Received 20 September 2022, accepted 1 October 2022, date of publication 5 October 2022, date of current version 28 November 2022.

Digital Object Identifier 10.1109/ACCESS.2022.3212145

A Dual-Domain Image Encryption Algorithm

Based on Hyperchaos and Dynamic
Wavelet Decomposition
QIUXIA QIN1 , ZHONGYUE LIANG1 , SHUANG LIU 1, XIAO WANG2 , AND CHANGJUN ZHOU 3
1 College of Computer Science and Engineering, Dalian Minzu University, Dalian 116600, China
2 Xingzhi College, Zhejiang Normal University, Jinhua 321004, China
3 College of Mathematics and Computer Science, Zhejiang Normal University, Jinhua 321000, China

Corresponding authors: Xiao Wang ([email protected]) and Changjun Zhou ([email protected])

This work was supported in part by the National Natural Science Foundation of China under Grant 62272418 and Grant 62102058, and in
part by the Basic Public Welfare Research Program of Zhejiang Province under Grant LGG18E050011.

ABSTRACT To ensure the safe and reliable transmission of images on public channels, this paper proposes
a dual-domain image encryption algorithm based on hyperchaos and dynamic wavelet decomposition. The
combination of dynamic wavelet decomposition and scrambling and diffusion operations is adopted in our
algorithm to realize the combination of spatial and frequency domain encryption. This not only ensures
the security of the encryption algorithm, but also ensures the robustness and operating efficiency of the
encryption, and at the same time reduces the risk of being attacked. First, divide the original image into
blocks, use a random number sequence to control the block scrambling process, and generate a scrambling
matrix; Then by calculating the Hamming distance related to the plaintext, dynamically selecting the wavelet
type, performing wavelet decomposition, and generating a wavelet coefficient matrix; Re-input the plaintext
image to the SHA-512 algorithm to generate the initial value of the hyperchaos. The chaotic system generates
the chaotic key matrix through iteration; Then the scrambling matrix is dynamically rotated, and then the
Zigzag transform is used to generate the key matrix; Finally, the wavelet coefficient matrix, the chaotic key
matrix, and the key matrix are subjected to bitwise XOR operation to realize the diffusion of pixel values and
obtain the final encrypted image. Simulation experiments and performance analysis experiments can show
that this algorithm can effectively encrypt and decrypt images, and has good encryption and decryption
quality, and the ability to resist various attacks.

INDEX TERMS Dynamic wavelet decomposition, hyperchaos, image encryption, SHA-512, Zigzag
transform.

I. INTRODUCTION cause huge losses and harm to the user. Therefore, ensur-
With the vigorous development of Internet technology, mul- ing the safe sharing and dissemination of images on public
timedia information such as text, voice, image, and video has channels has gradually become a vital topic in the field of
been shared and disseminated on the Internet platform [1], information security, and has attracted more and more schol-
[2]. Among them, image as an intuitive transmission medium ars’ attention and research. Among them, encryption is one of
has been widely used in the information age, and people use the most common means of protecting information security.
image data to convey information and exchange emotions [3], Compared with watermarking, hiding, steganography, etc.
[4]. However, this form of communication also brings many References [5], [6], [7], and [8], encryption can better protect
risks. Once the necessary information of the image is mali- the content of information, especially for the protection of
ciously obtained and tampered with by the attacker, it will private information.
Because chaos has inherent characteristics such as
The associate editor coordinating the review of this manuscript and initial sensitivity, ergodicity, pseudo-randomness, etc.
approving it for publication was Yeliz Karaca . Reference [9], it is very in line with the requirements of

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 License.

122726 For more information, see https://creativecommons.org/licenses/by-nc-nd/4.0/ VOLUME 10, 2022
Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

cryptography, and the chaotic system generates sequences implement scrambling and diffusion operations. For example,
very fast, so image encryption algorithms based on chaos Wang and Gao [23] proposed a chaos image encryption algo-
theory have been continuously proposed and improved [10], rithm based on matrix semi-tensor product and compound key
[11], [12]. The control parameters of a chaos system are often in 2020. This algorithm mainly performs semi-tensor prod-
used as the key of an encryption algorithm. In different uct operations on chaotic sequences and scrambled images,
chaos systems, the number of selected control parameters is which effectively improves the security of diffusion opera-
different. Classical one-dimensional chaotic systems, such as tions. In 2021, Wang et al. [24], [25] successively proposed
Logistic chaotic map and Sine map [10], have disadvantages two spatial image encryption algorithms. The first algorithm
such as a few control parameters, small generated key space, proposed [25] is an image encryption algorithm based on the
limited chaotic interval, and low security. The image encryp- chaotic diffusion value of the truth table. This algorithm uses
tion algorithms based on chaos theory use the pseudo-random the classic scrambling-diffusion framework to perform row,
sequence generated by chaos to operate the image according column, and diagonal bidirectional scrambling and diffu-
to the characteristics of the periodic window. In addition, the sion operations by using nonlinear chaotic sequences, which
image encryption effect is related to the periodic window. greatly improve the effects of scrambling and diffusion, and
For example, logistic, sine, and tent chaotic maps [10] have the ciphertext image is simultaneously affected by the chaos
short periotic windows, and it is difficult to guarantee the ran- system and the truth table rules, making it more secure. In the
domness of the encryption algorithm. In order to improve the same year, the second algorithm proposed by Wang et al. [24]
security and randomness of encryption algorithms, and solve is an image encryption algorithm based on dynamic row
the shortcomings of low-dimensional chaos in encryption, scrambling and Zigzag transform. The algorithm is based
researchers have proposed many high-dimensional chaotic on the idea of standard Zigzag scrambling, and the spe-
systems and hyperchaos. For example, Xiu et al. [11] com- cial traversal method was adopted to improve the scram-
bined the memory characteristics of cellular neural networks bling effect. The results show that the encryption algorithm
to design a five-dimensional memory characteristic CNN also has better security. In addition, in frequency domain
hyperchaos, which has the characteristics of large parame- encryption, the image is regarded as a two-dimensional sig-
ter space and good chaotic characteristics. Zhu et al. [12] nal with varying amplitude. The image can be processed
constructed a five-dimensional continuous hyperchaos. The by Fourier transform [21], discrete cosine transform [21],
chaos has two positive Lyapunov exponents, which is more wavelet decomposition and other methods [22] to generate
random than general high-dimensional chaotic systems. They frequency coefficient matrix, the matrix is processed to a
are used in encryption algorithms to improve the robustness certain extent, and finally the corresponding inverse trans-
of the algorithm and safety [12]. These encryption schemes formation can achieve the purpose of image encryption.
incorporate many new technologies in the algorithm design, For example, Wu et al. [26] proposed a new color image
improve the complexity of the algorithm, and also ensure lossless encryption scheme based on two-dimensional dis-
the security of ciphertext images. The above research results crete wavelet and six-dimensional hyperchaos. First, two-
show that the chaotic system is a very useful tool for gen- dimensional discrete wavelet is used to divide the image into
erating pseudo-random sequences for image encryption, and four subbands, and then a constant factor is used to change
the chaotic-based image encryption scheme is an effective the size of the subband, and finally the reconstruction of
image encryption method. In addition, because hyperchaos the subband. The experimental results show that the algo-
have at least two positive Lyapunov exponents, they have rithm is safe, fast, and capable of resisting various attacks,
stronger randomness. Compared with chaotic systems and which greatly enhances the performance of the encryption
high-dimensional chaotic systems, they have the character- algorithm.
istics of large key space, at least one control parameters, For the above two encryption methods, the spatial domain
and large chaotic intervals. In recent years, it has received encryption is more secure, and the frequency domain encryp-
continuous attention and research from researchers, and has tion can make images more chaotic and more efficient. Both
been widely used in chaotic image encryption. encryption methods can turn a visually meaningful image
Spatial domain encryption [13], [14], [15], [16], [17], [18], into a safe snowflake noise image, realizing effective image
[19], [20] and frequency domain encryption [21], [22] pro- encryption. The above references all use a single domain
vide different perspectives for image encryption. Between to encrypt images. In order to further improve encryption
them, in the spatial domain encryption, it is mainly carried efficiency and visual security, Wang et al. [27] proposed a
out by means of scrambling and diffusion. The scrambling three-image encryption and hiding algorithm based on chaos,
operation changes the position of the pixel and breaks the compressed sensing and three-dimensional discrete cosine
strong correlation between the adjacent pixels of the image. transform, which is based on dual-domain encryption, using
The diffusion operation changes the size of the pixel value, two-dimensional discrete wavelet to transform the image into
and the size of the pixel value is mutually diffused, making a sparse matrix, and then scrambling and compressing the
the image more chaotic. Chaos image encryption algorithms sparse matrix twice, and finally embedding the matrix into
based on spatial domain generally use a scrambling-diffusion the color carrier image. The results show the performance of
framework, and use sequences generated by chaotic maps to the algorithm on statistical information such as correlation

