A Comprehensive Survey of Covert Communication Techniques, Limitations and Future Challenges

Abstract

Data encryption aims to protect the confidentiality of data at storage, during transmission, or while in processing. However, it is not always the optimum choice as attackers know the existence of the ciphertext. Hence, they can exploit various weaknesses in the implementation of encryption algorithms and can thus decrypt or guess the related cryptographic primitives. Moreover, in the case of proprietary applications such as online social networks, users are at the mercy of the vendor’s security measures. Therefore, users are vulnerable to various security and privacy threats. Contrary to this, covert communication techniques hide the existence of communication and thus achieve security through obscurity and hidden communication channels. Over the period, there has been a significant advancement in this field. However, existing literature fails to encompass all the aspects of covert communications in a single document. This survey thus endeavors to highlight the latest trends in covert communication techniques, related challenges, and future directions.

Full text

A Comprehensive Survey of Covert Communication
Techniques, Limitations and Future Challenges
Imran Makhdoom1, Mehran Abolhasan2, Justin Lipman3
1,2,3University of Technology, Sydney, Australia
imran.makhdoom@uts.edu.au1, mehran.abolhasan@uts.edu.au2
justin.lipman@uts.edu.au3
Abstract
Data encryption aims to protect the confidentiality of data at storage, during transmission, or while in pro-
cessing. However, it is not always the optimum choice as attackers know the existence of the ciphertext.
Hence, they can exploit various weaknesses in the implementation of encryption algorithms and can thus
decrypt or guess the related cryptographic primitives. Moreover, in the case of proprietary applications such
as online social networks, users are at the mercy of the vendor's security measures. Therefore, users are
vulnerable to various security and privacy threats. Contrary to this, covert communication techniques hide
the existence of communication and thus achieve security through obscurity and hidden communication
channels. Over the period, there has been a significant advancement in this field. However, existing liter-
ature fails to encompass all the aspects of covert communications in a single document. This survey thus
endeavors to highlight the latest trends in covert communication techniques, related challenges, and future
directions.
Keywords: Covert communication, hidden messages, data hiding, data encapsulation.
1. Introduction
While cryptography provides data security by preventing unauthorized disclosure and modification of
data, it does not hide evidence of communication. Hence, attackers resort to various attacks to find secret
information. However, attackers tend to avoid direct attacks on cryptographic algorithms due to the computa-
tional complexity involved in brute-forcing the cryptographic primitives, such as finding encryption/decryption5
keys. Instead, they target weaknesses in the implementation and deployment of cryptographic algorithms.
By exploiting these vulnerabilities, attackers can gain valuable side-channel information [1], e.g., power con-
sumption, computing time, EM radiation, etc. As a result, attackers can completely bypass or significantly
weaken the theoretical strength of security solutions [2]. Therefore, encrypted data or cryptographic mod-
ules processing the ciphertext are vulnerable to numerous attacks [3] that aim to seek important information10
about secret messages or related cryptographic primitives, e.g., secret keys. Some of the significant side-
channel attacks include: FLUSH+RELOAD L3 cache [4], timing attacks [5], differential power analysis [6],
fault attacks [7], acoustic attacks [8], and frequency-based attacks [9].
On the contrary, covert communication techniques hide the existence of communication with a Low Proba-
bility of Detection (LPD). Covert communication channels have their uses in military, intelligence operations,15
online social networks for coordinating protests or ensuring user privacy. Accordingly, the art of hiding se-
cret messages/information in non-secret objects is also termed as “Steganography.” Historically, people
Preprint submitted to Journal of Network and Computer Applications May 28, 2022

used hidden tattoos or invisible ink to convey steganographic content. Today, computer and network tech-
nologies provide easy-to-use communication channels for steganography [10]. Irrespective of the type of
cover medium, the information-hiding process in a steganographic system starts by identifying redundant20
bits (those that can be modified without destroying respective cover medium's integrity). The embedding
process creates a stego medium by replacing these redundant bits with data from the secret message.
Modern steganographic techniques endeavor to keep the whole process undetectable.
The application of steganography in the scientific literature may be traced to the “Prisoners’ Problem” [11].
In this scenario, Alice and Bob are the prisoners and wish to plan an escape discretely. However, all their25
communications pass through the warden. If the warden detects any hint of communication, he will throw
them into solitary confinement. Hence, Alice and Bob must find a way to hide their secret messages in
an innocuous-looking cover medium. There are many other real-life applications of steganography. During
the 1980's, to prevent the leakage of confidential information to press by the cabinet ministers, Margaret
Thatcher instructed the word processors to encode the identity of ministers in the word spacing so that30
disloyal ministers could be traced. Similarly, there have been efforts to hide copyright information or serial
numbers in documents [12].
Corresspondingly, the academic community was approached to solve the verification of nuclear arms control
treaties problem between United States (US) and Union of Soviet Socialist Republics (USSR) [13]. The US
and the USSR wanted to place sensors in each others' nuclear facilities that would transmit certain informa-35
tion (such as the number of missiles) but not reveal other kinds of information (such as their location). This
forced a careful study of the ways in which one country's equipment might smuggle forbidden data past the
other country's monitoring facilities [14]. Correspondingly, Military organizations have also been interested
in unobtrusive communications. Their preferred mechanisms include spread spectrum (SS) and meteor
scatter radio with various combinations of resistance to detection, direction finding, and jamming [15].40
Apropos above, it can be deduced that the primary goal of steganography is to reliably and secretly send
confidential information, not merely to obscure its presence. In today's age of computers, steganogra-
phy is being considered a sub-discipline of data communication security. Hence, new directions based on
steganographic approaches have started to emerge. If combined with conventional communication security
techniques, these approaches can achieve better data secrecy. Modern steganography techniques exploit45
the characteristics of digital media by utilizing them as carriers (covers) to hold confidential information.
Covers can be of different types including image, audio, video, text, and IP datagram [16].
Nonetheless, the cover medium changes its statistical properties. As a result, eavesdroppers can detect the
distortions in the resulting stego medium's statistical properties. The process of finding abnormalities in the
stego medium is called statistical steganalysis. Hence, this paper aims to review most of the existing cover50
communication techniques and present a comparative analysis on their level of complexity, limitations, data
embedding capacity, and requisite challenges.
1.1. Related Work
Considerable research has been done in the last decade to unveil various covert communications tech-
niques and related challenges. However, no existing work elaborately covers all the aspects of covert55
communications or steganography in a single research article. For instance, in 2011, [17] presented a com-
prehensive study on digital audio steganographic techniques in the temporal domain, transform domain, and
encoder domain. The authors explored the potential and limitations of these techniques to ensure secure
communication. Similarly, [16] carried out a comparative study of digital audio steganographic techniques
2

Table 1: Research Focus of Existing Surveys
Reference Year of
Publication Research Focus
[17] 2011 Digital audio steganographic techniques
[16] 2012 Audio steganography
[18] 2013 Image-based steganographic methods
[19] 2014 Briefly highlights spatial domain, transform domain, and
SS techniques
[20] 2016
Concisely enumerates different techniques of steganogra-
phy including text, image, network, audio, and video. How-
ever, just explain spatial domain, transform domain and
spatial domain methods
[21] 2017 Compressed and raw video steganographic techniques
[22] 2018 Momentary discussion of audio steganography
[23] 2020
Addresses the problem of joint covert communication and
secure transmission in untrusted relaying networks when
multiple wardens exist in the network
[24] 2020
Provides an overview of state-of-the-art covert communi-
cation techniques and then introduces the fundamentals
of Intelligent Reflecting Surface (IRS). Also elaborates on
how an IRS can be integrated to benefit communication
covertness
[25] 2020 Text-based steganography
in 2012. The researchers evaluated the performance of various methods based on robustness, security,60
and embedding capacity indicators. Another contribution of this work is the provision of a robustness-based
classification of steganographic models depending on their occurrence in the embedding process.
Correspondingly, a critical analysis of various image steganographic techniques covering a brief overview of
each technique, major types, their classification, and applications was presented in [18] in 2013. In the same
way, in 2014, researchers in [19] briefly highlighted some spatial domain, transform domain, and SS domain65
steganography techniques. The authors also enumerated some factors affecting steganographic methods.
Likewise, [20] summarily enumerates different techniques of steganography, including text, image, network,
audio, and video. However, the article discusses just a few techniques in detail, such as spatial domain and
transform domain methods. In another work in 2017, [21] presented a comprehensive study and analysis
of numerous cutting edge video steganography methods and their performance evaluations. The article70
reviewed both compressed and raw video steganography. In the compressed domain, video steganography
techniques are categorized according to the video compression stages: intraframe prediction, inter-frame
prediction, motion vectors, transformed and quantized coefficients, and entropy coding. On the other hand,
raw video steganography methods are classified into spatial and transform domains. The authors also
highlighted some research directions and recommendations to improve on existing video steganography75
techniques.
3

