Quantum Information Processing (2023) 22:180

[Link]

Adaptive LSB quantum image watermarking algorithm

based on Haar wavelet transforms

Yiming Yu1 · Jie Gao2 · Xiaoyi Mu2 · Shumei Wang1

Received: 28 August 2022 / Accepted: 30 March 2023

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

Abstract
In this paper, a novel scheme for quantum image watermarking based on novel
enhanced quantum representation of digital images (NEQR) is proposed which can
embed a 2n−1 × 2n−1 binary watermark image into a 2n × 2n grayscale carrier image.
Since only the least significant bits of the diagonal details of the carrier image are
embedded with the watermark information, the embedded image is highly consis-
tent with the carrier image after restoration. Again by reversing the embedding, the
copyright owner can simply extract the watermarked image. The simulation technique
confirms the invisibility and robustness of the proposed watermarking method. The
embedded watermarked image and the carrier image are highly relevant, with peak
signal-to-noise ratio (PSNR) above 48 dB, structural similarity index metric (SSIM)
above 0.997 and correlation coefficient (R) above 0.994. The robustness of the pro-
posal is demonstrated by checking the bit error rate (BER) count and R after it has
been attacked. Through the above embedding method, the watermarked image can
ensure its robustness and achieve better visual effects.

Keywords Quantum image processing · Image watermarking · Quantum wavelet

transform · LSB

1 Introduction

In recent years, due to the rapid development of machine vision and pattern recogni-
tion, the number of images that need computer processing has increased dramatically.
This corresponds to more complex algorithms and higher hardware configurations for
classical computers. Therefore, the advantages of quantum image processing (QIP)

B Shumei Wang
wangshumei@[Link]

1 School of Science, Qingdao University of Technology, Qingdao, Shandong Province, China

2 School of Information and Control Engineering, Qingdao University of Technology, Qingdao,
Shandong Province, China

0123456789().: V,-vol 123

180 Page 2 of 21 Y. Yu et al.

are gradually highlighted. QIP is an up-and-coming subdiscipline focused on extend-

ing traditional image processing tasks and operations into the quantum computing
framework [1, 2]. Unlike classical image processing, images are stored in qubits in a
quantum computer [3]. Using excellent parallel computing power of quantum com-
puting can greatly reduce the image storage space and the size of the circuit required
to process the image.
QIP currently has two research directions: quantum image representation and image
processing using a quantum algorithm. Many quantum image representation methods
have been developed so far [4–13]. Among them, novel enhanced quantum repre-
sentation (NEQR) [7] and flexible representation of quantum images (FRQI) [8] are
two widely used representations. Based on the above quantum image representation
methods, combined with classical image processing methods, many quantum image
processing methods have been presented. The main directions are: image translation,
image encryption, image scaling, image edge detection, image watermarking, com-
pression, etc. Compared to their classic versions, they have an exponential decrease
in complexity [14].
Steganography is a technique that allows a message to be hidden in a carrier so
that its presence cannot be detected. One can deliver the message to everyone except
the recipient without being noticed. As a result, the information is protected and the
existence of the communication itself is obscured [15]. Quantum watermarking is a
technique for embedding required message, usually the owner’s identity information,
into quantum cover data, such as that used for copyright protection [16]. In recent
years, many methods for quantum image watermark embedding have been presented.
In 2015, Yan et al. [17] proposed a double watermarking strategy for multi-channel
quantum images using a reserve of efficient transforms consisting of channels of inter-
est, channel swapping and quantum Fourier transforms. In 2016, Heidari et al. [16]
integrated the least significant bit (LSB) and quantum image scrambling into a water-
marking method. In 2017, Naseri et al. [18] proposed a new quantum watermarking
method by using LSB and the most significant bit (MSB). Heidari et al. [19] utilized
quantum discrete wavelet transforms to decompose the image and achieved watermark
embedding. Based on NEQR model, Zhou et al. [20] combined Arnold transforms and
LSB and proposed a new quantum watermarking algorithm.
In order to improve the visual effect of existing quantum watermarking schemes,
some studies have taken into account the features of human vision system. In general,
these schemes hide more data in regions of higher spatial variation, such as edges,
rather than smoother areas, because the human eye is more sensitive to pixel changes
in smooth areas than edges due to spatial occlusion [21, 22]. Luo et al. [23] proposed
a watermark scheme that only utilizes the edge pixels’ LSB as the embedding area.
Moreover, some watermark schemes [24–26] identify smooth and edge regions by a
classically inspired pixel value difference (PVD) approach [27], where more quantum
watermark bits are embedded in edge regions than in non-edge regions. In Ref. [28]
Iranmanesh experimented the LSB algorithm proposed in Ref. [24] on the IBM Q
and improved the quantum circuit. The invisibility of the watermarking algorithm
proposed above is not very ideal, however, and can still be improved.
In image processing, the discrete wavelet transform (DWT) is a common tool often
used in image compression and signal processing [29]. DWT-based image water-

