Archives of Computational Methods in Engineering (2022) 29:737–761

https://doi.org/10.1007/s11831-021-09599-2

REVIEW ARTICLE

Review of Quantum Image Processing

Zhaobin Wang1 · Minzhe Xu1 · Yaonan Zhang2

Received: 24 November 2020 / Accepted: 27 April 2021 / Published online: 14 May 2021
© CIMNE, Barcelona, Spain 2021

Abstract
As an interdisciplinary between quantum computing and image processing, quantum image processing provides more pos-
sibilities for image processing due to the powerful parallel computing capabilities of quantum computers. In recent years,
quantum image processing attracts more and more researcher’s attention. In order to allow researchers to better understand
quantum image processing technology, we have reviewed relevant literature in recent years in the paper. First, the background
and mathematical concepts of quantum computing are introduced. Then, the research progress of quantum image processing
is sorted out and summarized in the fileds of quantum image representation, geometric transformation, image encryption,
edge detection, image segmentation, filtering and compression. Finally, we have discussed the advantages and disadvantages
of quantum image processing, and pointed out the potential future research.

1 Introduction more attention, and appeared in various fields of computer

science.
In the past few decades, the computing power of electronic In order to solve the problem of image processing on
computers has increased exponentially in accordance with quantum computers, classical image processing algorithms
Moore’s Law [1]. However, due to the limitations of many are accelerated by quantum computing, and quantum image
objective factors, the computing power of single-core CPUs processing algorithms are gradually emerging.
has not increased significantly in recent years. Therefore, Digital image processing is an important part of infor-
it is necessary to explore other ways to increase comput- mation science and the core program of many applications.
ing power. In 1982, Feynman proposed a new computing Image processing algorithms have the characteristics of
model, named quantum computer, which is based on quan- many parallel operations. In classical algorithms, a large
tum mechanics and can use the superposition and entangle- number of operations are required. In recent years, with the
ment characteristics of quantum mechanics to store, process rapid development of pattern recognition, image understand-
and transmit information, and its computing power is much ing, and image sensors, the number of images has become
higher than classical computer [2]. In 1994, Shor proposed more and more, and the size of images has increased, the
a quantum algorithm for prime factor decomposition [3], algorithms become more and more complex. It makes clas-
and in 1996, after Grover proposed a quantum search algo- sical image processing algorithms often require a lot of
rithm [4], the computing power of quantum computers was time, and the hardware requirements of image processing
proved, and quantum computing began to be paid more and algorithms are gradually increasing. Therefore, it is neces-
sary to find high-performance methods to store and process
these images. Unlike classical image processing, quantum
computers store images in qubits. Utilizing the superposition
* Zhaobin Wang state and entangled state of quantum, quantum computers
[email protected] have good parallel processing capabilities. The complex-
Yaonan Zhang ity of storing n-bit long sequences on a classical computer
[email protected] requires O(n × 2n ), but only O(n) on a quantum computer.
1
School of Infomation Science and Engineering, Lanzhou
When performing operations on sequences, such as bitwise
University, Lanzhou 730000, China inversion, classical computers need to calculate sequentially,
2
National Cryosphere Desert Data Center, Northwest Institute
with a complexity of O(2n ), or a large number of repeated
of Eco‑Environment and Resources, Chinese Academy arithmetic circuits, and quantum computers can operate
of Sciences, Lanzhou 730000, China

13
Vol.:(0123456789)
738 Z. Wang et al.

on each bit at the same time. Only O(1) is needed, which with their classical versions, they have an exponential reduc-
is helping to speed up image processing algorithms with tion in complexity. Many algorithms have been proposed for
many parallel calculations. The space required to store the image translation, image scaling, morphological transforma-
image and the circuit scale required to process the image are tion, edge detection, image encryption, watermarking, filter-
greatly reduced, so the efficiency of the algorithm is greatly ing, compression, and more complex image matching and
improved. quantum machine learning.
Quantum image processing roughly requires three parts. Quantum computing can bring huge space savings and
The digital image is represented on a quantum computer, speed improvements to digital image processing. Quantum
and then the image is processed on the quantum computer image processing is an emerging field that still needs to be
using a quantum algorithm, and finally the processed quan- developed and supplemented. With the gradual advancement
tum image is converted into a classical image. To process of quantum computers, the need for quantum image process-
images on a quantum computer, the images need to be stored ing is becoming more and more urgent. With the efforts of
in a quantum system. In order to solve such problems, many researchers, it is believed that more representation methods
quantum image representation methods have been proposed. and processing algorithms will be proposed.
First of all, in 2003, the Qubit Lattice representation method The remaining part of the paper proceeds as follows. A
was proposed [5], which is similar to the classical method brief introduction to basic concepts is provided in Sect. 2.
that treats the image as a matrix, and each qubit stores only And the survey will review the methods related to quantum
one pixel. It is the first attempt at quantum image representa- image representation in Sect. 3; geometric transformation
tion, which is closer to the classical way, but it does not use and morphological transformation in Sect. 4; quantum image
quantum superposition and is more complex. In 2005, Real protection in Sect. 5; edge detection and image segmentation
Ket’s representation method was proposed [6], which used in Sect. 6; similarity analyses and image matching in Sect. 7;
quantum superposition state for the first time. quantum image filtering and compression in Sect. 8; quan-
The entangled image model proposed in 2010 uses the tum machine learning in Sect. 9. And other algorithms will
entangled state [7]. In the same year, a flexible represen- be reviewed in Sect. 10. An analysis of possible future work
tation of quantum images (FRQI) was proposed [8], which is presented in Sect. 11. Finally, we conclude in Sect. 12.
represented grayscale information as an angle, stored in the
probability amplitude of a qubit, used n qubits to represent
the pixel number of a square image, and represented the gray-
scale qubits are entangled with them so that the coordinates 2 Basic Concepts
correspond to the grayscale. The multi-channel representa-
tion for quantum images (MCRQI), which extends it to RGB, 2.1 Vector
was proposed the following year [9]. This method makes full
use of quantum superposition and entanglement, and can In quantum computing, we pay attention to the state of quan-
perform the same operation on all pixels at the same time, tum. A quantum state can be represented by a vector in Hil-
thereby solving the real-time calculation problem in image bert space. The standard quantum mechanical notation for a
processing applications. Since then, many image processing vector in Hilbert space is as follows:
algorithms have also been proposed based on it. However,
⎡𝛼1 ⎤
FRQI only uses one qubit to store the grayscale information ⎢𝛼 ⎥
of each pixel in the image, so some complex color operations �𝜓⟩ = ⎢ 2 ⎥ (1)
are not easy to perform on the basis of FRQI. The novel ⎢⋮⎥
⎣𝛼n ⎦
enhanced quantum representation (NEQR) [10], which was
proposed in 2013, uses eight qubits to represent grayscale where the symbol �⋅⟩ is called Dirac symbol, which is a
information. It stores the grayscale information in an eight- standard symbol of state in quantum mechanics, used to
bit binary format in the qubit sequence. Although the use indicate that an object is a vector. 𝜓 is the label of the vec-
of qubits is increased, it solves the disadvantage that FRQI tor. �𝜓⟩ is called ket, and the corresponding bra is ⟨𝜓�, which
grayscale information cannot be accurately measured within is the conjugate transpose of the ket.
a limited number of times, and facilitates color operations.
It has also become a more widely used image representa-
tion model in quantum image processing. Since then, more 2.2 Tensor Product
improvements have also been proposed. On the basis of these
image representation methods, many quantum image pro- The tensor product is a way to put vector spaces together to
cessing algorithms have been proposed. Quantum versions form a larger vector space. It satisfies the following basic
of some classical image algorithms are proposed. Compared properties:

13
Review of Quantum Image Processing 739

� �
(1) For any scalar z, any element �v⟩ in V space, any element 𝛼
�w⟩ in W space �𝜓⟩ =𝛼�0⟩ + 𝛽�1⟩ = (9)
𝛽
z(�v⟩ ⊗ �v⟩) = (z�v⟩) ⊗ �w⟩ = �v⟩ ⊗ (z�w⟩) (2) Suppose there are two qubits. A two-qubit system has four
(2) For any �v1 ⟩ and �v2 ⟩ in V space, any �w⟩ in W space calculation ground states, which are �00⟩, �01⟩, �10⟩, and �11⟩.
A pair of qubits can exist in the superposition of these four
(�v1 ⟩ + �v2 ⟩) ⊗ �w⟩ = �v1 ⟩ ⊗ �w⟩ + �v2 ⟩ ⊗ �w⟩ (3) states, so the state vector describing the two qubits is
(3) For any �v⟩ in V space, any �w1 ⟩ and �w2 ⟩ in W space �𝜓⟩ = 𝛼00 �00⟩ + 𝛼01 �01⟩𝛼10 �10⟩𝛼11 �11⟩ (10)
�v⟩ ⊗ (�w1 ⟩ + �w2 ⟩) = �v⟩ ⊗ �w1 ⟩ + �v⟩ ⊗ �w2 ⟩) (4) where 𝛼x is the probability amplitude corresponding to state
�x⟩ and satisfies:
Assuming that A is a m × n matrix and B is a p × q matrix,
then A ⊗ B is shown in Eq.5 |𝛼00 |2 + |𝛼01 |2 + |𝛼10 |2 + |𝛼11 |2 = 1 (11)
⎡ A11 B A12 B … A1n B ⎤ Similar to the case for a single qubit, for a two qubit system:
⎢ A B A22 B … A2n B ⎥ � �
A ⊗ B ≡ ⎢ 21
⋮ ⋱ ⋮ ⎥⎥ (5)
⎢ ⋮ � � � � ⎡1 × 1 ⎤ ⎡1⎤
⎣Am1 B Am2 B … Amn B⎦ 1 1 ⎢ ⎥ ⎢ ⎥
�00⟩ =�0⟩ ⊗ �0⟩ = ⊗ =⎢ �0�⎥ = ⎢0⎥ (12)
0 0 ⎢0 × 1 ⎥ ⎢0⎥
In addition, A⊗k means that matrix A has done k tensor prod- ⎣ 0 ⎦ ⎣0⎦
uct operations on itself.
� �
2.3 Qubit � � � � ⎡1 × 0 ⎤ ⎡0⎤
1 0 ⎢ ⎥ ⎢ ⎥
�01⟩ =�0⟩ ⊗ �1⟩ = ⊗ =⎢ �1�⎥ = ⎢1⎥ (13)
Bit is the basic concept of classical computing. Quantum com- 0 1 ⎢0 × 0 ⎥ ⎢0⎥
puting is based on a similar concept, namely qubit. In most ⎣ 1 ⎦ ⎣0⎦
cases, qubits are treated as abstract mathematical objects. In
� �
this way, the theory of quantum computing can be constructed
� � � � ⎡0 × 1 ⎤ ⎡0⎤
freely, without relying on specific systems for implementation. 0 1 ⎢ ⎥ ⎢ ⎥
�10⟩ =�1⟩ ⊗ �0⟩ = ⊗ =⎢ �0�⎥ = ⎢0⎥ (14)
A classical bit has two states, 0 and 1. The two possible 1 0 1
⎢1 × ⎥ ⎢1⎥
states of a qubit are �0⟩ and �1⟩, which correspond to states 0 ⎣ 0 ⎦ ⎣0⎦
and 1 of the classical bit. The difference between a bit and a
qubit is that a qubit can be in a state other than �0⟩ or �1⟩. It � �
can exist in the continuous state between �0⟩ and �1⟩. Called � � � � ⎡0 × 0 ⎤ ⎡0⎤
0 0 ⎢ ⎥ ⎢ ⎥
superposition state: �11⟩ =�1⟩ ⊗ �1⟩ = ⊗ =⎢ �1�⎥ = ⎢0⎥ (15)
1 1 ⎢1 × 0 ⎥ ⎢0⎥
�𝜓⟩ = 𝛼�0⟩ + 𝛽�1⟩ (6) ⎣ 1 ⎦ ⎣1⎦

where, 𝛼 and 𝛽 are called probability amplitudes. The state There is a special case called entanglement. For example,
of a qubit is a vector in a two-dimensional complex vector two two-quantum states �𝜓1 ⟩ and �𝜓2 ⟩ are shown below.
space. The states �0⟩ and �1⟩ are called calculation base states,
and constitute a set of orthonormal basis of this vector space. �𝜓1 ⟩ = 𝛼1 �00⟩ + 𝛽1 �11⟩ (16)
When measuring the superposition state �𝜓⟩, only one
component state can be gotten and other states disappear. The |𝛼1 |2 + |𝛽1 |2 = 1 (17)
probability of �0⟩ being measured is |𝛼|2, the probability of �1⟩
being measured is |𝛽|2. And:
�𝜓2 ⟩ = 𝛼2 �10⟩ + 𝛽2 �11⟩ = �1⟩ ⊗ (𝛼2 �0⟩ + 𝛽2 �1⟩) (18)
|𝛼| + |𝛽| = 1
2 2
(7)
|𝛼2 |2 + |𝛽2 |2 = 1 (19)
In quantum computing, �0⟩ and �1⟩ are usually expressed as:
� � � � In �𝜓1 ⟩, if the first quantum is in state �0⟩, the second quan-
1 0
�0⟩ = , �1⟩ = (8) tum also is in state �0⟩; if the first quantum is in state �1⟩, the
0 1
second quantum also is in state �1⟩. They entangle togther.
But in �𝜓2 ⟩, the first quantum is in state �1⟩, the second quan-
tum also is in state (𝛼2 �0⟩ + 𝛽2 �1⟩).

13
740 Z. Wang et al.