Similarly, a study conducted in 2018 outlined various techniques of concealing “mystery facts” in sound doc-
uments utilizing sound information concealing systems [22]. The authors also highlighted different strategies
and the embedding capacity of various audio steganography methods. In addition, numerous survey papers
published in the last two years have explored some state-of-the-art covert communications technologies.80
In such an endeavor, researchers in [23] addressed the problem of joint covert communication and se-
cure transmission in untrusted relaying networks when multiple wardens exist in the network. The primary
aim of this research is to prevent the untrusted relay from decoding the source signal. To satisfy these
requirements, the study proposes that the destination and the source inject jamming signals during the
source-to-relay and relay-to-destination transmission phases, respectively. The researchers also suggest a85
power allocation strategy to maximize the secrecy rate and satisfy the covert requirements.
Furthermore, [24] provides an overview of the state-of-the-art covert communication techniques. However,
the main objective of this research is to introduce the fundamentals of Intelligent Reflecting Surface (IRS)
and elaborate on how an IRS can be integrated to benefit communication covertness. Correspondingly,
researchers in [25] presented a technique for the reconstruction of printed documents using text-based90
steganography. This technique can be used if the original documents (e.g., bank checks, legal documents,
and certificates) are torn apart, and the important information is lost. Extracting the information from the re-
maining part of the document and recreating the document will help in regaining the lost information. Hence,
the proposed method may help in the reconstruction of printed documents from their part or whole.
As shown in Table-1, although the existing literature individually covers most of the basic steganography95
techniques, there is a requirement to uncover the latest trends in covert communications and outline related
limitations, future challenges, and measures to increase covertness. Hence, this study presents a compre-
hensive review of current techniques that covertly embed secret messages in the cover medium. The main
contributions of this research include:
• A comprehensive survey of RF-based covert communication techniques.100
• An extensive review of image, video, audio, and text steganography.
• Review of numerous non-conventional covert communication techniques.
• Highlight limitations of reviewed techniques and illustrate future research challenges.
2. Covert Communication Techniques
2.1. Radio Frequency (RF) based Techniques105
As shown in Table-2, some eminent RF-based covert communication methods encompass:
a. Directional Transmission: Directional Transmission using multiple antennas techniques can increase
the stealthiness of wireless communications [26]. In this context, a beamformer is used to adjust
the phase and amplitude of the signals on each element of an antenna array so that the signal is
radiated in the desired direction only. Consequently, the transmitted signal is received unstintingly by110
the intended recipient. At the same time, the same signal has a null value to achieve a LPD in all the
other directions. However, the performance of beamforming techniques depends on the availability
of the Channel State Information (CSI). An estimation error can result in inaccurate CSI at a multi-
antenna transmitter, which can further cause signal leakage with high probability, thus degrading the
effectiveness of covert communication. Nonetheless, the adverse effects of inaccurate CSI can be115
reduced by increasing the number of antennas at the transmitter side, i.e., more than a hundred [27].
4

b. Artificial Noise Generation: Artificial Noise (AN) can be generated to disguise the existence of covert
channels [28]. The key idea here is to divert the jamming signals from the legitimate covert channels.
Correspondingly, the transmitter uses its multiple antennas to transmit AN in the null space of the
desired receiver's channel, and thus, not affecting the receiver, while at the same time degrading the120
adversary's channel [29]. There are numerous means of generating AN, e.g., cooperative jamming,
full-duplex jamming, and AN injection. In cooperative jamming, one or more parties can coordinately
jam adversary's channel while causing no harm to the legitimate covert transmission. However, coop-
erative jamming incurs communication and synchronization overhead for transmitting power control.
Nonetheless, the issues of cooperative jamming can be addressed by full-duplex jamming.125
The AN injection technique involves the simultaneous transmission of the AN signal and the informa-
tion signal. The AN signal, in this case, is orthogonal to the information signal so that only adversary's
channel is affected. The intention here is to balance the tradeoff between covertness and information
rate by optimizing the transmit powers of information and jamming signals.
c. Cooperative Relaying: It employs numerous relay nodes to forward a low-powered transmitted signal130
over longer distances [24]. The main objective is to avoid the detection of high-powered covert signals
transmitted over longer ranges. However, the use of third-party relaying nodes makes the deployment
scalability of this technique relatively low in comparison to other approaches.
d. SS: To make it difficult for an adversary to identify information signals, a SS approach can be used [24].
In this technique, an information signal is modulated on a pseudo-noise sequence spread over a wide135
frequency band. Hence, the Spectral Power Density (SPD) of the information signal is suppressed
below the noise floor level, which lowers the signal detection probability. The two widely used modula-
tion techniques in this regard include direct sequence and frequency hopping. The direct sequence is
considered more secure against malicious detection as it continuously keeps the SPD of the transmit-
ted signal at a low level. On the other hand, frequency hopping employs narrowband signals with high140
SPD at any given frequency hop; as a result, it is vulnerable to detection by the adversary. Broadly, the
SS approach with high deployability ensures the robustness of covert communication against fading.
e. Millimeter-Wave Communications: Millimeter-Wave (mmWave) communications feature steerable
narrow beams that operate in the frequency band of 30-300 GHz [24]. The mmWave frequency band
offers ultra-fast communication but at short ranges. Nonetheless, the directionality of narrow beams145
facilitates covert communications as to intercept a mmWave communication, the adversary can only
detect the intermittent misaligned beams. The on and off behavior of the misaligned beams and the
overhead for signal detection in a wide frequency band make the adversary's detectability very difficult.
However, there is a limitation associated with mmWave communication. Due to reduced scattering
and diffraction abilities, the narrow beams attenuate rapidly and are also vulnerable to obstacles. In150
addition, short-wavelength covert communication mainly depends on the availability of Line of Sight
(LoS). It is also affected by mobility, as the Doppler shift of mmWave is strong even at walking speed.
f. IRS: Researchers in [24] proposed an IRS-based covert communication to tackle the issues of high
bandwidth and energy usage. Based on a tunable metasurface, whose electromagnetic (EM) charac-
teristics can be re-engineered without re-fabrication, IRS can intensify/strengthen specific information155
signals by reflecting them in phase with the intended receiver. On the contrary, IRS reflects the de-
sired signals in the opposite phase as that of the unintended receivers to prevent information leakage
and interference. Hence, an increased transmission rate is achieved in the predetermined direction
5

with a LPD in the unwanted direction. Moreover, IRS can also be configured to reduce the effects
of co-channel interference and eavesdropping. IRS features low-cost fabrication with nearly passive160
elements. As a result, it can be powered through microwave energy harvesting. Besides, the devel-
opers of this technique claim it to be free of self-interference, with reduced energy requirements and
computational complexity. IRS also satisfies heterogeneous Quality of Service (QoS) requirements
such as stable connectivity, improved data rate, and higher spectral efficiency.
Certain factors may affect the performance of the IRS. E.g., the wave manipulation of an IRS is de-165
pendent on the CSI to enhance covert communication. However, instantaneous CSI of the reflection
channels is difficult to acquire due to an IRS's nearly passive operation. In addition, being lightweight
and having an ultra-thin footprint, IRSs can be deployed on the facade of various environmental ob-
jects, including walls, vehicles, and smart clothing. However, these objects tend to produce perplexing
spatial patterns. Resultantly, the propagation environment is jointly shaped by multiple IRSs.170
g. Wavelet Packet Transform (WPT) based OFDM System over Rayleigh Fading Channel. Re-
searchers, in [30] proposed a covert communication technique integrated into the physical layer of
the WPT-OFDM system. The crucial secret information is first encrypted and then transmitted along
with the baseband carrier signal through OFDM. The researchers used WPT-OFDM due to its better
handling of signal disturbance, more tolerance to inter-carrier interference, efficient use of bandwidth,175
improved Bit Error Rate (BER), and higher data rate than Fast Fourier Transform OFDM.
WPT is implemented using low-pass and high-pass filters followed by down sampling to make the
transform efficient. The WPT of a signal is calculated by passing it through a series of filters with
the impulse response. Similarly, the encrypted and embedded data streams are up-sampled at the
transmitter and passed through an Integral Wavelet Packet Transform (IWPT) filter bank. Whereas180
at the receiver end, the data is passed through a filter-bank to extract secret data. On the contrary,
the covert communication of conventional FFT-OFDM has disadvantageous in terms of reduced spec-
tral containment due to added Cyclic Prefix (CP). Moreover, FFT-OFDM subcarriers' orthogonality is
severely affected by the fading channel. Nonetheless, WPT-OFDM is more resilient against these
channel variations. Correspondingly, covert communication in WPT-OFDM is more robust to narrow-185
band interference and multipath propagation loss.
The experimental results have shown that the WPT-OFDM-based covert communication performs bet-
ter than the FFT-OFDM technique in terms of improved BER of the carrier data, Peak Signal-to-Noise
Ratio (PSNR), Mean Square Error (MSE), and accuracy of secret information. In this regard, covert
communication using WPT-OFDM provides an improved Signal-to-Noise (SNR) of 3-6 dB compared to190
FFT-OFDM for Rayleigh fading scenarios. However, WPT-OFDM is considered to be computationally
complex than conventional FFT-OFDM.
h. Covert Communication and Secure Transmission over Untrusted Relaying Networks: Authors in
[23] presented a joint covert and secure communication over untrusted relaying networks in the pres-
ence of multiple wardens (adversaries). The wardens are always on the lookout to detect any message195
being sent from the source to a destination. If the wardens detect any information being transmitted
from the source to the destination, they can launch further hostile network attacks. Correspondingly,
the actors in the experiment scenario include a multi-antenna source, a single antenna destination
station, 3-4 Amplify-Forward relay stations, and 3-4 wardens (adversaries). The ultimate objective of
the proposed framework is to hide the communication process from the wardens and prevent the de-200
6

coding of source station's secret message by the untrusted relays.
Hence, to achieve the desired objectives, the researchers suggested a jamming strategy. The pro-
posed scheme operates in two phases. In Phase-1, the source transmits secret information with
maximum-ratio transmission beamforming towards the untrusted AF-Relay. To decrease the probabil-
ity of detection by the wardens and decrease the achievable Signal to Interference and Noise (SINR)205
ratio at the relay, the destination station injects jamming signals into the network. The goal is to keep
the wardens guessing whether the source transmitted data or not. Whereas, in Phase-2, the source
introduces jamming signals to prevent the wardens from detecting whether the AF-Relay is forwarding
the signal to the destination or not. Moreover, the covert communication between the source and the
destination is performed through a discrete-time channel with T timeslots, i.e., Time Division Multiple210
Access (TDMA) protocol. The source transmits in each timeslot a symbol of length n. Where n is the
length of the codeword, hence, based on a shared secret, the source, AF-Relay, and the destination
determine the exact timeslots in which the source would transmit data. However, the critical factor
here is that all the nodes should be accurately synchronized.
The researchers performed a thorough numerical analysis of the proposed scheme, and the key find-215
ings are that the secrecy rate increases with an increase in the number of AF-Relays. On the contrary
secrecy rate decreases with the increase in the number of wardens.
i. Covert Communication by Exploiting Node Multiplicity and Channel Variations: To avoid detec-
tion of secret messages in wireless transmissions, researchers in [31] exploited the multiplicity of users
scattered across the wireless network and the channel variations caused by their mobility. Moreover,220
the proposed scheme does not require a shared secret among the network nodes. The basic idea is
that the transmitter hides the secret message in the non-covert public message by exploiting the node
multiplicity and channel variations in wireless broadcast networks. The covert message is superim-
posed on the public message and transmitted in such a way that the total transmit power remains the
same as it was without the covert message. Hence, it is difficult for an adversary to detect covert com-225
munication unless the non-covert public message is decoded. The researchers keep the throughput
loss on the non-covert message to the minimum by transmitting the hidden message only when the
transmission rate of the non-covert public message is low.
The authors also prove that the probability of total detection error approaches one with increasing the
number of non-covert users. Also, the detection error probability rises with the increase in the transmit230
power. Hence, the hidden message is transmitted at a higher rate. On the contrary, the existing LPD
techniques achieve a low covert rate due to transmission requirements at low power. Nonetheless, the
proposed scheme has a high complexity level.
j. Covert Communication in Downlink NOMA Systems with Random Transmit Power. According to
the latest study in [32], Non-orthogonal Multiple Access (NOMA) technique can be used for downlink235
covert communication in Fifth Generation (5G) wireless networks. The researchers suggest that Alice
can transmit a covert message at low power to a user D1 under cover of another message transmitted
to a user D2 at random power. The random power message sent to D2 prevents the adversary W from
detecting covert communication between Alice and D1 based on received power. This scheme has
not been tested for multiple user covert communication scenarios.240
k. Keyless Covert Communication via CSI: The researchers in [33] consider the problem of covert
communication when the channel state is available either non-causally, causally, or strictly causally,
7