VOLUME 10, 2022 122727

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

coefficients and information entropy, all are better than single

domain encryption. Hua et al. [28] proposed a new vision-
safe image encryption scheme based on compressed sensing,
which designed a new parallel compressed sensing technol-
ogy and realized adaptive threshold sparsification using SWT.
The results show that the algorithm has more high recon-
struction quality and encryption. Yan et al. [29] proposed
a chaotic image encryption algorithm based on fractional
scrambling wavelet transform and three-dimensional cyclic
shift operation. The original image is decomposed by three-
FIGURE 1. Bifurcation diagram of 2D-LSM.
level fractional wavelet to obtain low-frequency compo-
nents and high-frequency components. The high-frequency
components are scrambled by chaotic sequence, the low- decryption algorithms proposed. Section IV carries on the
frequency components are scrambled by three-dimensional experiment simulation. Section V performs performance
cyclic shift, and finally reconstructed to obtain the encrypted analysis. The last section concludes the paper.
image. Experimental results show that the algorithm has a
good encryption effect. Shakir et al. [30] proposed an image II. RELATED THEORY OF THE ENCRYPTION ALGORITHM
encryption method based on the combination of Logistic A. 2D-LSM
chaotic mapping and wavelet transform. Firstly, Haar wavelet The 2D-LSM [31] is obtained by diffusing the phase space of
transform is performed on the plaintext image to obtain differ- the one-dimensional Logistic map and the one-dimensional
ent frequency domains of the image, and then Logistic chaotic Sine map into two dimensions. The definition of this two-
mapping is used to scramble the generated matrix, the test dimensional hyperchaos is as follows:
results show that the algorithm has achieved good encryption (
effect. It can be seen from the above analysis that the cur- xi+1 = cos(4axi (1 − xi ) + bsin(πyi ) + 1);
(1)
rent research on dual-domain encryption schemes is neither yi+1 = cos(4ayi (1 − yi ) + bsin(πxi ) + 1).
sufficient nor perfect. Although combining the scrambling
operation in the spatial with the frequency domain makes the where x, y are the system variables of the hyperchaotic sys-
encryption algorithm more resistant to attacks and has a better tem, a, b are the control parameters of the hyperchaos. Since
encryption effect, it ignores the diffusion operation that is as the cosine transform is bounded for any input, the control
important as the scrambling operation in the spatial domain parameters a, b can be arbitrarily large [31]. When x(1) =
encryption. Therefore, the algorithm based on dual-domain 0.1, y(1) = 0.2, a = b = 60, the resulting bifurcation
encryption needs to be further improved. diagram is shown in Fig. 1. Obviously, the trajectory of the
Based on the above analysis, in order to further improve the 2D-LSM is distributed in the whole phase space, indicating
encryption effect of dual-domain encryption, the frequency that the chaotic system has complex chaotic characteristics.
domain can not only be combined with the scrambling oper- Besides, Fig. 2(a) gives the case of Lyapunov exponent (LE)
ation in the spatial domain, but also can be combined with when a ∈ (1, 100), b = 60, Fig. 2(b) shows the case of
the diffusion operation in the spatial domain, fully integrating Lyapunov exponent (LE) when a = 60, b ∈ (1, 100). It can be
the advantages of the two encryption methods. Therefore, this seen that 2D-LSM has a continuous chaotic range, the larger
paper proposes a dual-domain image encryption algorithm positive Lyapunov exponent, and is always in a hyperchaotic
based on hyperchaos and dynamic wavelet decomposition, state.
which not only ensures the security but also takes into account
the efficiency of encryption. The chaotic map selected in
this paper is a hyperchaos, and the complex chaotic char-
acteristics ensure the security of the algorithm. As for the
architecture, the combination of scrambling and diffusion in
the spatial domain and wavelet decomposition in the fre-
quency domain improves the performance of the algorithm.
In terms of efficiency, the improved Zigzag can achieve fast
scrambling without increasing complexity. Additionally, the
keystream used in the encryption process all relies on the FIGURE 2. Lyapunov index spectrum of 2D-LSM (a) LEs(b = 60),
original image. The experimental results compared with most (b) LEs(a = 60).

existing encryption algorithms show that the proposed algo-

rithm achieves the dual goals of security and efficiency. In this paper, the chaotic sequence generated by 2D-LSM
The rest of the paper is organized as follows. Section II is transformed into a chaotic key matrix to participate in the
introduces the basic principles and knowledge of the algo- XOR diffusion process, which increases the randomness and
rithm. Section III introduces the specific encryption and security of the algorithm.

122728 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

FIGURE 3. Three-layer wavelet decomposition diagram.

B. DYNAMIC WAVELET DECOMPOSITION FIGURE 4. Traditional Zigzag transform.

Wavelet decomposition is an important form of frequency
domain encryption [27], [32]. In wavelet decomposition,
the image is segmented through analysis filters. Each stage It can be seen from Figure 4 that the pixel value of some
of the analysis filter is composed of high-pass and low- positions in the scrambled matrix obtained by the traditional
pass filters. After each level is filtered, an image subband Zigzag transform of the 3*3 matrix is fixed, and the position
will be generated. The generated image sub-bands include of the pixel cannot be completely scrambled, and there are
two types, namely low-pass sub-bands and high-pass sub- limitations in scanning. In addition, the traditional Zigzag
bands. Among them, there are three types of high-pass sub- transform works on a square matrix, which has a limited
bands, which are three high-pass sub-bands in the horizontal, scope of application. In order to solve the above problems,
vertical, and diagonal directions. Generally, LL represents this paper improves the traditional Zigzag transform to form
the low-pass sub-band, LH represents the vertical high- an improved Zigzag transform. Firstly, the object of the
pass sub-band, HL represents the horizontal high-pass sub- improved Zigzag transform is no longer limited to square
band, and HH represents the diagonal high-pass sub-band. matrices, as shown in Fig. 5. Secondly, before the image
If it is a multi-layer wavelet decomposition, the LL low- matrix is scanned, the matrix is rotated first, and the specific
pass subband is decomposed in the same way. Until the end rotation angle is determined by the key related to the plaintext,
of multi-layer wavelet decomposition, a wavelet coefficient and then the matrix is transformed. This method makes the
matrix is obtained. The following takes three-layer wavelet scanning result no longer single and fixed, and has stronger
decomposition as an example to show the process of wavelet randomness, and does not greatly increase the computational
decomposition, as shown in Fig. 3. complexity. Finally, when the traditional Zigzag transform
Generally speaking, LL contains an approximate represen- transforms a square matrix with an odd side length, there is a
tation of the input image, while LH, HL, and HH only contain problem that the most central element of the matrix cannot
edge information and almost no energy. The process of three- change its position after one round transform. Therefore,
layer wavelet decomposition can be clearly and intuitively the improved Zigzag transform increases the backward shift
seen from Figure 3. The LL of the upper level can be used processing of the center element and deal with this problem
as the input of the next level, and the relevant detail signals effectively. It should be noted that the improved Zigzag trans-
can be obtained through the filtering of each level. To increase form does not need to deal with non square matrices and even
the randomness of the algorithm, a dynamic wavelet decom- side length square matrices except for the special treatment
position method is proposed in this paper. This method uses of the central elements of odd side length square matrices.
the key related to the plaintext to select the specific wavelet Taking a 3*3 square matrix as an example, an improved
type used in each wavelet decomposition, and calculates the Zigzag transform is performed, as shown in Fig. 6.
specific number of wavelet decomposition layers according Figure 5 shows the process of the improved Zigzag
to the size of the input image matrix. Finally, the frequency transform of non-square matrix, which shows that the
coefficient matrix representing the original image is obtained. improved Zigzag transform has a wider scope of application.
In Figure 6, the generated Hamming distance related to the
C. ZIGZAG TRANSFORM plaintext is used as a key to select a specific rotation angle,
1) IMPROVED ZIGZAG TRANSFORM and then the matrix is rotated counterclockwise. The matrix
The traditional Zigzag transform is often used to implement rotation operation makes a matrix have four possible outputs.
image scrambling due to its simplicity and ease of implemen- In addition, since the matrix is a square matrix with odd side
tation and low time complexity [33]. Starting from the upper lengths, the most central element of the output sequence is
left corner of the image matrix, scan in a zigzag shape, put the moved to the end of the sequence, and other elements remain
scanned values into a one-dimensional sequence, and finally unchanged, which solves the problem that some elements
turn the one-dimensional sequence into a two-dimensional remain fixed. In this paper, the improved Zigzag transform
matrix to complete the scrambling process. The following is mainly used to generate the key matrix, and the gener-
takes a 3*3 matrix as an example to show the traditional ated key matrix will participate in the subsequent diffusion
Zigzag transform process, as shown in Fig. 4. operation.

VOLUME 10, 2022 122729

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 1. Correlation coefficients of images in each round transform.

It can be seen from Figure 7 that as the number of rounds

increases, the images obtained by the three transform meth-
ods become more and more random, but the scrambling effect
cannot be directly judged visually. It can be seen from the
correlation coefficients in Table 1 that in one round trans-
FIGURE 5. Improved Zigzag transform of non-square matrix.
form, the correlation coefficients of the images generated
by the improved Zigzag in some directions are better than
other transform methods. As the number of rounds increases,
the ability of the improved Zigzag transform to reduce the
correlation becomes more and more prominent. Therefore,
it can be shown that the improved Zigzag transform proposed
in this paper can effectively reduce the correlation between
adjacent pixels of the image.

III. PROPOSED IMAGE ENCRYPTION AND DECRYPTION

ALGORITHM
This paper proposes a dual-domain image encryption algo-
rithm based on hyperchaos and dynamic wavelet decompo-
sition, which is mainly divided into two stages: scrambling
and diffusion. It is worth mentioning that this algorithm uses
frequency-domain wavelet decomposition for the diffusion
process in the spatial domain. It is an effective fusion of
spatial-domain encryption algorithm and frequency-domain
encryption, which improves the security and real-time per-
formance of encryption. In the scrambling stage, the image
FIGURE 6. Improved Zigzag transform of odd-sided square matrix. matrix is divided into blocks, and then a random number
sequence is used to control the scrambling process of each
2) ZIGZAG PERFORMANCE EVALUATION block; In the diffusion stage, the scrambling matrix first
In addition to the problems of fixed elements and limited generates a frequency coefficient matrix through dynamic
scope of use in traditional Zigzag transform, the generated wavelet decomposition, and then hyperchaos generates a
scrambling matrix often has the problems of strong corre- chaotic key matrix, and then uses an improved Zigzag trans-
lation between adjacent pixels and relatively concentrated form to generate a key matrix. Finally, the three kinds of
distribution. In order to solve the above problems, Ramasamy matrices are XORed to complete the diffusion process in
et al. [34] also proposed an improved Zigzag transform, a simple and efficient way, and realize the encryption of
which firstly exchanges the first and second elements of the the image. Section III-A-III-D will describe the encryption
matrix with the last two elements, and then performs trans- process in detail, and the encryption flow is shown in Fig. 8.
formation, which distorts the relationship between image
pixels. Fig. 7 shows the images obtained by performing three A. KEY GENERATION
rounds transform on the Rice image by the traditional Zigzag The key of this algorithm mainly contains three parts: the
transform, the Zigzag transform proposed by Ramasamy, initial value keys of the hyperchaos, the key used in the
and the improved Zigzag transform proposed in this paper. dynamic wavelet decomposition process to select the wavelet
In addition, the correlation coefficients of adjacent pixels of type, and the key used in the improved Zigzag transform to
each scrambled image in Figure 7 are also calculated using select the rotation angle of the image matrix. The specific
the Eq. (39), as shown in Table 1. process is shown in section III-A1-III-A3.