2.4 Gate and Circuit Table 1  Commonly used single quantum gates

Gate type Circuit Matrix
Similar to logic gates in classical digital circuits, logic � �
gates are also used in quantum circuits to manipulate Hadamard H 1 1 1
H= √
1 −1
information, and logical transformations are achieved by 2
[ ]
unitary transformations of quantum states. The quantum Pauli-X X 0 1
X=
gate can be represented in matrix form. 1 0
[ ]
A quantum logic gate that only requires one qubit to NOT 0 1
NOT =
participate is called a single quantum gate. As shown in 1 0
the Fig. 1, the horizontal line represents a wire, which [ ]
Pauli-Y Y 0 −i
represents a qubit, and U represents a single quantum gate. Y=
i 0
Input a qubit �𝜓⟩ and output U�𝜓⟩. [ ]
Pauli-Z Z 1 0
Expressed in matrix form: Z=
0 −1
� � � � [ ]
𝛼 a b Phase S 1 0
�𝜓⟩ = ,U = (20) S=
0 i
𝛽 c d
[ ]
Identity I 1 0
��� � � � I=
0 1
a b 𝛼 a𝛼 + b𝛽
U�𝜓⟩ =
c d 𝛽
=
c𝛼 + d𝛽 (21)

Table 1 shows the commonly used single quantum gates. Table 2, there are commonly used multiple quantum gates.
Hadamard gate: When multiple qubits participate, there is a relationship
� � between control and being controlled, such as CNOT gate.
1 1 1
H=√ (22)
2 1 −1 ⎡1 0 0 0⎤
⎢0 1 0 0⎥
CNOT = ⎢
1⎥⎥ (27)
If applying it to �0⟩ or �1⟩: ⎢0 0 0
� �� � � � ⎣0 0 1 0⎦
1 1 1 1 1 1 1 1
H�0⟩ = √ =√ = √ �0⟩ + √ �1⟩
2 1 −1 0 2 1 2 2
(23)
� �� � � �
1 1 1 0 1 1 1 1
H�1⟩ = √ =√ = √ �0⟩ − √ �1⟩ Table 2  Commonly used multiple quantum gates
2 1 −1 1 2 −1 2 2
(24) Gate type Circuit Matrix

Therefore, the H gate has the ability to convert �0⟩ into a CNOT • ⎡1 0 0 0⎤
superposition state with equal probability of �0⟩ and �1⟩. ⎢0
CNOT = ⎢
1 0 0⎥
Pauli-X: 0 0 0 1⎥
⎢ ⎥
⎣0 0 1 0⎦
[ ]
01 0CNOT ⎡0 1 0 0⎤
X=
10 (25) ⎢1 0 0 0⎥
0CNOT = ⎢
0 0 1 0⎥
⎢ ⎥
It can flip the state of a single qubit, so it is also called a ⎣0 0 0 1⎦
NOT gate: Swap × ⎡1 0 0 0⎤
×
[ ] ⎢0 0 1 0⎥
swap = ⎢
01 0 1 0 0⎥
NOT = (26) ⎢ ⎥
10 ⎣0 0 0 1⎦
Toffoli • ⎡1 0 0 0 0 0 0 0⎤
Corresponding to a single quantum gate, there are also gates • ⎢0 1 0 0 0 0 0 0⎥
that require multiple qubits to participate. As shown in the ⎢0 0 1 0 0 0 0 0⎥
⎢ ⎥
0 0 0 1 0 0 0 0⎥
Toffoli = ⎢
⎢0 0 0 0 1 0 0 0⎥
⎢0 0 0 0 0 1 0 0⎥
⎢0 0 0 0 0 0 0 1⎥⎥
Fig. 1  Single quantum gate ⎢
|ψ U U |ψ ⎣0 0 0 0 0 0 1 0⎦

13
Review of Quantum Image Processing 741

The first qubit is a control qubit, and the second qubit is The inverse transformation mapping relationship is
controlled by the first qubit and is called a target qubit. When
1 � −2𝜋jfF∕2n
2 −1 n

the first qubit is �1⟩, the second qubit is reversed. As shown

�f ⟩ ⟶ √ e �F⟩ (35)
in Eqs.(28-31): 2n F=0
⎡1 0 0 0⎤ ⎡1⎤ ⎡1⎤ Express F = F1 2n−1 + F2 2n−2 + ⋯ + Fn 20 as F = F1 F2 ⋯ Fn,
⎢0 1 0 0⎥ ⎢0⎥ ⎢0⎥
CNOT�00⟩ = ⎢ = �00⟩ (28) a n d e x p r e s s F𝜚 ∕2 + F𝜚+1 ∕22 + ⋯ + Fm ∕2m−𝜚+1 a s
1⎥⎥ ⎢⎢0⎥⎥ ⎢⎢0⎥⎥
=
⎢0 0 0 0.F𝜚 F𝜚+1 ⋯ Fm , then the quantum Fourier transform of
⎣0 0 1 0⎦ ⎣0⎦ ⎣0⎦ �F⟩ = �F1 F2 ⋯ Fn ⟩ is expressed as:
1 � �
⎡1 0 0 0⎤ ⎡0⎤ ⎡0⎤ �F1 F2 ⋯ Fn ⟩ → �0⟩ + e2𝜋j0.Fn �1⟩
⎢0 1 0 0⎥ ⎢1⎥ ⎢1⎥ 2n∕2
� �
CNOT�01⟩ = ⎢ = �01⟩ (29) (36)
1⎥⎥ ⎢⎢0⎥⎥ ⎢⎢0⎥⎥ ⋅ �0⟩ + e2𝜋j0.Fn−1 Fn �1⟩ ⋯
=
⎢0 0 0 � �
⎣0 0 1 0⎦ ⎣0⎦ ⎣0⎦ ⋅ �0⟩ + e2𝜋j0.F1 F2 ⋯Fn �1⟩

⎡1 0 0 0⎤ ⎡0⎤ ⎡0⎤
⎢0 1 0 0⎥ ⎢0⎥ ⎢0⎥
CNOT�10⟩ = ⎢ = �11⟩ (30)
1⎥⎥ ⎢⎢1⎥⎥ ⎢⎢0⎥⎥
=
⎢0 0 0
3 Quantum Image Representations
⎣0 0 1 0⎦ ⎣0⎦ ⎣1⎦
To process images on quantum computers, the images need
⎡1 0 0 0⎤ ⎡0⎤ ⎡0⎤ to be stored in a quantum system. For this reason, various
⎢0 1 0 0⎥ ⎢0⎥ ⎢0⎥ quantum image representations have been proposed. First
CNOT�11⟩ = ⎢ = �10⟩ (31)
1⎥⎥ ⎢⎢0⎥⎥ ⎢⎢1⎥⎥
=
⎢0 0 0 in 2003, Qubit Lattice was proposed [5], similar to the
⎣0 0 1 0⎦ ⎣1⎦ ⎣0⎦ classical approach, the image is stored as a matrix, each
qubit stores only one pixel. It is the first attempt at quan-
Such properties can be used to implement XOR operations.
tum image representations. In 2005, Real Ket was pro-
On the contrary, there are also 0CNOT gates:
posed [6], which used quantum superposition for the first
⎡0 1 0 0⎤ time. Entangled image proposed in 2010 uses quantum
⎢1 0 0 0⎥ entanglement [7]. In the same year, FRQI was proposed by
0CNOT = ⎢
0⎥⎥ (32) Le et al. [8], it stores the color information as an angle, and
⎢0 0 1
⎣0 0 0 1⎦ stores the coordinate information in an n-qubit sequence.
This model represents the image as follows:
If the control qubits of the CNOT gate are increased to two,
this kind of gate is called Toffoli gate. If the number of con- 1 ��
2n
2 −1
�
trol qubits continues to increase to n, it is called an n-CNOT �I(𝜃)⟩ = n
cos 𝜃i �0⟩ + sin 𝜃i �1⟩ ⊗ �i⟩ (37)
2 i=0
gate.
In addition, there is a swap gate that can exchange the states
𝜋
of two qubits: 𝜃i ∈ [0, ], i = 0, 1, ⋯ , 22n − 1 (38)
2
⎡1 0 0 0⎤ where �i⟩(i = 0, 1, ⋯ , 22n − 1) are 22n computational basis
⎢0 0 1 0⎥
swap = ⎢ (33) quantum states, and 𝜃 = (𝜃0 , 𝜃1 , ⋯ , 𝜃22n −1 ) is the vector of
⎢0 1 0 0⎥⎥ angles encoding colors. Gray scale information is encoded
⎣0 0 0 1⎦
with cos 𝜃i �0⟩ + sin 𝜃i �1⟩, and coordinate information is
expressed with �i⟩. This allows the grayscale information
2.5 Quantum Fourier Transform and coordinate information to correspond accurately. The
superposition of each pixel information reduces the use of
The quantum Fourier transform is defined as a linear transfor- storage space.
mation on n qubits [11], and its mapping is as follows The following year, Sun et al. [9] extended it to the
RGB space and MCRQI was proposed. Three qubits are
1 � 2𝜋jfF∕2n √ used to store RGB color information and opacity informa-
2 −1 n

�F⟩ ⟶ √ e �f ⟩, j= −1 (34) tion. Represent the image as:

2n f =0

13
742 Z. Wang et al.

1 �
22n −1
�I(𝜃)⟩ = �ciRGB𝛼 ⟩ ⊗ �i⟩ (39)
2n+1 i=0

�ciRGB𝛼 ⟩ = cos 𝜃Ri �000⟩ + cos 𝜃Gi �001⟩

+ cos 𝜃Bi �010⟩ + cos 𝜃𝛼i �011⟩
(40)
+ sin 𝜃Ri �100⟩ + sin 𝜃Gi �101⟩
+ sin 𝜃Bi �110⟩ + sin 𝜃𝛼i �111⟩

where 𝜃Ri , 𝜃Gi , 𝜃Bi , 𝜃𝛼i are the vectors encoding the colors of
the R,G,B and 𝛼 channels respectively. In a 2n × 2n image, Fig. 2  The original image
the color information is encoded with the probability ampli-
tude of three qubits, and the coordinate information is rep-
enhanced quantum representation (INEQR) changed the
resented with 2n qubits.
horizontal and vertical coordinates to unequal lengths [12].
FRQI uses a normalized superposition to store all the
The generalized quantum image representation (GQIR) uses
pixels in an image, the same operations can be performed
logarithmic coordinates to the point of being able to repre-
simultaneously on all pixels, and therefore FRQI can resolve
sent rectangular images of arbitrary size [13], but this inevi-
the real-time computation problem of image-processing
tably leads to redundancy. In addition, The novel quantum
applications. Since then, many image processing algorithms
representation for log-polar images (QUALPI) realizes the
have been proposed based on it. However, because FRQI
representation of polar coordinates [14]. In 2014, Li et al.
uses only a single qubit to store the gray-scale information
proposed a new representation method for multidimensional
for each pixel in an image, some complex color operations
color images, called n-qubit normal arbitrary superposition
cannot be done easily on the basis of FRQI. The NEQR pro-
state (NASS). They store grayscale in angles and use states
posed in 2013 uses two entangled qubit sequences to store
to represent multiple dimensions. For the first time on a
the grayscale and position information, and stores the whole
quantum computer to achieve a multi-dimensional color
image in the superposition of the two qubit sequences. It
image compression [15]. In 2016, Li et al. [16] improved
stores the grayscale information in binary form in q qubits
the FRQI, a flexible representation for quantum color image
[10]. Because it binary encodes grayscale, this model is more
(FRQCI) is proposed. This model uses probabilistic ampli-
convenient to use. The range 2q binary grayscale information
tude (Eqs. 42–45) or phase (Eqs. 46–47) to store the RGB
is stored in a q-qubit sequence, and the coordinate informa-
information of the pixels.
tion of the 2n × 2n image is stored in a 2n-qubit sequence.
The expression for a quantum image can be written as:
1 �
2 −1 2n

�I(𝜃, 𝜙)⟩ = n �i⟩�ci ⟩ (42)

1 � �� i
n n
2 −1 2 −1 q−1 2 i=0
�I⟩ = n �c ⟩�yx⟩ (41)
2 y=0 x=0 i=0 yx
𝜃i 𝜃
In the preparation procedure for NEQR, the first step is to �ci ⟩ = cos �0⟩ + ej𝜙i sin i �1⟩ (43)
2 2
prepare q + 2n qubits in �0⟩, and apply I-gates and H-gates to
the initial state. Then, set the gray-scale value for all pixels ( )
cRi × 216 + cG × 28 + cBi 𝜋
by 2n-CNOT gates. For example, when preparing a gray- 𝜃i = i
(44)
scale image with a size of 2 × 2 as shown in Fig. 2 into 224 − 1
NEQR model. First apply the H-gate to two of the ten �0⟩
qubits to store the coordinate information, and then use the 𝜙i =0 (45)
2-CNOT gate to set the grayscale information in turn. The
circuit is shown in Fig. 3. 𝜃i =0 (46)
Although more qubits are used, NEQR solves the problem
that the FRQI grayscale information cannot be accurately ( R )
ci × 216 + cG × 28 + cBi 2𝜋
measured within a limited number of times, and it facili- 𝜙i = i
(47)
tates color operations. It has also become a more commonly 224 − 1
used model in quantum image processing. Unfortunately, In the same year, Sang et al. [17] improved MCRQI and
these models have a disadvantage that they can only store NEQR and proposed a novel quantum representation of
square images. To solve this problem, The improved novel color digital images (NCQI). This model changes the q