either at the transmitter alone or at both transmitter and receiver. Accordingly, the researchers infer a
covert capacity region when the channel state is available at the transmitter and receiver. On the other
hand, if the channel state is only available at the transmitter, they formulate inner and outer bounds on245
the extent of covert communication. Subsequently, the researchers prove that covert communication
is possible once CSI is known, but it has zero probability without it.
Correspondingly, the authors propose a strategy for covert communication over a state-dependent
channel without any external key shared between the transmitter and the desired receiver. Shared
randomness is obtained from the CSI to form a secret key between the transmitter and the receiver in250
such a way that it remains hidden from the warden. However, the limitation of the proposed strategy
is that the warden should not have any knowledge of the CSI. Moreover, the proposed technique has
high complexity when deployed in real-world information systems.
l. CloakLoRa - A Covert Channel over LoRa Physical Layer: Long Range (LoRa) is a proprietary
Chirp Spread Spectrum (CSS) modulation technique that is designed to enable low-power IoT de-255
vices to communicate with each other at long ranges. Correspondingly, researchers in [34] introduced
CloakLoRa, a covert channel over LoRa physical layer. CloakLoRa embeds secret messages infor-
mation into a regular LoRa packet by modulating the amplitudes of LoRa chirps while keeping the
frequency intact. Where the Amplitude Modulation (AM) is orthogonal to CSS. Hence, a legitimate
receiver can easily demodulate the original CSS information. On the other hand, the covert receiver260
just focuses on the variation of received signal strength and extracts the covert information.
The authors also identified that the AM modulated LoRa chirps are transparent and covert to existing
security mechanisms. But, at the same time, they can be detected and decoded by a receiver. One
of the reasons is that LoRaWAN only protects the end-to-end communication at the network layer and
above, while many physical layer parameters, including the amplitude variations, are largely ignored.265
However, keeping in view the receiver's capability to decode the information correctly, the effective
range of the covert communication is approximately 250 m with the BER of 0.43%. Moreover, it is also
observed that the transmission time of secret data decreases with the increase in the transmission
rate of regular LoRa packets. In addition, the transmission time of hidden data also decreases with the
increase in the payload size and by increasing the alternating rate of on-off states of covert packets.270
Moreover, to achieve effective covertness, there needs to be a balance between modulation depth and
the BER, such that the receiver easily decodes the message with a reduced possibility of detection.
Table 2: RF-based Covert Communication Techniques
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
1.
Directional
Transmission
[26]
Multiple antennas can be
used to increase
stealthiness of transmission
by means of beamforming to
produce spacial selectivity
Inaccuracy of CSI at a
multi-antenna transmitter
due to estimation error could
result in a high probability of
signal leakage
High
Continued on next page
8

Table 2 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
2. AN generation
[28],[29]
Multiple-antenna techniques
can be employed to avert
adversary's detection
capability by directional
jamming
Cooperative jamming infers
synchronization and
communication overhead
(for TX power control), and
also provides low
deployment scalability.
Self-interference in case of
full-duplex jamming
Low for
cooperative
jamming
and high for
full duplex
and AN
injection
3. Cooperative
Relaying [24]
Utilizes intermediate nodes
to relay low power
transmissions in short hops
Depends upon third party
nodes to act as relay.
Hence, it has low
deployment scalability
Low
4. SS [24]
The covert communication
signal is modulated on a
pseudo-noise sequence
using direct sequence
technique, thus spreading
the transmission bandwidth
as compared to the
bandwidth required by
normal narrow-band signals
The frequency hopping
based SS technique is not
suitable for covert
communication as it makes
use of narrow-banded
signals with high PSD on
any frequency hop
High
5.
mmWave
Communications
[24]
Covert signals are
transmitted over high
frequency (30-300 GHz mm
Waves) steerable narrow
beams, realized by small
antennas. The directionality
of the narrow beam naturally
benefits covertness by
suppressing signal leakage
due to imperfect beam
patterns
The mm Waves experience
weakened scattering and
diffraction abilities which
make them attenuate acutely
and are also susceptible to
obstacles
High
6.
IRS-enabled
covert
communications
in wireless
networks [24]
Based on re-programmable
metasurfaces, IRS can
strengthen or cancel the
information signals in the
intended directions
The wave manipulation of an
IRS is dependent on the
CSI. Whereas,
instantaneous CSI of the
reflection channels is difficult
to be acquired due to the
nearly passive operation of
an IRS
Low
Continued on next page
9

Table 2 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
7.
WPT OFDM
system over
Rayleigh fading
channel [30]
The WPT of a signal is
calculated by passing it
through a series of low-pass
and high-pass filters with
impulse response
Computationally complex High
8.
Covert
communication
and secure
transmission
over untrusted
relaying
networks in the
presence of
multiple wardens
[23]
The source and the
destination stations
introduce jamming signals in
the network to keep the
communication process
hidden from the adversaries
(wardens), and also to
prevent AF-Relays from
decoding the secret data
The proposed framework
has been analyzed based on
simulation results and
numerical analysis. Besides,
the nodes sharing the secret
(codeword) need to be
accurately synchronized
High
9.
Covert
communication
by exploiting
node multiplicity
and channel
variations [31]
The transmitter hides the
secret message in the
non-covert public message
by exploiting the node
multiplicity and channel
variations in wireless
broadcast networks
The detectability of covert
communication depends
upon the number of
non-covert users in the
wireless network
High
10.
Covert
communication
in downlink
NOMA systems
with random
transmit power
[32]
Alice transmits a covert
message at low power to a
user D1 under the cover of
another message
transmitted to a user D2 at
random power
This scheme has not been
tested for multiple user
covert communication
scenario
High
11.
Keyless covert
communication
via CSI [33]
Covert communication is
made possible over a
state-dependent channel
without any external key
shared between the
transmitter and the desired
receiver
Warden should not have
knowledge of the CSI High
Continued on next page
10

Table 2 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
12. CloakLoRa [34]
A covert message is sent
over LoRa Physical layer by
modulating the amplitude of
the LoRa chirps by keeping
the frequency intact
Communication range of just
250 m Low
Figure 1: Image Steganography.
2.2. Image Steganography
In image steganography, a “Host Image” also known as a “Cover Image” is used to carry the secret275
message. As shown in Figure-1, the process/algorithm of hiding a message in an image is called an “Em-
bedding Technique”, and the mechanism of recovering the original message from the stego-image is called
as “Extraction Technique”. The use of a secret shared key between the communicating parties is optional.
Correspondingly, there are numerous image steganography methods which differ from each other based
on their embedding capacity (the size of message they can carry), the image quality, i.e., the difference280
between stego-image and the original image, and lastly, the resilience of the method to various steganalysis
(detection) attacks [35]. Similarly, the secret message and the cover image may be used in different forms
depending upon the level of security required, e.g., raw, compressed, or encrypted form. Accordingly, as
shown in Table-3, some of the significant image-based steganographic methods include:
a. Least Significant Bit (LSB) Technique: This is the most widely used image steganography technique285
[36]. In this method, the LSB of a selected (may be randomly selected) pixel is replaced with one bit
of the secret message. The complete process is shown in Figure-2. This technique can embed larger
messages without affecting the visual quality of the stego-image as compared to the original cover
11

Figure 2: LSB Image Steganography.
image. To improve the security and the visual quality of the stego-images, numerous variations of
the LSB technique have been proposed. Such as adaptive LSB [37], stego-key directed adaptive290
LSB [38], three-level-encrypted algorithm using cyclic18 LSB substitution [39], highly undetectable
embedded method (HUGO) [40], and multi-cover adaptive LSB [41]. However, in the LSB technique,
the size of the embedded message is directly related to the visual quality of the stego-image. Hence,
if more secret bits are embedded in the cover image, the quality of the stego-image would be low.
b. Pixel Value Difference (PVD) based Steganography: To address the issue of message size and the295
stego-image quality, researchers in [42] proposed a PVD based technique. In this method, the differ-
ence of value between two adjacent pixels dictates the number of secret bits that can be embedded
in the cover image. As shown in Figure-3, the difference between two neighboring pixels is 48. The
number of secret bits that can be embedded depends on the difference value of the pixels in the cover
image and the respective group of a range of these differences. E.g., in this case, the difference value300
of 48 falls in the group of [32, 63]. Now, based on the range of difference values in this group, we can
compute the number of secret bits that can be embedded by using the formula log2(63-32+1). Here
we get 5 bits; hence, we embedded 01100 (1210 with a new difference value of 44, which falls in the
same range of difference values as that of the original cover image.
Besides, pixel pairs in edge areas should be used as they can accommodate more message bits.305
Moreover, pixel adjustment as per octonary pixel pairing scheme [43] helps in achieving statistical
undetectability. However, PVD methods are considered to have low stego-image quality.
c. Exploiting Modification Direction (EMD): In this method, the cover image is partitioned into multi-
ple groups containing n pixels to insert the secret digit in (2n+1)-ary encoding system [44]. During
the embedding process, the rate of a certain pixel inside a group is adjusted by +
-1. However, the310
disadvantage of this scheme is the low image quality because the size of the group has two pixels.
Accordingly, the embedding capacity of the initial EMD technique was 1.16 bits per pixel (bpp). Hence,
12