122730 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

FIGURE 7. Image comparison of three rounds of transform of Rice image by three Zigzag
transform methods (a) Rice original image, (b) traditional Zigzag one round transform,
(c) Ramasamy-Zigzag one round transform, (d) improved Zigzag one round transform,
(e) traditional Zigzag two rounds transform, (f) Ramasamy-Zigzag two rounds transform,
(g) improved Zigzag two rounds transform, (h) traditional Zigzag three rounds transform,
(i) Ramasamy-Zigzag three rounds transform, (j) improved Zigzag three rounds transform.

image using the 512-bit message digest of the SHA-512 hash

function. The specific process of the initial value key of the
hyperchaos is as follows:
Step 1: The original image P is used as a variable of
the hash function to generate a 512-bit message digest,
as follows:
Info = hash(P,0 SHA − 5120 ) (2)
Among them, hash() is the hash function, P is the original
image, and Info is the message digest generated.
Step 2: Since the generated message digest is given in the
form of a 128-bit hexadecimal string, it is necessary to first
convert the 128-bit hexadecimal string into 128 decimal, and
further into 64 decimal DInfo1 . The process is as follows:
DInfo = hex2dec(Info(i)), i = 1, 2, . . . , 128 (3)
DIfo1 = DInfo(j) × 10 + DInfo(j + 1), j = 1, 2, . . . , 64
FIGURE 8. Encryption flowchart.
(4)
Among them, hex2dec() is a function that transforms hex-
1) INITIAL VALUE KEY OF THE 2D-LSM
adecimal into decimal.
Step 3: Generate initial values x(1), y(1) of 2D-LSM hyper-
The biggest feature of the hash function is that the forward
chaotic system using 64 decimal numbers DInfo1 , take the
operation is relatively easy, and the reverse operation is very
value from the front to back interval, the interval is 4, a total
difficult. On the one hand, however large the input value is,
of 17 values are generated, and the first 16 values are used to
it can be converted to 512 binary bits after using the SHA-
generate the initial chaotic value. The generation process is
512 function to output, and different inputs produce differ-
as follows:
ent results through this function, and the results are unique,
so this part of the key is also unique. On the other hand, x(1) = DInfo1 {1} ⊕ DInfo1 {4} ⊕ . . . ⊕ DInfo1{28} (5)
the keys generated by the hash function are related to the y(1) = DInfo1 {32} ⊕ DInfo1 {36} ⊕ . . . ⊕ DInfo1 {60} (6)
plaintext, which can effectively resist plaintext attacks. The
initial value key of the 2D-LSM is generated by the original Among them, ⊕ is a bitwise XOR operation.

VOLUME 10, 2022 122731

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

Step 4: There will be transient effects during the iteration

of the 2D-LSM. To eliminate the transient effects, the random
values generated by the previous iteration of the system need
to be discarded. This algorithm uses the last decimal value of
DInfo1 to determine the specific iterative discard that needs
to be discarded, as follows:
discard = 2000 + DInfo1 {64} (7)

2) SELECTION KEY OF WAVELET TYPE

In the process of dynamic wavelet decomposition, which
wavelet type is used for each image matrix is generated by
the original image through the relevant theory of Hamming
distance, as follows:
FIGURE 9. Schematic diagram of block scrambling.
Step 1: The number of rows of the original image P is M
and the number of columns is N , and a certain pixel value
P(i, j) of the original image is randomly selected using a so that the value of the generated selection angle RotAng is
random number generation function, as shown below: between 1 and 4. The processing process is as follows:
(
i = randperm(M , 1) RotAng = mod(Min, 4) + 1 (11)
(8)
j = randperm(N , 1) Among them, mod() is the remainder function.
Among them, randperm(n, k) means to generate k random
numbers from 1 − n. B. BLOCK SCRAMBLING STAGE
Step 2: Change the pixel value P(i, j) of the selected origi- This stage is mainly to divide the original image P into blocks,
nal image into 8-bit binary form, as follows: and then generate a random number sequence, use the random
number sequence to scramble each block, and complete the
P1 = dec2bin(P(i, j), 8) (9) process of generating the matrix P_scramble after the scram-
bling. The principle of block scrambling is shown in Fig. 9.
Among them, dec2bin() is a function that converts a decimal
Among them, the ‘‘Index={3,1,5,6,2,9,4,7,8}’’ in Fig. 9 is
number into a binary form. The second variable of the func-
a pseudo-random number sequence obtained by a random
tion represents the specific number of bits in the generated
number generator, and the pseudo-random number sequence
binary.
is used to control block scrambling. The specific process is
Step 3: The system gives an 8-bit binary sgn value, and
as follows:
calculates the Hamming distance HD between sgn and P1 .
Step 1: The plaintext image P is divided into 16*16 small
Because the two variables are both 8-bit binary values, they
blocks, the specific process is as follows:
are compared one by one starting from the first bit. If the two (
are equal, HD increases by 1. If the two are not equal, HD r = size(P, 1)/16
remains unchanged. After the comparison of the 8-bit values (12)
c = size(P, 2)/16
is completed, the final Hamming distance HD is obtained.
P_scramble = mat2cell(P, ones(1, r)×16, ones(1, c)×16)
Use the Hamming matrix HD to select one of the eight
wavelet types, and continue the dynamic wavelet decompo- (13)
sition operation. Among them, size() is a function to calculate the size of the
matrix. When the second variable of the function is 1, the
3) SELECTION KEY OF MATRIX ROTATION ANGLE output result is the total number of rows in the matrix. When
In the improved Zigzag transform, the matrix needs to be the second variable of the function is 2, the output result is
rotated first. The specific rotation angle is generated by the the total number of columns in the matrix. The mat2cell()
original image and the minimum concept, as shows: can divide a matrix into blocks of different sizes. ones() is a
Step 1: Calculate the minimum value of all pixel values of function to generate a matrix of all ones.
the original image P as follows: Step 2: Use the random number generation function to
generate a random number sequence with the same size as the
Min = min(min(P)0 ) (10)
total block number bsum, the generation process is as follows:
Among them, min() is a function to find the minimum value. bsum = r × c (14)
Since P is a two-dimensional matrix, it is necessary to call the
seq_scramble = randperm(bsum) (15)
min() twice to calculate the minimum value of all elements.
Step 2: Since there are 4 possible rotation angles, it is Among them, randperm(n, k) means to generate k random
necessary to perform certain processing on the Min value numbers from 1 − n.

122732 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

Step 3: Use the seq_scramble sequence to perform block 2) GENERATING CHAOTIC MATRIX OF 2D-LSM
scrambling on each block, and obtain the scrambling matrix Using the initial value of the 2D-LSM generated in
after completion. The process is as follows: section III-A1, the 2D-LSM is iterated many times to generate
the chaotic key matrix CM , details as follows:
P_scramble = cell2mat(reshape(P_scramble Step 1: The length of the original image P is M , the width
(seq_scramble), r, c))) (16) is N , and discard is the number of iterations that need to be
discarded in the early stage of the chaotic system generated in
Among them, cell2mat() converts a cell array composed of section III-A1. First, determine the total number of iterations
multiple matrices into a matrix function, and reshape() is of the chaos. The process is as follows:
a function to transform a specified matrix into a matrix of
specific dimensions. sum_discard = ceil(M × N /2) + discard (20)

where ceil() is the round-up function.

C. DIFFUSION STAGE
Step 2: In order to avoid transient effects, using the initial
The wavelet coefficient matrix generated by dynamic wavelet chaos value in III-A1, the 2D-LSM is iterated for discard
decomposition, the chaotic matrix generated by the 2D-LSM, times, and the generated chaotic value is discarded.
and the key matrix generated by the improved Zigzag trans- Step 3: Then continue to iterate. Since the chaotic system
form are sequentially XORed to realize the diffusion pro- is two-dimensional, each iteration will generate two chaotic
cess. Although the diffusion process is simple and easy to random values. The two chaotic random values generated
achieve, the diffusion effect is good. The details are shown each time are placed in sequence seq1 until the number of
in section III-C1-III-C4. iterations reaches sum_discard, generating M ∗N size chaotic
sequence seq1 .
1) DYNAMIC WAVELET DECOMPOSITION Step 4: Perform certain processing on the chaotic sequence
Select the number of wavelet decomposition layers according seq1 , first control the value of the chaotic sequence between
to the size of the original image P, use the HD key generated 1-255, and then convert the chaotic sequence into a chaotic
in section III-A2 to select the specific wavelet type used for matrix CM a length of M and a width of N , details as follows:
each encryption, and then perform wavelet decomposition on
the matrix generated after the scrambling, and the decompo- seq1 = mod(floor(seq1 × 1010 ), 256) (21)
sition is completed to generate a wavelet coefficient matrix CM = reshape(seq1 , M , N ) (22)
P_W , details as follows:
Step 1: Calculate the specific level level of wavelet decom- where floor() is a round down function, mod() is a remain-
position according to the original image P, the process is as der function, and the reshape() is a matrix reconstruction
follows: function.

