Multimedia Tools and Applications

https://doi.org/10.1007/s11042-023-14601-z

An efficient meaningful double-image encryption

algorithm based on parallel compressive sensing
and FRFT embedding

Donghua Jiang 1 & Lidong Liu 1,2 & Liya Zhu 3 & Xingyuan Wang 4 & Yingpin Chen 5 &
Xianwei Rong 6

Received: 7 June 2021 / Revised: 1 June 2022 / Accepted: 3 February 2023

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

Abstract
The transmission of images via the Internet has grown exponentially in the past few
decades. However, the Internet considered as an insecure method of information trans-
mission may cause serious privacy issues. To overcome such potential security issues, a
novel visually meaningful double-image encryption (VMDIE) algorithm conjugating
quantum cellular neural network (QCNN), compressive sensing (CS) and fractional
Fourier transform (FRFT) is proposed in this paper. First, the wavelet coefficients of
two plain images are scrambled by the Fisher-Yates confusion algorithm, and compressed
by key-controlled partial Hadamard matrix. The final meaningful cipher image is gener-
ated by embedding the encrypted images into a same-scale host image via the FRFT-based
embedding approach. Besides, the eigenvalues of plain images are utilized to generate
secret key streams to improve the ability of proposed VMDIE algorithm to withstand
various plaintext attacks. Afterward, the plaintext eigenvalues are hidden into the alpha
channel of meaningful cipher image under control of keys to relieve unnecessary storage
space and transmission cost. Ultimately, simulation results and security analyses indicate
that the proposed VMDIE algorithm is effective and can withstand multiple attacks.

Keywords Double-image compression and encryption . Visually meaningful encrypted image .

Quantum cellular neural network . Compressive sensing . Fractional Fourier transform

1 Introduction

With the advent of big data era, a large number of digital images are manufactured and
transmitted on the Internet every moment, which will also be accompanied with various

* Lidong Liu
[email protected]

Donghua Jiang
[email protected]

Extended author information available on the last page of the article

 Multimedia Tools and Applications

security issues [9, 46]. Therefore, how to efficiently and securely transmit digital images
without being intercepted and tampered is of great significance. As an effective protection
method, image encryption (IE) technology can encrypt the natural image into a noise-like
image, from which any useful information cannot be visually attained by eavesdroppers, thus
playing a role in protecting image information. Additionally, compressive sensing (CS)
technology, proposed by Candes and Donoho in 2006 [3, 7], is a technology that can
simultaneously under-sample, compress and encrypt the sparse signals. Hence, it is not an
unattractive option to design an efficient image cryptosystem by combining IE technology and
CS technology.
The CS technology can be seen as a variant of symmetric encryption, in which the sparse
signal, measurement matrix and sensor matrix respectively correspond to the plaintext, secret
key and ciphertext in cryptosystem from the perspective of cryptography. Moreover, Rachlin
et al. [30] pointed out that the CS-based cryptosystem has sufficient computational security to
withstand the only-ciphertext attacks and brute force attacks. However, since the sensor matrix
is obtained by performing linear projection on the sparse signal, it is vulnerable to the known-
and chosen-plaintext attacks [11, 43]. Thus, to attain a higher security level, researchers have
proposed varied image protection algorithms by combining compressive sensing with other
encryption techniques, such as chaos theory [4, 15, 28], coding technology [33, 37], cellular
automata [21, 24], neural network [34, 38] and so on.
For example, a parallel image compression-encryption algorithm is presented by Huang
[17]. In his scheme, the plain image is first divided into several sub-images and then linearly
measured by 1-D CS. Afterward, a series of operations, such as permutation, substitution and
block-wise XOR, are performed on the quantized measurement value matrixes to generate the
final cipher image. However, the large-scale Gaussian random matrix used as the secret keys in
Huang’s scheme requires additional storage space and transmission cost. Subsequently, to
eliminate this issue, key-controlled partial Hadamard matrix [48, 50], partial random block
weighing matrix [45], structurally random matrix [31] and chaos-based measurement matrix
[11, 47, 49] are introduced to compress the plain image. Besides, in the encryption phase, the
counter mode [13, 14], hash function [11, 29] and plaintext eigenvalue [10] are applied to
withstand the plaintext analysis-based security attack models, since different plain images
correspond to different key streams. Nevertheless, the above-mentioned CS-based image
encryption algorithm can prevent image data from leakage, but it cannot provide protection
in appearance.
Therefore in [1], Bao et al. proposed a feasible framework for simultaneous encryption and
steganography, that is encryption-embedding framework. In Bao’s scheme, the plain image is
first encrypted by an existing encryption algorithm to obtain a noise-like or texture-like cipher
image. After that, the meaningless cipher image is decomposed and embedded into a publicly
accessible host image by lifting wavelet transform (IWT). In the absence of compression stage,
the resolution of meaningful cipher image is four times that of the plain image, which increases
the unnecessary cost in storage and transmission. Later, many improved visually meaningful
image encryption algorithms [5, 6, 18, 40, 41] have been proposed one after another. Such as
[50], where the plain image is first encrypted and compressed through the coefficient random
scrambling strategy and compressive sensing with block-wise manner. Then the robust
singular value decomposition embedding approach is employed to embed the meaningless
cipher image into a host image with the same resolution of the plain image. Besides, the
counter mode is utilized to update the encryption keys to against the chosen-plaintext attacks.
Multimedia Tools and Applications

In this paper, we put forward a visually meaningful double-image encryption scheme based
on parallel compressive sensing and FRFT embedding. It mainly consists of two stages: pre-
encryption process and embedding process. To prevent image information from leakage, in the
first stage, the Fisher-Yates confusion and compressive sensing are utilized to encrypt and
compress the two plain images in discrete wavelet domain to attain the meaningless cipher
images. Afterward, in the second stage, the meaningless cipher images are embedded into a
host image via the FRFT embedding approach, so that their appearance is protected. Besides,
the 4-D quantum cellular neural network and improved 2-D Henon map, whose initial values
are generated from the plaintext eigenvalues, are applied to construct secret key streams and
key-controlled measurement matrix in encryption stage.
The innovations and contributions of this paper are summarized as follows.

(1) An efficient and visually meaningful encryption algorithm for double images based on
quantum cellular neural network, parallel compressive sensing and FRFT embedding is
designed to improve transmission efficiency.
(2) A key-controlled double embedding approach, that is FRFT-based embedding approach,
is proposed to improve the security of embedding phase.
(3) A novel “One cipher image corresponds to one key” mechanism is proposed to withstand
the plaintext analysis-based security attack models.
(4) Simulation analysis and comparison results indicate that the proposed encryption scheme
has high efficiency and can withstand multiple attacks.

The rest of this paper is arranged as follows. The background related to the proposed image
algorithm is described in Section 2. The third section and fourth section respectively introduce
the specific steps of proposed VMDIE algorithm and corresponding decryption algorithm in
detail. Moreover, simulation results and performance analysis are given in Section 5. After
that, our encryption algorithm is compared with the existing related algorithms, and the results
are listed in the sixth section. Then, a brief summary and future work are shown in the final
section.

2 Background

2.1 Chaotic system

In this subsection, two chaotic systems are introduced which are 4-D hyperchaotic quantum
cellular neural network and improved 2-D Henon map. Among them, the quantum cellular
neural network is used to construct the secret key streams with high unpredictability in
encryption. Additionally, to save storage space and reduce transmission costs, a key-
controlled measurement matrix is generated according to the improved Henon chaotic map.