Figure 3: PVD Embedding Technique.
to address the issue of image quality, researchers in [45] proposed a Huffman Coding-based tech-
nique. In this method, the group of pixels used for embedding secret messages is segmented into
two subgroups containing 2 and 3 pixels sequentially. Thus increasing the payload without affecting315
the image quality. However, EMD is vulnerable to the histogram and Regular Singular (RS) detection
analysis.
d. Multi-Base Notation System (MBNS): This technique involves the conversion of the secret message
into symbols that are expressed in a multiple-base notational system, i.e., binary, decimal, and octal
number system. Later, these symbols are embedded into pixels intensities. Generally, the larger no-320
tational base symbol indicates the larger embedding rate. There have been many improvements pro-
posed to the basic MBNS scheme to increase the visual imperceptibility, and the embedding capacity
[46], [47]. However, there is a need for strong synchronization between the sender and the receiver to
select bases. Moreover, it is susceptible to SPAM feature-based steganalysis for embedding capacity
>1 bpp, and it also has a high level of complexity.325
e. Pixel Pair Matching (PPM): In this method a pixel pair (Pj1, Pj2) is used as a reference coordinate
to look for another pixel coordinate (P’j1, P’j2) within a specific neighborhood of φ(Pj1, Pj2) . This
process must satisfy the expression f(P’j1, P’j2) = SB, where f is an extraction function and SB is
the secret message in B-ary notational system. Accordingly, data embedding involves replacing (Pj1,
Pj2) with (P’j1, P’j2). However, this method has less embedding capacity and low visual quality.330
Therefore, improving upon the previous work, a PVD-based Patch Reference Table (PRT) technique
was developed [48]. In this method, a unique embedding sequence was designed, and the number of
secret bits to be concealed was estimated by the pixel-value difference. It could keep the histogram
with a limited embedding capacity of 0.5 bpp and weak security against SPAM-based steganalysis.
Similarly, researchers in [49] proposed a PVD and PPM-based technique to achieve higher payload335
capacity and better visual quality.
f. Gray Level Modification (GLM): Researchers in [50] suggested an approach in which a secret mes-
sage is embedded in the cover image by modifying the gray level of selected pixels. Correspondingly,
GLM maps secret data bits to the odd (shaded blocks) and even values of the selected image pixels.
13

Figure 4: Gray Level Modification based Embedding.
In this procedure, first, the pixels are selected based on a mathematical function. Then the values340
of the pixels in the set are made even. E.g., in Figure-4, we have 8 bits of secret data. Based on
this, any linear equation can be used to select a set of 8 pixels from the cover image. Now the set
of these 8 pixels' have both even and odd numbers. Hence, a slight adjustment is made to turn the
odd numbers into even by adding one. After this step, the mapping phase starts. Now for each bit in
secret data, the value of the respective pixel is made odd by subtracting one. Hence, the odd numbers345
represent 1s, and even numbers represent 0 bits in the final set of embedded pixel values. Though
this method yields high visual quality however it is vulnerable to statistical steganalysis attacks. For
additional security, An advanced version of GLM [51] employed Multi-Level Encryption (MLE) before
GLM mapping.
g. Pixel Value Prediction (PVP): Direct altering of pixel values is believed to be not very effective as it350
results in distorted images while increasing the embedding capacity. Hence, the PVP technique was
proposed by [52]. In this technique, the pixel intensities are predicted through predictors instead of
direct modifications. Later, the predicted error values are modified to compensate for the secret mes-
sage. Correspondingly, researchers in [53] and [54] suggested prediction error and Modified Prediction
Error (MPE) based approaches. The proposed method modifies the histogram of the prediction errors355
to find an appropriate location for embedding secret data. In addition, [55] recommended a predictive
coding-based reversible embedding method.
In this technique, the secret data is embedded into compress codes that are used during lossless
image compression coding. Later, at the predictive coding stage, the secret data is embedded into
error values by referring to a hiding tree. On the other hand, the hidden data can be recovered by360
14

Figure 5: Histogram of an Image [56].
referring to the hiding tree at the entropy coding stage. Nonetheless, the proposed method is limited
to an embedding capacity of 0.0992 bpp.
h. Histogram based Steganography: Histogram shifting is considered to be one of the most effective
methods of secret data embedding. In this context, as shown in Figure-5, the histogram of an image
is the gray-scale value distribution showing the frequency of occurrence of each gray-level value [56].365
Secret data is embedded in histograms in two stages. In stage one, the peak and zero points are
located in the cover image. Then, the bins are shifted with one level between the zero and peak points
for emptying peak points. In the second stage, the secret bits are hidden by specific alterations in the
new peak point and the empty point.
To improve embedding capacity, researchers in [57] proposed a reversible data embedding approach.370
In this method, based on histogram shifting, the neighboring points of the peaks were used to hide
the secret data without affecting the peak values. Moreover, to increase the embedding capacity, the
researchers used the concept of localization to generate more peak points. Hence, more neighboring
points were available to conceal data. It is further suggested that the embedding capacity can be
further improved by incorporating localization with multi-layer embedding. However, still, this technique375
has achieved the embedding capacity of only 1 bpp.
i. Edge based Steganography: Generally, the image-based steganographic techniques experience vi-
sual distortion of stego-images as a result of direct modification of pixels in the cover image, especially
in the smooth areas. Hence, to achieve minimum image distortion and high imperceptibility, edge
adaptive embedding schemes were created [58]. Accordingly, confidential data is embedded in the380
edges of the image, i.e., complex and light edge textures regions are used to hide larger and smaller
payloads, respectively. Though the proposed approach was considered secure, the initial version of
this method suffered from low embedding capacity.
15

Hence, researchers in [59] proposed a semi-reversible embedding technique in which the cover im-
age is interpolated and divided into the edge and non-edge regions. Subsequently, the secret data is385
embedded in the edge areas based on a threshold value. Besides, in the case of non-edge regions,
the difference in values of two neighboring pixels was used to estimate the payload capacity. The
proposed method did maintain the visual quality while increasing the embedding capacity. However,
this scheme was not analyzed from the security point of view. Therefore, later [60] suggested a canny
edge-based approach. In this method, a canny edge detector was used to identify the edge of the390
cover image. Subsequently, the secret data was embedded only in the edge pixels. In addition, the
researchers utilized Huffman coding and a unique sorting technique to randomize the selection of
edge pixels for secret data embedding. Though this technique yields an intact histogram, it has high
computational complexity and low embedding capacity, i.e., less than 1 bpp.
j. Mapping based Steganography: Mapping-based image steganography is a technique in which a395
pixel, block, or bit plane of secret data is matched with the cover image data. Accordingly, [61] pro-
posed a two-way block mapping technique by dividing the secret data and the cover image into m∗n
blocks. These blocks are selected based on a high level of similarity between the secret image and
the cover image. After the block selection, the index information of matched and unmatched blocks
is compressed and embedded through distributed LSB to provide a less distorted image. Further im-400
proving on the mapping concept, researchers in [62] devised a mechanism to map English alphabets
to the pixel values using a mapping table. This approach maintained the mapping pattern as the same
is required for extraction as well. The proposed scheme is believed to have low computational com-
plexity.
Similarly, to improve the security and the visual quality of the stego-image, authors in [63] suggested405
a bit-plane mapping approach. This scheme comprised two phases; in phase 1, the secret data was
compressed by using a size reduction algorithm. And in phase 2, the compressed data was embedded
through Fibonacci representation in pixel intensities. This technique helped in reducing the embedding
distortion of stego pixels. Later, the researchers also proposed a virtual bit-plane mapping technique
[64], with the aim to improve the visual quality and embedding payload capacity.410
k. Pixel/Block Indicator based Approach: An RGB (Red Green Blue) based color image consists of 3
bytes of red, blue, and green color intensities. One of these channels is used as an indicator, and the
rest of the two are the data channels. Accordingly, researchers in [65] proposed a block-based RGB
indicator data embedding method. In this scheme, a user-defined key decides the one-to-one mapping
of secret data and the cover image. Accordingly, the cover image and the secret data are divided into415
eight blocks each. During the embedding process, one channel is considered as an indicator and
others as data channels. The secret data is embedded on LSB by the maximum matching portion to
reduce the visual distortion in the stego-image.
To further improve the payload embedding capacity and security, [66] introduced a technique in which
the cover image is scrambled and employed by PVD and adaptive LSB according to blue pixel indicator.420
In comparison, the red and green planes are utilized as data channels. The proposed method improved
the payload and security by scrambling the red and green planes. However, the visual quality is
dependent on embedding capacity.
l. Color Model-based Steganography: Research scientists have recently devised a mechanism to
use correlation of color spaces in RGB images for embedding secret data. Likewise, [67] introduced425
16

Figure 6: QR Image.
an Adaptive LSB (ALSB) embedding technique that used uncorrelated color spaces. The proposed
method endeavors to increase security by minimizing the chances of detection by human vision. To
embed secret data, firstly, the cover image was scrambled and then converted into Hue Saturation
Value (HSV) color space. After that, an ALSB method was used to conceal secret data inside the
V-plane of the HSV color model.430
Correspondingly, to improve the visual quality and security [68] suggested a unique Magic LSB Sub-
stitution Method (M-LSB-SM) for RGB-based colored images. The proposed method used a neutral
component (I-Plane) of the Hue-Saturation-Intensity (HSI) color model with MLE to embed secret data.
However, there is a question about the security of this scheme against steganalysis.
m. Machine Learning (ML) based Steganography Technique: To enhance the embedding capacity of435
image-based steganographic techniques, certain intelligent optimization techniques have also been
developed. Correspondingly, [69] introduced a three-phase approach to improve the visual quality and
payload capacity in color images. In the first phase, a Learning System (LS) comprising Adaptive
Neural Network (ANN) and a Genetic Algorithm (GA) was applied to estimate the number of secret
data bits to be embedded inside the cover image pixels. Whereas, other two phases of the proposed440
method were applied after the embedding. The proposed method achieved an embedding payload
of around 12 bpp with good visual quality. Moreover, researchers also made use of chaotic maps to
improve data embedding techniques based on GA [70].
n. QR Code Image-based Covert Communication: Based on the ubiquitous use of Quick Response
(QR) codes in our daily life, researchers in [71] introduced a QR code image based secret data em-445
bedding technique. As shown in Figure-6, QR code is a two-dimensional bar code that comprises
numerous black and white squares (also called modules) which are of the size of 4x4 pixels. Initially
designed by Denso Wave to track vehicles and parts during the manufacturing process, it can store
up to 7089 digits, or 4296 characters [72]. However, the more the data is added, the more the size
increases, and its structure becomes further complex. One of the remarkable features of the proposed450
17

