel in
the plain image to change it. The results are NPCR = 99.6095,
UACI = 33.4470. These results are very close to the expected
value NPCRE = 99.6094 and UACIE = 33.4635 [58].
This shows that the algorithm has excellent sensitivity to
plaintext.

D. ROBUSTNESS ANALYSIS
Noise and clipping attacks on images are a common attack
way, and image encryption algorithms usually require a cer- FIGURE 16. Image anti-noise test. (a) Decrypted image with 5% noise.
(b) Decrypted image with 15%. (c) Decrypted image with 25%.
tain degree of resistance to them, that is, certain robust-
ness [59]. By using 5%, 15%, and 25% noise interference or
shear loss to conduct experiments, respectively, on the cipher
image. On the one hand, Figure 16 and Figure 17 are used for
qualitative measurement. On the other hand, the indicators
of the Mean Square Error (MSE), the Peak Signal-to-Noise
Ratio (PSNR), and the Structural Similarity (SSIM) are used
for quantitative measurement. The calculation formulas of FIGURE 17. Anti-shear test during image decryption. (a) Cipher image cut
MSE, PSNR, and SSIM are Eqs. (18) and (19). 5%. (b) The decrypted result of Figure (a). (c) Cipher image cut 25%.
 (d) The decrypted result of Figure (c).
1 P M PN
MSE = [C1 (i, j) − C2 (i, j)]2


MN i=1 i=1 (18) size of the image increases or the key space increases, the
 255 × 255 time-consuming algorithm will also increase significantly,
PSNR = 10 × log (

)
10
 MSE but since the key space is variable, the encryption party can
 1
 c = (K1 L)2 , c2 = (K 2 L)
2 weigh security and efficiency according to its own needs. This
(2µx µy + c1 )(2σxy + c2 ) (19) paper proposes an encryption scheme rather than its efficient
SSIM(x, y) = (µ2 + µ2 + c )(σ 2 + σ 2 + c )
 implementation. Focusing on the article of Wang et al. [25],
x y 1 x y 2
they proposed an image encryption algorithm based on asso-
In Eq. (18), C1 represents the plain image, C2 represents ciated plain image, and this algorithm uses PWLCM, but it
the decrypted image after the cipher image is attacked, and M does not use optimization algorithm. This algorithm has a
and N respectively represent the number of rows and columns number of similarities with the proposed algorithm, in the
of the cipher image. In Eq. (19), µx , µy , represent the mean case of both using MATLAB programming, the proposed
gray value of image x, y, respectively, σx , σy denote the stan- algorithm even when m = 80, also achieved much better
dard deviation of the gray value of x, y, respectively, and σxy encryption speed than [25] 0.055MB/s, which shows that the
is the covariance of the gray value of x and y. In addition, proposed algorithm is competitive. Besides, since the key of
K1 = 0.01, K2 = 0.03, L = 255, the experimental results are the decryption process retains the SCD search results of the
shown in Table 6. The results show that this algorithm has encryption process, only one cycXOR(.) operation is required
better robustness. to be performed during decryption, which greatly reduces the
decryption time.
E. SPEED PERFORMANCE
The security and efficiency of encryption algorithms are often F. COMPARISON WITH META-HEURISTIC BASED
contradictory. A superior encryption algorithm must ensure ALGORITHMS
the security of encryption and consider the speed of encryp- We compared the encryption time, entropy, mean correla-
tion. The size of the image and the key space are variable in tion, NPCR and UACI with Ref. [29] (Wang et al. based on
this experiment. The test results are shown in Table 7. As the the simulated annealing algorithm.), [30] (Enayatifar et al.