n = size(P, 1) (17) 3) IMPROVED ZIGZAG TRANSFORM TO GENERATE KEY

level = log2(n) (18) MATRIX
The improved Zigzag transform consists of two parts. First,
Among them, the size() function is used to calculate the size use the RotAng key generated in section III-A3 to control the
of the matrix, and log2() is a logarithmic function. angle of the matrix rotation and perform the rotation of the
Step 2: According to the key HD, select the specific matrix. Then, the rotated matrix is scanned by Zigzag to gen-
wavelet Wavelet to be used, and calculate the four filters erate a one-dimensional sequence, and the one-dimensional
associated with this type of wavelet. The process is as follows: sequence is transformed into a two-dimensional matrix to
obtain the key matrix KM , details as follows:
[L_D, H _D, L_R, H _R] = wfilters(Wavelet) (19) Step 1: Rotate the scrambled matrix P_scramble according
to RotAng, as follows:
Among them, wfilters() is the function used to calculate the
wavelet filter, L_D is the decomposition low-pass filter, H _D RM = rot90(P_scramble, RotAng) (23)
is the decomposition high-pass filter, L_R is the reconstruc-
tion low-pass filter, and H _R is the reconstruction high-pass Among them, the rot90() is a function that makes a matrix
filter. rotate counterclockwise.
Step 3: The scrambled matrix P_scramble and L_D are Step 2: Perform a Zigzag scan on the matrix RM after
the decomposition low-pass filter, H _D is the decomposition matrix rotation to generate a one-dimensional sequence seq2 ,
high-pass filter, and the wavelet decomposition layer number as follows:
level is used as the variable parameter of the wavelet decom- seq2 = zigzag(RM ) (24)
position to perform the multi-layer wavelet decomposition
operation. After the multi-layer decomposition is completed, Among them, the zigzag() is a function for performing Zigzag
the frequency wavelet coefficient matrix P_W is generated. scanning.

VOLUME 10, 2022 122733

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

Step 3: Turn the one-dimensional sequence seq2 into a key

matrix KM of size M ∗ N , as follows:
KM = reshape(seq2 , M , N ) (25)
Among them, the reshpae() is a matrix reconstruction func-
tion.

4) SPECIFIC DIFFUSION PROCESS

The wavelet coefficient matrix P_W generated in
section III-C1, the 2-D chaotic key matrix CM generated
in section III-C2, and the key matrix KM generated in
section III-C3 are XORed to achieve diffusion, and the final
encrypted image C is obtained, details as follows:
Step 1: First, process the P_W matrix and control the value
of the matrix between 1-255. The process is as follows:
P_W = mod(floor(P_W × 1015 ), 256) (26) FIGURE 10. Decryption flowchart.

where floor() is a round down function, and mod() is a

E. DECRYPTION ALGORITHM
remainder function.
The decryption algorithm and the encryption algorithm have
Step 2: The P_W matrix, the CM matrix and the KM
a reversible relationship [15], [17], [20]. Among them, the
matrix are bitwise XORed to obtain the encrypted matrix C,
encryption key and the decryption key are consistent, and the
as follows:
chaotic sequence used in the decryption process is also con-
C(i, j) = P_W (i, j) ⊕ CM (i, j) ⊕ KM (i, j) (27) sistent with the encryption algorithm. The decryption party
i = 1, 2, . . . .M (28) uses the decryption key to generate the chaotic key matrix and
j = 1, 2, . . . .N (29) the key matrix according to section III-C2 and section III-C3.
Also note the following: First of all, the purpose of the scram-
Among them, ⊕ is a bitwise XOR operation. bling operation in the encryption and decryption algorithm
is to disturb the position of the pixel value and break the
D. ENCRYPTION ALGORITHM strong correlation between the pixels. This is a reversible
The dual-domain image encryption algorithm proposed in process. Second, the purpose of the diffusion operation in the
this paper based on hyperchaos and dynamic wavelet decom- encryption and decryption algorithm is to change the size of
position, the specific steps are as follows: the pixel value, and this process is also reversible. In addition,
1) The original image P is divided into blocks first, and the wavelet decomposition operation and wavelet reconstruc-
then each block is scrambled under the control of a random tion operation in the frequency domain are also reversible.
number sequence, as shown in section III-B. Therefore, the decryption algorithm in this paper is mainly
2) The key obtained in section III-A2 is used to select the divided into three parts: the inverse process of the diffusion
wavelet type, and then the dynamic wavelet decomposition operation, the wavelet reconstruction process, and the inverse
operation is performed on the matrix P_scramble obtained process of the scrambling operation. The flowchart of the
after the scrambling to obtain the wavelet coefficient matrix entire decryption algorithm is shown in Fig. 10.
P_W , as shown in section III-C1. The concrete steps of the decryption algorithm are as
3) In section III-A1, the initial value key of the 2D-LSM is follows:
obtained as the initial value of chaos, and the chaos is iterated. 1) The encrypted image C, the key matrix KM , and the
In order to avoid transient effects, the number of discarded key chaotic matrix CM are sequentially performed the inverse
previous iterations and the total number of iterations are operation of the bitwise XOR operation to obtain the wavelet
calculated in advance. When all the system iterations are coefficient matrix C_W .
over, the chaotic key matrix CM is generated, as shown in 2) Input the wavelet coefficient matrix C_W , L_R
section III-C2. reconstruction low-pass filter, H _R reconstruction high-pass
4) The Hamming distance HD key obtained in filter, and wavelet decomposition layer level level as param-
section III-A3 is used to select the angle of matrix rotation. eters into the wavelet reconstruction function, reconstruct the
The matrix P_scramble after the scrambling is first subjected wavelet coefficient matrix, and get the result after scrambling
to matrix rotation, and then Zigzag transform is performed to matrix C_scramble.
generate the key matrix KM , as shown in section III-C3. 3) The matrix C_scramble obtained in 2) is divided into
5) The P_W matrix, the KM matrix and the CM matrix are blocks, and the process is as follows:
sequentially subjected to the bitwise XOR operation to realize
the diffusion operation, complete the image encryption, and C_scramble = mat2cell(C_scramble,
obtain the encrypted image C, as shown in section III-C4. ones(1, r) × 16, ones(1, c) × 16) (30)

122734 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 2. Key table.

Among them, mat2cell() is a function to transform a matrix

into a cell array, and ones() is a function to generate a matrix
of all ones.
4) Continue to reverse scrambling each block of
C_scramble to obtain the decrypted image D. The specific
process is as follows:

iseq1 = 1 : length(seq1 ) (31) FIGURE 11. 512*512 size image encryption and decryption experiment
results (a) Lena_512 original image, (b) Lena_512 encrypted image,
seq = [seq1 , iseq1 ]
0 0
(32) (c) Lena_512 decrypted image, (d) Building original image, (e) Building
encrypted image, (f) Building decrypted image, (g) Insect original image,
seq = sortrows(seq, 1) (33) (h) Insect encrypted image, (i) Insect decrypted image, (j) Flower original
image, (k) Flower encrypted image, (l) Flower decrypted image.
D = cell2mat(reshape(C_scramble(seq(:, 2)), r, c))
(34)

Among them, seq1 is the random number sequence used

when the encryption algorithm is scrambling, length() is the
function to calculate the length, sortrows() is the function to
sort the matrix according to the specified row, reshape() is
the matrix reconstruction function, cell2mat() is a function
that converts a cell array composed of multiple matrices into
a matrix.

IV. EXPERIMENTAL SIMULATION

This section shows the experimental simulation results of
image encryption and decryption. The specific operating
environment is: personal notebook computer, window10
operating system, Intel(R) Core(TM) i7-1065G7 CPU @
1.30GHz 1.50GHz, 16G memory, Matlab2020a simulation
platform. The keys used in this simulation experiment are
shown in Table 2.
The experimental simulation in this section is mainly
divided into three parts: the encryption and decryption exper- FIGURE 12. Experimental results of encryption and decryption of images
of different sizes (a) Leaf original image, (b) Leaf encrypted image,
iments on four groups of 512*512 images with differ- (c) Leaf decrypted image, (d) Insect original image, (e) Insect encrypted
ent contents, and the experiments on 256*256, 512*512, image, (f) Insect decrypted image, (g) Butterfly original image,
(h) Butterfly encrypted image, (i) Butterfly decrypted image.
1024*1024 images with different sizes and different contents
to perfome image encryption and decryption experiments.
Perform image encryption and decryption experiments on
all black and white images. Experimental images are algorithm. Fig. 13 shows the experimental results of using
from https://unsplash.com. Fig. 11 shows the experimen- White and Black images for encryption and decryption.
tal results of encryption and decryption of four groups It can be seen from Fig. 13 that even if a all white image or
of 512*512 images. a all black image is used, secure and effective encryption and
As can be seen from Fig. 11, 512*512 images can be decryption can still be achieved, indicating that the algorithm
effectively encrypted and decrypted. Fig. 12 shows the exper- has a strong ability to resist malicious attacks. From the
imental results of the encryption and decryption experiment experimental results, all the images after encryption become
of the algorithm on three groups of images of different sizes. noise-like images, which can hide the effective information
It can be seen from Fig. 12 that the size of the image of the plaintext image well and meet the requirements of
does not affect the normal encryption and decryption of this image encryption. And all the images can get the correct

VOLUME 10, 2022 122735

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

other hand, this algorithm uses the SHA-512 hash algorithm

to generate a key. In this process, the hash algorithm can
generate a 512-bit message digest, representing a key space of
at least 2512 , which is far greater than 2100 . In short, the key
space of this algorithm is large enough to resist brute force
attacks.