13
Review of Quantum Image Processing 743

Fig. 3  Quantum circuit for

NEQR preparation [10]

qubits representing gray scale to 3q on the basis of NEQR, In 2018, Li et al. [21] proposed a bitplane representation
which is used to represent the three channels of RGB. The of quantum images (BRQI), and based on BRQI proposed
time complexity of preparing the NCQI quantum image complement of colors, reverse of bitplanes, and translation
experiences an approximately quadratic decrease compared of bitplanes based on BRQI. This model improves NEQR,
to MCRQI, and many color operations can be executed con- divides the gray value into 8 binary bits, decompose the
veniently based on NCQI, especially some complex color grayscale image into 8 binary images. Three qubits are used
transformation. But this model has the disadvantage of using to represent the number of the bitplanes, n qubits are used to
too many qubits. In order to solve this drawback, in 2018, represent the coordinate information, and one qubit is used
Liu et al proposed an optimized quantum representation for to store the binary grayscale information of each bitplane.
color digital images (OCQR) [18]. Compared with NCQI, When representing grayscale images:
OCQR uses nearly one-third times the qubits to store the
pixel value. The time complexity of preparing these two 1 � � �
23 −1 2n−k −1 2k −1
�𝛹B8 ⟩ =√ �g(x, y)⟩�x⟩�y⟩�l⟩ (50)
models are almost the same. Not only does OCQR use fewer 2n+3 l=0 x=0 y=0
qubits to save computing resources, but it also allows some
color transformations to be performed more efficiently. In where �x⟩ and �y⟩ represent coordinates, �l⟩ represents the
2017 Abdolmaleky et al. [19] expanded NEQR to RGB, a number of the bitplanes, �g(x, y)⟩ represents the gray level at
new quantum multi-channel Red–Green–Blue (RGB) repre- this coordinate of the bitplane.
sentation (QMCR) is proposed. This model uses more qubits When representing RGB images:
than MCRQI, but reduces the complexity of preparation
procedure and can accurately retrieve images. Jiang et al. 1 � � �
23 −1 2n−k −1 2k −1
�𝛹BR ⟩ =√ �gR (x, y)⟩�x⟩�y⟩�l⟩ (51)
[20] proposed a three-dimensional quantum representation
2n+3 l=0 x=0 y=0
for digital images, named quantum point cloud. This model
stores 3D images as:
1 � � �
23 −1 2n−k −1 2k −1
�𝛹BG ⟩ = √ �gG (x, y)⟩�x⟩�y⟩�l⟩ (52)
1 �
N−1
�P⟩ = mi �i⟩ ⊗ �xi yi zi ⟩ (48) 2n+3 l=0 x=0 y=0
2 2 i=0

� 1 � � �
23 −1 2n−k −1 2k −1
⌈log2 N⌉, N > 1 �𝛹BB ⟩ =√ �gB (x, y)⟩�x⟩�y⟩�l⟩ (53)
mi =
1, N=1 (49) 2n+3 l=0 x=0 y=0

where, i is the counting number, xi , yi , zi are the three dimen-

sional coordinates.

13
744 Z. Wang et al.

1
�𝛹B24 ⟩ = √ (�𝛹BR ⟩�01⟩ + �𝛹BG ⟩�10⟩ + �𝛹BB ⟩�11⟩) 1 �� �
2b −1 2n −1 2m −1
(54) �I⟩ = √ �f (𝜆, y, x)⟩ ⊗ �𝜆⟩ ⊗ �yx⟩ (56)
3
2b+n+m 𝜆=0 y=0 x=0

The preparation procedure for BRQI is similar to the NEQR.

The number of qubits used in MCQI model is less than
BRQI requires n + 4 qubits with a state of �0⟩. First apply
in the QRMW model for fewer channels. However, as the
Hadamard gates to �x⟩, �y⟩ and �l⟩, then use (n + 3)−CNOT
number of channels increases, the number of qubits used in
gates to operate on each pixel.
the QRMW model is less than in the MCQI models, more
When preparing RGB images, as shown in Fig. 4. Two
suitable for multi-spectral images with a large number of
qubits are added to distinguish the three planes of R, G,
channels. Li et al. [24] proposed an improved FRQI model
and B, �01⟩ represents R, √ �10⟩ represents G, and �11⟩ rep-
called the improved FRQI model of representing the color
resents B. The Rx [arctan( 2)] gate and the 0CNOT gate
image (FRQCI), and provided several simple image process-
and the controlled H gate set the probabilities of the three
ing operators for color and pixel position.
states of �01⟩ and �10⟩ and �11⟩ to be equal. UR UG UB is com-
In 2019, after analyzing FRQI and NEQR, Khan pro-
posed of (n + 3)−CNOT gates, which are used to operate
posed an improved flexible representation of quantum
the color information of the RGB plane. Compared with
images (IFRQI) [25], which expressed each two bits as an
NEQR and NCQI, this model can reduce the qubits of stored
angle in the probability amplitude, and successfully used
images. When performing some special image processing
a qubit to encode 2-bit grayscale information. It assists in
algorithms, such as inverting and operating on the entire
accurate retrieval of original image information. In the
bitplane, it can greatly reduce the complexity. Compared
same year, Wang et al. [26] proposed a quantum indexed
with NEQR and INEQR, the storage capacity of BRQI
image representation method (QIIR). A quantum indexed
improves 16 times. For color images, compared with NCQI,
image consists of a quantum data matrix and a quantum
it omproves 218 times.
palette matrix. Each data structure is based on the basic
Coincidentally, Wang et al. [22] also utilized a bitplane in
states of qubit sequence to represent information, includ-
the quantum representation model of color digital images in
ing pixel positions and pixel values in the data matrix, and
2019, called QRCI, where the color information is encoded
indexes and color values in the palette matrix. Indexed
by the basis states of qubit sequences.
images are usually used to store images with simple color
���
23 −1 2n −1 2n −1 requirements, and images with complex colors require true
1
�I⟩ = √ �Rlyx Glyx Blyx ⟩ ⊗ �lyx⟩ (55) colors. Xu et al. [27] proposed order-encoded quantum
22n+3 l=0 y=0 x=0 image model (OQIM). Li et al. [28] proposed four models
of quantum signal representation (QSR) for integer, real,
QRCI model utilizes 2n + 6 qubits to store a color digi-
and complex signals. These models provided the founda-
tal image with size 2n × 2n . Compared with the existing
tion of more complex applications in quantum signal pro-
NCQI representation model, the storage capacity of QRCI
cessing. In 2020, Grigoryan et al. [29] proposed an Fourier
improves 218 times. Compared with the way of storing RGB
transform representation (FTQR), which stored images in
separately in BRQI, fewer qubits are used and the prepara-
a new way. Wang et al. [30] proposed a double quantum
tion procedure is more concise. In 2018, in order to solve
color images representation model (DQRCI) which can
the storage of multi-spectral images on quantum computers,
store two color digital images simultaneously into a quan-
Sahin et al. [23] proposed quantum representation of multi-
tum superposition state.
wavelength images (QRMW). This model uses a quantum
sequence to specifically store the channel number.

Fig. 4  The implementation

circuit of BRQI for RGB color
images [21]

13
Review of Quantum Image Processing 745

4 Geometric Transformation including global translation and local translation. Global

and Morphological Transformation translation is implemented by adder modulo N, and Gray
code is employed in local tranlstion [34], global translation
Geometric transformation and morphological transformation is presented in Fig. 5, local translation is presented in Fig. 6.
are relatively mature operations in classical image process- Image scaling is one of the most frequently used and basic
ing, and almost all image processing software has these fea- operations in image processing, can adjust the size of digital
tures. For example, image translation, image scaling. Many images. In image scaling, it is necessary to use interpola-
image processing also use two basic morphological trans- tion methods to create new pixels when the image is scaled
formations, erosion and dilation. up, or to delete redundant pixels when the image is scaled
In 2011, Le et al. [31] proposed three design strategies down. The commonly used interpolation methods include
for constructing new geometric transformations on quan- nearest-neighbor interpolation, bilinear interpolation and
tum images from other transformations. The strategies focus bicubic interpolation [35]. Scaling up the image can make
on the affected areas in the images, the separability, and the details clearer, and scaling down the image can save
smoothness of the transformations by exploiting a repre- image storage space.
sentation of images on quantum computers extensively. In 2017, Zhou et al. [36] proposed quantum multidimen-
Image translation, which maps the position of each image sional color image scaling based on nearest-neighbor inter-
element into a new position, is a basic image transformation. polation and the bilinear interpolation method for NEQR
Although it has been deeply researched and widely used in [37], which extended FRQI to multi-dimensional color
classical image processing, its quantum version is a vacancy. model. Expressed as
In 2015, Wang et al. studies the quantum image translation
1 ��
n
2 −1
for the first time. Two types of quantum image translation �
�I(𝜃)⟩ = √ n cos 𝜃i �0⟩ + sin 𝜃i �1⟩ ⊗ �i⟩ (57)
entire translation and cyclic translation based on NEQR was 2 i=0
proposed [32]. In 2016, Fan et al. [33] proposed a quantum
algorithm for geometric transformation based on NASS
�i⟩ =�d1 ⟩�d2 ⟩ ⋯ �dk ⟩ (58)
(two-point switchings, symmetric flip, local flip, orthogonal
rotation and translation). They provide global operators and
𝜋 c
local operators respectively, and prove that, contrary to clas- 𝜃= × , c = R × 216 + G × 28 + B (59)
2 224 − 1
sical images, global operators are faster than local operators
in quantum image processing. In 2017, Zhou et al. designed Zhou et al. [35] proposed a quantum algorithm to scale
two kinds of quantum image translation based on FRQI, up and scale down quantum images based on bilinear

Fig. 5  Global translation [34]

Fig. 6  Local translation [34]

13
746 Z. Wang et al.

This algorithm breaks the previous situation where

quantum image scaling ratio was just an integer. Change
the Eq.62 in the previous algorithm to Eq.63.

⎧[ry −� (y − y × ry )] × [rx −� (x − x × rx )] × �cy,x ⟩⎫

� �

⎪ +(y − y × ry ) × [rx − (x − x × rx )] × �cy+1,x ⟩ ⎪

⎨ +[r − (y� − y × r )] × (x� − x × r ) × �c ⎬
⎪ y y x y,x+1 ⟩ ⎪
⎩ +(y − y × ry ) × (x − x × rx ) × �cy+1,x+1 ⟩ ⎭
� �

�cy� ,x� ⟩ =
(ry × rx )
(62)
− h y)] × [w − (wx − w x)] × �cy,x ⟩⎫
� � � � � �
⎧[h − (hy
Fig. 7  Scaling up [38]: a Original image (90 × 80); b The image after ⎪ +(hy − h y) × [w − (wx − w x)] × �cy+1,x ⟩ ⎪
� � � � �

Scaling up (192 × 224) ⎨ +[h − (hy − h y)] × (wx − w� x) × �c ⎬

y,x+1 ⟩ ⎪
� � � �

⎪
⎩ +(hy − h y) × (wx − w x) × �cy+1,x+1 ⟩ ⎭
� � � �

�cy� ,c� ⟩ =
(h� × w� )
(63)
In 2018, Li et al. designed a new method based on quantum
Fourier transform for bilinear interpolation of images. It uses
QFT to implement addition and multiplication operations
[39]. In the same year, Zhou also proposed the nearest neigh-
bor value (NNV) interpolation algorithm for INEQR [40].
The current image scaling research mainly focuses on
nearest neighbor interpolation and bilinear interpolation,
and there is a lack of related research on bicubic interpo-
lation. The bilinear interpolation method of FRQI and its
related research are also lacking.
Morphological image processing is a relatively mature
image processing method in classical image processing,
Fig. 8  Scaling down [38]: a Original image (256 × 256); b The image dilation and erosion operations are fundamental to mor-
after Scaling down (60 × 100) phological operations. Commonly used in gradient calcu-
lations, edge detection and other algorithms. But in quan-
tum image processing, there are relatively few studies on
morphology. In 2016, Yuan et al. [41] proposed two kinds
interpolation with integer scaling ratio based on GQIR. of improved quantum dilation and erosion operations to
Quantum images with arbitrary size h × w can be scaled reduce the time complexity of quantum morphology opera-
to h × w . Compared with nearest-neighbor interpolation,
� �
tions. Utilizing quantum position shifting transformation,
bilinear interpolation and bicubic interpolation have better and make the neighborhood information store in a quan-
effect on target image. In 2019, Zhou et al. improved their tum image set. The time complexity is greatly reduced
algorithm. A quantum algorithm based on bilinear interpola- compared with the previous quantum dilation and ero-
tion with arbitrary scaling ratio is proposed to resize quan- sion operations as show in Fig. 9. In 2019, Li et al. [42]
tum images [38]. Firstly, a quantum image with arbitrary designed the quantum circuits of the two basic operations
size h × w is described by GQIR. Then, utilizing bilinear of dilation and erosion for binary images and grayscale
interpolation, new pixels are created when scaling up, or images. On this basis, three morphological algorithms
redundant pixels are deleted when scaling down as shown for quantum circuits are proposed for binary images. The
in Figs. 7 and 8. noise removal, boundary extraction and skeleton extrac-
⌊ �
h
⌋ ⌊ �
w
⌋ tion are designed in detail. In the same year, Fan et al.
y = y × � ,x = x × � (60) proposed the flat grayscale dilation and erosion operations
h w
are proposed for NEQR. Furthermore, through combining
these two morphology operations, they further realize the
h � w �
dy = y − y, dx = � x − x (61) morphological gradient operation [43].
h� w

13
Review of Quantum Image Processing 747

Fig. 9  Morphological transformation [41]: a The original image; b The flat structuring element for dilation and erosion; c Image after dilation
operation; d Image after erosion operation