123
Adaptive LSB quantum image watermarking algorithm based on... Page 3 of 21 180

marking takes advantage of the multi-resolution characteristics of images and is more

compatible with the human visual system [30]. This is because the human eye is not
as sensitive to small changes in high-frequency components as it is to changes in
low-frequency characteristics. In other words, humans are more susceptible to notic-
ing edges and textures in an image rather than the average lightness of the image. In
DWT-based image analysis, the edges and textures are included in the high-frequency
sub-band. Therefore, hiding the watermark information in the high-frequency sub-
band can greatly improve the invisibility of the watermark. On this basis, Hu et al.
[31] proposed a new watermarking scheme for FRQI representation and achieved good
results in terms of invisibility.
In this paper, a new strategy of quantum watermarking based on NEQR is proposed
to combine quantum Harr wavelet and LSB techniques in specific regions to further
reduce the invisibility of watermarked images, and design a feasible quantum circuit
based on multi-quantum bit controlled gates to implement the process of watermark
embedding and extraction. The watermarked image after watermark embedding and
the extracted watermarked image are obtained by simulation on IBM Q [32]. The
simulation analysis not only confirms that the scheme achieves a good performance
in concealment while ensuring robustness.

2 Background

2.1 NEQR

The novel enhanced quantum image representation (NEQR) is a widely used model.
A image of size 2n × 2n can be represented by 2n + q quantum bit. And q is the image
color depth to represent the image up to 2q colors. According to the NEQR, a 2n × 2n
image with grayscale range [0, 2q − 1] can be written as the form shown below.

2 −1
2n
1 
|I  = n |ci  ⊗ |i
2
i=0 (1)
q−1
|ci  = |ci · · · ci ci ,ci ∈
1 0 k
{0, 1}, k = q − 1, · · · , 1, 0
|i = |y|x = |yn−1 yn−2 · · · y0 |xn−1 xn−2 · · · x0 , xi , yi ∈ {0, 1}

where |y|x are position information. The n qubits |yn−1 , |yn−1 , . . . , |y0  store
the vertical axis information and |xn−1 , |xn−1 , . . . , |x0  store the horizontal axis
information. Therefore, the whole image can be represented by 2n + q qubits.
Figure 1 shows a 2 × 2 image and its NEQR representation, where 8 qubits are used
to represent the gray value, and corresponding position information are represented
by 2 quantum bits.

123
180 Page 4 of 21 Y. Yu et al.

Fig. 1 A 2 × 2 NEQR image and expressions

Fig. 2 LSB of a pixel

2.2 LSB substitution

LSB substitution is a common technique in image steganography, which implants the

information into the least significant bit of the image pixel and replaces the original
information. The least significant bits (LSB) are bits that contain the least amount of
information for a binary image, which means that changes in LSB are negligible for
the overall image. For example, Fig. 2 shows a LSB of one of the pixels for an 8 bit
grayscale image. We change the LSB according to the message bits. If the message
bit is “1,” then we change the LSB from “0” to “1,” and color value 200 should be
replaced by 201. This ensures that the required information is delivered with only
slight changes to the original pixel information and that only the LSB bits are read
when the information is extracted.

2.3 HWT

As a kind of DWT, the use of Haar filters on DWT is often referred to as Haar wavelet
transform (HWT). As for a 2D image of applying HWT [33]. Figure 3 shows the first
level and second level image decomposition with HWT. Sub-band containing image
approximation is called LL. Correspondingly, the sub-bands containing the image
horizontal downsampling values, vertical downsampling values and image diagonal
downsampling are called LH, HL and HH, respectively [34]. The second level HWT
is to decompose the LL sub-band again to get four second level sub-bands: LL , LH ,
HL and HH . Watermark can be inserted into one sub-band or some sub-bands.

123
Adaptive LSB quantum image watermarking algorithm based on... Page 5 of 21 180

Fig. 3 Sample image decomposition process with HWT

2.4 QHWT

In 1999, Fijany and Williams constructed quantum Haar wavelet transform circuit
using permutation operator (2n ) and Hadamard  gate (H ) [35], where the matrix
√ 1 1
form of H gate can be expressed as H = 1/ 2 . When the H gate operates
1 −1
on a single qubit |ψ = α|0 + β|1, the result is:
    
√ 1 1 α √ α+β
H |ψ = 1/ 2 = 1/ 2 (2)
1 −1 β α−β

The permutation operator 2n is made up of one or more Swap gates. When n is the
minimum of 2, 4 is the Swap gate:
⎡ ⎤
1 0 0 0
⎢0 0 1 0⎥
4 = ⎢
⎣0
⎥ (3)
1 0 0⎦
0 0 0 1

For quantum computation, 4 is the "qubit swap operator",

4 |a1 a0  = |a0 a1  (4)

The n-qubit permutation operation 2n can be written as:

2n = 4 ⊗ I2⊗n−2 I2 ⊗ 4 ⊗ I2⊗n−3

(5)
· · · I2⊗n−3 ⊗ 4 ⊗ I2 I2⊗n−2 ⊗ 4

123
180 Page 6 of 21 Y. Yu et al.