B. KEY SENSITIVITY ANALYSIS

Key sensitivity analysis is the analysis of the sensitivity of the
image key, and is an important indicator for evaluating the
security of the encryption system [39], [40]. Key sensitivity
mainly includes encryption key sensitivity and decryption key
sensitivity. In a secure encryption and decryption algorithm,
when the encryption key is slightly changed, the correspond-
FIGURE 13. Experimental results of encryption and decryption of all ing encrypted image will change to a greater extent; When
white and all black images (a) White original image, (b) White encrypted
image, (c) White decrypted image, (d) Black original image, (e) Black the decryption key is slightly changed, the corresponding
encrypted image, (f) Black decrypted image. decrypted image will also change to a larger degree. The
greater the degree of change in the corresponding encrypted
image or decrypted image, the stronger the key sensitivity.
decryption effect, which further proves the effectiveness of The pixel change rate NPCR and the uniform average
this algorithm. In short, the encryption algorithm proposed in change intensity UACI are usually used to measure the sensi-
this paper can encrypt and decrypt images of different sizes, tivity of the key [40]. Among them, the ideal value of NPCR is
or all black and all white images, and has a wide range of 99.6054%, and the ideal value of UACI is 33.4653% [41]. The
applications and has certain practical application value. stronger the key sensitivity is, the closer the two index values
obtained are to the ideal value. The definitions of NPCR and
V. PERFORMANCE ANALYSIS UACI are as follows:
Security analysis is the most important performance indica- PM −1 PN −1
i=0 j=0 C(i, j)
tor of encryption algorithm [34], [35], [36]. Therefore, this NPCR(D1 , D2 ) = × 100% (35)
section evaluates the security of encryption algorithms in key M ×N
PM −1 PN −1 |D1 (i,j)−D2 (i,j)|
space and sensitivity, security analysis of chaotic key gener- i=0 j=0 255
ation functions, and statistical analysis. In addition, a good UACI (D1 , D2 ) = × 100%
M ×N
encryption algorithm must be able to resist all known types (36)
of attacks, so this section analyzes the ability of the proposed
(
0, D1 (i, j) = D2 (i, j)
algorithm to resist all known attacks. In addition, a good C(i, j) = (37)
encryption algorithm must have a certain timeliness and can 1, D1 (i, j) 6= D2 (i, j)
accomplish the effect of fast encryption and decryption in a Among them, D1 and D2 are the two encrypted images before
short enough time, so this section gives an analysis of time and after changing the key respectively. Their length is M and
efficiency. their width is N .
This experiment uses the 2D-LSM initial value key x(0) as
A. KEY SPACE ANALYSIS the key to change. The specific key change method: x(0) =
The key space represents the sum of various keys used in 0.4688 becomes x(0) = 0.4688 + 10−16 . Perform NPCR and
the encryption system, and the size of the key space can UACI calculations on the encrypted images of different sizes
measure the ability of the encryption algorithm to resist brute and different contents and the encrypted images generated
force attacks [37], [38]. When the key space of the encryption after the keys are changed. The NPCR and UACI values
algorithm is greater than 2100 , the security of the encryption obtained are shown in Table 3.
system can be proved. Table 3 shows the NPCR and UACI values of the encrypted
The key space of this algorithm will be analyzed from images before and after the image key changes of different
two aspects below. On the one hand, the key of this algo- sizes and different contents. Through numerical comparison,
rithm mainly includes two initial value keys of 2D-LSM, it can be seen that the seven groups of values are very close
the number of iterations discarded by chaotic system, the to the ideal value. Therefore, the quantitative analysis of the
selection key of matrix rotation angle in improved Zigzag formula can show that the key sensitivity of this algorithm
transform, the selection key of wavelet type, and the given key is strong. In addition to using data to illustrate the degree of
of the system when calculating Hamming distance. Assuming difference between encrypted images obtained by changing
that the calculation accuracy of the double-precision value is the key, in addition to the Lena_512 image as an example,
10−16 , the key space of this encryption algorithm is (1016 )4 + this section gives an intuitive image comparison situation.
(28 )2 , which is much larger than the required 2100 . On the Fig. 14 shows the encrypted image before and after the key is

122736 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 3. NPCR, UACI table. TABLE 4. Information entropy table.

the generated sequence needs to have strong statistical char-

acteristics. In this paper, a 2D-LSM is used in the encryption
system to generate a pseudo-random sequence, the test results
show that the entropy source has strong statistical charac-
teristics. The second requirement of encryption randomness
generates a subset of random numbers that cannot be used to
predict other random numbers. The third requirement and the
fourth requirement are specific manifestations of the second
requirement. As we all know, the one-way function is very
suitable for this requirement. One of the uniqueness of the
encryption algorithm proposed in this paper is the use of
the proven secure encryption primitive SHA-512 algorithm,
which can provide a single functional requirement. In other
words, provided the SHA-512 algorithm is safe, the random-
ness of this algorithm can be guaranteed.

D. INFORMATION ENTROPY ANALYSIS

Information entropy is an indicator used to describe the
randomness of disordered information. Through information
FIGURE 14. Simulation results of key sensitivity (a) Lena_512 original entropy, the level of confusion in pixel distribution can be
image, (b) Lena_512 encrypted image, (c) Lena_512 original image,
(d) Lena_512 encrypted image with a key change of one bit, (e) difference
indicated [7]. For a closed system, information entropy is a
image of two encrypted images, (f) histogram of the difference image. finite number. The value of information entropy in an ideal
state is 8 [43], and the mathematical definition formula of
information entropy is as follows:
M
changed, the difference image of the two encrypted images, X−1
i=2
and the histogram of the difference image. H (m) = − P(mi )log2 P(mi ) (38)
As can be seen from the above images, although the two i=0
encrypted images are both in the shape of snowflake noise, Among them, mi represents the i-th gray value of the image
the difference between the two cannot be seen by the naked m, M represents the number of binary digits required to
eye, but the difference image of the two encrypted images and represent the gray value mi , and P(mi ) is the probability of
the histogram of the difference image can be proved that there mi appearing. Under normal circumstances, the closer the
is indeed quite different between the two encrypted images, information entropy is to 8, the stronger the ability of the
so the image comparison can prove that the key sensitivity encrypted image to hide the original image information, and
of this algorithm is strong. In short, whether it is through the higher the security of the algorithm. Table 4 shows the
numerical quantitative analysis or image comparison, it can information entropy of images with different content and
be proved that this algorithm is very sensitive to keys and has different sizes after being encrypted by this algorithm and
high security. Table 5 shows the comparison with the information entropy
of the algorithm [38], [42], [43], [44], [45], [46].
C. SECURITY ANALYSIS OF THE CHAOTIC KEY It can be seen from Table 4 that the information entropy
GENERATION FUNCTION value of the ciphertext image get from images of different
Randomness is a vital indicator to measure a encryption sizes and different contents or all black and all white images
system [42]. When the encryption system does not meet the are very approach to the ideal value of 8, which means that the
requirements of randomness, the security of the system will ciphertext image get after the original image is encrypted by
be threatened, affecting the normal use of the system. the encryption algorithm hardly reveals any useful informa-
In an encryption system, there are four basic requirements tion. In addition, compared with the information entropy val-
for randomness. Among them, the first requirement is that ues of references in Table 5, the information entropy values of

VOLUME 10, 2022 122737

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 5. Lena information entropy comparison. TABLE 6. Correlation coefficient table.

TABLE 7. Lena_256 correlation coefficients comparison.

this algorithm are slightly above than that of other references,

and are closer to the ideal value. Therefore, the algorithm can
effectively turn the image into a chaotic state, and the security
is very high. TABLE 8. Lena_512 correlation coefficients comparison.

E. STATISTICAL ANALYSIS
The experiment in this section mainly includes two aspects:
correlation statistical analysis and histogram statistical
analysis. Correlation analysis is used to analyze the degree
of correlation between adjacent pixels in plaintext images
and adjacent pixels in ciphertext images, and to determine
whether the proposed encryption algorithm has the ability Among them, x and y are two adjacent pixels, N represents
to eliminate strong correlation [9]. The histogram statistical the total logarithm of randomly selected adjacent pixels,
analysis is to analyze the distribution of the pixel values of the cov(x, y) represents covariance, D(x) represents variance, and
original image and the encrypted image, by comparing the E(x) represents expectation. This paper selects 1500 pairs of
original image and the encrypted image histogram to deter- adjacent pixels in the horizontal, vertical, and diagonal direc-
mine whether the proposed algorithm can make the histogram tions of the original image and encrypted image of different
pixel values evenly distributed [18]. The details are shown in sizes and all black and all white images, and calculates the
section V-E1-V-E2. correlation coefficient by eq. (39). The obtained correlation
coefficient is shown in Table 6. In addition, Table 7 and
1) CORRELATION ANALYSIS Table 8 show the comparison with related references [38],
The correlation coefficient is an index used to evaluate the [42], [43], [44], [45], [46].
degree of correlation between adjacent pixels [21]. The cor- It can be seen from Table 6 that no matter what type
relation of adjacent pixels includes horizontal correlation, of image it is, the correlation coefficient of adjacent pixels
vertical correlation, and diagonal correlation. For plaintext before encryption is approach to 1, and the correlation is very
images, the correlation between adjacent pixels is very strong, strong. However, the correlation coefficient between adjacent
and the attacker can decrypt the ciphertext image through pixels of the encrypted image is approach to 0, and the corre-
statistical attacks. However, this strong correlation is not lation is significantly reduced. Moreover, Table 7 and Table 8
obvious in the ciphertext image, and the correlation of the show the comparison with the latest references and also show
encrypted image is low, which can effectively crack the sta- that the algorithm has superior performance in reducing the
tistical attack. The correlation of adjacent pixels is defined as correlation of adjacent pixels, can resist statistical attacks
follows: very effectively, and ensure the security of the algorithm.
cov(x, y) The points with the same or similar pixel values in the orig-
rxy = √ √ (39) inal image will be distributed in a concentrated manner. This
D(x) D(y)
N situation is most obvious in the three directions of horizon-
1 X tal, vertical, and diagonal, while the pixels of the encrypted
cov(x, y) = × (xi − E(x))(yi − E(y)) (40)
N image will be uniformly distributed. This section draws the
i=1
N correlation images of adjacent pixels in the three directions
1 X
of the Lena_512 image, and intuitively shows the distribution
D(x) = × (xi − E(x)2 ) (41)
N of adjacent pixels through the images. The specific situation
i=1
N is shown in Fig. 15.
1 X
It can be seen from Fig. 15 that the pixels in the three direc-
E(x) = × xi (42)
N tions of the original image are concentrated in the diagonal
i=1