MF = (F ⊕ B)(s, t) − (F ⊖ B)(s, t) strategy can embed the watermarked image into both spatial
= [max F(s + x, t + y) − min F(s − x, t − y)], (x, y) ∈ DB and frequency domains. When preprocessing the watermark,
(64) two keys are generated utilizing quantum measurements and
owner assignation respectively. They are used for scrambling
the watermark in the embedding procedure and descram-
bling in the extraction procedure.
5 Quantum Image Protection In 2016, Miyake et al. [46] proposed a new quantum
gray-scale image watermarking scheme by using simple and
In classical image processing, image protection has been small-scale quantum circuit. First, expand the watermark
very mature. It is to protect image information from theft of n × n size 8-bit grayscale to 2n × 2n size 2-bit grayscale
or unauthorized use. Similar to classical image process- image. Then, the expanded image is scrambled by the SWAP
ing, quantum images may also be attacked in many ways, gate controlled by the keys. The scrambled image is embed-
so image protection is also essential for quantum images. ded into the carrier image by the XOR operation. In 2016,
Image protection is divided in two categories. One is image Heidari et al. [44] proposed a new quantum watermarking
encryption, and the other is quantum data hiding including protocol including quantum image scrambling based on
quantum watermarking and quantum steganography. Image Least Significant Bit (LSB).
encryption is to transform the images into meaningless form, In 2017, in order to solve the problem of embedding the
and the watermark is to hide image information by embed- watermark into the quantum color image, Li et al. [47] pro-
ding it into the carrier images. In addition, there is an image posed an improved scheme of using small-scale quantum
encryption technology similar to watermark, called steg- circuits and color scrambling. First, utilize a controlled
anography, whose purpose is to communicate secretly. In the rotation gates to scramble the color of pixels in the water-
quantum image steganography process, the original secret mark image, and then expand the scrambled watermark with
quantum image is first encrypted, and then the encrypted 2n × 2n image size and 24-qubit grayscale to an image with
image is embedded into the quantum cover image. In this 2n+1 × 2n+2 image size and 3-qubit gray scale. Finally, the
section, quantum image encryption, quantum watermark- expanded watermark image is embedded into the carrier
ing and quantum steganography are introduced respectively. image by the CNOT gate. In 2017, Naseri et al. proposed a
new watermark strategy for quantum images. In this scheme
5.1 Quantum Data Hiding and with the aim of data hiding, in addition to using the least
significant bit (LSB), the most significant bit (MSB) is also
Quantum watermarking is a technique which embeds the employed [48]. Compared with the previous protocol,this
invisible quantum signal such as the owners identification scheme indicates not only the better resistance against
into quantum multimedia data (such as audio, video and attacks and noises, but also less resources and complexity,
image) for copyright protection [44]. because it exhibits a better PSNR, and it is more secure.
In recent years, many watermarking strategies have been In the same year, Qu et al. [49] proposed a new quantum
proposed. In 2015, Yan et al. [45] utilizing a stockpile of effi- watermark algorithm based on quantum log-polar image
cient transformations consisting of channel of interest, chan- (QUALPI). This algorithm utilized LSB modification tech-
nel swapping, and quantum Fourier transforms, proposed nique for embedding. The least significant bit (LSB) replace-
a duple watermarking strategy onmulti-channel quantum ment method provides an easy way to hide secret data in the
images. The objective in most of watermarking strategy is cover image. In order to realize quantum watermark embed-
accomplished by embedding some secret messages, and this ding, the LSB of the quantum carrier image is replaced by

13
748 Z. Wang et al.

the quantum watermark image. Compared with other quan- of color digital images model (INCQI) [52]. This algorithm
tum watermarking algorithms, this algorithm effectively employs a fast bitplane scrambling method, which can trans-
utilizes two important properties of log-polar sampling, form the secret image into a disordered form. Both the car-
i.e., rotation and scale invariances. By combining QUALPI rier image and the watermark image are non-square color
representation with the LSB modification technique, the new images, the embedding procedure adopts dual embedding
algorithm make quantum watermark image extracted have a algorithm.
good robustness when it was subjected to geometric attacks. In 2019, Luo et al. [53] proposed a adaptive LSB quantum
Unfortunately, this algorithm does not consider more watermarking method using tri-way pixel value differenc-
complex attacks such as filtering and compression. In 2018, ing. First, the quantum cover image is partitioned into 2 × 2
Luo et al. [50] proposed an enhanced quantum watermarking blocks with four non-overlapping pixels. In order to classify
scheme on the basis of INEQR. In contrast to previous quan- the block as a smooth area or an edge area, the tri-way pixel
tum watermarking scheme, this scheme utilizes the edge pix- value differences are calculated and compared with a prede-
els of a carrier image, which cannot be noticed visually as fined threshold. Then, according to the level k of each block,
the embedding region. This watermarking scheme is keyless the expanded and encrypted quantum watermark image is
and blind, with higher visual quality and better robustness. embedded into the quantum cover image by the k-bit LSB
As human vision is insensitive to the edge regions of an replacement method. In addition, the algorithm utilizes the
image, the watermark is embedded into the edge region of a pixel value difference (PVD), therefore, the method embeds
carrier image by LSB modification technique. This quantum more secret data inedge areas to acquire more payload, and
watermarking scheme has high visual effects. However, the embeds less data in the smooth area to preserve visual qual-
payload for this scheme is very low, which only has 1/16 ity. And in order to enhance the robustness and security of
bits per pixel, and the watermark image is not scrambled, the method, parity bits are utilized in the embedding process.
which is not conducive to enhancing the robustness of the As show in Fig. 10, they do not show any visible marks to
watermarked images. be suspected.
To improve the payload and the robustness of the quan- Steganography is the study and application of techniques
tum watermarked image, in the same year Hu et al. proposed to hide messages within carriers, so that people cannot per-
a LSBs-based quantum color images watermarking algo- ceive their existence. one may pass the message unnoticed
rithm in edge region. First, utilize the nearest neighbor inter- to everyone except for the receiver. Thus, the information
polation method to resize the original quantum watermark is protected and the existence of the communication itself
image to an appropriate size and then implement Gray code is concealed [54]. Steganography is similar to watermark-
transformation to scramble the resized watermark image. In ing but has a different purpose. Quantum image watermark-
order to scatter the watermark image qubits embedded into ing aims to use the signal embedded in the cover image to
the edge regions of the quantum color image LSB and sec- protect the copyright of the image. In the case of quantum
ond LSB, two quantum embedded key images are generated. steganography, its objective is to securely communicate the
And in the same year, in order to break through the limi- secret data by embedding the secret data in the quantum
tations of quantum encryption algorithms based on square cover image without triggering any suspicions from the
images, Zhou et al. [51] presented a watermarking scheme unauthorized third parties [55]. Moreover, steganography
based on INEQR, and a quantum color image watermarking is a mode of communication, so large amounts of data needs
scheme based on the improved novel quantum representation to be embedded. And it is important to make sure the image

Fig. 10  Watermarking [53]: a Watermark image; b Original image; c The image with embedded watermark; d The extracted watermark image

13
Review of Quantum Image Processing 749

does not exhibit any traces of hidden data. On the contrary, reflected Gray code for colored quantum images, and the
the watermark has no limitation on the amount of embed- embedding capacity of this scheme is up to 4 bits per pixel
ded data. And the watermark only embeds the copyright [61].
information of the image. In 2018, Zhou et al. [62] proposed a novel quantum
In 2016 Laurel et al. [54] proposed a steganography on image steganography scheme based on the NEQR and LSB
quantum pixel images using shannon entropy. They embed scheme. This model uses bit-plane scrambling method to
secret information in the least significant bit (LSB) from the scramble the original secret image, expands the scrambled
code of the most significant bit information (MSBI). The image and then scrambling with the Arnold scrambling. In
Shannon entropy is used to determine the optimal LSB from 2019, Qu et al. proposed a novel quantum image steganogra-
the MSBI, which improves the hiding security of the system. phy algorithm based on an efficient embedding technique of
Images altered in the LSB from the MSBI, display a better exploiting modification direction. This embedding technique
performance than simply altering the LSB. Jiang et al. [56] is referred to the exploiting modification direction (EMD)
proposed two blind LSB steganography algorithms in the embedding [63]. The merit of the EMD embedding is that it
form of quantum circuits based on the novel enhanced quan- provides good image quality. However, the EMD embedding
tum representation (NEQR). One algorithm is plain LSB has room for further improvement of its embedding capacity.
which uses the message bits to substitute for the pixels’ LSB Luo et al. [55] proposed an quantum steganography
directly. The other is block LSB which embeds a message bit scheme using inverted pattern approach. Before embedding
into a number of pixels that belong to one image block. The the secret image, each pixel of the quantum secret image will
algorithm has good invisibility, and the balance between the be inverted or not. The decisions are recorded by a quantum
capacity and the robustness can be adjusted according to the key image. Then, the quantum secret image is embedded in
needs of applications. the k least significant bits of the quantum cover image. The
In 2017, Heidari et al. [57] suggested three quantum steg- computational complexity of this scheme is lower than its
anography procedures which utilize all RGB channels in the classical counterpart and other quantum shorthand schemes.
cover image. The first algorithm employs only one channel And this scheme has outperformed other methods in terms
to cover secret data. The second algorithm is based on LSB of the visual quality and embedding capacity. In the same
XORing technique, and the last algorithm utilizes two chan- year, they proposed two-level information hiding for quan-
nels to cover the color image to hide secret quantum data. tum images using optimal LSB. The first level is to hide
In 2018, El-latif et al. [58] proposed a highly secure quan- the encrypted quantum secret image into a quantum water-
tum image steganography scheme based on the logistic cha- mark image, and the second level is to embed the quantum
otic map. The quantum secret image is encrypted using a watermark image into a quantum cover image. Using the
CNOT gate. Then the encrypted secret image is embedded optimal LSB-based algorithm, the double embedding can
into the quantum cover image utilizing the two most and make the position of embedding have a certain randomness,
least significant qubits. In addition, they presented a quan- thus increasing security [64]. Under the same embedding
tum image watermarking approach. The quantum watermark capability, it can provide better visual effects than existing
image is scrambled by utilizing Arnold’s cat map, is then quantum steganography solutions, as show in Fig. 11. In
embedded into the quantum carrier image using the two 2020, Su et al. [65] proposed a novel information hiding
least and most significant qubits. Sahin et al. [59] proposed method based on the NEQR quantum image and Bell states.
a novel quantum steganography algorithm based on LSBq
for multiwavelength quantum images. Compared with other 5.2 Quantum Image Encryption
methods, this method does not need to use the quantum
comparator operations. Thus, this allows us to reduce time Image encryption can transform the image into a meaning-
complexity. And they proposed that both position and chan- less form according to the key by a variety of encryption
nel qubits are considered as a position and the pixels in this models, so that third parties cannot obtain the image content.
new position are used for embedding by a certain modulo The bitplane consists of 8 binary images formed by
value, not in order. separating the grayscale information, and the image can be
Li et al. [60] proposed a novel quantum steganography scrambled by operating on the bitplane. In 2018, Heidari
scheme using color images as cover images. First, the secret et al. [66] proposed a quantum representation of a digital
information is divided into 3-bit segments, and then each scrambling algorithm for quantum NCQI color images. This
3-bit segment is embedded into the LSB of one color pixel method is divided into two sub-algorithms. First, algorithm
in the cover image according to its own value and using Gray manipulates qubits of the RGB channels by XOR and XNOR
code mapping rules. In the same year, in order to improve operations for bitplane scrambling. Then, the pixel-plane is
the embedding capacity of quantum image steganography, scrambled to disorder pixels of the image. Zhou et al. used
they proposed a novel LSB-based steganography using the quantum channel swapping operation and hyper-chaotic

13
750 Z. Wang et al.

Fig. 11  Steganography [64]:

a The image that needs to be
hidden; b Watermark image;
c Cover image; d The image
embedded with hidden informa-
tion; e The extracted results

the inter–intra bit-level permutation strategy. This method

utilizes both intra and inter bit permutation to scramble
pixels, and chaotic diffusion with logistic map is also per-
formed. Li et al. [71] proposes a block-based image scram-
bling scheme for the generalized model of novel enhanced
quantum representation (GNEQR). This scheme employs
geometric transformations and an operation of bitplane
scrambling to perform position and pixel scrambling, respec-
tively. And it suit for rectangular images, so this scrambling
Fig. 12  Encrypting the region of Interest of the medical image by scheme is more general.
operating on the bitplane [64] The quantum rotation gate can rotate the qubit on the
Bloch sphere at any angle, it can be used in image encryp-
tion. In 2017, Li et al. [72] proposed a simple encryption
system at the same time to enhance the scrambling effect. A scheme for quantum color image. Qubits are respectively
bit-level quantum color image encryption scheme by exploit- transformed from a basic state into a balanced superposi-
ing quantum cross-exchange operation and a 5D hyper-cha- tion state according to the key by employed the controlled
otic system is proposed [67]. First, the algorithm exchanges rotation gates. In 2018, Liu et al. [73] proposed a novel color
the bitplanes and RGB channels, and then uses 5D hyper- image encryption/decryption method based on random rota-
chaotic system and XOR operation to encrypt. El-Latif et al. tion of qubit and quantum Fourier transform (QFT). Qubit
[68] proposed a novel approach for the efficient quantum rotation operates once in spatial and frequency domains
image encryption of healthcare media. The quantum image respectively, with the help of quantum Fourier transform.
is scrambled by bitplane and Gray code. Then, the scrambled The encryption effect is shown in Fig. 13. In the same year,
quantum image is encrypted by a XOR operation based on a novel double quantum image encryption approach based
a key generated by the logistic-sine map. In 2019, Heidari on quantum Arnold transform (QAT) and qubit random rota-
et al. [69] proposed a novel quantum selective encryption tion is proposed [74]. The algorithm is extended to double
method for medical images based on BRQI images. Region quantum on the basis of the previous. And before the qubit
of interest of the medical image can be effectively encrypted rotation operates in spatial and frequency domains, QAT
by operating on the bitplane, as show in Fig. 12. Liu et al. operations would be performed in the spatial domain. In
[70] proposed a quantum image encryption schemeby using 2018, Khan et al. [75] proposed an innovative encryption

13
Review of Quantum Image Processing 751

Fig. 14  Encryption algorithm utilizing QFT [79]