Figure 7: Horizontal Secret Data Bits Embedding [71].
Figure 8: Application of Zigzaggedness Remedy Scheme [71].
method is that a normal QR reader can read the facial data from the stego-image as usual. Whereas
a specially designed extraction algorithm can be used to extract the hidden data.
The proposed approach conceals secret data into the QR image by making minor alterations in the
boundary positions at the junctions of black and white modules without destructing the original module
structure of the QR image. As shown in Figure-7, the secret data bits are embedded by adjusting the455
content of every two neighboring modules of different colors to embed a message bit of ‘1’; or leave it
untouched if the bit to be embedded is ‘0’ or if the two modules are of an identical color. The scheme is
claimed to be resilient to noise attacks, and it also employs a zigzaggedness remedy scheme (shown
in Figure-8) to avoid destruction of the original module structure in the stego-image. However, the
18

stego-image is resistant to attacks of up to 8% pepper-and-salt noise [73]. However, the proposed460
technique yields an embedding capacity of up to 14000 bits based on the version of the QR code
image.
Table 3: Image-based Steganography
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
1. LSB technique
[41]
The LSBs of the selected
pixels in the cover image are
replaced by the secret
message bits.
The size of embedded
message is directly related
to the visual quality of the
stego-image.
Low
2.
PVD based
image
steganography
[42]
The difference of value
between two adjacent pixels
dictates the number of secret
bits that can be embedded in
the cover image
Low image quality Low
3. EMD based
methods [44]
In this method, the cover
image is partitioned into
multiple groups containing n
pixels to insert the secret
digit in (2n+1)-ary encoding
system
Vulnerable to histogram and
RS detection analysis High
4. MBNS [46]
Secret data is converted into
symbols, that are further
expressed in multibase
system such as binary,
decimal and octal.
Susceptible to SPAM
feature-based steganalysis
for embedding capacity >
1bpp
High
5. PPM [49]
A pixel pair is used as a
reference coordinate to find
another pair in a
neighborhood of a
predefined set of pixel pairs
Vulnerable to modern
steganalysis and also there
is a reference table overhead
High
6. GLM [50]
The gray level values of a set
of selected pixels is adjusted
according to the secret data
bit. Such that, the odd
values represent 1s and the
even values represent o bits
Vulnerable to statistical
steganalysis Low
7. PVP [55]
The pixel intensities are
modified based on a
predictive error based
steganography
Low embedding payload and
weak security high
Continued on next page
19

Table 3 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
8.
Histogram based
Data Embedding
[57]
First the histogram of an
image is generated. Then
the peaks and low points
(zeros) are located in the
histogram. The secret data
is embedded by
altering/shifting the peaks
and zeros
Performs better with smooth
images than heavy textured
ones. However still it has low
embedding capacity, i.e., 1
bpp
Low
9. Edge based
Technique [60]
Canny edge detector is used
to locate the edges in the
cover image. Later, secret
data is hidden in the
randomly selected edges
using Huffman coding
Low embedding capacity High
10.
Mapping based
Steganography
[63]
The secret data is
compressed, and then
mapped to cover image
through Fibonacci
representation in pixel
intensities
Computation overhead, and
low embedding capacity High
11.
Pixel/Block
Indicator based
Approach [66]
RGB-based images are
used as cover images. And
secret data is embedded in
data channels based on
pixel/block mapping. In
some cases a user-defined
key is used to divide the
cover image and the secret
data into desired number of
blocks
Limited payload capacity of
less than 1 bpp, vulnerable
to statistical steganalysis
due to LSB
Low
12.
Color Model
Based Method
[67], [68]
Using correlation of color
spaces in RGB images for
embedding secret data
Not evaluated against
steganalysis
High due to
MLE option
13.
ML-based
Techniques [69],
[70]
Use of ML algorithms such
as ANN and GA, to estimate
the number of secret data
bits to be embedded inside
the cover image pixels
Extensive computations are
involved High
Continued on next page
20

Figure 9: Data Hiding Opportunities in H.264 Codec [21].
Table 3 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
14.
Covert
communication
via the QR code
image [71]
Secret data is embedded
into the QR image by making
small alterations in the
boundary positions at the
junctions of black and white
modules without destructing
the original module structure
of the QR image
Secret message payload
depends upon the
zigzaggedness remedy.
Stego-image is resistant to
attacks of up to 8%
pepper-and-salt noise only
High
2.3. Video Steganography
Due to the ubiquitous use of the internet and digital media technologies, and redundancy in video data,465
digital videos are increasingly being used for secret message hiding. Video steganography techniques
(shown in Table-4) usually fall into two categories, compressed and raw (uncompressed) domains [21].
21

a. Compressed Video Steganography Techniques: Whenever video compression is under discus-
sion, H.264 video compression codec comes to the spotlight. H.264 video codec incorporates flexible
macroblock ordering, quarter-pixel interpolation, intra prediction in intraframe, deblocking filtering post-470
processing, and multiple frames reference capability. Moreover, H.264 comprises several Groups of
Pictures (GOP). Every GOP includes three types of frames: intraframe, predicted frame, and bidirec-
tional frame. Accordingly, video steganography approaches in the compressed domain are categorized
in line with video coding stages, including intra-frame prediction, inter-frame prediction, motion vec-
tors, transformed and quantized coefficients, and entropy coding. Correspondingly, some possibilities475
of secret data hiding in H.264 video codec standard are highlighted in Figure-9.
•Intra-Frame Prediction: As shown in Figure-9, during the video compression process, the input
video signal is split into macroblocks. These macroblocks are further encoded using a number
of intra-prediction modes. In the H.264 codec, the numbers of intra prediction modes are nine of
4×4blocks and four of 16 ×16 blocks, respectively. On the other hand, a High-Efficiency Video480
Coding (HEVC) codec can incorporate up to thirty-five intra-prediction modes. To hide secret
data, these modes are mapped to one or more secret message bits. Accordingly, [74] proposed
a four-stage data hiding approach in the compressed video domain using intra-frame prediction.
As illustrated in Figure-10, in stage 1, firstly, the video sequences are coded, and then Discrete
Cosine Transform (DCT) coefficients are computed. It is followed by the acquisition of motion485
vectors and intra-coded macro-blocks. Whereas the second stage focuses on the identification of
fluctuation scenes by performing scene detection on the consecutive intra-frames. The fluctuation
scenes are observed using a histogram variation of discrete cosine coefficients within intraframe
DCT coefficients.
Similarly, stage 3 involves the embedding of secret message bits using intra-frames of fluctuation490
scenes. Finally, in stage 4, statistical steganalysis techniques are employed on stego-video to
assess its security. However, the proposed method has low payload embedding capacity be-
cause it utilizes only the fluctuation scenes for data hiding. Correspondingly, researchers in [75]
introduced a modified version of H.264 codec based data embedding technique.
The proposed method embeds the secret messages into 4×4luminance block during the in-495
traframe coding process. Subsequently, a bit of secret message is concealed per block using
the 4×4intra-prediction modes. Accordingly, based on the pre-defined mapping rules between
the secret message and intra-prediction modes, the 4×4intra-prediction modes are divided into
two subsets to hide secret message bits, i.e., 0 or 1. However, it is believed that, generally, the
intra-frame prediction-based steganography techniques have low payload embedding capacity.500
•Inter-Frame Prediction: In reference to H.264 codec, inter-frame prediction for blocks of size
16×16,16×8,8×16,8×8,8×4,4×8, and 4×4are mostly used to conceal secret data by mapping
each block to a number of secret message bits. In such an endeavor, [76] introduced a secret
data hiding method by utilizing scene change detection in H.264 codec for four different block
sizes. Where each block is mapped to a pair of secret data respectively. Accordingly, the secret505
message comprises scene change frames that are further embedded into the encoded videos.
However, the payload embedding capacity of the intra-frame prediction is also very limited. E.g.,
let us suppose WA (Western Australia) is the secret information that must be embedded into the
inter-frame prediction blocks in H.264 codec. The goal of hiding secret message bits can be
22

Figure 10: Four Stage Process of Data Hiding using Intra-Frame Prediction.
achieved by using mapping rules of different block sizes. In this regard, Figure-11 illustrates the510
embedding process using mapping rules.
•Motion Vector-based Technique: This type of video steganography makes use of some pre-
defined motion vector characteristics to embed secret messages. These characteristics include
horizontal and vertical components, amplitude, or phase angles. Accordingly, [77] introduced a
steganography scheme that relies on intraframe, predicted frame, and bidirectional frame. This515
method conceals secret data in selected high magnitude motion vectors of the predicted frame
and bidirectional frames. Where each macroblock has a motion vector, on the other hand, the
control information, including capacity payload and segment range of each GOP, is embedded
into intra-frames. From the GOP point of view, every GOP has one intra-frame that carries the
control information required for the data extraction phase. In addition, every GOP contains pre-520
dicted and bidirectional frames that carry secret message bits in their motion vectors. Nonethe-
less, the proposed scheme has low payload embedding capacity as it relies only on the high
magnitude motion vectors.
Hence, improving upon the said scheme, researchers in [78] proposed a motion vector and lin-
ear block codes-based technique. In this approach, groups of motion vectors in each video525
inter-frame are selected based on a pre-defined threshold. Then phase angles are calculated to
obtain 0 or 1 values of the selected motion vectors. After this, the secret message bits are hidden
into the motion vector array using linear block coding. Here, the linear block codes are used to
increase the payload embedding capacity while minimizing the alteration rate of motion vectors.
Subsequently, the proposed scheme can hide four bits of secret message in every six bits of the530
motion vector array. However, this is still believed to be a low embedding capacity.
•Transform Coefficients based Technique: Because of their varying frequency coefficients, i.e.,
low, middle, and high-frequency coefficients, DCT, Quantized Discrete Cosine Transform (QDCT),
and Wavelet Transform coefficients of luminance component are suitable for secret data embed-
ding. Correspondingly, [79] proposed a DCT and communication theory-based reliable secret535
message hiding technique. In this method, the secret data is hidden into the DC (high energy)
23