Fig. 4 Quantum circuit for 2n

Fig. 5 Quantum circuit for full levels QHWT (W2n )

The n-qubit operator 2n is composed of multiple 4 gates as shown in Fig. 4.

When 2n applied to an n-qubit quantum registers |an−1 an−2 · · · a1 a0 , it yields:

2n |an−1 an−2 · · · a1 a0  = |a0 an−1 an−2 · · · a1  (6)

Based on the two operator above, the circuit for n-qubit QHWT is implemented by a
series of matrix multiplication operations, direct sum and tensor product as illustrated
in Fig. 5.
Therefore, the matrix form of the QHWT in Fig. 5 is:
  
W2n =2n 2n−1 ⊕ I2n −2n−1 2n−2 ⊕ I2n −2n−2 · · · 2n−i ⊕ I2n −2n−1
· · · (8 ⊕ I2n −8 ) (4 ⊕ I2n −4 ) (H ⊕ I2n −2 ) (I2 ⊗ H ⊕ I2n −4 )
  (7)
· · · I2n−i ⊗ H ⊕ I2n −2n−i+1 · · · I2n−3 ⊗ H ⊕ I2n −2n−2
 
I2n−2 ⊗ H ⊕ I2n −2n−1 I2n−1 ⊗ H

3 Quantum image watermarking algorithm

In this section, a quantum watermarking image embedding algorithm is proposed to

embed a 2n−1 × 2n−1 binary watermark image into a 2n × 2n NEQR carrier image,
the watermark embedding and extraction algorithms and the corresponding quantum
lines are described below.

123
Adaptive LSB quantum image watermarking algorithm based on... Page 7 of 21 180

Fig. 6 Quantum watermark image embedding and extracting procedures

3.1 Quantum watermark image embedding

Based on QHWT image decomposition and LSB substitution, the procedures for quan-
tum image embedding are illustrated in Fig. 6.
The 2n−1 × 2n−1 binary watermark image W I in quantum image representation
has the following form:

1 −1
22n−2
|W I  = |wi  ⊗ |i, wi ∈ {0, 1}
2n−1 (8)
i=0
|i = |y|x = |yn−2 yn−3 · · · y0 |xn−2 xn−3 · · · x0 , xi , yi ∈ {0, 1}

The steps of the quantum watermark image embedding algorithm are as follows:
Step 1: At first, transform classic images into NEQR quantum image represent
model.

2 −1 2n
1 
|C = n |ci  ⊗ |i
2
i=0 (9)
q−1
|ci  = {0, 1}, k = q − 1, · · · , 1, 0
|ci · · · ci ci ,ci ∈
1 0 k

|i = |y|x = |yn−1 yn−2 · · · y0 |xn−1 xn−2 · · · x0 , xi , yi ∈ {0, 1}

123
180 Page 8 of 21 Y. Yu et al.

Fig. 7 Circuit of full levels QHWT transformation for NEQR image

After obtaining the NEQR model, we need to decompose image based on QHWT.
Figure 7 shows the circuit for full levels QHWT for rows and columns, respectively,
to decompose the NEQR image |I .
Row decomposition is realized by applying W2n operator to the NEQR image’s coor-
dinate qubits |x = |xn−1 xn−2 · · · x1 x0  in the horizontal direction. The same goes for
column. The matrix form W2D of full levels QHWT for NEQR image decomposition
can be given by:

⊗q ⊗q
W2D = Wcolumn Wrow = I2 ⊗ W2n ⊗ I2⊗n I2 ⊗ I2⊗n ⊗ W2n (10)

For NEQR model, watermark embedding can be directly completed in the first level
QHWT image decomposition. The circuit of first level QHWT is shown in Fig. 8.
Fig. 9 is an example of a 2×2 NEQR image decomposition using first level QHWT.
As shown in Fig. 1, the 2 × 2 NEQR image is expressed as follows:

1
|I  = [|00000000 ⊗ |00 + |01100110 ⊗ |01
2 (11)
+ |10011001 ⊗ |10 + |11001100 ⊗ |11]

The matrix of first level QHWT for a 2×2 NEQR can be expressed as a 1024×1024
matrices based on Eqs. 7 and 10:

123
Adaptive LSB quantum image watermarking algorithm based on... Page 9 of 21 180

Fig. 8 Circuit of first level QHWT transformation for NEQR image

Fig. 9 Quantum circuit for a 2 × 2 NEQR image decomposition

W2×2 = I2⊗8 ⊗ H ⊗ I2 I2⊗8 ⊗ I2 ⊗ H

⎡ ⎤
1/2 1/2 1/2 1/2 0 0 0 0
⎢ 1/2 −1/2 1/2 −1/2 . . . 0 0 0 0 ⎥
⎢ ⎥
⎢ 1/2 1/2 −1/2 −1/2 0 0 0 0 ⎥
⎢ ⎥
⎢ 1/2 −1/2 −1/2 1/2 0 0 0 0 ⎥
⎢ ⎥ (12)
⎢ .. .. .. ⎥
=⎢ . . . ⎥
⎢ ⎥
⎢ 0 0 0 0 1/2 1/2 1/2 1/2 ⎥
⎢ ⎥
⎢ 0 0 0 0 . . . 1/2 −1/2 1/2 −1/2 ⎥
⎢ ⎥
⎣ 0 0 0 0 1/2 1/2 −1/2 −1/2 ⎦
0 0 0 0 1/2 −1/2 −1/2 1/2

123
180 Page 10 of 21 Y. Yu et al.