2018, Gong et al. [80] proposed a new single channel quan-

tum color image encryption algorithm based on HSI model
and QFT, where the color components are converted to HSI
and the logistic map is employed to diffuse the relationship
Fig. 13  Encryption algorithm employing rotation gate [73]: a, b of pixels in color components and use quantum Fourier
Original images; c, d Encrypted images transform to complete encryption.
Arnold transform can effectively scramble the image,
expand the key space, and is widely used in image encryp-
scheme for digital data based on quantum spinning and rota- tion. In 2015, Zhou et al. [81] proposed a novel quantum
tion operators. image encryption algorithm based on generalized Arnold
Fourier transform is a commonly used operation in signal transform and double random-phase encoding. The image
processing, and QFT can also be applied in quantum image is scrambled by the generalized Arnold transform, and the
encryption. The algorithm proposed by Liu et al. [74] in gray level information is encoded by the double random-
2018 uses QFT . In 2013, Yang et al. [76] proposed a novel phase operations.
gray-level image encryption/decryption scheme which is In 2017, Hu et al. proposed a novel quantum multi-image
based on quantum Fourier transform and double random- encryption algorithm based on iteration Arnold transform
phase encoding technique. This algorithm applies two phase with parameters and image correlation decomposition [82].
codings in spatial and frequency domains to perform dou- Perform 2-D discrete wavelet transform on each image
ble quantum image encryption, and the two random-phase respectively, randomly splice the corresponding low-fre-
encodings are used as the keys. They generalized the double quency image to one image. scramble the new image by the
random-phase encoding technique to quantum scenarios for iteration Arnold transform with parameters, and then use the
the first time. quantum image correlation decomposition to encode. Zhou
In 2014, Yang et al. [77] expanded the algorithm to RGB ey al. suggested a new quantum image encryption scheme
on this basis, and proposed a quantum cryptographic algo- by using the iterative generalized Arnold transforms and the
rithm for color images using quantum Fourier transform and quantum image cycle shift operations [83]. The image pix-
double random-phase encoding. El-Latif et al. proposed a els are scrambled by the iterative generalized Arnold trans-
new color image encryption scheme based on quantum cha- form, and the quantum image cycle shift operations is used
otic system is proposed. Firstly, scramble only the Y (Lumi- to altere the values of the pixels, where the times of shift
nance) component of low frequency subband. Then mix the operations are controlled by a new 4D hyperchaotic system.
features of horizontally and vertically adjacent pixels with In 2018, Zhou et al. [84] devised a novel quantum multi-
the help of adopted quantum chaotic map. Finally, generate image encryption scheme by combining quantum 3D Arnold
an intermediate chaotic key stream image with the help of transform and quantum XOR operations with scaled Zhong-
quantum chaotic system [78]. In 2016, Tan et al. [79] pro- tang chaotic system. This encryption scheme could encrypt
posed a quantum color image encryption algorithm based multiple images simultaneously. In 2019, Luo et al. [85]
on a hyper-chaotic system and quantum Fourier transform. propose a novel quantum secret image-sharing scheme. The
A sequence is generated using the Chen’s hyper-chaotic images are scrambled through the Arnold transform, and
system, which is scrambled and diffused with the original then the quantum shared images are constructed by the swap
image, and use quantum Fourier transform to complete operations and the CNOT gates. Liu et al. [86] proposed a
encryption, the encryption effect is shown in Fig. 14. In quantum block image encryption scheme based on quantum

13
752 Z. Wang et al.

Arnold transform. First, a quantum block image represen- to realize a one-time running-key stream, and reduce the
tation (QBIR) model for block image is proposed. Then, dynamical degradation of digital chaos. In 2016, Liang et al.
QAT is applied to scramble the position of image blocks. [94] devised a novel quantum image encryption algorithm
Finally, the pseudo-random sequence is generated by SCM, combining the generalized affine transform with logistic
and the encrypted image is obtained through quantum XOR map. The key is generated by the logistic map, and the gray
operations. level information is encrypted by XOR operation, while
In addition, there are many other encryption algorithms, the position information is encoded by generalized affine
which use various chaotic systems to help encryption. The transform.
behavior of chaotic systems is very complex and highly sen- Gong et al. [91] proposed a novel encryption algo-
sitive to initial conditions, making it difficult to predict. In rithm for quantum images based on quantum image XOR
2019, Abd-El-Atty et al. proposed a encryption and decryp- operations. The hyper-chaotic sequences generated with
tion algorithms for NEQR images based on discrete quantum the Chen’s hyper-chaotic system is utilized to control the
walks on a circle. Discrete quantum walk can be regarded CNOT gate to encode gray-level information, the encryp-
as a nonlinear mapping between quantum state and position tion effect is shown in Fig. 16. Yang et al. proposed a novel
probability distribution, so it can be used to generate encryp- quantum gray-scale image encryption algorithm based on
tion keys [88]. The algorithm uses the key image generated one-dimensional quantum cellular automata. The quantum
by QWs to XOR the original image. In 2020, EL-Latif et al. image encryption algorithm is realized by subtly construct-
[87] presented a method to build pseudorandom number ing the evolution rules of one-dimensional quantum cellular
generators based on quantum walks and used this result to automata [95]. In 2018, Naseri et al. [96] proposed a new
encrypt quantum color images. This algorithm utilize con- bi-step quantum image cryptography algorithm. The scheme
trolled alternate quantum walk to create PRNG. The image is consisted of four different coding algorithms. According
is encrypted by the CNOT gate based on the key sequence to the pixels and key, select the algorithm.
generated by the pseudo random number generator (PRNG). In 2019, Li et al. [97] proposed an algorithm of quantum
The encryption effect is shown in Fig. 15. image encryption based on NASS by using quantum geo-
In 2017, Kadir et al. [89] proposes a color image encryp- metric transform, phase-shift transform, and quantum Haar
tion scheme based on coupled hyper chaotic Lorenz systems. wavelet packet transform. Firstly, pixels are exchanged along
This algorithm randomly inject impulse signals into coupled
Lorenz system during iterations to enhance the complexity
of trajectory. In 2018, Ran et al. [90] proposed a quantum
color image encryption scheme based on coupled hypercha-
otic Lorenz system with three impulse injections. In order to
prevent the behavior of the short-period chaotic system after
too many iterations from causing degeneration of dynamics,
three impulse signals values are injected into coupled hyper-
chaotic Lorenz system during iterations.
In 2012, Akhshani et al. [92] proposed an implementation
of image encryption scheme based on the quantum logistic
map, it is the first attempt to apply quantum map in the con-
struction of chaotic cryptographic systems. In 2014, Cao
et al. [93] designed a new image encryption scheme based Fig. 16  The original image is encrypted by Chen’s hyper-chaotic sys-
on quantum chaos. An disturbing mechanism is introduced tem [91]

Fig. 15  The effect of QWs

encryption scheme [87]: a
The original image lena; b
The encrypted image c The
decrypted image

13
Review of Quantum Image Processing 753

diagonal lines. Next, keys randomly distributed are encoded as the optimal threshold to obtain the maximum amount of
by using phase-shift transform. Then, image information is information. In 2017, Yao et al. proposed a highly efficient
stored in low frequency using quantum Haar wavelet packet quantum algorithm for detecting the boundary between
transform (QHWPT). Liu et al. [98] proposed a new algo- different regions of a image. It requires one single-qubit
rithm of image encryption based on random selection of gate in the processing stage, independent of the size of the
crossover operation and mutation operation. In order to image [106]. In order to try to resolve the realtime prob-
obtain the high complexity and unpredictability further, lem of image edge extraction in practice image processing,
quantum chaotic map is coupled with nearest-neighboring in 2018 Fan et al. [107] proposed a quantum image edge
coupledmap lattices (NCML). After performing XOR opera- extraction for NEQR based on classical Sobel operator, and
tion, adjacent pixels are carried out bit-level crossover oper- an enhanced quantum edge detection algorithm based on
ation. Then two different bits of each pixel are employed to NEQR, which combines the classical Laplacian operator and
perform mutation. zerocross method [108].
Xu et al. [99] proposed a novel encryption algorithm And in 2019, Zhou et al. [109] proposed a quantum image
based on quantum chaotic map and four-wing complex sys- edge extraction algorithm based on improved sobel operator
tem. Firstly, quantum logistic map and Arnold transform is for the generalized quantum image representation (GQIR).
utilized to disrupt position information. Then, a complex This scheme can achieve more accurate edge extraction,
hyperchaotic system is used to diffuse the value of each especially for diagonal edges. In the same year, in order
pixel. Wang et al. [100] proposed a improved QISS scheme, to solve the coarse edge detection and false edge detection
which is implemented on both quantum gray image and caused by artificial selection of threshold in the traditional
quantum color image. Wang et al. [101] presented a novel Prewitt edge detection algorithm, Zhou et al. [110] pro-
quantum image encryption algorithm based on quantum key posed a quantum image edge extraction for NEQR based
image. The secret image does the XOR operations with a on improved Prewitt operator, which combines the non-
quantum key image bit by bit, which is generated by a cryp- maximum suppression method and adaptive threshold value
tographic algorithm. Liu et al. [102] proposed a novel n out method. As shown in Fig. 17. However, these algorithms are
of n quantum visual secret sharing (QVSS) scheme. Jiang still sensitive to noise.
et al. [103] proposed a quantum image encryption based on In 2020, Li et al. [111] proposed a quantum scheme of
Henon mapping. In 2020, Luo et al. [104] proposed an image classical Marr–Hildreth edge detection. This scheme is
encryption scheme that is based on hyper-chaos and quan- insensitive to the noise of mixed images.
tum coding. First, perform the bit-level adjacent-exchange
operation. Then, the image is diffused by the quantified 6.2 Image Segmentation
sequences where the self-adapting parameters are applied to
quantify the sequences that are generated by Chen’s hyper- Image segmentation is the process of separating the fore-
chaotic system. Finally obtain the final cipher-image by a ground of one or more objects from the background in a
novel scrambling method that is based on quantum location digital image [112]. It is a key step in image processing and
coding. plays an important role in computer vision.
In 2015, Youssry et al. [112] proposed a novel and
generic framework based on quantum mechanics for image
6 Edge Detection and Image Segmentation processing and applied it to image segmentation. Caraiman
et al. [113] discussed the development of a quantum
6.1 Edge Detection

Edge detection is an important part of image processing.

It refers to sharp discontinuity localization process in an
image, the classical edge detection algorithms are mostly
based on the numerical derivative near the pixel of the image
[105]. Edge detection can extract most of the important fea-
tures of the image, which is very useful on image segmenta-
tion and object identification.
In 2016, Abdel-Khalek et al. [105] proposed a novel edge
detection algorithm based on quantum entropy. This algo-
rithm utilizes quantum entropy to take correlations among
quantum bases into the calculation of entropy and uses the Fig. 17  Edge detection algorithm [110]: a The original image; b The
quasi-threshold that leads to the maximum quantum entropy result image of this quantum image edge extraction

13
754 Z. Wang et al.

version for the image segmentation operation. In 2016,

1 ��
n n
2 −1 2 −1
Karmakar et al. presented the application of quantum dot �I⟩ = �yx⟩�fA (y, x)⟩�fB (y, x)⟩ (65)
2n y=0 x=0
gate nonvolatile memory (QDNVM) in image segmenta-
tion. Zhao et al. [114] proposed a novel side scan sonar These schemes only consider exact matching without con-
image segmentation algorithm integrating neutrosophic set sidering fuzzy matching of quantum images. Luo et al. [122]
(NS) with quantumbehaved particle swarm optimization proposed a fuzzy quantum image matching scheme based on
(QPSO). In order to implement image segmentation pre- gray-scale difference. This scheme evaluate the gray-scale
cisely and efficiently, Wang et al. [115] present an image difference between two quantum images by thresholding. If
thresholding method based on the criteria of global quan- all of the obtained gray-scale differences are not greater than
tum entropy maximization (GQEM), and the quantum the threshold value, it indicates a successful fuzzy matching
lossy-encoding based entropy maximization (QLEEM) of quantum images. In 2019, Liu et al. [123] applied quan-
approach is used to deal with the time consumption prob- tum counting to five algorithms for assessing the similarity
lem of thresholding. In 2017, in order to get a reasonable of quantum images of binary image gray image and color
threshold for threshold segmentation, Huo et al. [116] image respectively, brings an advantage to the number of
proposed an improved algorithm of Bloch spherical coor- quantum measurement.
dinates for quantum artificial bee colony, and applied it to
image threshold segmentation. In 2020, Huo et al. [117]
proposed an improved Bloch quantum artificial bee colony 8 Quantum Image Filtering
by combining the Bloch spherical coordinates of qubit
and Compression
with artificial bee colony (ABC) algorithm, and applied it
in multilevel image threshold segmentation. 8.1 Quantum Image Filtering

In classical image processing, image filtering is a common

preprocessing operation, which is generally achieved by cor-
7 Similarity Analyses and Image Matching relating the image with a filter mask. In computer vision, it
is often used to eliminate noise in images. Although it is dif-
Image matching is the process of searching for a small ficult to perform convolution operations in quantum image
image in a big image [118]. It is widely used in computer processing and bring difficulties to filtering operations, many
vision, face detection, and so on. As the fundamental feasible solutions have been proposed.
research of quantum image matching, similarity analyses In 2017, Yuan et al. [124] proposed a framework of quan-
between quantum images are so essential. In 2016, Iliyasu tum image filtering in the spatial domain. In order to avoid
et al. [119] proposed an enhanced quantum-based image the quantum multiplication, they employed quantum addi-
fidelity metric, the enhanced quantum-based image fidelity tion operation. But this method is only suitable for integer
metric (QIFM) as a tool to assess the “congruity” between filter coefficients, and before each filtering behavior the
two or more quantum images. Jiang et al. [118] provided a value of the filter coefficients should be knew exactly. The
quantum image matching solution which does not compute filtering effect of salt and pepper noise is shown in Fig. 18.
similarity. This scheme can get answer by being processed In 2018, in order to overcome these shortcomings, Yuan
and measured only once, which helps to drop the scheme’s et al. [125] proposed an improved version which employs
complexity. But the paper only matched one pixel, instead
of an area. If more than one pixels in the big image are
the same as the one at the upper left corner of the small
image, the algorithm will randomly measure one of them.
In 2017, Dang et al. [120] presented an improved ver-
sion which takes full advantage of the whole matched area
to locate a small image in a big image. In 2018, Zhou
et al. [121] proposed a similarity analyses scheme based
on a novel quantum image representation and quantum
amplitude amplification algorithm. Quantum amplitude
amplification algorithm can avoid measuring many times.
And in order to store two quantum images with internal
relation more conveniently, two images are stored into a
novel quantum image representation that shares coordinate Fig. 18  Quantum image filtering [124]: a Images with salt and pepper
information. noise; b Images after filtering