VOLUME 10, 2022 114609

 X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

TABLE 7. Encryption speed test. [2] X. Chai, X. Zheng, Z. Gan, D. Han, and Y. Chen, ‘‘An image encryption
algorithm based on chaotic system and compressive sensing,’’ Signal Pro-
cess., vol. 148, pp. 124–144, Jul. 2018.
[3] D. Ravichandran, P. Praveenkumar, and J. B. B. Rayappan, ‘‘DNA chaos
blend to secure medical privacy,’’ IEEE Trans. Nanobiosci., vol. 16, no. 8,
pp. 850–858, Dec. 2017.
[4] G. Chen, Y. Mao, and C. K. Chui, ‘‘A symmetric image encryption
scheme based on 3D chaotic cat maps,’’ Chaos, Solitons Fractals, vol. 21,
pp. 749–761, Jul. 2004.
[5] N. K. Pareek, V. Patidar, and K. K. Sud, ‘‘Image encryption using chaotic
logistic map,’’ Image Vis. Comput., vol. 24, no. 9, pp. 926–934, 2006.
[6] Y. Wang, K.-W. Wong, X. Liao, T. Xiang, and G. Chen, ‘‘A chaos-based
image encryption algorithm with variable control parameters,’’ Chaos,
Solitons Fractals, vol. 41, no. 4, pp. 1773–1783, 2009.
[7] X. Chai, Y. Chen, and L. Broyde, ‘‘A novel chaos-based image encryption
algorithm using DNA sequence operations,’’ Opt. Lasers Eng., vol. 88,
TABLE 8. Comparison with meta-heuristic based algorithms. pp. 197–213, Jan. 2017.
[8] C. Li, G. Luo, K. Qin, and C. Li, ‘‘An image encryption scheme based on
chaotic tent map,’’ Nonlinear Dyn., vol. 87, no. 1, pp. 127–133, 2017.
[9] Q. Xu, K. Sun, C. Cao, and C. Zhu, ‘‘A fast image encryption algorithm
based on compressive sensing and hyperchaotic map,’’ Opt. Lasers Eng.,
vol. 121, pp. 203–214, Oct. 2019.
[10] M. Zhou and C. Wang, ‘‘A novel image encryption scheme based on con-
servative hyperchaotic system and closed-loop diffusion between blocks,’’
Signal Process., vol. 171, Jun. 2020, Art. no. 107484.
[11] S. M. Ismail, L. A. Said, A. G. Radwan, A. H. Madian, and
based on weighted discrete imperialist competitive algo- M. F. Abu-ElYazeed, ‘‘A novel image encryption system merging
rithm.) and [47] (Enayatifar et al. based on DNA coding fractional-order edge detection and generalized chaotic maps,’’ Signal Pro-
cess., vol. 167, Feb. 2020, Art. no. 107280.
and genetic algorithms.). The results are shown in Table 8.
[12] P. Sneha, S. Sankar, and A. S. Kumar, ‘‘A chaotic colour image
It can be seen that the proposed algorithm has the fastest encryption scheme combining Walsh–Hadamard transform and maps,’’
speed, the highest entropy value and the lowest adjacent pixel J. Ambient Intell. Humanized Comput., vol. 11, no. 3, pp. 1289–1308,
correlation. This shows that the proposed SCD method based 2020.
[13] Z. Hua, Z. Zhu, S. Yi, Z. Zhang, and H. Huang, ‘‘Cross-plane colour image
on coordinate descent is suitable for the optimization task of encryption using a two-dimensional logistic tent modular map,’’ Inf. Sci.,
image encryption. vol. 546, pp. 1063–1083, Feb. 2021.
[14] Y. Xian and X. Wang, ‘‘Fractal sorting matrix and its application on chaotic
image encryption,’’ Inf. Sci., vol. 547, pp. 1154–1169, Feb. 2021.
IV. CONCLUSION [15] C. Li, D. Lin, J. Lü, and F. Hao, ‘‘Cryptanalyzing an image encryption algo-
In this paper, SEA and SCD methods are proposed, and rithm based on autoblocking and electrocardiography,’’ IEEE MultiMedia,
PWLCM is improved to FI-PWLCM. This scheme works vol. 25, no. 4, pp. 46–56, Oct./Dec. 2018.