2.1.1 Hyperchaotic quantum cellular neural network

Quantum cellular neural network (QCNN) is constructed by several quantum cellular automata
(QCA) [19]. And it has complex dynamic characteristics due to quantum interaction between
the quantum dots. For the two-cell QCNN, its state equation is defined as
 Multimedia Tools and Applications

8 pffiffiffiffiffiffiffiffiffiffi
>
> ẋ1 ¼ −2w1 1−x21 sinx2
>
> x1
>
> pffiffiffiffiffiffiffiffiffiffi
<ẋ2 ¼ −w2 ðx1 −x3 Þ þ 2w1 1−x2 cosx2
>
qffiffiffiffiffiffiffiffiffiffi 1
ð1Þ
>
> ẋ3 ¼ −2w3 1−x3 sinx4
2
>
>
>
> x3
>
: 4 ¼ −w4 ðx3 −x1 Þ þ 2w3 pffiffiffiffiffiffiffiffiffi2ffi cosx4
ẋ
1−x3

where x1 and x3 are polarizations. x2 and x4 are the quantum phase. Moreover, w1 and w3 are
the proportional coefficients of inter-dot energy in each cell. w2 and w4 are the weighted impact
coefficients of the difference in the polarizations of adjacent cells. Additionally, when
w1 = w3 = 0.28, w2 = 0.7 and w4 ∈ (0.1, 1], Eq. (1) enters the hyper-chaotic state, and its corre-
sponding hyperchaotic trajectories are plotted in Fig. 1.

2.1.2 Improved 2-D Henon map

Since the classical 2-D Henon map [12] has a small key space and its chaotic trajectories are
simple, an improved Henon map (IHM) is proposed in [44]. Its system equation is as follows.
  
unþ1 ¼ 1−au2n sinvn
ð2Þ
vnþ1 ¼ sinbvn þ un

where a and b are system control parameters. Furthermore, un + 1 and vn + 1 are the generated
pseudo-random numbers, belonging to [−1, 1]. Figure 2 displays the bifurcation diagram of
variable u under the condition that a is set to 2.1 and the initial value is [0.53, 0.89]T. As Fig. 2
shows, when b ∈ (−∞, −0.74] ∪ [0.74, +∞), the improved Henon map is in a chaotic state.

2.2 Compressive sensing

Compressive sensing [3, 7] refers to using a measurement matrix unrelated to sparse transform
basis to linearly project the high-dimensional natural signals into a low-dimensional space.
And then, reconstructing the original signals with high probability from these few projections.
Suppose that the natural signal y = {y1, y2, y3, …, yN} can be expressed linearly by a group of
orthonormal basis, as displayed in Eq. (3).
N
y ¼ ΨS ¼ ∑ S i Ψ i ð3Þ
i¼1

(a) x1-x2-x3 (b) x2-x3-x4 (c) x1-x3-x4

Fig. 1 The hyperchaotic trajectories of this quantum cellular neural network
Multimedia Tools and Applications

Fig. 2 The bifurcation diagram of improved Henon map

In Eq. (3), Ψ = [Ψ1, Ψ2, Ψ3, …, ΨN] is the basis matrix. And the column vector S sized of N × 1 is
the sparse representation coefficient of y in Ψ domain. Besides, if ‖S‖0 = k, y is said to be k-
sparse on the orthonormal basis Ψ. Then the process of linearly measuring the signal y with
sparsity through the measurement matrix Φ ∈ ℝM × N can be expressed as
z ¼ Φy ¼ ΦΨS ¼ ΘS ð4Þ
where z = {z1, z2, z3, …, zM} is the observed vector, and Θ = ΦΨ is called as the sensing matrix.
Since Eq. (4) is an underdetermined equation system, other regular constraints need to be
added to restore the natural signal y. Related studies [2] indicate that if the signal y is sparse
enough on the orthogonal basis Ψ, and when the matrices Φ and Ψ are irrelevant, the sparse
coefficient vector S can be recovered from the vector z with high probability by solving the
convex optimization problem, shown in Eq. (5). Finally, the inverse transform of sparse
representation is performed on the vector S to restore the natural signal y.

b
S ¼ argmin kS k0 ; s:t: z ¼ ΘS ð5Þ

In Eq. (5), ‖S‖0 refers to the l0-norm of vector S, which is equal to the number of non-zero
elements in the vector.

2.3 Fractional Fourier transform

Fractional Fourier transform (FRFT) is a generalized form of Fourier transform [8, 35]. It is
obtained by rotating the natural signal counterclockwise at any angle on the time axis.
Therefore, the FRFT of a signal contains both its time- and frequency-domain features. The
p1, p2-order FRFT of a 2-D signal f(x, y) is defined as Eq. (6).
þ∞ þ∞
F p1;p2 ðu; vÞ ¼ ∫−∞ ∫−∞ f ðx; yÞk p1;p2 ðx; y; u; vÞdxdy ð6Þ
 Multimedia Tools and Applications

pffiffiffiffiffiffiffiffiffiffiffipffiffiffiffiffiffiffiffiffiffiffi  
1−icotα 1−icotβ iðx2 þu2 Þ
In Eq. (6), the kernel function k p1;p2 ðx; y; u; vÞ ¼ 2π exp 2tanα − sinαixu

 
iðy2 þv2 Þ
exp 2tanβ − sinβiyv
, where α
and β represent the rotation angles on the x and y axes, respectively.

2.4 Fisher-yates confusion

Fisher-Yates algorithm [23], also known as Knuth random scrambling algorithm, is utilized to
scramble the sparse coefficients of plain image to reduce the strong correlation between
adjacent elements. Its scrambling process is illustrated in Fig. 3. In operation, the chaotic
sequence generated by the 4-D QCNN is employed to replace each randomly generated
number, effectively controlling the elements exchanged each time. Meanwhile, the initial
conditions of 4-D QCNN are calculated by extracting partial pixels of plain image with the
secret keys. Therefore, different plain images correspond to different scrambling sequences.

3 The proposed encryption scheme

3.1 Generation of some important parameters

3.1.1 Computing the initial values for QCNN

Logistic map is utilized to extract partial pixels of plain images P1 and P2 to generate the
initial values of 4-D QCCN, assuming that their resolutions are both m × n. Its definition is
displayed in Eq. (7). When the system parameter μ is between 3.57 and 4, this non-linear map
exhibits chaotic characteristics.
rn ¼ μ  rn−1 ð1−rn−1 Þ; n ¼ 2; 3; … ð7Þ

The detailed approach for generating the initial values of QCNN is displayed as follows.

Randomly select a number among the

The original sequence former six elements to exchange with 7

1 2 3 4 5 6 7 1 2 7 4 5 6 3

Randomly select a number among the Randomly select a number among the
former four elements to exchange with 5 former five elements to exchange with 6

6 2 5 4 7 1 3 6 2 7 4 5 1 3
...

The final sequence

5 4 6 2 7 1 3
Fig. 3 The scrambling process of Fisher-Yates algorithm
Multimedia Tools and Applications

First, the Logistic map is iterated (mn + T0) times with the initial value r0, and the first T0
elements are discarded to obtain a chaotic sequence r = {r1, r2, r3, …, rmn}. Then the sequence r
is sorted in ascending order to generate a new sequence Tr.

Step 1: Calculate the sum of [P1Tr(1), P1Tr(2), …, P1Tr(mn/2)] and [P2Tr(mn/2 + 1), P2Tr(mn/2 +
2), …, P2Tr(mn)] and its average value av1 is determined. The mathematical descrip-
tion of this step is shown as follows.

mn=2   
su ¼ ∑ P1ðTri Þ þ P2 Trmn=2þi ð8Þ
i¼1

su
av1 ¼ ð9Þ
mn

Step 1: Step 3. Finally, the initial values of 4-D QCNN are calculated by performing Eq.
(11), where the symbol “mod” means modular operation, ⌊x⌋ represents an integer
not greater than x, and ki (i = 1, 2) are the internal key parameters.

fv ¼ bðav1 −bav1 cÞ  255c  255−1 ð10Þ

8
>
> ẋ1 ¼ fv þ k 1 mod 1
>
<
ẋ2 ¼ fv þ ẋ1 mod 2 þ 2
ð11Þ
> ẋ3 ¼ fv  103 þ k 2 mod 1
>
>
:
ẋ4 ¼ fv  103 þ ẋ3 mod 2 þ 2

3.1.2 Obtaining the initial values for IHM

Storing the entire measurement matrix directly requires a lot of memory space, and sufficient
bandwidth needs to be used to transmit it to the decoder. Thus, in this paper, the measurement
matrix is determined by the chaotic sequence that is generated by the improved 2-D Henon
map. In the meantime, to improve the anti-attack ability of proposed algorithm, partial content
of two plain images will be used to generate the initial values of IHM.
The process of generating the initial values of improved 2-D Henon map is as follows.