13
Review of Quantum Image Processing 755

quantum multiplication. In 2017, Li et al. [126] investi- 9 Quantum Machine Learning

gated design method of quantum weighted averaging filter
and its application in image de-noising. In 2018, Li et al. Machine learning is a field with broad applications. It can
[127] improved this method, investigated the use of quantum perform some complex tasks that only the human brain can
Fourier transform (QFT) in the field of image processing, do, such as image recognition, language translation, and
developed a quantum version for the color image filter- decision making, and it can process tasks faster. With the
ing operation. In the same year, they proposed a quantum advent of quantum technology, designing quantum neural
image median filtering method in the spatial domain [128]. networks for quantum machine learning is a crucial task.
In this method, the original image is first translated by one Quantum machine learning is a very promising emerging
unit in eight directions, then the median of nine pixels with research field. To form a deep neural network, the selection
the same position is calculated. In 2019, Jiang et al. [129] of learning algorithms and its parameters is very important.
improved the method, utilizing the extremum detection There have been many attempts in this field in recent years.
approach to distinguish between noises and normal signal In 2013, Pudenz et al. [133] developed an approach to
points proposed a improved quantum image median filtering machine learning and anomaly detection via quantum adi-
in the spatial domain. The filtering effect of salt and pepper abatic evolution. This method uses classical preprocessing.
noise is shown in Fig. 19. This scheme stores only the color In 2016, Dunjko et al. [134] propose an approach for the
information of the neighborhood pixels entangled with posi- systematic treatment of machine learning, from the per-
tion information of the original image. spective of quantum information, which covers supervised,
unsupervised, and reinforcement learning. Konar et al. [135]
proposed an efficient technique for binary object extrac-
8.2 Quantum Image Compression tion in real time from noisy background using quantum
bi-directional self-organizing neural network (QBDSONN)
Image compression algorithm is a widely used algorithm in architecture. Lau et al. [136] generalized quantum machine
classical image processing. In recent years, several quantum learning to infinite-imensional systems, and proposed the
image compression methods have been proposed. critical subroutines of quantum machine learning algorithms
In 2018, Jiang et al. [130] proposed a novel quantum for an all-photonic continuous-variable quantum computer.
image compression method based on JPEG. This method In 2017, Montanaro et al. [137] suggested a quantum
inputs the quantized JPEG coefficients into qubits and then algorithm which achieves a super-polynomial separation
convert them into pixel values. Because the data amount from classical computation for the basic problem of pattern
of JPEG coefficients are less than the data amount of pixel matching on average case inputs. Benedetti et al. [138] pro-
values, the JPEG scheme can reduce the number of quan- posed the quantum-assisted Helmholtz machine, which is a
tum gates used in the GQIR model. Li et al. [131] devised a hybrid quantum–classical framework with the potential of
quantum gray image encryption-compression scheme based tackling high-dimensional real-world machine learning data-
on quantum cosine transform and 5-dimensional hypercha- sets on continuous variables. This method uses deep learning
otic system. Image is compressed by the quantum cosine to extract low-dimensional binary representations of data,
transform and Zigzag scan coding, the compressed image and then uses quantum deep learning to train unsupervised
is encrypted by the 5-dimensional hyperchaotic system. In generative model.
2019, Pang et al. [132] presented a quantum discrete cosine In 2018, Patel et al. [139] proposed a quantum-inspired
transform algorithm (QDCT), and used it to develop and stacked autoencoder-based deep neural network learning
realize a quantum image compression technique. algorithm, which uses stacked auto-encoder to form a deep

Fig. 19  Quantum image median

filtering [129]: a The original
image; b Images with salt and
pepper noise; c Images after
filtering

13
756 Z. Wang et al.

neural network. In order to avoid improper selection of parameterized circuit for image recognition. Illustrates the
learning algorithm parameters, a representation and evalua- architecture of a quantum convolutional layers sequence fol-
tion function characterized by the representation of popula- lowed by a quantum classified layer. Inspired by the varia-
tion dynamic is used. Liu et al. [140] presented a practical tional quantum algorithms, a quantum classical hybrid train-
gradient-based learning scheme to train quantum circuit ing scheme is proposed for the parameter updating in the
Born machine as a generative model of discrete data. QDCNN. First, the input image is prepared as the quantum
In order to optimize the efficiency of image classification, state with basis encoding. Then a sequence of parameter-
Dang et al. proposed an image classification scheme based dependent unitary transformations is employed to realize
on quantum K Nearest Neighbor algorithm [141]. After the corresponding quantum evolution, which can be divided
extracting the feature vectors on a classical computer, the into quantum convolution layer and quantum classified layer.
feature vectors are input into quantum computer for parallel Finally, a quantum measurement is performed on the speci-
computing of similarity. Piat et al. [142] proposed a frame- fied qubit to obtain the category label.
work to accomplish the task of processing large scale data Kerstin et al. [149] proposed a truly quantum analogue of
on small quantum devices. First, a classical autoencoder is classical neurons, which form quantum feedforward neural
trained to compress image to a size that can be loaded on a networks capable of universal quantum computation, and
quantum device. Then, use the compressed data to train the describe the efficient training of these networks using the
RBM on the D-Wave device, and then use the weights from fidelity as a cost function. This method allows for fast opti-
the restricted Boltzmann machine (RBM) to initialize the misation with reduced memory requirements. They proposed
neural network for image classification. a natural quantum perceptron and define it to be a general
Potok et al. [143] proposed a new deep learning architec- unitary operator acting on the corresponding input and out-
ture based on the unique capabilities of the quantum, high put qubits, and proposed a training algorithm for this quan-
performance, and neuromorphic approaches. The high per- tum neural network. Mixed state 𝜌out for the output qubits is:
formance computer is used to create a well performing CNN � � � �
on image type data. The final layer or two is processed by 𝜌out ≡ trin,hid U 𝜌in ⊗ �0 ⋯ 0⟩hid,out ⟨0 ⋯ 0� U† (66)
quantum computers using the limited Boltzmann machine
where U ≡ U out U L U L−1 ⋯ U 1 is the quantum neural network
(LBM) network. Employing the spiking neural network
quantum circuit, Ul are the layer unitaries, comprised of a
(SNN) to model the temporalaspects of the data, then merge
product of quantum perceptrons acting on the qubits in lay-
the ensemble models and produce output. Wiebe et al. [144]
ers l − 1 and l.
suggested a number of ways that quantum information can
be used to help make quantum classifiers more secure or
private. In 2019, Huggins et al. [145] proposed quantum
computing approaches to both discriminative and generative 10 Other Algorithms
learning, with circuits based on tree and matrix productstate
tensor networks. In addition to the previously introduced algorithms, many
Because of the intractability of deep quantum circuits, it other excellent algorithms have been proposed.
is difficult to simulate classical deep learning models. Thus, In order to extend image stabilization to the quantum
in 2020, Chen et al. [146] designed a quantum algorithm computing domain. In 2016, Yan et al. [150] explored a
for quantum machine learning for noisy intermediate scale novel quantum video framework and proposed a method
quantum (NISQ) devices. This work explores variational based on it to perform the image stabilization by utilizing the
quantum circuits for deep reinforcement learning. This work quantum comparator and quantum image translation opera-
is the first to prove that the variational quantum circuits to tions. This method can estimate the camera motion during
approximate the deep q-value function for decision making the exposure process and compensate for the video jitter
and policy selection reinforcement learning with experi- based on it. By Schmidt decomposition and symmetric states
ence replay and target network. Yang et al. [147] presented permutation, Yue et al. [151] proposed a method can got the
a quantum deep learning scheme based on multi-qubit entan- quantum image with high retrieval performance.
glement states, including computation and training of neural For binary images, or bitplanes of non-binary images,
network in full quantum process. The distance between the Chapeau-Blondeau et al. [152] investigated a quantum image
unknown unit vector and the known unit vector is calculated coding method with a reference frame independent scheme.
by the measurement based on the Greenberger-Horne-Zel- Jiang et al. [153] presented a quantum image location algo-
linger entanglement. It also provided the quantum scheme rithm, which modifies the probability of pixels to make
corresponding to the multilayer feedforward neural network. the target pixel to be measured with higher probability. In
Li et al. [148] investigated a quantum deep convolutional order to dealing with the fusion task of infrared and visible
neural network (QDCNN) model based on the quantum images, Kong et al. [154] proposed a novel fusion method

13
Review of Quantum Image Processing 757

based on improved quantum theory model. Compared with For morphological transformation, erosion and expansion
quantum theory model (QTM), a new qubit state 0.5 is added operations have been proposed, but quantum implementa-
in improved quantum theory model (IQTM). It’s helpful to tions of more complex morphological transformations need
represent the gray information more accurately. Yang et al. more research. And more applications can be enriched based
discussed the application of quantum Hash function to image on these basic operations. In quantum scale operations, algo-
encryption. It is found that quantum Hash function can act rithms using bilinear interpolation are still relatively rare,
as a hash function for the privacy amplification process of and algorithms using bicubic interpolation have not yet been
quantum key distribution systems. And they proposed a proposed.
novel image encryption algorithm [155]. Although image encryption models have been greatly
In 2017, Naseri et al. [156] proposed a novel quantum developed, most algorithms are for grayscale images. The
version of Hilditch algorithm for quantum image thinning. encryption of color images, double-quantum images, and
Skeletonization or thinning process is an important step in multiple images is still relatively rare. In terms of watermark-
pre-processing phase. This algorithm can be considered ing. The robustness of watermarking under noise needs to be
as the first quantum image thinning algorithm. Liu et al. improved. And frequency domain watermarking algorithms
proposed three noise removal algorithms based on NEQR. are still relatively few.
These algorithms are the first attempts to detect and deal In the field of edge detection and image segmentation, rela-
with noise using quantum image representation model in tively few algorithms have been proposed. Although quantum
spatial domain [157]. algorithms provide lower complexity, there is not enough edge
In 2019, Du et al. [158] realized the synthesis of two continuity in the proposed methods and the algorithms are
quantum images theoretically by constructing a specific not robust enough to noise. There are also few algorithms for
phase rotation transform. Xia et al. [159] designed a novel image filtering. Since convolution is not easily used for quan-
multi-bit quantum comparators, and realized quantum image tum computation, there is a need to find quantum image spatial
binarization based on it. This multi-bit quantum compara- filtering methods that do not depend on convolution. In addi-
tors can compare more bits with only one auxiliary bits in a tion, in image matching research, some exact matching meth-
less quantum cost. Liu et al. [160] proposed a novel multi- ods have been proposed, but fuzzy matching algorithms are
modality image fusion algorithm based on QWT and quan- rare. And algorithms for image classification can be designed
tum version of SML. Quantum images are transformed with based on quantum machine learning.
QWT to capture salient features of source images. Fusing the In addition, the research on quantum image processing
wavelet coefficients by sum-modified-laplacian (SML) rule. is still in its infancy. Although many quantum image pro-
The final fused image is obtained by using inverse quantum cessing algorithms have been proposed, it is still not rich
wavelet transform. Heidari et al. proposed a new general enough. Relatively complex algorithms such as edge detec-
model for quantum image histogram, which is based on tion, image segmentation, filtering, frequency domain pro-
NEQR and NCQI [161]. cessing, image compression are relatively few. There are
also few algorithms for image synthesis, binarization, and
image grayscale histograms. Therefore, more algorithms
11 Discussion are needed to increase the richness of quantum image pro-
cessing. Moreover, the algorithms that have been proposed
In recent years, quantum image processing has developed use auxiliary qubits, and it is always the question that how
rapidly. Many quantum image processing algorithms have to reduce the use of auxiliary qubits to reduce the com-
been proposed. These algorithms make use of the superiority plexity. Also, quantum image processing has potential for
of quantum computing to parallel computing, exponentially machine learning, pattern recognition, etc. In addition, the
accelerate and simplify the classical image processing algo- development of algorithms requires the support of quantum
rithms. However, compared with classical image processing, computers. As algorithms increase in size and complexity,
quantum image processing is a brand new research field, and algorithm simulations will be limited by classical computers
existing algorithms still urgently need to be improved. and it will be more efficient to simulate quantum systems on
For quantum image representations, the proposed repre- quantum computers.
sentation model stores images whose side length can only
be an integer power of 2. Otherwise, images of arbitrary size
can be stored at the cost of added redundancy. Therefore, 12 Conclusion
quantum images need to have more efficient representation
methods. And the image representation models contain- Quantum processing algorithms proposed in recent years
ing error correction codes can also be proposed to improve are sorted and summarized in the paper. On the basis of
robustness. mathematical concepts of quantum computing, we have

13
758 Z. Wang et al.