Then decompose the image using W2×2 :

W2×2 (|I )
1
= [|00000000 ⊗ (|00 + |01 + |10 + |11)
4
+ |01100110 ⊗ (|00 − |01 + |10 − |11)
+ |10011001 ⊗ (|00 + |01 − |10 − |11)
+ |11001100 ⊗ (|00 − |01 − |10 + |11]

1 1
= (|00000000 + |01100110 + |10011001 + |11001100) ⊗ |00 (13)
2 2
1
+ (|00000000 − |01100110 + |10011001 − |11001100) ⊗ |01
2
1
+ (|00000000 + |01100110 − |10011001 − |11001100) ⊗ |10
2 
1
+ (|00000000 − |01100110 − |10011001 + |11001100) ⊗ |11
2

In decomposition image, the LL is given by the color information 41 (|00000000 +

|01100110 + |10011001 + |11001100) in position |00, and the color information
in position |01, |10 and |11 represents the HL, LH and HH, respectively.
Step 2: Using first level QHWT to decompose the quantum carrier image |C,
shown in Fig. 8.

⊗q ⊗q+n−1
|C D  = I2 ⊗ 2n ⊗ I2⊗n I2 ⊗ H ⊗ I2⊗n
(14)
⊗q+n ⊗q+2n−1
× I2 ⊗ 2 n I2 ⊗ H |C

where the middle quantum state |C D  is divided into four parts: LL, HL, LH and
HH as shown in Fig. 3.
Step 3: Based on the color information qubit of the watermark image and the
corresponding positions of the watermark image and the carrier image, we design the
unitary transforms U EM to embed watermark information in LSB of carrier image HH
sub-band. U EM is constructed from 2n−1 × 2n−1 multiple-qubit control gates UiC , and
each UiC implements the embedding of one bit of watermark information. Apply U EM
to decomposed image |C D  as follows:

 
UiC = X ⊗ I⊗q−1 ⊗ |1|Yn−1
i
|1|X n−1
i
|1|Y in−1 |X in−1 
 
X in−1 |Y in−1 |1|X n−1
i
|1|Yn−1
i
|1| (15)
⎛ 2n−2 ⎞
2 −1
|C E |W I  = U E M |C D |W I  = ⎝ UiC ⎠ |C D |W I  (16)
i=0

123
Adaptive LSB quantum image watermarking algorithm based on... Page 11 of 21 180

Fig. 10 Circuit for watermark embedding process

where |y = |yn−1 yn−2 · · · y1 y0  and |x = |xn−1 xn−2 · · · x1 x0  have the highest
qubits |yn−1  = |1 and |xn−1  = |1 to restrict watermark information embed in the
area of HH sub-band. In other words, when the information needs to be embedded is
bit 0, the LSB of corresponding position in HH sub-band is unchanged. On the other
hand, the LSB will flip via multi-qubit controlled NOT.
Step 4: The watermarked image |C W  is reconstructed by first level QHWT of the
quantum state |C E .
 †
⊗q+n ⊗q+2n−1
|C W  = I2 ⊗ 2 n I 2 ⊗H
 † (17)
⊗q ⊗q+n−1
I2 ⊗ 2n ⊗ I2⊗n I2 ⊗ H ⊗ I2⊗n |C E 

where † denotes the Hermite conjugate matrix.

Step 5: Finally, utilizing quantum measurement for |C W  to obtain the classical
embedded image C W .
In Fig. 10, the integrated watermark image quantum circuit is demonstrated.

3.2 Quantum watermark image extraction

In the process of watermark extraction, both the original image |C and embedded
image |C W  are required. Therefore, it is a non-blind watermarking algorithm. The
detailed procedures is shown in Fig. 6.

123
180 Page 12 of 21 Y. Yu et al.

Step 1: At first, utilizing first level QHWT to decompose embedded image |C W 

and carrier image |C:

⊗q ⊗q
|C W D |C D  = I2 ⊗ 2n ⊗ I2⊗n ⊗ I2 ⊗ 2n ⊗ I2⊗n
⊗q+n−1 ⊗q+n−1
I2 ⊗ H ⊗ I2⊗n ⊗ I2 ⊗ H ⊗ I2⊗n
(18)
⊗q+n ⊗q+n
I2 ⊗ 2 n ⊗ I 2 ⊗ 2 n
⊗q+2n−1 ⊗q+2n−1
I2 ⊗ H ⊗ I2 ⊗ H |C W |C

Step 2: A set of multiple-qubit controlled unitary operators U Cj is constructed and

applied to the middle quantum state |C W D |C D .