Figure 11: Inter-Frame Prediction to Hide ’WA’ Characters.
and AC (low frequency) coefficients of the DCT coefficient. Further, to increase the scheme's
robustness, the researchers used the Bose, Chaudhuri, and Hocquenghem (BCH) codes and
soft-decision decoding. In addition, the said technique is tested to be resilient against StirMark
attack [80].540
Similarly, researchers in [81] presented an H.264 codec-based data hiding method that selected
some blocks as carriers for concealing secret message bits. The proposed method prevented de-
formation proliferation from neighboring blocks by relying on the prediction of intra-frame modes
of adjacent blocks. The researchers enhanced the performance of the proposed method by en-
coding the secret message bits using BCH codes before the embedding phase. The encoded545
information is then hidden into the 4×4QDCT coefficients using only a luminance plane of the
intra-frame. However, DCT/QDCT coefficients-based techniques have a low average embedding
capacity ratio of 0.09%.
•Entropy Coding based Steganography: Concerning H.264 video compression process, Con-
text Adaptive Variable Length Coding (CAVLC) and Context Adaptive Binary Arithmetic Coding550
(CABAC) entropy coding can be used as a secret data carrier. Accordingly, [82] used CAVLC
mapping for watermarking of MPEG-2 standard in a compressed domain. During the CAVLC
encoder, there are some run-level pairs that cannot systematically meet each other in intra-frame
blocks called unused pairs. The secret message is concealed into the codewords of unused
run-level pairs of the CAVLC entropy coding. This method yielded a low modification rate of the555
selected run-level pairs, which keeps the watermarked video's visual quality and bit-stream size
nearly unchanged.
24

Correspondingly, [83] utilized CABAC features to achieve real-time watermarking in H.264 codec.
The CABAC encoder uses a unary binarization, i.e., a process of concatenating all binary values
of syntax elements. Moreover, a certain number of motion vectors for both predicted and bidirec-560
tional frames are utilized for the secret message embedding using the CABAC properties. The
secret watermark is concealed by displacing the binary sequence of the selected syntax elements
orderly. This approach results in low visual quality degradation because of the minute difference
between the original code and the replacement code. In this context, only 1 bit is altered out of
8-bits of the selected motion vector. However, CAVLC or CABAC based steganography has low565
embedding capacity. E.g., [82] embeds 1 bit per 8×8intra block.
b. Raw Video Steganography Techniques: In the case of raw video steganography, the digital video
is first converted into a sequence of frames (still images), and then each frame is individually utilized
to embed secret data. Finally, the stego-video is generated by merging all the frames together. Raw
video steganography methods usually exist in spatial and transform domain both [68].570
•Spatial Domain-based Video Steganography: Currently, there are many spatial domain stegano-
graphic techniques that make use of pixel intensities to hide secret message bits. Some of these
methods include LSB substitution, Bit-Plane Complexity Segmentation (BPCS), SS, Region of
Interest (ROI), histogram manipulation, matrix encoding, and mapping rule. Accordingly, [84] pro-
posed an LSB steganography scheme using the reversible histogram transformation function. In575
this scheme, the secret message bits are concealed into the LSB pixels of the cover data. This
algorithm is robust against two x2-detection and regular-singular attacks, which are classified as
statistical steganalysis. The average embedding capacity ratio of this technique is 12%.
Similarly, [85] utilized BCH coding to hide secret message bits into a block of carrier object. Secret
data hiding is achieved by modifying different coefficients in the input block to set the syndrome580
values null. This technique yields enhanced embedding capacity and low computational com-
plexity as it reduces the complexity of the algorithm from exponential to linear. Moreover, this
approach is not robust against signal processing operations.
In another endeavor, researchers in [86] embedded secret data bits using a computer forensic
process. The secret message is first authenticated and then encrypted with a secret key, and585
then it is hidden into the 4 LSB of each pixel of the video frames. Moreover, to share the en-
cryption/decryption key with the other communicating party, the key is concealed into one of the
frames recognized by the sender and the recipient. The purpose of using the computer foren-
sic process is the validity of the obtained stego-videos. However, this technique is not robust
against signal processing, noises, and compressions due to the utilization of the spatial domain.590
Nonetheless, this approach has an embedded capacity ratio of 12.5%.
In the same way, [87] introduced a BPCS based steganography method. The proposed method
initially converts the video frame into 8-bit planes, and then each plane is further divided into
simple and noise-like regions. The critical difference between BPCS and LSB techniques is that
BPCS utilizes all of the 0-7 bit planes. On the other hand, LSB uses only a 0-bit plane. Hence,595
BPCS has more embedding capacity than LSB techniques, with an embedding capacity ratio of
41%. However, it is not robust against signal processing and noises.
Correspondingly, [88] proposed an improved BPCS data hiding technique. In contrast to the pre-
vious BPCS methods that compute the complex regions based on the total length of the black
25

and white border, this technique measures the complexity level of the selected regions by uti-600
lizing Canonical Gray Coding (CGC), run-length irregularity, and border noisiness. The secret
message is then concealed in the noise-like areas based on their degree of complexity. Though
this approach increases the average embedding capacity ratio to 45%, it is not resilient to signal
processing, noises, and compression.
•Video Steganography Techniques in Transform Domain: Video steganographic techniques605
operating in the transform domain transform each video frame into the frequency domain using
DCT, DWT, and Discrete Fourier Transform (DFT). Subsequently, the low, middle or high fre-
quency transformed coefficients are used to embed secret data. In this regard, [89] introduced
a BPCS steganography method using an Embedded Zerotree Wavelet (EZW) lossy compres-
sion. The proposed technique applies the BPCS steganography to DWT coefficient sub-bands610
containing different features by representing the original video frame's pixels using the DWT's co-
efficients. Moreover, each DWT sub-band is divided into bit-planes, and then the secret message
bits are concealed in the quantized coefficients. This approach achieves an average embedding
capacity ratio of 25%, and it is robust against lossy compression.
Similarly, [90] proposed a BPCS and Wavelet compression-based video steganography method.615
This method first divides each bit-plane of the secret message and the video frame into blocks of
8×8. It is followed by the selection of noise-like bit-plane blocks based on the degree of complexity
(a threshold value). Then secret message bits are concealed into the quantized DWT coefficients
by utilizing the BPCS method and applying two Wavelet compression techniques. This method
achieves a high embedding capacity ratio of 18% for 1 bit-plane and 28% for 2 bit-planes.620
Likewise, researchers in [91] presented another data hiding method based on Lazy Wavelet
Transform (LWT). In the proposed method, each video frame is first divided into four sub-bands
(separating the odd and even coefficients), and then the secret message is embedded into the
RGB LWT coefficients. For accurate extraction of embedded data, the length of hidden data is
concealed into the audio coefficients. Although this approach yields an embedding capacity ratio625
of 12.5%, this type of wavelet is not an actual mathematical wavelet operation. Consequently, it
is not robust against signal processing, noises, and compression. It is believed that the stegano-
graphic methods based on the transform domains improve resilience to signal processing, noises,
and compression. However, these techniques are more complex than the spatial domain meth-
ods.630
Table 4: Comparison of Video Steganographic Techniques
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
Video Steganography Techniques in Compressed Domain
1. Intra-frame
Prediction [74]
Low embedding capacity
with high impact on video
quality
Moderate
2. Inter-frame
Prediction [76] Low embedding capacity Low
Continued on next page
26

Table 4 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
3.
Motion Vector
based
Steganography
[77], [78]
Secret data bits are
embedded using motion
vectors and linear block
codes
Limited embedding capacity
(4 bits of secret data in every
6 bits of motion vector array)
with high impact on video
quality
Moderate
4.
Transform
Coefficients
based Technique
[81]
Use of DCT, QDCT, or
Wavelet Transform
coefficients of luminance
component are used for data
hiding. Further, DC and AC
coefficients of DCT
coefficient can also be used
to embed secret data bits
Low embedding capacity Low
5.
Entropy Coding
based
Steganography
[82], [83]
Use of CAVLC, and CABAC
entropy coding in H.264
codec to hide secret
message bits
Low data embedding
capacity. Quality of
stego-video is very
deteriorated
Low
Video Steganography Techniques in Raw Domain
1.
Reversible
Histogram
Transformation
Function [84]
Secret message bits are
concealed into the LSB
pixels of the cover data
Average embedding capacity
ratio of just 12% Low
2.
BCH Coding
based Technique
[85]
Utilizes BCH coding to hide
secret message bits into a
block of carrier object
Not robust against signal
processing operations Low
3.
Secret Data
Embedding
using Computer
Forensic
Process [86]
Authenticated and encrypted
secret data is hidden into the
4 LSB of each pixel of the
video frames
It is not resilient to signal
processing, noises, and
compression
Low
4.
BPCS based
steganography
method [87]
Video frame is converted
into 8 bit planes, where,
each plane is further divided
into complex regions. The
secret data is then
embedded into these
complex regions
Vulnerable to signal
processing and noises High
Continued on next page
27