emphatically reviewed the research progress in the fields 15. Li HS, Zhu Q, Zhou RG, Li MC, Song L, Ian H (2014) Mul-
of quantum image representation, geometric transformation, tidimensional color image storage, retrieval, and compression
based on quantum amplitudes and phases. Inf Sci 273:212–232
morphological transformation, quantum image protection, 16. Li P, Xiao H, Li B (2016) Quantum representation and water-
edge detection and image segmentation, image matching, mark strategy for color images based on the controlled rotation
filtering, compression, quantum machine learning. Based of qubits. Quant Inf Process 15(11):4415–4440
on the current research progress, we discussc and point out 17. Sang J, Wang S, Li Q (2017) A novel quantum representation
of color digital images. Quant Inf Process 16(2):42
several exisitng problems and challenges in the field of quan- 18. Liu K, Zhang Y, Lu K, Wang X, Wang X (2018) An optimized
tum image processing, which will also be one of the future quantum representation for color digital images. Quant Inf Pro-
works. cess 57(10):2938–2948
19. Abdolmaleky M, Naseri M, Batle J, Farouk A, Gong LH (2017)
Red-green-blue multi-channel quantum representation of digi-
tal images. Opt Int J Light Electron Opt 128:121–132
Funding This research was funded by National Natural Science Foun- 20. Jiang N, Hu H, Dang Y, Zhang W (2017) Quantum point cloud
dation of China (Grant No. 61201421), National cryosphere desert and its compression. Int J Theor Phys 56(10):3147–3163
data center (grant No.E01Z7902 ) and Capability improvement pro- 21. Li H-S, Chen X, Xia H-Y, Liang Y, Zhou Z (2018) A quan-
ject for cryosphere desert data center of the Chinese Academy of tum image representation based on bitplanes. IEEE Access
Sciences(Grant No. Y9298302). 6:62396–62404
22. Wang L, Ran Q, Ma J, Yu S, Tan L (2019) Qrci: a new quantum
Declaration representation model of color digital images. Opt Commun
438:147–158
Conflict of interest The authors declare that they have no conflict of 23. Sahin E, YILMAZ I (2018) Qrmw: quantum representation
interest. of multi wavelength images. Turk J Elect Eng Comput Sci
26(2):768–779
24. Li P, Liu X (2018) Color image representation model and
its application based on an improved frqi. Int J Quant Inf
16(1):1850005
References 25. Khan RA An improved flexible representation of quantum
images. Quant Inf Process 18(7)
1. Moore G (1998) Cramming more components onto integrated 26. Wang B, Hao M-Q, Li P-C, Liu Z-B (2019) Quantum represen-
circuits. Proc IEEE 86(1):82–85 tation of indexed images and its applications. Int J Theor Phys
2. Feynman R-P (1982) Simulating physics with computers. Int J 59(2):374–402
Theor Phys 21(6/7):467–488 27. Xu G, Xu X, Wang X, Wang X (2019) Order-encoded quantum
3. Shor PW (1994) Algorithms for quantum computation: discrete image model and parallel histogram specification. Quant Inf
logarithms and factoring. In: Proceedings 35th Annual Sympo- Process 18(11):1–26
sium on Foundations of Computer Science, pp 124–134 28. Li HS, Fan P, Xia HY, Peng H, Song S (2019) Quantum
4. Grover LK (1996) A fast quantum mechanical algorithm for implementation circuits of quantum signal representation and
database search. In: Proceedings of the 28th Annual ACM Sym- type conversion. IEEE Trans Circuits Syst I Regular Papers
posium on the Theory of Computing, ACM, pp 212–219 66(1):341–354
5. Venegas-Andraca S-E, Bose S (2003) Storing processing and 29. Grigoryan AM, Agaian SS New look on quantum represen-
retrieving an image using quantum mechanics. SPIE Conf Quant tation of images: fourier transform representation. Quant Inf
Inf Comput 5106(01):137–147 Process 19(5)
6. Latorre JI (Oct 2005) Image compression and entanglement, 30. Wang L, Ran Q, Ma J (2020) Double quantum color images
Tech. Rep. quant-ph/0510031, University of Barcelona encryption scheme based on DQRCI. Multimed Tools Appl
7. Venegas-Andraca SE, Ball JL (2010) Processing images in entan- 79(9–10):6661–6687
gled quantum system. Quant Inf Process 9(1):1–11 31. Le PQ, Iliyasu AM, Dong F, Hirota K (2011) Strategies for
8. Le PQ, Dong F, Hirota K (2011) A flexible representation of designing geometric transformations on quantum images.
quantum images for polynomial preparation, image compression Theoret Comput Sci 412(15):1406–1418
and processing operations. Quant Inf Process 10(1):63–84 32. Wang J, Jiang N, Wang L (2015) Quantum image translation.
9. Sun B, Iliyasu AM, Le P, Dong F, Hirota K (2011) A multi- Quant Inf Process 14(5):1589–1604
channel representation for images on quantum computers using 33. Fan P, Zhou RG, Jing N, Li HS (2016) Geometric transforma-
the rgb𝛼 color space. In: IEEE 7th International Symposium on tions of multidimensional color images based on nass. Inf Sci
Intelligent, Signal Processing, Malta, Floriana, pp 160–165 340:191–208
10. Zhang Y, Lu K, Gao Y, Wang M (2013) Neqr: a novel enhanced 34. Zhou RG, Tan C, Ian H (2017) Global and local translation
quantum representation of digital images. Quant Inf Process designs of quantum image based on frqi. Int J Theor Phys
12(8):2833–2860 56(4):1382–1398
11. Nielsen MA, Chuang IL (2007) Quantum computation and quan- 35. Zhou RG, Liu X, Luo J (2017) Quantum circuit realization of
tum information. Math Struct Comput Sci 17(6):1115–1115 the bilinear interpolation method for gqir. Int J Theor Phys
12. Jiang N, Wang L (2015) Quantum image scaling using nearest 56(9):2966–2980
neighbor interpolation. Quant Inf Process 14(5):1559–1571 36. Zhou R-G, Tan C, Fan P (2017) Quantum multidimensional
13. Jiang N, Wang J, Mu Y (2015) Quantum image scaling up based color image scaling using nearest-neighbor interpolation based
on nearest-neighbor interpolation with integer scaling ratio. on the extension of frqi. Mod Phys Lett B 31(17):1750184
Quant Inf Process 14(11):4001–4026 37. Zhou R-G, Hu W, Fan P, Ian H (2017) Quantum realization of
14. Zhang Y, Lu K, Gao Y, Xu K (2013) A novel quantum represen- the bilinear interpolation method for neqr. Sci Rep 7:2511
tation for log-polar images. Quant Inf Process 12(9):3103–3126

13
Review of Quantum Image Processing 759

38. Zhou R-G, Cheng Y, Liu D (2019) Quantum image scaling based 61. Li P, Lu A (2018) Lsb-based steganography using reflected
on bilinear interpolation with arbitrary scaling ratio. Quant Inf gray code for color quantum images. Int J Theor Phys
Process 18(9):267 57(5):1516–1548
39. Li P, Liu X (2018) Bilinear interpolation method for quantum 62. Zhou R-G, Luo J, Liu XA, Zhu C, Wei L, Zhang X (2018) A
images based on quantum fourier transform. Int J Quant Inf novel quantum image steganography scheme based on lsb. Int J
16(4):1850031 Theor Phys 57(6):1848–1863
40. Zhou R, Hu W, Luo G, Liu X, Fan P (2018) Quantum realiza- 63. Qu Z, Cheng Z, Liu W, Wang X (2019) A novel quantum image
tion of the nearest neighbor value interpolation method for ineqr. steganography algorithm based on exploiting modification direc-
Quant Inf Process 17(7):166 tion. Mult Tools Appl 78(7):7981–8001
41. Yuan S, Mao X, Chen L, Wang X (2016) Improved quantum 64. Luo G, Zhou R-G, Mao Y (2019) Two-level information hid-
dilation and erosion operations. Int J Quant Inf 14(7):1085–1564 ing for quantum images using optimal lsb. Quant Inf Process
42. Li P, Shi T, Lu A, Wang B (2019) Quantum circuit design for 18(10):297
several morphological image processing methods. Quantum Inf 65. Su C-F, Chen C-Y (2020) Information hiding method based on
Process 18(12):364 quantum image by using bell states. Quant Inf Process 19(1):36
43. Fan P, Zhou R-G, Hu W, Jing N (2019) Quantum circuit realiza- 66. Heidari S, Vafaei M, Houshmand M, Tabatabaey-Mashadi N
tion of morphological gradient for quantum grayscale image. Int (2018) A dual quantum image scrambling method. Quant Inf
J Theor Phys 58(2):415–435 Process 18(1):9
44. Heidari S, Naseri M (2016) A novel lsb based quantum water- 67. Zhou N, Chen W, Yan X, Wang Y (2018) Bit-level quantum color
marking. Int J Theor Phys 55(10):4205–4218 image encryption scheme with quantum cross-exchange opera-
45. Yan F, Iliyasu AM, Sun B, Venegas-Andraca SE, Dong F, Hirota tion and hyper-chaotic system. Quant Inf Process 17(6):137
K (2015) A duple watermarking strategy for multi-channel quan- 68. El-Latif B, Abd-El-Atty Ahmed A, Talha M (2018) Robust
tum images. Quantum Inf Process 14(5):1675–1692 encryption of quantum medical images. IEEE Access
46. Miyake S, Nakamae K (2016) A quantum watermarking scheme 6(1):1073–1081
using simple and small-scale quantum circuits. Quantum Inf Pro- 69. Heidari S, Naseri M, Nagata K (2019) Quantum selective encryp-
cess 15(5):1849–1864 tion for medical images. Int J Theor Phys 58(11):3908–3926
47. Li P, Zhao Y, Xiao H, Cao M (2017) An improved quantum 70. Liu X, Xiao D, Xiang Y (2019) Quantum image encryption
watermarking scheme using small-scale quantum circuits and using intra and inter bit permutation based on logistic map. IEEE
color scrambling. Quantum Inf Process 16(5):127 Access 7(1):6937–6946
48. Naseri M, Heidari S, Batle J, Baghfalaki M, Fatahi N, Gheibi R, 71. Li HS, Chen X, Song S, Liao Z, Fang J (2019) A block-
Farouk A, Habibi A (2017) A new secure quantum watermarking based quantum image scrambling for gneqr. IEEE Access
scheme. Opt Int J Light Elect Opt 139:77–86 7(1):138233–138243
49. Qu Z, Cheng Z, Wang M (2017) A robust quantum watermark 72. Li P, Zhao Y (2017) A simple encryption algorithm for quantum
algorithm based on quantum log-polar images. Int J Theor Phys color image. Int J Theor Phys 56(6):1961–1982
56(11):3460–3476 73. Liu X, Xiao H, Li P, Zhao Y (2018) Design and implementation
50. Luo G, Zhou R-G, Hu W, Luo J, Liu X, Ian H (2018) Enhanced of color image encryption based on qubit rotation about axis.
least significant qubit watermarking scheme for quantum images. Chin J Elect 27(4):799–807
Quant Inf Process 17(11):299 74. Liu X, Xiao D, Liu C (2018) Double quantum image encryption
51. Zhou R-G, Zhou Y, Zhu C, Wei L, Zhang X, Ian H (2018) Quan- based on arnold transform and qubit random rotation. Entropy
tum watermarking scheme based on ineqr. Int J Theor Phys 20(11):867
57(4):1120–1131 75. Khan M, Waseem HM (2018) A novel image encryption scheme
52. Zhou R-G, Yang PL, Liu XA, Ian H (2018) Quantum color image based on quantum dynamical spinning and rotations. PLoS ONE
watermarking based on fast bit-plane scramble and dual embed- 13(11):e0206460
ded. Int J Quant Inf 16(07):1850060 76. Yang Y-G, Xia J, Jia X, Zhang H (2013) Novel image encryption/
53. Luo G, Zhou R-G, Luo J, Hu W, Zhou Y, Ian H (2019) Adaptive decryption based on quantum fourier transform and double phase
lsb quantum watermarking method using tri-way pixel value dif- encoding. Quant Inf Process 12(11):3477–3493
ferencing. Quant Inf Process 18(2):49 77. Yang YG, Jia X, Sun SJ, Pan QX (2014) Quantum cryptographic
54. Laurel CO, Dong S-H, Cruz-Irisson M (2016) Steganography algorithm for color images using quantum fourier transform and
on quantum pixel images using shannon entropy. Int J Quant Inf double random-phase encoding. Inf Sci 277:445–457
14(5):1650021 78. El-Latif AAA, Li L, Ning W, Qi H, Niu X (2013) A new
55. Luo G, Zhou R-G, Hu W (2019) Efficient quantum steganogra- approach to chaotic image encryption based on quantum chaotic
phy scheme using inverted pattern approach. Quant Inf Process system, exploiting color spaces. Sig Process 93(11):2986–3000
18(7):222 79. Tan R-C, Lei T, Zhao Q-M, Gong L-H, Zhou Z-H (2016) Quan-
56. Jiang N, Zhao N, Wang L (2016) Lsb based quantum image steg- tum color image encryption algorithm based on a hyper-cha-
anography algorithm. Int J Theor Phys 55(1):107–123 otic system and quantum fourier transform. Int J Theor Phys
57. Heidari S, Pourarian MR, Gheibi R, Naseri M, Houshmand M 55(12):5368–5384
(2017) Quantum red–green–blue image steganography. Int J 80. Gong LH, He XT, Tan RC, Zhou ZH (2018) Single channel quan-
Quant Inf 15(5):1750039 tum color image encryption algorithm based on hsi model and
58. Abd El-Latif AA, Abd-El-Atty B, Hossain M, Rahman M, Alamri quantum fourier transform. Int J Theor Phys 57(1):59–73
A, Gupta B (2018) Efficient quantum information hiding for 81. Zhou NR, Hua TX, Gong LH, Pei DJ, Liao QH (2015) Quantum
remote medical image sharing. IEEE Access 6(1):21075–21083 image encryption based on generalized arnold transform and dou-
59. Sahin E, Yilmaz I (2018) A novel quantum steganography algo- ble random-phase encoding. Quant Inf Process 14(4):1193–1213
rithm based on lsbq for multi-wavelength quantum images. Quant 82. Hu Y, Xie X, Liu X, Zhou N (2017) Quantum multi-image
Inf Process 17(11):319 encryption based on iteration arnold transform with param-
60. Li P, Liu X (2018) A novel quantum steganography scheme for eters and image correlation decomposition. Int J Theor Phys
color images. Int J Quant Inf 16(2):1850020 56(7):2192–2205

13
760 Z. Wang et al.