122738 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

FIGURE 15. Correlation scatter diagram in three directions (a) original

FIGURE 16. Histogram of Lena_512 image (a) original image, (b) original
image horizontal direction, (b) encrypted image horizontal direction,
image histogram, (c) encrypted image, (d) encrypted image histogram,
(c) original image vertical direction, (d) encrypted image vertical
(e) decrypted image, (f) decrypted image histogram.
direction, (e) original image diagonal direction, (f) encrypted image
diagonal direction.

distribution of encrypted images more uniform and has the

ability to resist statistical attacks.
and the areas on both sides of the diagonal, indicating that
there is a strong correlation. However, the pixels in the F. PSNR ANALYSIS
three directions of the encrypted image are evenly distributed PSNR is an objective index to measure whether there is dis-
within the coordinate range, indicating that the correlation is tortion in an image [43]. In order to quantify the visual quality
significantly weakened. Therefore, both quantitative analysis of the decrypted image in quantitative form and compare
and qualitative analysis show that this algorithm has the abil- the difference between the decrypted image and the original
ity to reduce the correlation of encrypted images and resist image, it is usually measured by PSNR. Its mathematical
statistical attacks. definition is as follows:
MAXD 2
2) HISTOGRAM ANALYSIS PSNR = 10 × log10 ( )
MSE
The histogram of an image is used to describe the distribution MAXD
of image pixel values. It is generally used to indicate the = 20 × log10 ( √ ) (43)
MSE
number of pixels with the same gray value or the frequency 1
of the pixel [6]. The abscissa of the histogram represents the MSE = PM PN (44)
M × N i=1 j=1 (D(i, j) − P(i, j))2
gray level of the pixel, and the ordinate is used to represent the
number of pixels that appear in the gray level [46]. Normally, Among them, D(i, j) and P(i, j) represent the decrypted image
natural images reflect the real and observable world, so the and the original image respectively, M and N are the length
histogram distribution of natural images presents a inhomo- and width of the image respectively, and MAXD is the quan-
geneous state. But for the image obtained after encryption, tized gray level of the decrypted image.
since the visual requirements must be chaotic, the histogram The evaluation result is expressed in dB (decibel). Gen-
distribution should show a uniform state, so as to ensure that erally, the greater the PSNR value between two images, the
no effective information can be get from the encrypted image. higher the degree of restoration. When the PSNR value is
Fig. 16 shows the original image, encrypted image, decrypted close to 0 dB, the image distortion is the largest. In addition
image and their respective histograms using the Lena_512 to visually judging the difference between the original image
image as an example. and the decrypted image through the results of image simula-
As can be seen from the above images, the histogram of tion, this section also judges the difference between the two
the original image is distributed in the middle range, with quantitatively. Table 9 shows the PSNR values of the original
obvious statistical law, and the distribution of pixel values image and the decrypted image of different sizes and all black
fluctuates; The histogram of the encrypted image shows an and white. Table 10 and Table 11 show the comparison with
approximately uniform distribution, which is obviously dif- related references [38], [42], [43], [44], [45], [46].
ferent from the histogram of the original image. Therefore, It can be seen from Table 9 that the PSNR of the original
it can be shown that this algorithm can make the histogram image measured in this section and the respective decrypted

VOLUME 10, 2022 122739

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 9. MSE, PSNR table.

TABLE 10. Lena_256 MSE, PSNR comparison.

TABLE 11. Lena_512 MSE, PSNR comparison.

FIGURE 17. Simulation experiment of differential attack (a) encrypted

image E, (b) encrypted image E1, (c) E-E1 difference image, (d) E-E1
difference image histogram, (e) encrypted image E, (f) encrypted image
E2, (g) E-E2 difference image, (h) E-E2 difference image histogram,
(i) encrypted image E, (g) encrypted image E3, (k) E-E3 difference image,
(i) E-E3 difference image histogram.

TABLE 12. Differential attack NPCR and UACI comparison table.

image is relatively high, indicating that the algorithm can
effectively restore the information of the plaintext image.
Moreover, the comparison of Table 10 and Table 11 with the
latest references can also show that decrypted image obtained
by this algorithm has the lowest distortion and has a strong
ability to restore.

G. DIFFERENTIAL ATTACK ANALYSIS

The meaning of the differential attack is to attack the cryp- It can be seen intuitively from Fig. 17 that the encrypted
tographic system by analyzing the change and propagation image generated by the original image without changing and
of the plaintext images with specific differences after being the encrypted image generated by the plaintext image change
encrypted [3], [11], [35]. In order to resist the differential both show the state of snowflake noise, and there is no obvi-
attack and eliminate the correlation between the plaintext ous difference visually, but through the difference images and
image and the encrypted image, small changes in the plain- the histogram of the difference images, it can be explained
text image should cause prominent changes in the encrypted that although the ciphertext images appear to be snowflake
image. The pixel change rate NPCR and the uniform average noise images, in fact, the degree of difference between the
change intensity UACI are usually used to test the correla- encrypted images is very large. In addition, it can be seen
tion between the plaintext image and the ciphertext image. from the data analysis in Table 12 that a certain or a few
The specific definitions of the two formulas are given in pixel values of the original image change, which may cause
section V-B. The ideal value of NPCR is 99.6054% and the huge differences. This can further explain that subtle changes
ideal value of UACI is 33.4653%. in the plaintext image will cause significant changes in the
Take the Lena_512 image as a original image as an exam- ciphertext image, and this algorithm has strong resistance to
ple to carry out three groups of experiments, mainly includ- differential attacks.
ing: randomly changing one pixel value of the plaintext image
to obtain the encrypted image E1, randomly changing the two H. TYPICAL ATTACK ANALYSIS
pixel values of the plaintext image to obtain the encrypted Generally speaking, only the correct key can be used to
image E2, and randomly changing the three pixel values of restore the ciphertext to the correct plaintext information.
the plaintext image to obtain the encrypted image E3. Fig. 17 However, according to cryptography, an attacker can use a
shows the encrypted image E generated without changing the special technique to destroy the encryption algorithm through
plaintext image, the encrypted images generated by changing repeated experiments, and then obtain plaintext information.
the original images, the difference images of the encrypted If the attacker does not know the key, he can use other
images, and the histogram of the difference images. Table 12 effective information to attack the encryption system. Then,
calculates the NPCR values and UACI values of the encrypted according to the type of information the attacker has, the
images E1, E2, E3 and E obtained from the three groups of attack types can be divided into four types. These four typical
experiments. attack types are shown in this section [2], [7], [28]:

122740 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

1) Ciphertext only attack: The attacker possesses the

cryptographic algorithm and statistical characteristics of the
plaintext image, and also intercepts one or more ciphertexts
encrypted with the same key. The plaintext or key can be
analyzed from this information. It can usually be understood
as a brute force attack when only the ciphertext is known.
2) Chosen ciphertext attack: The attacker has access to
the decryption system and can choose any ciphertext that is FIGURE 18. Speckle noise attack (a) 0.0001 speckle noise,
beneficial to the attack and its corresponding plaintext. (b) 0.0011 speckle noise, (c) 0.01 speckle noise.
3) Known plaintext attack: The attacker obtains a certain
part of the plaintext and the corresponding ciphertext, which
can be any non-empty subset of the known plaintext pair, and
uses this information to analyze and crack the cryptographic
system.
4) Chosen plaintext attack: Chosen plaintext attack is the
most typical attack method. The attacker has access to the
decryption system and can select the plaintext and its corre-
sponding ciphertext, that is, know the encrypted ciphertext
FIGURE 19. Gaussian noise attack (a) 0.0001 Gaussian noise,
and the decrypted plaintext. (b) 0.0011 Gaussian noise, (c) 0.01 Gaussian noise.
Obviously, chosen plaintext attack is the strongest attack
method. If the system can resist a chosen plaintext attack,
it must be able to resist the other three typical attack methods.
In this algorithm, the plaintext image uses the SHA-512 hash
function to generate a message digest, and the message digest
is used to generate the initial value of the 2D-LSM and the
number of iterative discards. The chaotic system iteratively
generates a chaotic sequence. There is a kind dependency
of chaotic sequence and the original image. Therefore, the
algorithm can resist chosen plaintext attacks, which further FIGURE 20. Salt and pepper noise attack (a) 0.0001 salt and pepper
noise, (b) 0.0011 salt and pepper noise, (c) 0.01 salt and pepper noise.
shows that the algorithm can also resist the other three typical
attacks.