 
U Cj = X ⊗ I⊗q−1 ⊗ |1|Yn−1 |1|X n−1 |1|Y n−1 |X n−1 
j j j j

j  j  j j
X n−1 |Y n−1 |1|X n−1 |1|Yn−1 |1| (19)
⎛ 2n−2 ⎞
2 −1
|C W I |C D  = U E X |C W D |C D  = ⎝ U Cj ⎠ |C W D |C D  (20)
j=0

Step 3: Measure |c0yx , |yn−2 yn−3 · · · y1 y0  and |xn−2 xn−3 · · · x1 x0  in |C W I . In

step 2, the color information of the original image is separated into embedded image
through U E X , and then, the |C W I  containing the watermark image information can
be obtained. The watermark image information can be restored by measuring corre-
sponding circuits.
The effective circuit for watermark image extraction is shown in Fig. 11.

4 Simulation experiments and analysis

This section presents the simulation analysis results of the proposed algorithm in
this paper. Limited by the availability of present quantum computers, we simulated
watermarking schemes on classical computers using Qiskit [36], which can execute
quantum circuits using quantum simulators as well as quantum hardware supplied by
IBM Q Experience [32]. In this work, a group of four classical 128 × 128 grayscale
images were chosen for the carrier images and three binary 64 × 64 watermark images
are used in the simulations given in Fig. 12. In order to recover to the classical image,
300,000 measurements per quantum image are required [37]. The proposed visual-
ization of the watermarking process is shown in Fig. 13. The circuit diagram of the
algorithm is shown in Fig. 14 based on qiskit.
There are two critical characteristics to analyze for image watermarking, i.e.,
stealthiness and robustness.

123
Adaptive LSB quantum image watermarking algorithm based on... Page 13 of 21 180

Fig. 11 Circuit for watermark extracting process

Fig. 12 Original images and watermark images

123
180 Page 14 of 21 Y. Yu et al.

Fig. 13 Original images and embedded images. a–d are original images, e–h are embedded images with
watermark 1, i–l are embedded images with watermark 2, m–p are embedded images with watermark 3,
and q–s are watermark images

Fig. 14 Circuit diagram of the algorithm built with qiskit

4.1 Invisibility

The invisibility is evaluated by measuring the similarity between the carrier image and
the embedded image. As described below, several methods will be used to evaluate the
invisibility of the proposal: calculation of peak signal-to-noise ratio (PSNR), structural
similarity index metric (SSIM), universal image quality (UQI), correlation coefficient
(R) and histogram analysis.
PSNR is employed to gauge the distortion between the embedded image and the
carrier image. For PSNR, the mean square error (MSE) of the image is its main
component. The higher the value of PSNR, the lower the distortion and the better the

123
Adaptive LSB quantum image watermarking algorithm based on... Page 15 of 21 180

watermark concealment. The range of PSNR values is [0,100], when the PSNR value
is greater than 30, and the human visual system cannot perceive the difference between
the watermarked image and the original image. The formula of PSNR for two 2n × 2n
of grayscale images is calculated as follows.
n −1 2n −1
2 
MSE = 1/22n [C(i, j) − W (i, j)]2 (21)
i=0 j=0
 
MAXC
PSNR = 20 log10 √ (22)
MSE

where C indicates the original image and W is watermarked image. MAXC represents
the maximal value of pixels of the image C, and MSE is the average square error.
The SSIM method presents an index of quality measurement based on the calcu-
lation of three main dimensions of brightness, contrast and structure or related terms.
The UQI estimates the loss of correlation, brightness distortion and contrast distortion.
SSIM measures the degree of distortion of a picture and also the degree of similarity
between two pictures. Unlike MSE and PSNR, which measure absolute error, SSIM
and UQI are perceptual models, i.e., it is more in line with the intuition of the human
eye.
 
2μx μ y + c1 2σx y + c2
SSIM(x, y) = (23)
μ2x + μ2y + c1 σx2 + σ y2 + c2

4μx μ y 2σx y
UQI(x, y) = (24)
μ2x + μ2y σx2 + σ y2

where μx and μ y are the local means, σx and σ y are the standard deviations and σx y
is the cross-covariance for images x and y sequentially.
The correlation coefficient (R) is used in image processing to measure the correlation
of two images. It is described in Eq. 25.
  