[16] S. Li and X. Zheng, ‘‘Cryptanalysis of a chaotic image encryption
by randomly inserting pixel values and using SHA-256 to method,’’ in Proc. IEEE Int. Symp. Circuits Syst., vol. 2, May 2002,
associate the key with the plain image. The proposed SEA pp. II–II.
and FI-PWLCM realize the one-to-one mapping between the [17] E. Solak, C. Çokal, O. Yildiz, and T. Biyikoǧlu, ‘‘Cryptanalysis of
seed key and the encryption key stream, which is very consis- Fridrich’s chaotic image encryption,’’ Int. J. Bifurcation Chaos, vol. 20,
no. 5, pp. 1405–1413, 2010.
tent with the one-time pad. The SCD method can effectively [18] J. Chen, L. Chen, and Y. Zhou, ‘‘Cryptanalysis of image ciphers with
improve the histogram characteristics of the cipher image, permutation-substitution network and chaos,’’ IEEE Trans. Circuits Syst.
make the distribution of pixels at all levels of the cipher Video Technol., vol. 31, no. 6, pp. 2494–2508, Jun. 2021.
[19] J. Katz and Y. Lindell, Introduction to Modern Cryptography. Boca Raton,
image more uniform, and the information entropy is higher. FL, USA: CRC Press, 2020.
This makes the scheme have better statistical characteristics [20] X. Wang, Y. Wang, X. Zhu, and C. Luo, ‘‘A novel chaotic algorithm for
of cryptography. In addition, this approach can obtain bet- image encryption utilizing one-time pad based on pixel level and DNA
ter histogram properties, entropy values and correlation in level,’’ Opt. Lasers Eng., vol. 125, Feb. 2020, Art. no. 105851.
[21] S. Zhu, C. Zhu, and W. Wang, ‘‘A novel image compression-encryption
less time than existing meta-heuristic image encryption algo- scheme based on chaos and compression sensing,’’ IEEE Access, vol. 6,
rithms. Several experiments and security analysis show that pp. 67095–67107, 2018.
the algorithm has a large enough key space, and can effec- [22] Y.-Q. Zhang, J.-L. Hao, and X.-Y. Wang, ‘‘An efficient image encryption
tively resist selective text attack, brute force attack, statistical scheme based on S-boxes and fractional-order differential logistic map,’’
IEEE Access, vol. 8, pp. 54175–54188, 2020.
statistics, noise attack and clipping attack. This algorithm [23] X. Zhang, Z. Zhou, and Y. Niu, ‘‘An image encryption method based on
is less efficient for image encryption with large size, so in the feistel network and dynamic DNA encoding,’’ IEEE Photon. J., vol. 10,
the future, we can reduce the amount of image data through no. 4, pp. 1–14, Aug. 2018.
[24] X. Chai, Y. Chen, and L. Broyde, ‘‘A novel chaos-based image encryption
compression, or speed up the encryption process through the algorithm using DNA sequence operations,’’ Opt. Lasers Eng., vol. 88,
technology of block and parallel computing. pp. 197–213, Jan. 2017.
[25] X. Wang, X. Zhu, X. Wu, and Y. Zhang, ‘‘Image encryption algorithm
based on multiple mixed hash functions and cyclic shift,’’ Opt. Lasers Eng.,
REFERENCES vol. 107, no. 1, pp. 370–379, Aug. 2017.
[1] S. Wang, C. Wang, and C. Xu, ‘‘An image encryption algorithm based on a [26] C. Li, B. Feng, S. Li, J. Kurths, and G. Chen, ‘‘Dynamic analysis of digital
hidden attractor chaos system and the Knuth–Durstenfeld algorithm,’’ Opt. chaotic maps via state-mapping networks,’’ IEEE Trans. Circuits Syst. I,
Lasers Eng., vol. 128, May 2020, Art. no. 105995. Reg. Papers, vol. 66, no. 6, pp. 2322–2335, Jun. 2019.