Step 1: First, the average value av2 is determined as follows.

1 mn=2   
av2 ¼ ∑ P1 Trmn=2þi þ P2ðTri Þ ð12Þ
m  n i¼1

Then the parameter gv is obtained by performing Eq. (13) on the average value av2, where
len(x) represents the number of integer parts of x. For example, len(153.72902) = 3.
j k
gv ¼ av2  10−lenðav2 Þ  255  255−1 ð13Þ
 Multimedia Tools and Applications

Step 2: Finally, the initial values of IHM are computed according to the following Eq. (14),
where k3 and k4 are the internal key parameters. Additionally, sign(·) means sign
function.

  
u0 ¼ logk 3 gv mod 1  signðgv−k 3 Þ
ð14Þ
v0 ¼ u0 þ k 4 mod 1

3.2 The VMDIE algorithm

The flow chart of proposed VMDIE algorithm is displayed in Fig. 4. As Fig. 4 shows, it
mainly consists of two stages. In the first stage, the secret information carried by two plain
images is encrypted and compacted by the Fisher-Yates confusion and the key-controlled
partial Hadamard matrix, respectively. Then, in the second stage, on the one hand, the
encrypted data is randomly embedded into the host image through the 2-D fractional Fourier
transform embedding, and this process is controlled by the index sequence generated from the

Plain image P1 Plain image P2

Parameters μ, r0 Logistic map

Secret keys k1, Sparsity processing Sparse orthogonal matrix

Parameters fv, gv
k2, k3, k4

Initial values Threshold processing Threshold value Ts1, Ts2

Improved 2D Quantum cellular

Fisher-Yates confusion
Henon map neural network

key-controlled partial
Parallel measurement
Hadama matrix Stage 1: Pre-encryption

Index confusion

Host image HI 2D DCT Encrypted image EI1 Encrypted image EI2 Secret keys α, β

Fractional Fourier Inverse fractional

Secret keys α, β Real Part RP1 Imaginary part IP1
transform Fourier transform

Parameters fv, gv

Alpha channel

Secret keys h1, h2

Cipher image CI Real Part RP2 Imaginary part IP2 Inverse 2D DCT

Stage 2: Embedding

Fig. 4 The schematic of the proposed VMDIE algorithm

Multimedia Tools and Applications

4-D QCNN. In addition, some important parameters are also hidden in the alpha channel of
final visually meaningful cipher image.

3.2.1 Pre-encryption process

First, an orthogonal sparse representation matrix Ψ ∈ ℝn × m (n = m) is constructed using the

Daubechies wavelet. And then, the sparse processing is performed on two plain images P1 and
P2 through Eq. (15), where the symbol ΨT represents the transpose matrix of Ψ.

P3 ¼ Ψ  P1  Ψ T
ð15Þ
P4 ¼ Ψ  P2  Ψ T

Step 1: To further improve the sparsity of coefficient matrices P3 and P4, the elements
whose absolute values are less than or equal to the threshold values Ts1 and Ts2 are
forced to be set to zero. And the matrices after threshold processing are denoted as
P5 and P6 respectively.
Step 2: After sparse processing, most of the energy of two plain images P1 and P2 is mainly
concentrated in the upper left corner of the matrices P5 and P6, which is not
conducive to parallel compression. Thus, the Fisher-Yates confusion is

utilized to evenly distribute the energy to the entire matrix. The 4-D QCNN is iterated (mn +
h i
T0) times with the initial value ẋ1 ; ẋ2 ; ẋ3 ; ẋ4 T. Then four chaotic sequences with size of
1 × mn are obtained by abandoning the former T0 values, as shown in Eq. (16).
8
>
> X ¼ fx1 ; x2 ; x3 ; …; xmn g
<
Y ¼ fy1 ; y2 ; y3 ; …; ymn g
ð16Þ
> Z ¼ fz1 ; z2 ; z3 ; …; zmn g
>
:
W ¼ fw1 ; w2 ; w3 ; …; wmn g

Step 3: Then, the random sequences X and Y are processed according to the following
equation.


Tx ¼ X  1010 mod ½mn : −1 : 1 þ 1
ð17Þ
Ty ¼ Y  1010 mod ½mn : −1 : 1 þ 1

Step 4: As described in Section 2.4, the Fisher-Yates algorithm is used to scramble the
matrices P5 and P6, which is controlled by the sequences Tx and Ty. After
confusion, the resulting matrices are named P7 and P8 respectively.
Step 5: Next, the measurement matrix Φ ∈ ℝcn × m is constructed using the hardware-friendly
Hadamard matrix, where cn = CR × n. CR is the preset compression rate. The 2-D
 Multimedia Tools and Applications

IHM is iterated (mn + T0) times under the condition that the system parameter and the
initial value are set to [2.1, 6.74] and [u0, v0]T, respectively. Then, the chaotic
sequence U ¼ fuT 0 þ1 ; uT 0 þ2 ; uT 0 þ3 ; …; uT 0 þmn g is determined.
Step 6: Construct a Hadamard matrix H sized of m × m. It is calculated by the Kronecker
product of two low-order matrices, and its recursion equation is shown in Eq. (18).
Additionally, considering the constraints of the Hadamard matrix, we assume that m
can be divisible by four.

 
H 2n−1 H 2n−1
H ¼ H 2 ⊗H 2n−1 ¼ ð18Þ
H 2n−1 −H 2n−1
pffiffi
 
þ1 þ1
where H 2 ¼ 2
.
2 þ1 −1

Step 7: Sort the sequence U in the ascending order to generate the index sequence Tu. Then,
the key-controlled partial Hadamard matrix Φ is obtained by the following equation.

2 3
H ðTu1 ; 1 : endÞ
6 H ðTu2 ; 1 : endÞ 7
6 H ðTu3 ; 1 : endÞ 7
Φ¼6 7 ð19Þ
4 ⋮ 5
H ðTucn ; 1 : endÞ

Step 8: The matrices P7 and P8 are compressive in parallel by the key-controlled measure-
ment matrix Φ to generate the encrypted matrices P9 and P10. This process can be
described by Eq. (20).

Pði þ 2Þ ¼ Φ  Pi; i ¼ 7; 8 ð20Þ

3.2.2 Embedding process

Step 1: Select host image HI ∈ ℕm × n and perform the 2-D discrete cosine transform on it,
which is formulated in Eq. (21).

2 m−1 n−1 ð2p þ 1Þiπ ð2q þ 1Þjπ

H1ði; jÞ ¼ pffiffiffiffiffiffi cðiÞcð jÞ ∑ ∑ HI ðp; qÞ cos cos ð21Þ
mn p¼0 q¼0 2m 2n
8
< 1
pffiffiffi ; i ¼ 0; j ¼ 0
where the transform kernel function cðiÞ ¼ cð jÞ ¼ .
: 2
1; otherwise
Multimedia Tools and Applications

Step 2: The sub-matrix H2 is determined by Eq. (22). And then, apply the 2-D fractional
Fourier transform on it with the rotation angle [α, β] to get the complex matrix
H3 = RP1 + IP1ζ. ζ is the imaginary unit.