Table 4 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
5.
Improved BPCS
Data Hiding
Technique [88]
Secret message bits are
hidden into noise-like
regions based on their level
of complexity
Not resilient to signal
processing, noises, and
compression
High
6.
BPCS
steganography
method using
EZW Lossy
Compression
[89]
BPCS steganography is
applied to DWT coefficient
sub-bands. Where, the
original video frame's pixels
are represented by DWT's
coefficients, and each DWT
sub-band is divided into
bit-planes. Subsequently,
the secret message bits are
concealed in the quantized
coefficients.
High complexity High
7.
BPCS and
Wavelet
Compression
based Video
Steganography
[90]
Secret message bits are
concealed into the quantized
DWT coefficients by utilizing
BPCS method and applying
two Wavelet compression
techniques.
High complexity High
8.
LWT-based Data
Hiding Technique
[91]
Each video frame is first
divided into four sub-bands
(separating the odd and
even coefficients), and then
secret message is
embedded into the RGB
LWT coefficients. For
accurate extraction of
embedded data, the length
of hidden data is concealed
into the audio coefficients
Not robust against signal
processing, noises, and
compression
High
2.4. Audio Steganography
Audio steganography is a technique of concealing confidential data in audio files. Currently, secret
messages are covertly embedded mostly in Wav, au, and mp3 audio file formats [22]. Compared to other
steganographic techniques such as image steganography, embedding secret messages in audio signals635
is an arduous task. Usually, information concerning copyright protection is embedded in audio signals.
28

Figure 12: Parity Coding Technique.
Moreover, the characteristics of Human Audible Sound (HAS) are utilized to hide secret data in audio signals.
Correspondingly, as shown in Table-5, audio steganography methods are broadly categorized as temporal
domain, transform domain, compressed, and coded domain techniques.
a. Temporal Domain Techniques640
•LSB Coding: LSB coding is considered to be the simplest way of hiding secret data in an audio
signal. In this method, each secret message bit is replaced by the LSB of each sample of the
digitized audio file [92]. E.g., suppose the English letter “Z” is to be hidden in an audio file, in
which every sample is represented by 16 bits. In that case, each LSB of 7 consecutive audio
samples is replaced with a respective bit of the binary equivalent of the letter Z, i.e., 1011010.645
Correspondingly, in this method, the embedding capacity is proportional to the audio file sampling
rate that usually ranges from 8 kbps to 44.1 kbps. However, the direct LSB substitution scheme
is not robust because the embedded secret bits in the LSBs of the audio samples in a silent or
near-silent audio segment can produce strong hissing noise when played. Hence, modification to
the original audio signal can be easily detected [93].650
•Parity Coding: It is one of the robust audio steganographic techniques. In this method, the cover
signal is divided into separate samples. Then each bit of the secret message is concealed by
utilizing a parity bit from each sample. As shown in Figure-12, if the parity bit of a selected region
does not match the secret bit to be encoded, then the LSB of one of the samples in the region is
inverted [94]. In this way, the sender has more flexibility in encoding the secret bit.655
29

Figure 13: Phase Coding.
•Echo Hiding: Echo data hiding deals with the embedding of secret message bits in an audio file
by introducing an echo to the cover signal. The data is then hidden by varying three parameters
of the echo:
–Initial amplitude
–Decay rate660
–Offset
b. Transform Domain
•SS: In its simplest form, the SS technique spreads the secret information across the frequency
spectrum of the cover audio signal using a code that is independent of the actual signal [94]. The
two versions of SS that are mostly used in audio steganography include; direct sequence SS and665
frequency hopping [95].
Direct-sequence SS attempts to spread out the secret message by a constant known as chip
rate, and then it is modulated with a pseudo-random signal and interleaved with the cover signal.
Whereas, in frequency-hopping SS, the frequency spectrum of audio files is changed so that it
hops rapidly between frequencies. Though SS methods provide better robustness, these are670
vulnerable to time scale modification.
•Phase Coding: Phase coding addresses the disadvantages of the noise-inducing methods [95].
It works by substituting the phase of an initial audio segment with a reference phase that repre-
sents the secret message (shown in Figure-13). This technique relies on the fact that the phase
components of sound are not as distinguishable to the human ear as compared to the noise.675
Hence, instead of introducing perturbations, this technique encodes the message bits as phase
shifts in the phase spectrum of a digital signal. Thus achieving an inaudible encoding in terms of
Signal-to-Perceived Noise Ratio (SPNR).
•Discrete Wavelet Transform (DWT): In DWT, the secret message bits are concealed in the
LSBs of the wavelet coefficients of the cover audio signals. Correspondingly, to improve the680
imperceptibility of hidden data, researchers in [16] employed a hearing threshold when concealing
data in the integer wavelet coefficients. On the other hand, [96] avoided data hiding in silent parts
of the audio signal. Even though data embedding in wavelet domain yields a high embedding
rate of up to 70kbps, data extraction at the receiver side is not always accurate.
•Tone Insertion: This approach relies on the inaudibility of lower power tones in the presence of685
significantly higher ones [96],[97]. Correspondingly, to conceal one bit of a secret message in an
30

audio frame, a pair of tones generated at two chosen frequencies f0 and f1 is selected. The power
level of the two masked frequencies pf0and pf1is set to a known ratio of the standard power
of each audio frame pi where: i = 1,...n and n is the frame number. Inserting tones at known
frequencies and at low power levels results in secure data embedding and correct extraction by690
the receiver. To detect the tones and respective embedded information from the stego- audio
frames, the power pi for each frame is computed as well as the power pf0and pf1for the chosen
frequencies f0and f1. If the ratio, pi
pf0>pi
pf1, then the hidden bit is “0”, otherwise it is “1”. The tone
insertion technique is robust against low-pass filtering and bit truncation attacks. However, it has
low embedding capacity, and the hidden data can also be maliciously extracted since inserted695
tones are easy to detect. This vulnerability can be addressed by varying four or more pairs of
frequencies in a keyed order.
•Amplitude Coding: The HAS characteristics are more sensitive to frequency variations than the
amplitude components of the sound. Based on this principle, researchers in [98] proposed a
steganographic method that embeds high-capacity data (up to 20 kbps) in the magnitude speech700
spectrum while controlling the distortion of the cover signal and also ensuring concealed data's
security. The secret data to be hidden can be in different forms, including encrypted data, com-
pressed data, groups of data (LPC, MP3, AMR, CELP, parameters of speech recognition, etc.).
Moreover, the proposed technique finds secure spectral embedding areas in a wideband mag-
nitude speech spectrum using a frequency mask defined at 13 dB below the cover audio signal705
spectrum. Correspondingly, The embedding locations and hiding capacity in magnitude compo-
nents are defined according to a tolerated distortion level defined in the magnitude spectrum.
Since the frequency components within the range of 7 kHz to 8 kHz contribute minimally to wide-
band speech intelligibility, authors in [99] presented a technique to hide data in the said range
by replacing the original frequencies with the secret message. This approach results in high710
embedding capacity without degrading the speech quality of the cover audio signal.
c. Compressed Domain Technique
•Vector Quantization: Vector quantization is a technique that is used to hide mystery records in
the compressed cover document.
•Fractal Compression: Researchers in [100] introduced a data embedding technique based on715
fractal coding and a chaotic LSB a.k.a High Availability Service Function Chain (HASFC). The
proposed method aims to increase the embedding capacity and preserve statistical transparency
and security. The HASFC model manages to embed secret audio into cover audio of the same
size. To achieve desired results, fractal coding is adopted, which produces a high compression
ratio with the acceptable reconstructed signal. Moreover, the chaotic map is used to randomly720
select the cover samples for embedding. Its initial parameters are utilized as a secret key to
augment the framework's security. Contrary to the existing audio steganography schemes, the
HASFC model outperforms related studies by improving the embedding capacity up to 30% and
maintaining stego audio's transparency with average SNR values at 70.4 dB, PRD at 0.0002,
and SDG at 4.7. Moreover, the proposed technique is robust against brute-force and statistical725
analysis attacks.
d. Coded Domain Techniques: Data hiding for real-time communications mostly utilizes voice encoders
such as AMR, ACELP, and SILK at their respective encoding rate. While being processed by one of
31

such encoders, the transmitted audio signal is coded according to the encoder rate then decompressed
at the decoder end [16]. Hence, the audio signal at the receiver side is different as compared to730
the signal at the sender’s end. This difference makes these techniques challenging as it affects the
correct retrieval of hidden data from the encoded signal. These techniques are broadly categorized as
in-encoder and post-encoder techniques.
•In-Encoder: To ensure that the concealed data is robust against audio codec, compression,
reverberations, and background noises, researchers in [101] introduced an in-encoder technique.735
The proposed method hides data into speech and music signals of various types using sub-band
amplitude modulation. Similarly, [102] proposed a technique to embed data in the LPC vocoder.
The researchers performed a voiced/unvoiced segmentation of the cover signal by utilizing an
auto-correlation-based pitch tracking algorithm. They replaced the linear prediction residual in
the unvoiced segments with a data sequence. Once the residual's power is matched, the SNR740
does not deteriorate. Correspondingly, the signal is conceived using the unmodified LPC filter
coefficients, and the hidden data is retrieved by performing a linear prediction analysis of the
received signal. This method yields an embedding rate of 2kbps.
In the same way, researchers in [103] concealed data in the audio codecs by exploiting the LSB
technique. The proposed methodology hides data in the LSB of the Fourier Transform in the745
prediction residual of the host audio signal. An LPC filter is then used to automatically shape the
spectrum of LSB noise. Consequently, the noise generated by data hiding is substantially less
audible.
•Post-Encoder: Post-encoder a.k.a in-stream, is an alternative to the in-encoder techniques.
Correspondingly, researchers in [104] concealed secret data in the bit-stream of an ACELP750
codec. The proposed method embeds secret data jointly with the analysis-by-synthesis code-
book search. Accordingly, the authors were able to conceal 2kbps of secret data in the bit-stream
by applying the said concept on the AMR encoder at a rate of 12.2 kbps.
Similarly, a lossless steganography technique for G.711-PCMU telephony encoder was presented
in [105]. Data, in this case, is represented by folded binary code which codes each sample with a755
value between -127 and 127, including -0 and +0. One bit is embedded in an 8-bits sample with
an absolute amplitude of zero. Depending on the number of samples with absolute amplitudes
of 0, a potential hiding rate ranging from 24 to 400 bps is achieved. To enhance the embedding
capacity, the researchers later introduced a semi-lossless technique for G.711-PCMU [106]. In
this method, the audio sample amplitudes are amplified with a pre-defined level “i”. The audio760
signal samples with absolute amplitudes varying from 0 to i are utilized in the data embedding
process. For a greater hiding capacity, authors in [107] suggested embedding data in the inactive
frames of low bit-rate audio streams (i.e., 6.3 kbps) encoded by G.723.1 source codec.
32