114610 VOLUME 10, 2022

X. Sun, Z. Chen: Novel Chaotic Image Encryption Algorithm Based on Coordinate Descent and SHA-256

[27] M. Kaur, D. Singh, K. Sun, and U. Rawat, ‘‘Color image encryption using [50] Y. Ding, F. Tan, Z. Qin, M. Cao, K.-K.-R. Choo, and Z. Qin, ‘‘Deep-
non-dominated sorting genetic algorithm with local chaotic search based KeyGen: A deep learning-based stream cipher generator for medical
5D chaotic map,’’ Future Gener. Comput. Syst., vol. 107, pp. 333–350, image encryption and decryption,’’ IEEE Trans. Neural Netw. Learn. Syst.,
Jun. 2020. vol. 33, no. 9, pp. 4915–4929, Sep. 2022.
[28] N. K. Sreelaja and G. A. Vijayalakshmi Pai, ‘‘Stream cipher for binary [51] Y. Zhang, ‘‘The fast image encryption algorithm based on lifting scheme
image encryption using ant colony optimization based key generation,’’ and chaos,’’ Inf. Sci., vol. 520, pp. 177–194, May 2020.
Appl. Soft Comput., vol. 12, no. 9, pp. 2879–2895, Sep. 2012. [52] B. Ge and H.-B. Luo, ‘‘Image encryption application of chaotic sequences
[29] X. Wang, C. Liu, and D. Xu, ‘‘Image encryption scheme using chaos incorporating quantum keys,’’ Int. J. Autom. Comput., vol. 17, no. 1,
and simulated annealing algorithm,’’ Nonlinear Dyn., vol. 84, no. 3, pp. 123–138, Feb. 2020.
pp. 1417–1429, 2016. [53] H. Xiang and L. Liu, ‘‘A novel image encryption algorithm based on
[30] R. Enayatifar, A. H. Abdullah, and M. Lee, ‘‘A weighted discrete imperial- improved key selection and digital chaotic map,’’ Multimedia Tools Appl.,
ist competitive algorithm (WDICA) combined with chaotic map for image vol. 80, no. 14, pp. 22135–22162, Jun. 2021.
encryption,’’ Opt. Lasers Eng., vol. 51, no. 9, pp. 1066–1077, Sep. 2013. [54] S. Sun, ‘‘Chaotic image encryption scheme using two-by-two deoxyri-
[31] K. Mirzaei Talarposhti and M. Khaki Jamei, ‘‘A secure image encryption bonucleic acid complementary rules,’’ Opt. Eng., vol. 56, no. 11, 2017,
method based on dynamic harmony search (DHS) combined with chaotic Art. no. 116117.
map,’’ Opt. Lasers Eng., vol. 81, pp. 21–34, Jun. 2016. [55] S. Mozaffari, ‘‘Parallel image encryption with bitplane decomposi-
[32] M. Kaur and V. Kumar, ‘‘A comprehensive review on image encryption tion and genetic algorithm,’’ Multimedia Tools Appl., vol. 77, no. 19,
techniques,’’ Arch. Comput. Methods Eng., vol. 27, no. 1, pp. 15–43, 2020. pp. 25799–25819, Oct. 2018.
[33] I. Necoara, Y. Nesterov, and F. Glineur, ‘‘Random block coordinate descent [56] M. Mahmud, M. Lee, and J.-Y. Choi, ‘‘Evolutionary-based image encryp-
methods for linearly constrained optimization over networks,’’ J. Optim. tion using RNA codons truth table,’’ Opt. Laser Technol., vol. 121,
Theory Appl., vol. 173, no. 1, pp. 227–254, Apr. 2017. Jan. 2020, Art. no. 105818.
[34] T. T. Wu and K. Lange, ‘‘Coordinate descent algorithms for lasso penalized [57] D. Martin, C. Fowlkes, D. Tal, and J. Malik, ‘‘A database of human seg-
regression,’’ Ann. Appl. Statist., vol. 2, no. 1, pp. 224–244, Mar. 2008. mented natural images and its application to evaluating segmentation algo-
[35] K. Fountoulakis and R. Tappenden, ‘‘A flexible coordinate descent rithms and measuring ecological statistics,’’ in Proc. 8th IEEE Int. Conf.
method,’’ Comput. Optim. Appl., vol. 70, no. 2, pp. 351–394, Jun. 2018. Comput. Vis. (ICCV), vol. 2, Jun. 2001, pp. 416–423.
[36] H.-J. Michael Shi, S. Tu, Y. Xu, and W. Yin, ‘‘A primer on coordinate [58] Y. Wu, J. P. Noonan, and S. Agaian, ‘‘NPCR and UACI randomness
descent algorithms,’’ 2016, arXiv:1610.00040. tests for image encryption,’’ Cyber J., Multidisciplinary J. Sci. Technol.,