83. Zhou N, Hu Y, Gong L, Li G (2017) Quantum image encryption 104. Luo Y, Tang S, Liu J, Cao L, Qiu S (2020) Image encryption
scheme with iterative generalized arnold transforms and quantum scheme by combining the hyper-chaotic system with quantum
image cycle shift operations. Quant Inf Process 16(6):164 coding. Opt Lasers Eng 124(1):105836
84. Zhou N, Yan X, Liang H, Tao X, Li G (2018) Multi-image 105. Abdel-Khalek S, Abdel-Azim G, Abo-Eleneen ZA, Obada ASF
encryption scheme based on quantum 3d arnold transform and (2016) New approach to image edge detection based on quantum
scaled zhongtang chaotic system. Quant Inf Process 17(12):338 entropy. J Russ Laser Res 37(2):141–154
85. Luo G-F, Zhou R-G, Hu W-W (2019) Novel quantum secret 106. Yao XW, Wang H, Liao Z, Chen MC, Pan J, Li J, Zhang K, Lin
image-sharing scheme. Chin Phys B 28(4):040302 X, Wang Z, Luo Z, Zheng W, Li J, Zhao M, Peng X, Suter D
86. Liu X, Xiao D, Huang W, Liu C (2019) Quantum block image (2017) Quantum image processing and its application to edge
encryption based on arnold transform and sine chaotification detection: theory and experiment. Phys Rev X 7(3):031041
model. IEEE Access 7(1):57188–57199 107. Fan P, Zhou RG, Hu W, Jing N (2019) Quantum image edge
87. EL-Latif AAA, Abd-El-Atty B, Venegas-Andraca SE (2020) extraction based on classical sobel operator for neqr. Quant Inf
Controlled alternate quantum walk-based pseudo-random num- Process 18(1):24
ber generator and its application to quantum color image encryp- 108. Fan P, Zhou RG, Hu WW, Jing NH (2019) Quantum image edge
tion. Phys A Stat Mech Appl 547(1):123869 extraction based on laplacian operator and zero-cross method.
88. Abd-El-Atty B, EL-Latif AAA, Venegas-Andraca SE (2019) An Quant Inf Process 18(1):27
encryption protocol for neqr images based on one-particle quan- 109. Zhou RG, Liu DQ (2019) Quantum image edge extraction based
tum walks on a circle. Quant Inf Process 18(9):272 on improved sobel operator. Int J Theor Phys 58(9):2969–2985
89. Kadir A, Aili M, Sattar M (2017) Color image encryption scheme 110. Zhou RG, Yu H, Cheng Y, Li FX (2019) Quantum image edge
using coupled hyper chaotic system with multiple impulse injec- extraction based on improved prewitt operator. Quant Inf Process
tions. Opt Int J Light Elect Opt 129:231–238 18(9):261
90. Ran Q, Wang L, Ma J, Tan L, Yu S (2018) A quantum color 111. Li P, Shi T, Lu A, Wang B (2020) Quantum implementation
image encryption scheme based on coupled hyper-chaotic lor- of classical marr-hildreth edge detection. Quant Inf Process
enz system with three impulse injections. Quant Inf Process 19(2):64
17(8):188 112. Youssry A, El-Rafei A, Elramly S (2015) A quantum mechan-
91. Gong L-H, He X-T, Cheng S, Hua T-X, Zhou N-R (2016) Quan- ics-based framework for image processing and its application to
tum image encryption algorithm based on quantum image xor image segmentation. Quant Inf Process 14(10):3613–3638
operations. Int J Theor Phys 55(7):3234–3250 113. Caraiman S, Manta VI (2015) Image segmentation on a quantum
92. Akhshani A, Akhavan A, Lim SC, Hassan Z (2012) An image computer. Quant Inf Process 14(5):4693–41715
encryption scheme based on quantum logistic map. Commun 114. Zhao J, Wang X, Zhang H, Hu J, Jian X (2016) Side scan sonar
Nonlinear Sci Numer Simul 17(12):4653–4661 image segmentation based on neutrosophic set and quantum-
93. Cao G, Zhou J, Zhang Y, Jiang Y, Zhang X (2014) Quantum behaved particle swarm optimization algorithm. Mar Geophys
chaotic image encryption with one time running key. Int J Sec Res 37(3):229–241
Appl 8(4):77–88 115. Wang X, Yang C, Xie GS, Liu Z (2018) Image thresholding
94. Liang HR, Tao XY, Zhou NR (2016) Quantum image encryption segmentation on quantum state space. Entropy 20(10):728
based on generalized affine transform and logistic map. Quant Inf 116. Huo F, Liu Y, Wang D, Sun B (2017) Bloch quantum artificial
Process 15(7):2701–2724 bee colony algorithm and its application in image threshold seg-
95. Yang YG, Tian J, Lei H, Zhou YH, Shi WM (2016) Novel quan- mentation. Sig Image Video Process 11(8):1585–1592
tum image encryption using one-dimensional quantum cellular 117. Huo F, Sun X, Ren W (2020) Multilevel image threshold seg-
automata. Inf Sci 345(1):257–270 mentation using an improved bloch quantum artificial bee colony
96. Naseri M, Abdolmaleky M, Laref A, Parandin F, Celik T, algorithm. Multimed Tools Appl 79(3–4):2447–2471
Farouk A, Mohamadi M, Jalalian H (2018) A new cryptogra- 118. Jiang N, Dang Y, Wang J (2016) Quantum image matching.
phy algorithm for quantum images. Opt Int J Light Elect Opt Quant Inf Process 15(9):3543–3572
171:947–959 119. Iliyasu AM, Yan F, Hirota K (2016) Metric for estimating con-
97. Li H-S, Li C, Chen X, Xia H (2019) Quantum image encryption gruity between quantum images. Entropy 18(10):360
based on phase-shift transform and quantum haar wavelet packet 120. Dang Y, Jiang N, Hu H, Zhang W (2017) Analysis and improve-
transform. Mod Phys Lett A 34(26):1950214 ment of the quantum image matching. Quant Inf Process
98. Liu H, Zhao B, Huang L (2019) A novel quantum image encryp- 16(11):269
tion algorithm based on crossover operation and mutation opera- 121. Zhou RG, Liu XA, Zhu C, Wei L, Zhang X, Ian H (2018) Simi-
tion. Multimed Tools Appl 78(14):20465–20483 larity analysis between quantum images. Quant Inf Process
99. Xu J, Li P, Yang F, Yan H (2019) High intensity image encryp- 17(6):121
tion scheme based on quantum logistic chaotic map and complex 122. GaoFeng Luo, Zhou R.-G., Liu X, Hu W, Luo J (2018) Fuzzy
hyperchaotic system. IEEE Access 7(1):167904–167918 matching based on gray-scale difference for quantum images. Int
100. Wang H-Q, Song X-H, Chen L-L, Xie W (2019) A secret sharing J Theor Phys 57(8):2447–2460
scheme for quantum gray and color images based on encryption. 123. Liu XA, Zhou RG, El-Rafei A, Li FX, Xu RQ (2019) Similarity
Int J Theor Phys 58(5):1626–1650 assessment of quantum images. Quant Inf Process 18(8):244
101. Wang J, Geng YC, Han L, Liu JQ (2019) Quantum image encryp- 124. Yuan S, Mao X, Zhou J, Wang X (2017) Quantum image filtering
tion algorithm based on quantum key image. Int J Theor Phys in the spatial domain. Int J Theor Phys 56(8):2495–2511
58(1):308–322 125. Yuan S, Lu Y, Mao X, Yuan J (2018) Improved quantum image
102. Liu W, Xu Y, Zhang M, Chen J, Yang C-N (2019) A filtering in the spatial domain. Int J Theor Phys 57(3):804–8013
novel quantum visual secret sharing scheme. IEEE Access 126. Li P, Liu X, Xiao H (2017) Quantum image weighted average
7(1):114374–114384 filtering in spatial domain. Int J Theor Phys 56(11):3690–3716
103. Jiang N, Dong X, Hu H, Ji Z, Zhang W (2019) Quantum 127. Li P, Xiao H (2018) An improved filtering method for quan-
image encryption based on henon mapping. Int J Theor Phys tum color image in frequency domain. Int J Theor Phys
58(3):979–991 57(1):258–278

13
Review of Quantum Image Processing 761

128. Li P, Liu X, Xiao H (2018) Quantum image median filtering in 146. Chen SY-C, Yang C-HH, Qi J, Chen P-Y, Ma X, Goan H-S (2020)
the spatial domain. Quant Inf Process 17(3):49 Variational quantum circuits for deep reinforcement learning.
129. Jiang SX, Zhou RG, Hu WW, Li YC (2019) Improved quantum IEEE Access 8:141007–141024
image median filtering in the spatial domain. Int J Theor Phys 147. Yang Z, Zhang X (2020) Entanglement-based quantum deep
58(7):2115–2133 learning. New J Phys 22(3):033041
130. Jiang N, Lu X, Hu H, Dang Y, Cai Y (2018) A novel quan- 148. Li YC, Zhou RG, Xu RQ, Luo J, Hu WW (2020) A quantum deep
tum image compression method based on jpeg. Int J Theor Phys convolutional neural network for image recognition. Quant Sci
57(3):611–636 Technol 5(4):044003
131. Li X-Z, Chen W-W, Wang Y-Q (2018) Quantum image com- 149. Beer K, Bondarenko D, Farrelly T, Osborne TJ, Salzmann R,
pression-encryption scheme based on quantum discrete cosine Scheiermann D, Wolf R (2020) Training deep quantum neural
transform. Int J Theor Phys 57(9):2904–2919 networks. Nat Commun 11(1):808
132. Pang CY, Zhou RG, Hu BQ, Hu WW, El-Rafei A (2019) Signal 150. Yan F, Iliyasu AM, Yang H, Hirota K (2016) Strategy for quan-
and image compression using quantum discrete cosine transform. tum image stabilization. Sci China Inf Sci 59(5):052102
Inf Sci 473:121–141 151. Ruan Y, Chen H, Liu Z, Tan J (2016) Quantum image with high
133. Pudenz KL, Lidar DA (2013) Quantum adiabatic machine learn- retrieval performance. Quant Inf Process 15(2):637–650
ing. Quant Inf Process 12(5):2027–2070 152. Chapeau-Blondeau F, Belin E (2016) Quantum image coding
134. Dunjko V, Taylor JM, Briegel HJ (2016) Quantum-enhanced with a reference-frame-independent scheme. Quant Inf Process
machine learning. Phys Rev Lett 117(13):130501 15(7):2685–2700
135. Konar D, Bhattacharyya S, Panigrahi BK, Nakamatsu K (2016) A 153. Jiang N, Dang Y, Zhao N (2016) Quantum image location. Int J
quantum bi-directional self-organizing neural network (qbdsonn) Theor Phys 55(10):4501–4512
architecture for binary object extraction from a noisy perspective. 154. Kong W, Lei Y, Ren M (2016) Fusion method for infrared and
Appl Soft Comput J 46(1):731–752 visible images based on improved quantum theory model. Neu-
136. Lau HK, Pooser R, Siopsis G, Weedbrook C (2017) Quan- rocomputing 212(SI):12–21
tum machine learning over infinite dimensions. Phys Rev Lett 155. Yang Y-G, Xu P, Yang R, Zhou Y-H, Shi W-M (2016) Quan-
118(8):080501 tum hash function and its application to privacy amplification
137. Montanaro Ashley (2017) Quantum pattern matching fast on in quantum key distribution, pseudo-random number generation
average. Algorithmica 77(1):16–39 and image encryption. Sci Rep 6:19788
138. Benedetti M, Realpe-Gómez J, Perdomo-Ortiz A (2017) Quan- 156. Naseri M, Heidari S, Gheibi R, Gong L-H, Ahmadzadeh Rajii
tum-assisted helmholtz machines: a quantum-classical deep M, Sadri A (2017) A novel quantum binary images thinning
learning framework for industrial datasets in near-term devices. algorithm: a quantum version of the hilditch’s algorithm. Optik
Quant Sci Technol 3(3):034007 131:678–686
139. Patel OP, Tiwari A, Bagade V (2018) Quantum-inspired stacked 157. Liu K, Zhang Y, Lu K, Wang X (2017) Restoration for noise
auto-encoder-based deep neural network algorithm (q-dnn). Arab removal in quantum images. Int J Theor Phys 56(9):2867–2886
J Sci Eng 43(12):6929–6943 158. Du S, Qiu D, Gruska J, Mateus P (2019) Synthesis of quantum
140. Liu JG, Wang L (2018) Differentiable learning of quantum circuit images using phase rotation. Quant Inf Process 18(9):286
born machine. Phys Rev A 98(6):062324 159. Xia H, Li H, Zhang H, Liang Y, Xin J (2019) Novel multi-bit
141. Dang Y, Jiang N, Hu H, Ji Z, Zhang W (2018) Image classifica- quantum comparators and their application in image binarization.
tion based on quantum k-nearest-neighbor algorithm. Quant Inf Quant Inf Process 18(7):229
Process 17(9):239 160. Liu X, Xiao D (2019) Multimodality image fusion based on
142. Piat S, Usher N, Severini S, Herbster M, Mansi T, Mountney P quantum wavelet transform and sum-modified-laplacian rule.
(2018) Image classification with quantum pre-training and auto- Int J Theor Phys 58(3):734–744
encoders. Int J Quant Inf 16(8):1840009 161. Heidari S, Abutalib MM, Alkhambashi M, Farouk A, Naseri M
143. Potok TE, Schuman C, Young S, Patton R, Spedalieri F, Liu (2019) A new general model for quantum image histogram (qih).
J, Yao KT, Rose G, Chakma G (2018) A study of complex Quant Inf Process 18(6):175
deep learning networks on high performance, neuromorphic,
and quantum computers. ACM J Emerg Technol Comput Syst Publisher’s Note Springer Nature remains neutral with regard to
14(2):19 jurisdictional claims in published maps and institutional affiliations.
144. Wiebe N, Kumar RSS (2018) Hardening quantum machine learn-
ing against adversaries. New J Phys 20:123019
145. Huggins W, Patil P, Mitchell B, Whaley KB, Stoudenmire EM
(2019) Towards quantum machine learning with tensor networks.
Quant Sci Technol 4(2):024001

13