C − C̄ − W − W̄
R= (25)
 2   2
C − C̄ W − W̄

C̄ is mean of pixels in cover image, and W̄ is also mean of pixels in stego images.
The results in simulation are shown in Table 1.
Based on Table 1, the proposed scheme has achieved an acceptable effect. PSNR
values average at 52 dB. Higher PSNR value indicates the higher similarity of embed-
ded image to original image, which means it is difficult for naked eye to distinguish
the difference between pre-embedded image and post-embedded image. For SSIM,
UQI and R, the proposed watermarking algorithm also performs well.
The histogram is an essential criterion for evaluating the quality of embedded image
fidelity compared with original image and it clearly shows the change in the frequency

123
180 Page 16 of 21 Y. Yu et al.

Table 1 Performance results for the proposed algorithm

Carrier Watermark PSNR SSIM UQI R

Cat Watermark 1 57.61174509 0.99977365 0.99977096 0.99970302

Watermark 2 52.48820979 0.99926328 0.99925451 0.99945057
Watermark 3 53.40962339 0.99940411 0.99939702 0.99851892
Lotus Watermark 1 50.78502018 0.99856148 0.99853807 0.99743458
Watermark 2 48.02897982 0.99728717 0.99724304 0.99624313
Watermark 3 49.52929835 0.99807995 0.99804872 0.99645382
Sea Watermark 1 58.55102578 0.99952683 0.99950872 0.99848922
Watermark 2 50.65921549 0.99708728 0.99697580 0.99419032
Watermark 3 52.03040046 0.99787608 0.99779480 0.99432472
Sunflower Watermark 1 52.08595115 0.99954114 0.99953811 0.99797273
Watermark 2 50.87158166 0.99939320 0.99938912 0.99840384
Watermark 3 53.01817304 0.99962983 0.99962734 0.99955281

of image pixels. The histogram of four carrier images and three watermark images
embedded in each is shown in Fig. 15. Above each histogram is the corresponding
original image, and on the left is the corresponding [Link], the his-
tograms of the embedded images are in high agreement with the histograms of the
original images they correspond to. Therefore, the destructive of watermark image is
negligible for original image.

4.2 Robustness

Robustness refers to the ability of message to survive in attacks [38]. In this paper, we
utilize bit error rate (BER) and correlation coefficient (R) as two criteria in robustness
analysis.

4.2.1 Bit error rate

This BER is defined as the inverse of PSNR:

1
BER = (26)
PSNR
The BER reflects the bits of the original image that were altered in the watermarking
process. Table 2 shows the results of the BER values calculated for the embedded
images in the simulation. From the table, we can see that the values of all images are
less than 0.021. This shows that the introduced algorithm has little influence on the
original image while embedding the information of the watermark image.

123
Adaptive LSB quantum image watermarking algorithm based on... Page 17 of 21 180

Fig. 15 Histogram of original images and embedded images. a–d are histogram of original images, e–h are
histogram of embedded images with watermark 1, i–l are histogram of embedded images with watermark
2, and m–p are histogram of embedded images with watermark 3

Table 2 Calculated BER for

Carrier Watermark 1 Watermark 2 Watermark 3
different images
Cat 0.01735757 0.01905189 0.01872321
Lotus 0.01969084 0.02082076 0.02019007
Sea 0.01707912 0.01973974 0.01921953
Sunflower 0.01919903 0.01965732 0.01886145

4.2.2 Correlation coefficient

The correlation coefficient (R) determines the correlation rate between the two images
as shown in Eq.30. To analyze the algorithm’s resistance against attacks, we con-
sider four different attacks in our simulations, namely brightness change, contrast
change, image compression and addition of noise, that are exercised on the water-
marked image. After the attack, the watermarked images were extracted and the R
was computed, which indicates the correlation between the original and extracted
watermarked images.
Given the results in Fig. 16, we can judge that the proposed watermarking scheme
has a good resistance to the attacks under consideration. A comparison of the proposed
watermarking algorithm in Ref. [19] shows that correlation is improved in the face of

123
180 Page 18 of 21 Y. Yu et al.

Fig. 16 a Visual display of extracted watermark after attacks. b Correlation rate of extracted watermarks
after attacks

all other attacks except for noise interference. The image of the watermark extracted
after the attack is shown in Fig. 16a. As can be seen in the figure, the watermarked
image is recognizable and the proposed algorithm shows good attack resistance.

4.3 Comparison of proposed algorithms

To confirm the effectiveness of the proposed quantum watermarking algorithm, the

comparison of performance with other simulation-based quantum watermarking meth-
ods is shown in Table 3. The watermarking algorithms proposed in this paper have
good comprehensive capabilities.
Table 3 shows that the HWT-based watermarking method has better performance
in invisibility and better quality of the watermarked embedded image compared to the
existing image watermarking methods. This is due to the fact that watermarking based
on HWT is more consistent with the human eye visual system for image processing,
and can embed the watermark information into the edges of the original image more
stealthily. After the image is processed by HWT, the watermark information is embed-
ded into the image by LSB method to further ensure that the image is not corrupted
and has good anti-attack capability.