2 m n m 3
H1 þ 1; þ 1 ⋯ H1 þ 1; n
6 2 2 2 7
H2 ¼ 6
4 ⋮ ⋱ ⋮ 7
5 ð22Þ
n
H1 m; þ 1 ⋯ H1ðm; nÞ
2
By sorting the chaotic sequences [Z1, Z2, Z3, …, Zcn × n] and [W1, W2, W3, …, Wcn × n] in ascend-
ing order, the index sequences Tz, and Tw are acquired. Then the index confusion is carried out
on the P9 and P10 according to Eq. (23), where i = 1, 2, …, cn × n.

P11ðiÞ ¼ P9ðTzðiÞÞ
ð23Þ
P12ðiÞ ¼ P10ðTwðiÞÞ

Step 3: Step 4. The matrices P11 and P12 are embedded into the real and imaginary parts of
the complex matrix H3, respectively, after adjusting their amplitudes by the gain
factor γ. These operations are formulated by Eq. (24).

RP2 ¼ RP1 þ ð1−γ Þ  P11
ð24Þ
IP2 ¼ IP1 þ ð1−γ Þ  P12

Step 4: Then, perform the inverse 2-D fractional Fourier transform on the complex matrix
H4 = RP2 + IP2ζ and replace the lower right corner of matrix H1 to obtain a new
complex matrix H5.
Step 5: Next, the complex matrix H6 = RP3 + IP3ζ is generated by applying the inverse 2-D
discrete cosine transform on the matrix H5.

To facilitate storage, the matrix IP3 is processed by the following equation. Then the Alp is
used as the alpha channel together with the matrix RP3 to generate the final visually
meaningful cipher image CI ∈ ℕm × n.
Alp ¼ 247 þ IP3 ð25Þ

Step 6: In the proposed scheme, there are different parameters (fv and gv) for different plain
images. Thus, considering the basic principles of symmetric encryption, we embed
them into the alpha channel of final cipher image. See Eq. (26) for the specific
operation. At this point, the entire encryption operations are complete.

  
Alp h1  1010 mod ½m; n þ 1 ¼ fv  255
  ð26Þ
Alp h2  1010 mod ½m; n þ 1 ¼ gv  255
 Multimedia Tools and Applications

4 The image decryption algorithm

The inverse process of VMDIE algorithm is the decryption algorithm, which also includes two
processes, namely the extraction and decryption process. To successfully recover the plain
images, some of internal secret keys need to be transmitted to the decoder through a secure
channel, including ki (i = 1, 2, 3, 4), α, β, γ and hi (i = 1, 2). Meanwhile, the internal secret keys
fv and gv are extracted from the alpha channel of visually meaningful cipher image. Besides,
the host image is indispensable for extracting the encrypted data from the cipher image. Thus,
it is recommended to randomly select the host image from the public image databases to avoid
increasing extra transmission costs. Then, the detailed decryption process is shown below.

4.1 Extraction process

Step 1: the parameters fv and gv are extracted from the alpha channel of cipher image
according to Eq. (27). Then, to avoid affecting the quality of the decrypted images,
the values of these two parameters in the Alp are set to 247 after extraction.

  
fv ¼ Alp h1  1010 mod ½m; n þ 1  255−1
  ð27Þ
gv ¼ Alp h2  1010 mod ½m; n þ 1  255−1

Step 2: By performing Eq. (11) and Eq. (14), the initial values of 4-D QCNN and the 2-D
IHM are obtained. And then, iterate them to generate the key streams X, Y, Z, W and
U.
Step 3: The modified matrix Alp is processed by the following equation to extract the IP3.

IP3 ¼ Alp−247 ð28Þ

The complex matrix H5 is determined by carrying out the 2-D discrete cosine transform on the
H6 = RP3 + IP3ζ.

Step 4: After performing the 2-D fractional Fourier transform on the H4, The matrices P11
and P12 are obtained according to the Eq. (29).


P11 ¼ ðRP2−RP1Þ  ð1−γ Þ−1
ð29Þ
P12 ¼ ðIP2−IP1Þ  ð1−γ Þ−1

Step 5: After inverse scrambling of matrices P11 and P12 with the index sequences Tz and
Tw respectively, the encrypted matrices P9 and P10 are generated.
Multimedia Tools and Applications

4.2 Decryption process

Step 1: The key-controlled measurement matrix Φ ∈ ℝcn × m is generated as described in the

step 8 of Section 3.2.1, And then utilized to recover the matrices P7 and P8 with the
SL0 algorithm. It can be denoted as Eq. (30).

Pði−2Þ ¼ SL0 ðΦ; PiÞ; i ¼ 7; 8 ð30Þ

Step 2: Inverse Fisher-Yates confusion (IFYC) is applied to the matrices P7 and P8 with the
Tx and Ty generated by sorting the sequences X and Y, returning the matrices P5 and
P6, respectively.
Step 3: By performing inverse sparse transform, displayed in Eq. (31), on the P5 and P6,
then the plain images P1 and P2 with resolution of m × n are reconstructed.


P1 ¼ Ψ T  P5  Ψ
ð31Þ
P2 ¼ Ψ T  P6  Ψ

5 Simulation results and performance analyses

5.1 Simulation environment setting

The simulation experiments are all carried out on the laptop computer equipped with 1.8GHz
i7-8550U CPU and 16G RAM. The operating system is Microsoft Windows 10, and the
simulation platform is selected as Matlab 2018b. Three groups of plain images and host images
with different resolutions are used for the test images. Among them, most of test images are
randomly selected from (https://openi.nlm.nih.gov/), (https://ccia.ugr.es/cvg/dbimagenes/) and
(https://sipi.usc.edu/database/). Meanwhile, the internal secret keys are set as follows: k1 = 0.
9723, k2 = 0.7875, k3 = 0.3746, k4 = 0.2481, h1 = 0.2849, h2 = 0.7128, α = 1.41, β = 0.6 and γ =
0.998. Additionally, the remaining encryption parameters are Ts1 = 35, Ts2 = 25, CR = 0.25,
r0 = 0.4751, μ = 3.99. And in the encryption phase, the chaotic systems first iterate 500 times to
eliminate the transient effect.

5.2 Encryption and decryption results

To demonstrate the effectiveness and practicability of proposed VMDIE algorithm, simulation

experiments are conducted in this subsection, and the results are drawn in the Figs. 5, 6 and 7.
As can be seen, the plain images are encrypted into the same-sized cipher images which are
meaningful and visually similar to the corresponding host images, indicating that the proposed
FRFT embedding approach is effective. Actually, when the visually meaningful cipher image
is transmitted or stored together with other natural images, it is less likely to be discovered and
attacked by the attackers, making it more secure. In other respects, the decrypted images are
visually identical to their respective plain images.
Next, to quantitatively analyze the imperceptibility of cipher images and the reconstructed
quality of decrypted images, the peak signal-to-noise ratio (PSNR) and mean structural
 Multimedia Tools and Applications

(a) Plain image Lena (b) Plain image Woman (c) Host image Lake (d) Cipher image

(e) Decrypted image Lena (f) Decrypted image Woman (g) Histogram of (c) (h) Histogram of (d)

Fig. 5 Simulation results with Lena and Woman as plain images (resolution 512 × 512)

similarity (MSSIM) will be employed, which are defined in Eq. (32) and Eq. (33) [16]
respectively. Then, the PSNR and MSSIM values of simulation results are elaborated in
Table 1. As illustrated in Table 1, regardless of the resolution, the values of PSNRcip are
greater than 34 dB, and MSSIMcip are approximately equal to 1. Therefore, these numerical
results indicate that the encrypted data carried by the cipher images is highly invisible.
Moreover, the PSNRdec are basically greater than 30 dB. And as the resolution of plain image
increases, the reconstructed quality of decrypted image also increases. To a certain extent, the
reconstruction quality is satisfactory. In short, the proposed meaningful encryption scheme can
provide double protection of image data and appearance, as well as has great potential for
application in other fields such as medicine and transportation.