I. NOISE ATTACK ANALYSIS noise, the interference of Gaussian noise has a greater impact
In addition to having good security to resist brute force crack- on the decrypted image, but its impact is within an acceptable
ing by the attacker, the image encryption algorithm should range and will not affect the recognition of the original image.
also have a certain ability to resist noise attacks [29], [39]. Generally speaking, the three types of different levels of noise
Because the transmission channel may sometimes have a have a certain impact on the decrypted image to some extent,
certain degree of instability, the image data may be disturbed but the impact is very small and can be ignored. Therefore, the
by noise factors. The encrypted image is often very sensi- proposed algorithm has a strong ability to resist noise attacks.
tive, and sometimes a little noise interference may cause the
receiver to receive the corrupted encrypted image, which in J. CROPPING ATTACK ANALYSIS
turn leads to the distortion of the decrypted image. In this Data loss will inevitably occur during the transmission of
section, by adding speckle noise, Gaussian noise and salt encrypted images, which requires the encryption and decryp-
and pepper noise of different intensities to the test image tion schemes to be able to resist cropping attacks well.
Lena_512 image [30], it is used to test the performance of the In order to prove that the proposed scheme has a certain abil-
proposed algorithm to resist noise attacks. The experimental ity to resist cropping attacks, this section tests the encrypted
results are shown in Fig. 18, Fig. 19, and Fig. 20. images and decrypted images that lose 1/16 and 1/4 of the
It can be seen from the experimental results that speckle data, as shown in Fig. 21.
noise has the least impact on the decrypted image, followed It can be seen from Fig. 21 that the decrypted image can
by salt and pepper noise, and Gaussian noise has the greatest still identify the valid information of the plaintext image,
impact on the decrypted image. As the degree of interference indicating that the encryption algorithm can resist cropping
of salt and pepper noise increases, the affected pixels in the attacks and has certain robustness.
decrypted image gradually increase, but it does not affect the
recognition of the image by the naked eye, and the rough K. RANDOMNESS ANALYSIS OF ENCRYPTED IMAGE
information of the original image can still be clearly distin- NIST SP800 is a series of information security guidelines
guished. Compared with the interference of salt and pepper issued by the National Institute of Standards and Technology

VOLUME 10, 2022 122741

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

TABLE 14. Time table.

TABLE 15. Lena_256 time comparison.

FIGURE 21. Cropping attack test results (a) 1/16 of the encrypted image
is cropped in the upper left corner, (b) 1/16 of the encrypted image is
cropped in the upper right corner, (c) 1/4 of the encrypted image is
cropped in the lower left corner, (d) 1/4 of the encrypted image is cropped
in the upper right corner, (e) decrypted image of (a), (f) decrypted image
of (b), (g) decrypted image of (c), (h) decrypted image of (d).
TABLE 16. Lena_512 time comparison.

TABLE 13. NIST test result of encrypted images.

L. TIME PERFORMANCE ANALYSIS

A good encryption algorithm should have high time effi-
ciency [38], [40]. For the proposed image encryption
algorithm, under the precondition of ensuring the image
transmission quality and security, if its operating efficiency is
equal to or even higher than the existing encryption algorithm,
the purpose of designing a new image encryption algorithm
to improve transmission efficiency has been achieved.
In this section, the time performance of the encryption
algorithm is analyzed. In order to verify that the proposed
algorithm can evidently improve the efficiency of encryption
(NIST), and has become a de facto standard and authoritative and decryption, the encryption and decryption time of three
guide widely recognized by researchers in the information groups of plaintext images of different sizes and all black
security field [13], [30]. The NIST test plan is a statistical and all white images are calculated. The test results are
package that includes 15 test methods. These test methods shown in Table 14. In addition, the comparison with related
can test the randomness of arbitrary length binary sequences references [38], [42], [43], [44], [45], and [46] is also given,
generated as random passwords or pseudo-random number as shown in Table 15 and Table 16.
generators based on software or hardware. These test methods It can be seen from Table 14 that the encryption and decryp-
are mainly used to determine the various pseudo-randomness tion speeds of the five groups of images are all higher than
that may exist in the sequence. This section is mainly used the qualified minimum standard. And even if the image with
to test the randomness of the encrypted image, convert each the largest size is encrypted, the time required is only about
pixel value of the encrypted image into 8 binary bits, and 0.85262 s. This test result is very satisfactory, indicating that
perform statistical tests on the bit sequence. Due to the num- the proposed algorithm has passed the operational efficiency
ber of cycles required to apply the random drift test and its test. Moreover, no matter which type and size of the image is
variants, this test is suitable for encrypted images with a size encrypted or decrypted, the algorithm takes a very short time,
of 512*512. The test results can be obtained from Table 13. and it also meets the requirements of real-time communica-
In Table 13, the P value of the NIST test applied to multiple tion of encrypted and decrypted images. In addition, the com-
encrypted images is given. Since all P values are greater parison in Table 15 and Table 16 also show that the efficiency
than 0.01, this means that the encrypted image has passed of image encryption using this scheme is significantly better
the randomness test, and the encrypted image has high secu- than that of the latest algorithm for encryption and decryption,
rity, indicating that the proposed algorithm is safe in actual indicating that this scheme has certain advantages in time
encryption applications. performance.

122742 VOLUME 10, 2022

Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

VI. CONCLUSION [7] Q. Li, X. Wang, B. Ma, X. Wang, C. Wang, S. Gao, and Y. Shi, ‘‘Con-
In this paper, a dual-domain image encryption algorithm cealed attack for robust watermarking based on generative model and
perceptual loss,’’ IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 8,
based on hyperchaos and dynamic wavelet decomposition is pp. 5695–5706, Aug. 2022.
proposed. A new encryption algorithm similar to ‘sandwich’ [8] C. Wang, B. Ma, Z. Xia, J. Li, Q. Li, and Y.-Q. Shi, ‘‘Stereoscopic image
is realized, which combines the block scrambling in the description with trinion fractional-order continuous orthogonal moments,’’
IEEE Trans. Circuits Syst. Video Technol., vol. 32, no. 4, pp. 1998–2012,
spatial domain, the wavelet decomposition in the frequency Apr. 2022.
domain, and the diffusion operation in the spatial domain. [9] Z. Hua, Y. Chen, H. Bao, and Y. Zhou, ‘‘Two-dimensional parametric poly-
Firstly, the original image is divided into several blocks, and nomial chaotic system,’’ IEEE Trans. Syst., Man, Cybern. Syst., vol. 52,
no. 7, pp. 6585–6599, Jul. 2021.
each matrix block is scrambled under the control of a random [10] Y. Khedmati, R. Parvaz, and Y. Behroo, ‘‘2D hybrid chaos map for image
number sequence to obtain a scrambled matrix. Secondly, security transform based on framelet and cellular automata,’’ Inf. Sci.,
use the plaintext Hamming distance to dynamically select vol. 512, pp. 855–879, Feb. 2020.
[11] C. Xiu, R. Zhou, S. Zhao, and G. Xu, ‘‘Memristive hyperchaos secure
the wavelet type, perform wavelet decomposition on the communication based on sliding mode control,’’ Nonlinear Dyn., vol. 104,
scrambled matrix, and generate a wavelet coefficient matrix. no. 1, pp. 789–805, Mar. 2021.
At the same time, the key matrix is obtained by performing [12] S. Zhu and C. Zhu, ‘‘Secure image encryption algorithm based on hyper-
chaos and dynamic DNA coding,’’ Entropy, vol. 22, no. 7, p. 772, Jul. 2020.
the improved Zigzag transformation on the scrambled matrix. [13] Z. Man, J. Li, X. Di, Y. Sheng, and Z. Liu, ‘‘Double image encryption
Then, the initial value of the chaotic system is generated algorithm based on neural network and chaos,’’ Chaos, Solitons Fractals,
by the original image and the SHA-512 algorithm, and the vol. 152, Nov. 2021, Art. no. 111318.
[14] Z. Man, J. Li, X. Di, Y. Sheng, and Z. Liu, ‘‘Double image encryption
chaotic key matrix is generated after the 2D-LSM iteration. algorithm based on neural network and chaos,’’ Chaos, Solitons Fractals,
Finally, the XOR operation is performed on the wavelet coef- vol. 152, Nov. 2021, Art. no. 111318.
ficient matrix, the chaotic key matrix, and the key matrix to [15] W. Feng, Y.-G. He, H.-M. Li, and C.-L. Li, ‘‘A plain-image-related chaotic
image encryption algorithm based on DNA sequence operation and dis-
obtain the encrypted image. Among them, the key is related crete logarithm,’’ IEEE Access, vol. 7, pp. 181589–181609, 2019.
to the plaintext image, which greatly improves the algo- [16] W. Feng and Y.-G. He, ‘‘Cryptanalysis and improvement of the hyper-
rithm’s ability to resist known plaintext and chosen plaintext chaotic image encryption scheme based on DNA encoding and scram-
bling,’’ IEEE Photon. J., vol. 10, no. 6, pp. 1–15, Dec. 2018.
attacks. The experimental results show that the algorithm has [17] W. Feng, Y.-G. He, H.-M. Li, and C.-L. Li, ‘‘Image encryption algorithm
a large key space, good encryption effect, strong randomness, based on discrete logarithm and memristive chaotic system,’’ Eur. Phys. J.
and high time efficiency. In addition, correlation analysis, Special Topics, vol. 228, no. 10, pp. 1951–1967, Oct. 2019.
[18] W. Feng, Z. Qin, J. Zhang, and M. Ahmad, ‘‘Cryptanalysis and improve-
information entropy analysis, differential attack, and noise ment of the image encryption scheme based on feistel network and dynamic
attack analysis are also carried out on the proposed encryption DNA encoding,’’ IEEE Access, vol. 9, pp. 145459–145470, 2021.
algorithm to prove that the algorithm is safe and reliable. [19] W. Feng and J. Zhang, ‘‘Cryptanalzing a novel hyper-chaotic image
encryption scheme based on pixel-level filtering and DNA-level diffu-
In future research, we will continue to study the image sion,’’ IEEE Access, vol. 8, pp. 209471–209482, 2020.
encryption method combining spatial domain encryption and [20] H. Li, T. Li, W. Feng, J. Zhang, J. Zhang, L. Gan, and C. Li, ‘‘A novel
frequency domain encryption, and strive to propose better image encryption scheme based on non-adjacent parallelable permutation
and dynamic DNA-level two-way diffusion,’’ J. Inf. Secur. Appl., vol. 61,
algorithms. At the same time, we will continue to improve the Sep. 2021, Art. no. 102844.
robustness of the algorithm and further improve the security [21] Y.-M. Li, D. Wei, and L. Zhang, ‘‘Double-encrypted watermarking algo-
level of the algorithm. In addition, we will generalize the rithm based on cosine transform and fractional Fourier transform in invari-
ant wavelet domain,’’ Inf. Sci., vol. 551, pp. 205–227, Apr. 2021.
encryption algorithm to practical applications, especially to [22] X. Yang, Q. Xue, X. Yang, H. Yin, Y. Qu, X. Li, and J. Wu, ‘‘A novel
apply it to information encryption systems that require high prediction model for the inbound passenger flow of urban rail transit,’’ Inf.
efficiency and high security. Sci., vol. 566, pp. 347–363, Aug. 2021.
[23] X. Wang and S. Gao, ‘‘Image encryption algorithm based on the matrix
semi-tensor product with a compound secret key produced by a Boolean
REFERENCES network,’’ Inf. Sci., vol. 539, pp. 195–214, Oct. 2020.
[24] X. Wang and X. Chen, ‘‘An image encryption algorithm based on dynamic
[1] Z. Liang, Q. Qin, and C. Zhou, ‘‘An image encryption algorithm based on
row scrambling and zigzag transformation,’’ Chaos, Solitons Fractals,
Fibonacci Q-matrix and genetic algorithm,’’ Neural Comput. Appl., vol. 34,
vol. 147, Jun. 2021, Art. no. 110962.
no. 21, pp. 19313–19341, Nov. 2022, doi: 10.1007/s00521-022-07493-x.
[25] X. Wang and M. Zhang, ‘‘An image encryption algorithm based on new
[2] Z. Liang, Q. Qin, C. Zhou, N. Wang, Y. Xu, and W. Zhou, ‘‘Medical image chaos and diffusion values of a truth table,’’ Inf. Sci., vol. 579, pp. 128–149,
encryption algorithm based on a new five-dimensional three-leaf chaotic Nov. 2021.
system and genetic operation,’’ PLoS ONE, vol. 16, no. 11, Nov. 2021, [26] X. Wu, D. Wang, J. Kurths, and H. Kan, ‘‘A novel lossless color image
Art. no. e0260014. encryption scheme using 2D DWT and 6D hyperchaotic system,’’ Inf. Sci.,
[3] Z. Liang, Q. Qin, C. Zhou, and S. Xu, ‘‘Color image encryption algo- vol. 349, pp. 137–153, Jul. 2016.
rithm based on four-dimensional multi-stable hyper chaotic system and [27] X. Wang, C. Liu, and D. Jiang, ‘‘A novel triple-image encryption and hiding
DNA strand displacement,’’ J. Electr. Eng. Technol., to be published, doi: algorithm based on chaos, compressive sensing and 3D DCT,’’ Inf. Sci.,
10.1007/s42835-022-01157-5. vol. 574, pp. 505–527, Oct. 2021.
[4] C. Zou, Q. Zhang, C. Zhou, and W. Cao, ‘‘A nonlinear neural network based [28] Z. Hua, K. Zhang, Y. Li, and Y. Zhou, ‘‘Visually secure image encryption
on an analog DNA toehold mediated strand displacement reaction circuit,’’ using adaptive-thresholding sparsification and parallel compressive sens-
Nanoscale, vol. 14, no. 17, pp. 6585–6599, May 2022. ing,’’ Signal Process., vol. 183, Jun. 2021, Art. no. 107998.
[5] X. Wang, X. Wang, B. Ma, Q. Li, and Y.-Q. Shi, ‘‘High precision error [29] X. Yan, X. Wang, and Y. Xian, ‘‘Chaotic image encryption algorithm
prediction algorithm based on ridge regression predictor for reversible data based on fractional order scrambling wavelet transform and 3D cyclic
hiding,’’ IEEE Signal Process. Lett., vol. 28, pp. 1125–1129, 2021. displacement operation,’’ IEEE Access, vol. 8, pp. 208718–208736, 2020.
[6] B. Ma and Y. Q. Shi, ‘‘A reversible data hiding scheme based on code [30] H. R. Shakir, ‘‘An image encryption method based on selective AES coding
division multiplexing,’’ IEEE Trans. Inf. Forensics Security, vol. 11, no. 9, of wavelet transform and chaotic pixel shuffling,’’ Multimedia Tools Appl.,
pp. 1914–1927, Sep. 2016. vol. 78, no. 18, pp. 26073–26087, Sep. 2019.