123
Adaptive LSB quantum image watermarking algorithm based on... Page 19 of 21 180

Table 3 Comparison of proposed algorithms

Studies Heidari et al. [19] Zeng et al. [26] Hu et al. [31] Proposed

Representation NEQR NEQR FRQI NEQR

Method DWT MPD HWT HWT
PSNR 51.1 38.2−41.8 51-69 48-58
SSIM 0.973−0.999 0.999-1 0.997−0.999
UIQ 0.87-1 0.996−0.999
R(after attacks) 0.94 0.95
Image size 256 × 256 64 × 64 128 × 128
Watermark size 128 × 128 32 × 32 64 × 64

5 Conclusion

In this work, we explore the integration of LSB and quantum Haar wavelet transform
into a watermarking method for quantum images based on NEQR representation. In
the proposed scheme, a quantum wavelet transform is used to embed a binary image
of size 2n−1 × 2n−1 into a grayscale image of size 2n × 2n . Then, the multi-qubit
controlled gate is used to implement the LSB technique to embed the watermarked
image into the carrier image. It is proved that the watermarked image can be simply
extracted by the copyright owner by reversing the embedding.
Simulation results confirm the effectiveness of the method and demonstrate many
advantages. Any applicable watermarking protocol must satisfy two key properties:
invisibility and robustness. The watermarked embedded image and the carrier image
are highly correlated with PSNR higher than 48 db, SSIM more than 0.997 and R higher
than 0.994. In addition, the robustness of the scheme is demonstrated by checking the
number of BER and R.
In conclusion, the advantage of the proposed scheme over previous watermarking
schemes is that by introducing the quantum Haar wavelet transform, the watermark
information is embedded in the diagonal details, which enhances the invisibility of the
watermark information and ensures robustness at the same time. However, as a non-
blind watermarking algorithm, it is still necessary to use the carrier image to extract
the watermark information, which is the direction of future algorithm improvement.
In terms of security, proposed algorithm does not design the encryption link, if the
need to ensure the security of the watermarked image, the watermarked image can
be encrypted in advance, which is also the part of the algorithm can be improved.
Currently and in the future, we will vigorously promote the combination of quantum
row and classical algorithms, and further apply them to watermarking algorithms for
image information in medical and military directions.
Funding This work is supported by the National Natural Science Foundation of China (Grant No.
61772295), Natural Science Foundation of Shandong Province, China (Grant Nos. ZR2021MF049,
ZR2019YQ01) and Project of Shandong Provincial Natural Science Foundation Joint Fund Application
(ZR202108020011).

123
 180 Page 20 of 21 Y. Yu et al.

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

Declarations
Conflict of interest All authors declare no competing interests.

References
1. Iliyasu, A.M.: Towards realising secure and efficient image and video processing applications on
quantum computers[J]. Entropy 15(8), 2874–2974 (2013)
2. Yan, F., Iliyasu, A.M., Le, P.Q.: Quantum image processing: a review of advances in its security
technologies[J]. Int. J. Quant. Inf. 15(03), 1730001 (2017)
3. Stajic, J.: The future of quantum information processing[J]. Science 339(6124), 1163–1163 (2013)
4. Venegas-Andraca, S.E., Bose, S.: Storing, processing, and retrieving an image using quantum mechan-
ics[C]. In: Quantum information and computation, pp. 137–147. SPIE (2003)
5. Latorre J.I.: Image compression and entanglement[J]. arXiv preprint arXiv: quant-ph/0510031, (2005)
6. Venegas-Andraca, S.E., Ball, J.L.: Processing images in entangled quantum systems[J]. Quant. Inf.
Process. 9(1), 1–11 (2010)
7. Zhang, Y., Lu, K., Gao, Y., et al.: NEQR: a novel enhanced quantum representation of digital images[J].
Quant. Inf. Process. 12(8), 2833–2860 (2013)
8. Le, P.Q., Dong, F., Hirota, K.: A flexible representation of quantum images for polynomial preparation,
image compression, and processing operations[J]. Quant. Inf. Process. 10(1), 63–84 (2011)
9. Li, H.S., Zhu, Q., Li, M.C., et al.: Multidimensional color image storage, retrieval, and compression
based on quantum amplitudes and phases[J]. Inf. Sci. 273, 212–232 (2014)
10. Şahin, E., Yilmaz, I.: QRMW: quantum representation of multi wavelength images[J]. Turk. J. Electr.
Eng. Comput. Sci. 26(2), 768–779 (2018)
11. Wang, L., Ran, Q., Ma, J., et al.: QRCI: a new quantum representation model of color digital images[J].
Opt. Commun. 438, 147–158 (2019)
12. Wang, B., Hao, M., Li, P., et al.: Quantum representation of indexed images and its applications[J].
Int. J. Theor. Phys. 59(2), 374–402 (2020)
13. Li, H.S., Fan, P., Xia, H.Y., et al.: Quantum implementation circuits of quantum signal representation
and type conversion[J]. IEEE Trans. Circuits Syst. I Regular Pap. 66(1), 341–354 (2018)
14. Wang, Z., Xu, M., Zhang, Y.: Review of quantum image processing[J]. Arch. Comput. Methods Eng.
29(2), 737–761 (2022)
15. Laurel, C.O., Dong, S.H., Cruz-Irisson, M.: Steganography on quantum pixel images using Shannon
entropy[J]. Int. J. Quant. Inf. 14(05), 1650021 (2016)
16. Heidari, S., Naseri, M.: A novel LSB based quantum watermarking[J]. Int. J. Theor. Phys. 55(10),
4205–4218 (2016)
17. Yan, F., Iliyasu, A.M., Sun, B., et al.: A duple watermarking strategy for multi-channel quantum
images[J]. Quant. Inf. Process. 14(5), 1675–1692 (2015)
18. Naseri, M., Heidari, S., Baghfalaki, M., et al.: A new secure quantum watermarking scheme[J]. Optik
139, 77–86 (2017)
19. Heidari, S., Naseri, M., Gheibi, R., et al.: A new quantum watermarking based on quantum wavelet
transforms[J]. Commun. Theor. Phys. 67(6), 732 (2017)
20. Zhou, R.G., Hu, W., Fan, P.: Quantum watermarking scheme through Arnold scrambling and LSB
steganography[J]. Quant. Inf. Process. 16(9), 1–21 (2017)
21. Atta, R., Ghanbari, M.: A high payload steganography mechanism based on wavelet packet transfor-
mation and neutrosophic set[J]. J. Vis. Commun. Image Represent. 53, 42–54 (2018)
22. Atta, R., Ghanbari, M., Elnahry, I.: Advanced image steganography based on exploiting modification
direction and neutrosophic set[J]. Multim. Tools Appl. 80(14), 21751–21769 (2021)
23. Luo, G., Zhou, R.G., Hu, W.W., et al.: Enhanced least significant qubit watermarking scheme for
quantum images[J]. Quant. Inf. Process. 17(11), 1–19 (2018)
24. Luo, G., Zhou, R.G., Luo, J., et al.: Adaptive LSB quantum watermarking method using tri-way pixel
value differencing[J]. Quant. Inf. Process. 18(2), 1–20 (2019)