2552
PSNR ¼ 10  log
1 m n  2 ðdBÞ ð32Þ
∑ ∑ I 1i; j −I i;2 j
m  n i¼1 j¼1

(a) Plain image Kidney (b) Plain image Brain (c) Host image Lung (d) Cipher image

(e) Decrypted image Kidney (f) Decrypted image Brain (g) Histogram of (c) (h) Histogram of (d)

Fig. 6 Simulation results with Kidney and Brain as plain images (resolution 512 × 512)
Multimedia Tools and Applications

(a) Plain image Plane (b) Plain image Boat (c) Host image Sailboat (d) Cipher image

(e) Decrypted image Plane (f) Decrypted image Boat (g) Histogram of (c) (h) Histogram of (d)

Fig. 7 Simulation results with Plane and Boat as plain images (resolution 512 × 512 × 3)

1 L 2μi μi þ C 1 2σi σi þ C 2 σi þ C 3
MSSIM ¼  ∑  2 I 1 I 2 2   2 I 1 I 2 2  Ii1 I 2 i ð33Þ
L i¼1 μi þ μiI 2 þ C 1 σiI 1 þ σiI 2 þ C 2 σI 1  σI 2
I1

where C1 = (0.01 × 255)2, C2 = (0.03 × 255)2 and C3 = C2 × 0.5. Besides, μiI1 and σiI1 respectively
represent the mean value and variance of the i-th non-overlapping block in the image I1.

5.3 Influence of gain factor on simulation results

Considering that the amplitude of encrypted data is regulated by the gain factor γ in the
embedding process. Therefore, the influence of gain factor on the simulation results is given in
this section, which is plotted in Fig. 8. Among them, the red curves represent the PSNR
between final meaningful cipher images and host images. Moreover, the blue curves represent
the PSNR between plain images and decrypted images. It can be seen from the figure that as
the value of the gain factor increases, the red curve gradually rises. While for the blue curve, it
rises in the first stage and then falls. Additionally, in the case of weighing the quality of the
decrypted image and the visual security of the cipher image, the optimal gain factor is different
for different plain images.

Table 1 PSNR and MSSIM values of simulation results

Resolution Plain image Host image PSNRdec (dB) PSNRcip (dB) MSSIMcip

256×256 Finger Baboon 30.2669 34.6253 0.9903

Cameraman 30.1529
Peppers Goldhill 29.7925 35.4728 0.9844
Zelda 30.1743
512×512 Lena Lake 32.9445 34.3542 0.9970
Woman 34.4590
Kidney Lung 34.3181 34.1042 0.9983
Brain 36.1321

*Where the symbols PSNRcip and MSSIMcip mean the PSNR and MSSIM values between host image and final
cipher image, respectively. Additionally, the symbol PSNRdec is the PSNR value between plain image and
decrypted image
 Multimedia Tools and Applications

Fig. 8 PSNR vs gain factor with different images

5.4 Influence of embedding position on decryption quality

In the proposed VMDIE algorithm, we introduce a FRFT-based embedding approach, which

can embed two plain images into the real and imaginary parts of complex matrix, respectively.
Therefore, this sub-section will evaluate the influence of embedding position on decryption
stage. First, two identical plain images are subjected to proposed encryption algorithm under
the condition that Ts1 = Ts2 = 35 and other encryption parameters are consistent with those
described in Section 5.1. Besides, the image “Baboon” is selected as the host image. The
experimental results are plotted in Fig. 9. As observed from Fig. 9, the embedding position
basically has no effect on the quality of decrypted image. Therefore, the plain images can be
encrypted and freely embedded into different locations.

Fig. 9 PSNR vs embedding position for different images with different resolutions
Multimedia Tools and Applications

5.5 Multiple attack analysis

5.5.1 Violent attack

The key space and key sensitivity together determine the ability of algorithm to resist violent attacks.
In this paper, the secret keys mainly consist of the following three parts. i.e., (a) the (k1, k2, k3, k4)
used for generating the initial values of two chaotic systems. (b) the (h1, h2) used for calculating the
location of internal parameters in the alpha channel and (c) the rotation angle (α, β). Suppose that the
step length of ki (i = 1, 2, 3, 4) and hi (i = 1, 2) are 10−14 and 10−10 respectively, while the step length
of rotation angle is 10−2. Thus, the total key space can be calculated by Eq. (34).
 4  2  2
keytotal ¼ 1014  1010  2π  102 ¼ 4π2  1080 ð34Þ

Furthermore, the key space for other algorithms is listed in Table 2. It can be seen that
compared with [6, 32, 41, 50], the proposed VMDIE algorithm has a larger key space.
To qualitatively analyze the key sensitivity, the images “Lena” and “Woman” are subjected
to the proposed algorithm. Then, the modified secret keys by adding a slight change to one of
the correct keys are used to decryption. Moreover, the decrypt-ed images are illustrated in
Fig. 10. It can be clearly seen that when one of the correct keys is slightly changed, the
decrypted image does not visually reveal any useful information of plain image, indicating that
the proposed encryption scheme is sensitive to the keys. To sum up, our scheme is sufficiently
resistant to violent attacks.

5.5.2 Statistical attack

The pixel distribution of an image can be reflected by the histogram. Since the strong
correlation between adjacent pixels in natural image, its histogram always presents an uneven
shape. However, for an effective meaningful image encryption algorithm, the histogram of final
visually secure cipher image should be as consistent as possible with that of the corres-ponding
host image. Next, the distance of histogram intersection [32] is introduced to measure the slight
differences among host images and cipher images, which can be calculated by Eq. (35).
 −1
2B 2B
H ð J; VÞ ¼ ∑ minð Ji ; Vi Þ  ∑ Vi ð35Þ
i¼1 i¼1

where (J, V) is a pair of histogram and B represents the bit depth of image. As shown in Eq. (35),
when the histograms J and V are equal, the H(J, V) reaches its maximum value, that is one. In
the experiment, the plain images with different resolutions are encrypted and embedded into
different host images. Then, the obtained results are listed in Table 3. It is observed that the
distance of histogram intersection between the host images and the cipher images is close to 1,
indicating that the proposed VMDIE algorithm has good visual security. Similarly, it can also
be seen that the host images have little impact on the value of H(J, V).

Table 2 Comparative results of key space

Algorithm Ours [50] [6] [41] [32]

Key space 4π2 ×1080 2.56×1059 1056 1056 1075

 Multimedia Tools and Applications

Fig. 10 Decrypted image “Lena” using incorrect keys

5.5.3 Noise attack

Considering that the meaningful cipher image transmitted over the channel will be inevitably
affected by various noises, resulting in loss of partial ciphertext data, which increases the
difficulty of recovering the plain image. We will perform several experiments to evaluate the
anti-noise performance of proposed VMDIE algorithm in this sub-section under the condition
that the parameters fv and gv hidden in the alpha channel are not destroyed. First, the plain
images “Lena” and “Woman” with resolution of 512 × 512 are encrypted and embedded into
the host image “Lake” via FRFT embedding approach. Then, multiple types of noise with
normalized intensity of 0.0001%, 0.0005%, 0.001% and 0.005% are added to the meaningful
cipher image respectively, including salt & pepper noise (SPN), speckled noise (SN) and
Gaussian noise (GN). The resultant decrypted images are plotted in Figs. 11, 12 and 13.
Besides, the values of PSNR between decrypted image and plain image under different noise
attacks are listed in Table 4.
As illustrated in Figs. 11, 12 and 13 and Table 4, when the attack intensity varies from
0.0001% to 0.005%, the quality of decrypted image drops significantly, and the maximum
drop reach 4.787 dB, but we can still visually identify the secret information carried by the
decrypted image. In another aspect, it can be seen from the experimental data that in the case of
the same noise attack intensity, GN has the greatest impact on our proposed scheme, while