VOLUME 10, 2022 122743

 Q. Qin et al.: Dual-Domain Image Encryption Algorithm Based on Hyperchaos and Dynamic Wavelet Decomposition

[31] Z. Hua, Z. Zhu, Y. Chen, and Y. Li, ‘‘Color image encryption using ZHONGYUE LIANG was born in Binzhou,
orthogonal Latin squares and a new 2D chaotic system,’’ Nonlinear Dyn., Shandong, China, in 1994. She is currently pursu-
vol. 104, no. 4, pp. 4505–4522, Jun. 2021. ing the master’s degree. She has published many
[32] T. Yan, P. Wu, Y. Qian, Z. Hu, and F. Liu, ‘‘Multiscale fusion and aggre- articles in the field of image encryption as the first
gation PCNN for 3D shape recovery,’’ Inf. Sci., vol. 536, pp. 277–297, author in NCA, Plos one, and other journals. Her
Oct. 2020. main research interests include image encryption
[33] D. Zhang, L. Chen, and T. Li, ‘‘Hyper-chaotic color image encryption and DNA computing.
based on transformed zigzag diffusion and RNA operation,’’ Entropy,
vol. 23, no. 3, p. 361, Mar. 2021.
[34] P. Ramasamy, V. Ranganathan, S. Kadry, R. Damaševičius, and
T. Blažauskas, ‘‘An image encryption scheme based on block scram-
bling, modified zigzag transformation and key generation using enhanced
logistic—Tent map,’’ Entropy, vol. 21, no. 7, p. 656, Jul. 2019.
[35] X. Xue, H. Jin, D. Zhou, and C. Zhou, ‘‘Medical image protection algo-
rithm based on deoxyribonucleic acid chain of dynamic length,’’ Frontiers
Genet., vol. 12, Mar. 2021, Art. no. 654663.
[36] A. A. Rezk, A. H. Madian, A. G. Radwan, and A. M. Soliman,
‘‘On-the-fly parallel processing IP-core for image blur detection, com-
pression, and chaotic encryption based on FPGA,’’ IEEE Access, vol. 9, SHUANG LIU was born in 1977. She is an Asso-
pp. 82726–82746, 2021. ciate Professor and the Doctor of engineering. She
[37] O. P. Singh and A. K. Singh, ‘‘A robust information hiding algorithm based is currently the Deputy Dean of the School of
on lossless encryption and NSCT-HD-SVD,’’ Mach. Vis. Appl., vol. 32, Computer Science and Engineering, Dalian Minzu
no. 4, p. 101, Jul. 2021. University. Her research interests include image
[38] V. Manikandan and R. Amirtharajan, ‘‘On dual encryption with RC6 and encryption, machine learning, and video informa-
combined logistic tent map for grayscale and DICOM,’’ Multimedia Tools tion processing.
Appl., pp. 23511–23540, May 2021.
[39] K. A. Korba, D. Abed, and M. Fezari, ‘‘Securing physical layer using new
chaotic parametric maps,’’ Multimedia Tools Appl., vol. 80, nos. 21–23,
pp. 32595–32613, Sep. 2021.
[40] N. Tsafack, J. Kengne, B. Abd-El-Atty, A. M. Iliyasu, K. Hirota, and
A. A. A. EL-Latif, ‘‘Design and implementation of a simple dynamical 4-
D chaotic circuit with applications in image encryption,’’ Inf. Sci., vol. 515,
pp. 191–217, Apr. 2020.
[41] S. Yu, N. Zhou, L. Gong, and Z. Nie, ‘‘Optical image encryption
algorithm based on phase-truncated short-time fractional Fourier trans-
form and hyper-chaotic system,’’ Opt. Lasers Eng., vol. 124, Jan. 2020,
Art. no. 105816. XIAO WANG was born in Wuyi, China, in 1977.
[42] T. Li, B. Du, and X. Liang, ‘‘Image encryption algorithm based on logistic She received the master’s degree in mechani-
and two-dimensional Lorenz,’’ IEEE Access, vol. 8, pp. 13792–13805, cal and electronic engineering from the Zhe-
2020. jiang University of Technology. She is a Lecturer.
[43] Y. Luo, X. Ouyang, J. Liu, L. Cao, and Y. Zou, ‘‘An image encryption She has published more than ten articles. Her
scheme based on particle swarm optimization algorithm and hyperchaotic research interests include image encryption, intel-
system,’’ Soft Comput., vol. 26, no. 11, pp. 5409–5435, Jan. 2022. ligent computing, and computer aided design.
[44] L. E. Reyes-López, J. S. Murguía, H. González-Aguilar, M. T. Ramírez-
Torres, M. Mejía-Carlos, and J. O. Armijo-Correa, ‘‘Scaling analysis of an
image encryption scheme based on chaotic dynamical systems,’’ Entropy,
vol. 23, no. 6, p. 672, May 2021.
[45] M. Zhang, X. Tong, Z. Wang, and P. Chen, ‘‘Joint lossless image compres-
sion and encryption scheme based on CALIC and hyperchaotic system,’’
Entropy, vol. 23, no. 8, p. 1096, Aug. 2021.
[46] Y. Zhao and L. Liu, ‘‘A bit shift image encryption algorithm based on
double chaotic systems,’’ Entropy, vol. 23, no. 9, p. 1127, Aug. 2021.

QIUXIA QIN was born in Jinan, Shandong, CHANGJUN ZHOU was born in 1977. He received
China, in 1996. She is currently pursuing the mas- the Ph.D. degree. He is currently a Distinguished
ter’s degree. Her research interests include image Professor of Shuanglong Scholar with the School
encryption and artificial intelligence. of Mathematics and Computer Science, Zhejiang
Normal University. His research interests include
image encryption, intelligent computing and pat-
tern recognition in new computing models, bio-
logical computing theory and its application, and
software system development.

122744 VOLUME 10, 2022