123
Adaptive LSB quantum image watermarking algorithm based on... Page 21 of 21 180

25. Luo, J., Zhou, R.G., Luo, G.F., et al.: Traceable quantum steganography scheme based on pixel value
differencing[J]. Sci. Rep. 9(1), 1–12 (2019)
26. Zeng, Q.W., Wen, Z.Y., Fu, J.F., et al.: Quantum watermark algorithm based on maximum pixel
difference and tent map[J]. Int. J. Theor. Phys. 60(9), 3306–3333 (2021)
27. Wu, D.C., Tsai, W.H.: A steganographic method for images by pixel-value differencing[J]. Pattern
Recognit. Lett. 24(9–10), 1613–1626 (2003)
28. Iranmanesh, S., Atta, R., Ghanbari, M.: Implementation of a quantum image watermarking scheme
using NEQR on IBM quantum experience[J]. Quant. Inf. Process. 21(6), 1–40 (2022)
29. Taubman, D.S., Marcellin, M.W., Rabbani, M.: JPEG2000: image compression fundamentals, stan-
dards and practice. J. Electron. Imag. 11, 286–287 (2002)
30. Nanmaran, R., Nagarajan, S., Sindhuja, R., et al.: Wavelet transform based multiple image watermark-
ing technique[C]. In: IOP conference series: materials science and engineering. IOP Publishing p.
012167 (2020)
31. Hu, W.W., Zhou, R.G., El-Rafei, A., et al.: Quantum image watermarking algorithm based on haar
wavelet transform[J]. IEEE Access 7, 121303–121320 (2019)
32. IBM Q Experience. [Link]
33. Al-Haj, A.: Combined DWT-DCT digital image watermarking[J]. J. Comput. Sci. 3(9), 740–746 (2007)
34. Sudibyo, U., Eranisa, F., Rachmawanto, E.H., et al. A secure image watermarking using Chinese
remainder theorem based on haar wavelet transform[C]. In: 2017 4th International Conference on Infor-
mation Technology, Computer, and Electrical Engineering (ICITACEE). IEEE, pp. 208–212 (2017)
35. Fijany, A., Williams, C.P.: Quantum wavelet transforms: Fast algorithms and complete circuits[C].
In: NASA international conference on quantum computing and quantum communications, pp. 10–33.
Springer, Berlin, Heidelberg (1999)
36. Abraham, H., et al.: Qiskit: an open-source framework for quantum computing. [Link]
Qiskit/qiskit
37. Yan, F., Le, P.Q., Iliyasu, A.M.: Assessing the similarity of quantum images based on probability
measurements[C]. In: IEEE Congress on Evolutionary Computation. IEEE pp. 1–6 (2012)
38. Jiang, N., Zhao, N., Wang, L.: LSB based quantum image steganography algorithm[J]. International
Journal of Theoretical Physics 55(1), 107–123 (2016)

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.

123