Table 3 The difference between the histograms of host images and cipher images

Resolution Plain image Host image Distance

256×256 Finger Baboon 0.9311

Cameraman Goldhill 0.9266
Peppers 0.9208
Zelda 0.9472
512×512 Lena Lake 0.9123
Woman Lung 0.9158
Barbara 0.9387
Baboon 0.9423
Multimedia Tools and Applications
Cipher image

(a) 0.0001% SPN (b) 0.0005% SPN (c) 0.001% SPN (d) 0.005% SPN
Decrypted image 1

(e) 0.0001% SPN (f) 0.0005% SPN (g) 0.001% SPN (h) 0.005% SPN
Decrypted image 2

(i) 0.0001% SPN (j) 0.0005% SPN (k) 0.001% SPN (l) 0.005% SPN
Fig. 11 Simulation results under SPN with different intensities
Cipher image

(a) 0.0001% SN (b) 0.0005% SN (c) 0.001% SN (d) 0.005% SN

Decrypted image 1

(e) 0.0001% SN (f) 0.0005% SN (g) 0.001% SN (h) 0.005% SN

Decrypted image 2

(i) 0.0001% SN (j) 0.0005% SN (k) 0.001% SN (l) 0.005% SN

Fig. 12 Simulation results under SN with different intensities
 Multimedia Tools and Applications
Cipher image

(a) 0.0001% GN (b) 0.0005% GN (c) 0.001% GN (d) 0.005% GN

Decrypted image 1

(e) 0.0001% GN (f) 0.0005% GN (g) 0.001% GN (h) 0.005% GN

Decrypted image 2

(i) 0.0001% GN (j) 0.0005% GN (k) 0.001% GN (l) 0.005% GN

Fig. 13 Simulation results under GN with different intensities

SPN and SN have the least impact. In general, the proposed VMDIE scheme has a good ability
to resist noise interference.

5.5.4 Cropping attack

The ability of proposed VMDIE algorithm to resist cropping attacks is also evaluated in this
sub-section. Similarly, it is assumed that the parameters fv and gv in the meaningful cipher
image are not corrupted. Then, different positions of the cipher images are cut with sizes of
128 × 128, 180 × 180 and 256 × 256, which are drawn in the first row of Fig. 14. And the
corresponding decrypted images are depicted in the second and third row of Fig. 14.

Table 4 PSNR values under different noise attacks

Noise type Intensity

0.0001% 0.0005% 0.001% 0.005%

SPN Psnr1 =32.9598 Psnr1 =32.9598 Psnr1 =31.1012 Psnr1 =29.8191

Psnr2 =34.4402 Psnr2 =34.4402 Psnr2 =31.9757 Psnr2 =30.2442
SN Psnr1 =32.9598 Psnr1 =32.1540 Psnr1 =31.6417 Psnr1 =30.1879
Psnr2 =34.4402 Psnr2 =33.3408 Psnr2 =32.5616 Psnr2 =30.7421
GN Psnr1 =32.5458 Psnr1 =30.9543 Psnr1 =30.3381 Psnr1 =28.8122
Psnr2 =33.8099 Psnr2 =31.6741 Psnr2 =30.8617 Psnr2 =29.0229
Multimedia Tools and Applications

Fig. 14 Decrypted results of the cipher image (512 × 512) suffering from different cropping attacks

It can be seen from the figure that the position of cropping blocks has basically no impact
on the visual quality of the decrypted image in our scheme. Besides, as the size of cropping
block increases, the reconstructed quality of decrypted image decreases. However, when a
quarter of cipher image data is lost, we can still find the information carried by the plain image
in decrypted image visually. And the PSNR of decrypted images are about 26.7 dB and
26.8 dB respectively, indicating that our VMDIE scheme can withstand the cropping attacks to
a certain extent, provided that the eigenvalues are not lost.

5.5.5 Known-plaintext and chosen-plaintext attacks

Up to now, many image encryption algorithms have been cracked by plaintext analysis-based
security attack models, such as [20, 25, 42]. The main reason is that different plain images
correspond to the stationary secret key streams in encryption stage. However, in this paper, the
pixel eigenvalue of two plain images is utilized to control the generated key stream. First, the
initial values of 4-D QCNN are determined by the eigenvalue fv. Then, it is iterated and sorted
to fabricate two sets of index sequences. One set of sequences is applied to control the Fisher-
Yates confusion performed on the image. And the remaining set of sequences is used to
randomly embed the encrypted data into the host image. In other aspects, the eigenvalue gv is
adopted to control the 2-D IHM to generate the key-controlled partial Hadamard matrix, which
is used to compress two plain images in parallel. In short, since the proposed VMDIE
algorithm can realize “one plain image corresponds to one key”, it can withstand known-
plaintext and chosen-plaintext attacks.
Additionally, in order not to violate the basic principles of symmetric cryptosystems, we
propose to embed the eigenvalues into the final meaningful cipher image and synchronously
transmit it to the decoders. Moreover, the scheme of hiding eigenvalues has two
 Multimedia Tools and Applications

characteristics. First, the location of eigenvalues is controlled by secret keys. Secondly, the
eigenvalues are embedded in the alpha channel of visual cipher image to reduce the probability
of being attacked.

5.6 Running efficiency analysis

Except for considering the security of algorithm, running efficiency is also an indispensable
indicator to evaluate the performance of algorithm in practical application scenarios. In the first
place, the time complexity for proposed meaningful image encryption algorithm mainly
includes the time cost of following three phases, that is, the important parameter generation
phase, pre-encryption phase and embedding phase. Generally speaking, the complexity of
pretreatment for double images can be ignored. Added up, the time complexity for proposed
  
encryption scheme is Θ 52 þ CR n2 þ n . What’s more, the execution times required for
encryption and decryption of images with different resolutions are listed in Tables 5 and 6. In
these two tables, “Pretreatment” represents the phase that generates the initial values of two
chaotic systems. And “Reconstruction” stands for the process of recovering two natural images
from encrypted data through the SL0 reconstruction algorithm and the inverse Fisher-Yates
scrambling.
As can be seen, embedding and reconstructing the encrypted data account for a large
proportion of total encryption and decryption time. Such as for the plain image with size of
512 × 512, the FRFT embedding phase takes up about 89.465% of total time in encryption.
Moreover, the reconstruction phase occupies around 89.966% of total decryption time. In
another aspect, when the resolution of plain image changes from 256 × 256 to 512 × 512, the
time spent in the encryption and decryption process increases about 10 times and 6 times,
respectively. Thus, it is suggested to divide the large-scale plain image into several small sub-
images, and then perform encryption and embedding operations in parallel to shorten the
execution time.

6 Comparison with the existing work

To highlight the proposed VMDIE algorithm, we first summarize the characteristics of existing
visually meaningful image encryption algorithm, and compare them with ours. The

Table 5 Encryption time for plain images with different resolutions (Unit: s)

Resolution Plain image Pretreatment Pre-encryption FRFT embedding Total

256×256 Cameraman 0.005864 0.028206 0.087978 0.122048

Goldhill
Finger 0.007858 0.021990 0.081224 0.111072
Peppers
Average 0.006861 0.025098 0.084601
512×512 Lena 0.024722 0.093751 1.004689 1.123162
Woman
Kidney 0.025451 0.089174 0.974852 1.089477
Brain
Average 0.025087 0.091463 0.989771
Multimedia Tools and Applications

Table 6 Decryption time for plain images with different resolutions (Unit: s)

Resolution Plain image Pretreatment Extraction Reconstruction Total

256×256 Cameraman 0.010422 0.037255 0.984013 1.031690

Goldhill
Finger 0.008713 0.050944 0.944407 1.004064
Peppers
Average 0.009568 0.044100 0.964210
512×512 Lena 0.029605 0.548270 5.285881 5.863756
Woman
Kidney 0.029082 0.590665 5.451860 6.071607
Brain
Average 0.029344 0.569468 5.368871

comparison results are displayed in Table 7. Besides, part of them is contrasted with our
proposed encryption algorithm from the following three aspects: visual security, anti-noise
performance and running efficiency. Moreover, the corresponding numerical results are
elaborated in Tables 8, 9 and 10. Note that for the sake of fairness, the experimental data of
[5, 32, 50] are obtained from their source articles or the related articles, and N/A means that the
value of corresponding indicator is not provided. It can be seen from the comparison results