[37] M. Gurbuzbalaban, A. Ozdaglar, P. A. Parrilo, and N. Vanli, ‘‘When cyclic J. Select. Areas Telecommu., vol. 1, pp. 31–38, Apr. 2011.
coordinate descent outperforms randomized coordinate descent,’’ in Proc. [59] Z.-J. Huang, S. Cheng, L.-H. Gong, and N.-R. Zhou, ‘‘Nonlinear opti-
Adv. Neural Inf. Process. Syst., vol. 30, 2017, pp. 1–9. cal multi-image encryption scheme with two-dimensional linear canonical
[38] A. Patrascu and I. Necoara, ‘‘Efficient random coordinate descent algo- transform,’’ Opt. Lasers Eng., vol. 124, Jan. 2020, Art. no. 105821.
rithms for large-scale structured nonconvex optimization,’’ J. Global [60] Z. Hua, Y. Zhou, and H. Huang, ‘‘Cosine-transform-based chaotic system
Optim., vol. 61, no. 1, pp. 19–46, Jan. 2015. for image encryption,’’ Inf. Sci., vol. 480, no. 1, pp. 403–419, Apr. 2019.
[39] Z. Q. Luo and P. Tseng, ‘‘On the convergence of the coordinate descent
method for convex differentiable minimization,’’ J. Optim. Theory Appl.,
vol. 72, no. 1, pp. 7–35, 1992.
[40] G. Alvarez and S. Li, ‘‘Some basic cryptographic requirements for
chaos-based cryptosystems,’’ Int. J. Bifurcation Chaos, vol. 16, no. 8,
pp. 2129–2151, 2006.
[41] M. Preishuber, T. Hütter, S. Katzenbeisser, and A. Uhl, ‘‘Depreciating XIYU SUN received the bachelor’s degree from
motivation and empirical security analysis of chaos-based image and the School of Mechanical Engineering, Hunan
video encryption,’’ IEEE Trans. Inf. Forensics Security, vol. 13, no. 9, Institute of Science and Technology, China. He is
pp. 2137–2150. Sep. 2018. currently pursuing the master’s degree in elec-
[42] F. Özkaynak, ‘‘Brief review on application of nonlinear dynamics in image tronic information with Hengyang Normal Univer-
encryption,’’ Nonlinear Dyn., vol. 92, no. 2, pp. 305–313, Apr. 2018. sity, Hunan, China. His research interests include
[43] T. Wang and M.-H. Wang, ‘‘Hyperchaotic image encryption algorithm chaotic encryption and image processing.
based on bit-level permutation and DNA encoding,’’ Opt. Laser Technol.,
vol. 132, Dec. 2020, Art. no. 106355.
[44] Z. Hua, B. Xu, F. Jin, and H. Huang, ‘‘Image encryption using Josephus
problem and filtering diffusion,’’ IEEE Access, vol. 7, pp. 8660–8674,
2019.
[45] T. S. Ali and R. Ali, ‘‘A new chaos based color image encryption algorithm
using permutation substitution and Boolean operation,’’ Multimedia Tools
Appl., vol. 79, nos. 27–28, pp. 19853–19873, Jul. 2020. ZHONG CHEN received the M.S. degree in
[46] M. Z. Talhaoui, X. Wang, and M. A. Midoun, ‘‘A new one-dimensional applied mathematics from the Department of
cosine polynomial chaotic map and its use in image encryption,’’ Vis. Applied Mathematics, Southwest Jiaotong Uni-
Comput., vol. 37, no. 3, pp. 541–551, 2021.
versity, China, in 2004, and the Ph.D. degree
[47] R. Enayatifar, A. H. Abdullah, and I. Isnin, ‘‘Chaos-based image encryp-
in mechanical engineering from Hunan Univer-
tion using a hybrid genetic algorithm and a DNA sequence,’’ Opt. Lasers
Eng., vol. 56, pp. 83–93, May 2014. sity, China, in 2018. He is currently an Associate
[48] M. Ghazvini, M. Mirzadi, and N. Parvar, ‘‘A modified method for image Professor at the School of Computer Science
encryption based on chaotic map and genetic algorithm,’’ Multimedia Tools and Technology, Hengyang Normal University,
Appl., vol. 79, nos. 37–38, pp. 26927–26950, Oct. 2020. Hunan, China. His research interests include dig-
[49] X. Zhang, L. Wang, Z. Zhou, and Y. Niu, ‘‘A chaos-based image encryption ital image encryption, nonlinear dynamics, and
technique utilizing Hilbert curves and H-fractals,’’ IEEE Access, vol. 7, deep learning.
p. 74734–74746, 2019.

VOLUME 10, 2022 114611