Table 7 Comparison of characteristics for different algorithms

No Other algorithms Our

1 The plain image with resolution of m ×n is Before the embedding phase, the compressive
pre-encrypted and embedded into a host image sensing is adopted to compress the plain image to
with resolution of 2m×2n to obtain a meaningful a quarter of its original resolution. Thus, the
cipher image, thus increasing the cost of storage meaningful cipher image has the same resolution
and transmission, such as [1, 41]. as the plain image in our scheme.
2 Only one plain image can be encrypted at a time, Two plain images can be simultaneously encrypted
such as [22, 27, 36, 39]. The encryption efficiency and embedded into the host image with the same
and transmission efficiency are limited. resolution as the plain image via the FRFT
embedding method in this paper. The transmission
efficiency has been improved, and there is no need
to provide additional transmission costs.
3 There are both quantization error and truncation error In our scheme, to reduce the influence of errors, the
in [22, 27, 36]. Because of the accumulation of encryp-ted data is directly embedded into a host
errors, the visual quality of the decrypted image is image under the control of gain factor, without
reduced. quantization operation.
4 The existing embedding methods, such as LSB [6], In the proposed FRFT-based embedding method, the
SVD [41, 50], IWT [1, 18], RCT [22], DCT [26] rota-tion angles of the fractional Fourier transform
and so on, cannot protect the hidden ciphertext. can be used as the keys to improve the security of
the hidden ciphertext.
5 Part of the existing visually meaningful image The characteristic information of the two plain
encryption algorithms are vulnerable to the images is utilized to generate the key stream in
chosen-plaintext and known-plaintext attacks, encryption, so that different plain images
since in their schemes, different plain images cor- correspond to different key streams. Additionally,
respond to the same key stream. in order not to violate the basic principles of the
symmetric encryption cryptosystem, the
eigenvalues are embedded in the alpha channel of
the meaningful cipher image under the control of
the key.
 Multimedia Tools and Applications

Table 8 Comparisons of the PSNR and MSSIM between the cipher and host images in different encryption
schemes

Plain image Host image [5] [32] [50] Ours

PSNR MSSIM PSNR MSSIM PSNR MSSIM PSNR MSSIM

Lena Peppers 18.5136 0.6726 32.3513 0.9257 31.7986 0.9903 36.0020 0.9963
Brain Cameraman 24.8700 0.6488 34.8967 0.9381 31.1582 0.9725 35.3293 0.9985
Jet Baboon 23.3967 0.6991 37.8967 0.9833 32.6033 0.9955 34.2920 0.9984
Barbara Bridge 25.2321 0.7337 35.5629 0.9783 31.7397 0.9946 34.1259 0.9982
Average 23.0031 0.6886 35.1769 0.9564 31.8250 0.9882 34.9373 0.9979

*To compare with the single-image visually meaningful encryption algorithm, in this experiment, two identical
plain images are placed in the proposed VMDIE algorithm

Table 9 Comparison of the resistance capability of noise attacks

Nosie Type PSNR

Noise Intensity=0.0001% Noise Intensity=0.0003% Noise Intensity=0.0005%

[32] [50] Our [32] [50] Our [32] [50] Our

GN 14.75 25.13 33.49 9.35 19.41 32.31 8.54 17.78 31.84

SN 33.44 31.56 33.97 22.50 27.68 33.97 17.30 24.39 33.97
SPN 33.44 31.56 33.97 33.44 30.22 33.97 33.44 30.02 33.92

*In the above experiment, the plain image and host image are selected as “Jet” and “Goldhill”, respectively

that our designed VMDIE algorithm has better anti-noise performance and higher running
efficiency, compared with that in [5, 32]. As for the visual security, the proposed scheme in
this paper is comparable to the [32], but both are significantly better than that of [5].

7 Conclusion

This paper introduces an efficient VMDIE algorithm combined with QCNN, CS and FRFT-
based embedding approach, which can simultaneously realize the image data security and
appearance security. Besides, for withstanding the plaintext attacks, the eigenvalues of plain
images are adopted to generate the cipher flows. And then, these eigenvalues are embedded
into the alpha channel of meaningful cipher image to reduce the probability of being destroyed.

Table 10 Comparisons results of encryption and decryption times in different encryption schemes (Unit: f/s)

Resolution [5] [32] [50] Our

Encryption Decryption Encryption Decryption Encryption Decryption Encryption Decryption

256×256 0.4049 1.4724 0.1533 2.2679 0.3546 0.5192 0.0583 0.5089

512×512 0.9279 10.5830 N/A N/A 0.6185 2.2219 0.5532 2.9838
Multimedia Tools and Applications

Moreover, compared with other existing related schemes, its transmission efficiency has been
improved twice. Eventually, a series of security analysis indicate that the proposed VMDIE
algorithm not only has high visual security and decryption quality, but also can resist the
diversified attacks, such as violent attack, noise attack, chosen-plaintext attack and so on. In
the following work, we will devote ourselves to further improving the sparsity of plain images
to attain the higher image compression rate.

Acknowledgments This work is supported by the National Natural Science Foundation of China [Grant
No.61701043, 41874140], the Shaanxi Province Science and Technology Program [Grant No.2020JM-220,
2020JQ-351], the Fundamental Research Funds for the Central Universities of China [Grant No.300102240205],
the Natural Science Foundation of Fujian Province [Grant No.2020 J05169] and the Natural Science Foundation
of Heilongjiang Province [Grant No.F2018022].

Data availability The image datasets generated and analyzed during the current study are available from the
corresponding author upon the reasonable request.

Declarations

Conflict of interest The authors report no conflicts of interest. The authors alone are responsible for the content
and writing of this paper.

References

1. Bao L, Zhou Y (2015) Image encryption: generating visually meaningful encrypted images. Inf Sci 324:
197–207
2. Baraniuk R (2007) Compressive sensing. IEEE Signal Process Mag 24:118–121
3. Candes E, Romberg J, Tao T (2006) Robust uncertainty principles: exact signal reconstruction from highly
incomplete frequency information. IEEE Trans Inf Theory 52:489–509
4. Cao W, Mao Y, Zhou Y (2020) Designing a 2D infinite collapse map for image encryption. Signal Process
171:107457
5. Chai X, Gan Z, Chen Y, Zhang Y (2017) A visually secure image encryption scheme based on compressive
sensing. Signal Process 134:35–51
6. Chai X, Wu H, Gan Z, Zhang Y, Chen Y, Nixon K (2020) An efficient visually meaningful image
compression and encryption scheme based on compressive sensing and dynamic LSB embedding. Opt
Laser Eng 124:105837
7. Donoho D (2006) Compressed sensing. IEEE Trans Inf Theory 52:1289–1306
8. Elhoseny H, Faragallah O, Ahmed H, Kazemian H, El-sayed H, El-Samie F (2016) The effect of fractional
Fourier transform angle in encryption quality for digital images. Optik 127:315–319
9. Fan H, Zhou K, Zhang E, Wen W, Li M (2020) Subdata image encryption scheme based on compressive
sensing and vector quantization. Neural Comput Applic 32:12771–12787
10. Gan Z, Chai X, Zhang J, Zhang Y, Chen Y (2020) An effective image compression-encryption scheme
based on compressive sensing (CS) and game of life (GOL). Neural Comput Applic 32:14113–14141
11. Gong L, Qiu K, Deng C, Zhou N (2019) An image compression and encryption algorithm based on chaotic
system and compressive sensing. Opt Laser Technol 115:257–267
12. Henon M (1976) A two-dimensional mapping with a strange attractor. Commun Math Phys 50(1):69–77
13. Hu G, Xiao D, Wang Y, Xiang T (2017) An image coding scheme using parallel compressive sensing for
simultaneous compression-encryption applications. J Vis Commun Image Represent 44:116–127
14. Hu G, Xiao D, Wang Y, Xiang T, Zhou Q (2017) Securing image information using double random phase
encoding and parallel compressive sensing with updated sampling processes. Opt Laser Eng 98:123–133
15. Hua Z, Zhou Y (2021) Exponential chaotic model for generating robust chaos. IEEE Trans Syst Man
Cybern -Syst 51(6):3713–3724
16. Hua ZY, Zhang KY, Li YM, Zhou YC (2021) Visually secure image encryption using adaptive-
thresholding sparsification and parallel compressive sensing. Signal Process 183:107998
 Multimedia Tools and Applications

17. Huang R, Rhee K, Uchida S (2014) A parallel image encryption method based on compressive sensing.
Multimed Tools Appl 72:71–93
18. Khan J, Boulila W, Ahmad J, Rubaiee S, Rehman A, Alroobaea R, Buchanan W (2020) DNA and plaintext
dependent chaotic visual selective image encryption. IEEE Access 8:159732–159744
19. Li J, Di X, Liu X, Chen X (2017) Image encryption based on quantum-CNN hyperchaos system and
anamorphic fractional Fourier transform. In 10th International Congress on Image and Signal Processing. J
Biomed Eng Inform:1–6
20. Li M, Wang P, Yue Y, Liu Y (2021) Cryptanalysis of a secure image encryption scheme based on a novel
2D sine-cosine cross-chaotic map. J Real-Time Image Proc 18:2135–2149
21. Mondal B, Singh S, Kumar P (2019) A secure image encryption scheme based on cellular automata and
chaotic skew tent map. J Inf Secur Appl 45:117–130
22. Musanna F, Dangwal D, Kumar S (2020) A novel chaos-based approach in conjunction with MR-SVD and
pairing function for generating visually meaningful cipher images. Multimed Tools Appl 79:25115–25142
23. Musanna F, Kumar S (2019) A novel fractional order chaos-based image encryption using fisher yates
algorithm and 3-D cat map. Multimed Tools Appl 78:14867–14895
24. Naskar P, Bhattacharyya S, Nandy D, Chaudhuri A (2020) A robust image encryption scheme using chaotic
tent map and cellular automata. Nonlinear Dyn. 100:2877–2898
25. Pak C, Huang L (2017) A new color image encryption using combination of the 1D chaotic map. Signal
Process 138:129–137
26. Pan C, Ye G, Huang X, Zhou J (2019) Novel meaningful image encryption based on block compressive
sensing. Secur Commun Netw:6572105
27. Ping P, Fu J, Mao Y, Xu F, Gao J (2019) Meaningful encryption: generating visually meaningful encrypted
images by compressive sensing and reversible color transformation. IEEE Access 7:170168–170184
28. Ping P, Xu F, Mao Y, Wang Z (2018) Designing permutation-substitution image encryption networks with
Henon map. Neuro-computing. 283:53–63
29. Ponuma R, Amutha R (2017) Compressive sensing based image compression-encryption using novel 1D-
chaotic map. Multimed Tools Appl 77:19209–19234
30. Rachlin, Y., Baron, D. The secrecy of compressed sensing measurements. in: Proceedings of the Allerton
Conference on Communi-cation, Control and Computing, 813–817 (2008).
31. Souyah A, Faraoun K (2016) Fast and efficient randomized encryption scheme for digital images based on
quadtree decomposition and reversible memory cellular automata. Nonlinear Dyn 84(2):715–732
32. Wang H, Xiao D, Li M, Xiang Y, Li X (2019) A visually secure image encryption scheme based on parallel
compressive sensing. Signal Process 155:218–232
33. Wang X, Guan N (2020) A novel chaotic image encryption algorithm based on extended zigzag confusion
and RNA operation. Opt Laser Technol 131:106366
34. Wang X, Li Z (2019) A color image encryption algorithm based on Hopfield chaotic neural network. Opt
Laser Eng 115:107–118
35. Wang X, Su Y (2020) Color image encryption based on chaotic compressed sensing and two-dimensional
fractional Fourier transform. Sci Rep 10:18556
36. Wen W, Hong Y, Fang Y, Li M, Li M (2020) A visually secure image encryption scheme based on semi-
tensor product compressed sensing. Signal Process 173:107580
37. Wen W, Wei K, Zhang Y, Fang Y, Li M (2020) Colour light field image encryption based on DNA
sequences and chaotic systems. Nonlinear Dyn. 99:1587–1600
38. Yang F, Mou J, Cao Y, Chu R (2020) An image encryption algorithm based on BP neural network and
hyperchaotic system. China Commun 17(5):21–28
39. Yang Y, Zou L, Zhou Y, Shi W (2020) Visually meaningful encryption for color images by using qi
hyperchaotic system and singular value decomposition in YCbCr color space. Optik 213:164422
40. Ye G, Pan C, Dong Y, Jiao K, Huang X (2021) A novel multi-image visually meaningful encryption
algorithm based on compressive sensing and Schur decomposition. Trans Emerg Telecommun Technol
32(2):e4071
41. Ye G, Pan C, Dong Y, Shi Y, Huang X (2020) Image encryption and hiding algorithm based on
compressive sensing and random numbers insertion. Signal Process 172:107563
42. Ye G, Pan C, Huang X, Zhao Z, He J (2018) A chaotic image encryption algorithm based on information
entropy. Int J Bifurcation Chaos 28(1):1850010
43. Zhang L, Wong K, Zhang Y, Zhou J (2016) Bi-level protected compressive sampling. IEEE T Multimedia
18(9):1720–1732
44. Zhao H, Xie S, Zhang J, Wu T (2020) Fast image encryption algorithm based on improved Henon map.
Application Research of Computers 37(12):3726–3730
45. Zhao H, Ye H, Wang R (2016) The construction of measurement matrices based on block weighing matrix
in compressed sensing. Signal Process 123:64–74
Multimedia Tools and Applications

46. Zhou K, Xu M, Luo J, Fan H, Li M (2019) Cryptanalyzing an image encryption based on a modified Henon
map using hybrid chaotic shift transform. Digit Signal Process 93:115–127
47. Zhou N, Pan S, Cheng S, Zhou Z (2016) Image compression-encryption scheme based on hyper-chaotic
system and 2D compressive sensing. Opt Laser Technol 82:121–133
48. Zhou N, Zhang A, Wu J, Pei D, Yang Y (2014) Novel hybrid image compression-encryption algorithm
based on compressive sensing. Optik 125:5075–5080
49. Zhu L, Song H, Zhang X, Yan M, Zhang L, Yan T (2019) A novel image encryption scheme based on
nonuniform sampling in block compressive sensing. IEEE Access 7:22161–22174
50. Zhu L, Song H, Zhang X, Yan M, Zhang T, Wang X, Xu J (2020) A robust meaningful image encryption
scheme based on block compressive sensing and SVD embedding. Signal Process 175:107629

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

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

Affiliations

Donghua Jiang 1 & Lidong Liu 1,2 & Liya Zhu 3 & Xingyuan Wang 4 & Yingpin Chen 5 &
Xianwei Rong 6
1
School of Computer Science and Engineering, Sun Yat-sen University, Guangzhou 510006, China
2
School of Information Engineering, Chang’an University, Xi’an, China
3
School of Electrical and Control Engineering, Chang’an University, Xi’an 710064, China
4
School of Information Science and Technology, Dalian Maritime University, Da’lian 116026, China
5
School of Physics and Information Engineering, Minnan Normal University, Zhang’zhou 363000, China
6
Physics and Electronic Engineering School, Harbin Normal University, Harbin 150025, China