Table 5: Techniques of Audio Steganography
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
Temporal Domain Techniques
1. LSB Coding
[92],[93]
LSBs of each audio sample
are replaced with the secret
message bits
Low robustness against
statistical and perceptual
analysis
Low
2. Parity Coding
The cover signal is divided
into separate samples. Then
each bit of the secret
message is concealed by
utilizing a parity bit from
each sample
Low quality Low
3. Echo Hiding
Conceals secret data by
introducing echo in the cover
audio signal
Low security and low data
hiding rate (50bps) Low
Transform Domain Techniques
4. SS Spreads the secret data bits
over all signal frequencies
Vulnerable to time scale
modification and has low
embedding rate (20bps)
High
5. Phase Coding
Modulating the phase of the
cover signal as per phase of
the secret message
Low data hiding rate
(333bps) Low
6. Discrete Wavelet
Transform
Altering wavelet coefficients
to hide secret data Lossy data retrieval High
7. Tone Insertion
Secret message is
embedded by inserting
inaudible tones at selected
frequencies
Lack of transparency and
security with low data
embedding rate (250bps)
High
8. Amplitude
Coding
Use frequency bands to hide
data
Low robustness to simple
audio manipulations low
Compressed Domain Techniques
9. Vector
Quantization
Hides mystery records in
compressed cover document Not given High
10. Fractal
Compression
Secret data is embedded in
the cover audio based on
fractal coding and a chaotic
LSB a.k.a HASFC
Resistance against
Reversed Psychoacoustic
and Derivative-Based
steganalysis is not tested
High
Codecs Domain Techniques
11. Codebook
Modification
Altering codebook
parameters
Low embedding capacity
(2kbps) Low
Continued on next page
33

Table 5 – Continued from previous page
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
12. Bitstream Hiding
LSB is applied on the bit-
stream resulting from the
encoder process
Low embedding capacity
(1.6kbps) Low
765
2.5. Text Steganography
In this steganography technique secret data is hidden within the text documents. However, it is the most
difficult method because it has negligible redundant bits to hide data. Moreover, it is also observed that
any change in the structure of the text can easily be observed by the human eye as compared to other
mediums like images, audio files, and videos. Nonetheless, different methods have been devised to embed770
secret data within the cover texts/documents. The usual applications of text-based steganography include
copyright protection, document authentication, prevention of e-document forging, and covert communication
[95].
Correspondingly, many researchers have explored the field of text-based steganography. They have used
various methods ranging from the use of white spaces to line spacing within a document to hide secret data.775
In the context of the use of white spaces, if there is one space between two words, then it is assumed that
bit 1 is hidden, and if there is double space, it is assumed that bit 0 is hidden. However, the major drawback
of this technique is that it is effective only for electronic documents. Once the documents are printed, the
information is lost. Similarly, some researchers have used the line spacing method. In this case, the secret
message is concealed by shifting the positions of lines up and down the cover document. Besides, there780
is also some research on linguistic methods. In this approach, some symbols are replaced with other pre-
defined symbols like commas, hyphens, and semicolons. However, the critical issue here is the embedding
capacity [25].
Figure 14: Text-based Steganography.
As illustrated in Figure-14, and Table-6, there are two main categories of text-based steganography: format-785
based and linguistic steganography. The format-based method is further divided into word shift encoding,
34

line shift encoding, white space encoding, and feature encoding methods. In contrast, the linguistic approach
is categorized into syntax method and semantic method.
a. Format-based Steganography: In this technique, physical features of text symbols are modified in
such a manner that the naked human eye cannot detect the changes. In one of such methods, lines790
are shifted up and down to hide information bits 0 and 1. In another case, words are shifted left or right
or just up or down. Moreover, white spaces between the words or between the lines or paragraphs are
also altered to embed secret data bits. In feature-based encoding, the physical features of the words
are modified as per the characteristics of different languages.
b. Linguistic Steganography: In this approach, Unicode symbols of different languages, e.g., Malayan795
language, are used to embed a secret message in the cover text. Various researches have shown that
linguistic steganography techniques have great potential to embed secret messages as it is very hard
for a naked human eye to detect minor changes in the symbols of the text.
Table 6: Text-based Covert Communication Methods
Ser Technique Basic Idea Limitations/ Performance
Overheads
Level of
Complexity
1. Format-based
steganography
Physical features of text
symbols are modified to embed
secret messages
Effective for electronic
documents only Low
2. Linguistic
steganography
Unicode symbols of different
languages are used to hide
secret data
Effective for electronic
documents only Low
2.6. Internet Protocol Based Covert Communication
a. Covert communication over VoIP streaming media with dynamic key distribution and authenti-800
cation: Researchers in [108] introduced a novel information-theoretical model of secure VoIP covert
communication. Accordingly, a new dynamic steganographic algorithm was designed that includes
one-way accumulation integrating into dynamic key updating and exchange. Such integration can pro-
tect steganographic systems from man-in-the-middle attacks, which threaten covert steganographic
communications.805
As shown in Figure-15, a secret message (M) (to be embedded) is encrypted with a secret key gener-
ated from a random number generator to form a ciphertext. The encrypted message is then segmented
into distinct parts, which are further embedded in a series of packets of media streams, namely cover
objects (C). S in the figure denotes the packet containing a hidden message. The proposed framework
integrates dynamic key distribution and authentication with data embedding and extraction. As far as810
the embedding capacity of this model is concerned, two bytes of the secret message can be hidden
into sixteen bytes of VoIP streams.
b. Covert Communication Using MODBUS Protocol in IoT Devices: Researchers in [109] introduced
a MODBUS protocol based covert communication in an IoT environment. Although MQ Telemetry
Transport (MQTT) is the protocol of choice for IoT devices, researchers employed the MODBUS pro-815
tocol because it is not very common in the IoT environment. Hence, it is believed to be harder to
35

Figure 15: VoIP Streaming Steganography [108].
implement and understand. As a result, it would be difficult for an adversary to detect the existence of
covert communication.
There are two types of communications in the MODBUS protocol; query/response and broadcast.
[109] uses the query/response communication for covert transmission. In the query/response, the820
communication is between the Master (user client) and the Slave (server/IoT device). The Master ini-
tiates the communication by requesting the current status of an IoT device (e.g., an air conditioner) by
using read commands. The Slave (IoT device) replies back with the status. The Master can also send
a request to change the current status of the air conditioner from On to OFF or vice-versa by sending
write commands. The Slave then follows the functionality and switches OFF or ON the air conditioner825
accordingly. The authors used the (address, count) fields in the read commands to embed secret data
because read commands can carry more information than the write commands that need to be pretty
specific.
Furthermore, in order to send a message comprising twenty-one characters, the proposed scheme
took 6-7 seconds with a bandwidth requirement of 3kbps. As far as the resilience of this scheme to830
detection is concerned, as the “Read” requests are flooded into the network during character mapping,
network packet analyzers such as Wireshark are unable to capture all the read requests. Because it is
very difficult for a packet analyzer to differentiate between the legitimate and fake read request pack-
ets. However, if the packets received from the Slave are analyzed, one can easily identify garbage
read requests since the “Count” value in (address, count) would be more than 1. Nonetheless, in835
order to analyze the packets, the packet analyzer must be physically connected to the Slave, which is
practically impossible in private IoT networks.
The significant limitation of the proposed framework is that the secret message can only include: block
letters from A to Z, digits from 0-9, and special keys including space and period. To increase the
range of letters and special symbols, the IoT network size has to be increased. Moreover, unlike the840
36

Figure 16: StegoTorus Data Flow.
MODBUS RTU model, the MODBUS protocol used in the proposed scheme is the TCP/IP frame which
does not include a Cyclic Redundancy Check (CRC) functionality. Hence, there is no mechanism to
verify if the received covert message is correct or not.
c. Pluggable Transport Protocols: To prevent censorship attacks, Appelbaum and Mathewson pro-
posed a framework for developing protocol-level obfuscation plugins for The Onion Router (ToR) called845
pluggable transports [110]. These transports appear as SOCKS proxies to the ToR client. Hence, the
ToR clients send data to the pluggable transport SOCKS proxy instead of a bridge. The proxy then
further transmits the data to the bridge in an obfuscated manner. Moreover, to hide ToR traffic in their
protocols, the developers can customize this framework to build their own transports. Basically, the
pluggable transports disguise ToR messages in a different form of traffic, such as HTTP. The ultimate850
goal of this framework is to provide resistance against client-to-bridge censorship attacks launched by
censoring ISPs.
d. StegoTorus: To circumvent censorship of ToR network traffic, researchers in [111] presented a
scheme to steganograph ToR traffic. The proposed technique mimics ToR traffic as http in order
to prevent fingerprinting attacks. StegoTorus is a pluggable transport [112] (an extension of SOCKS855
[113]) for ToR that masks its connection to a relay server. As shown in Figure-16, ToR uses Stego-
Torus as its SOCKS proxy. The ToR traffic is chopped into variable-length blocks by a chopper. The
StegoTorus client re-encrypts these blocks to pass through all known pattern filters. However, pattern
37

Figure 17: SkypeMorph Data Flow.
filters that only allow packets with known and recognizable headers would block this traffic. Besides,
the encrypted blocks are also disguised as the cover protocol, i.e., http traffic. Later, the StegoTorus860
server assembles and decodes the disguised traffic to the ToR relay network. The StegoTorus cover
protocols were initially designed to run over TCP.
e. SkypeMorph: To avoid censorship of publicly listed ToR relay nodes, researchers in [114] instead
used unlisted bridges as an entry point to the ToR network. Correspondingly, the communication
between clients and bridges and also in between bridges is disguised as Skype video calls. Protection865
mechanisms and code obfuscation techniques used in the Skype software do not disclose anything
about its internal mechanism. Moreover, there is no open-source variant of the Skype application.
Hence, it is challenging for an adversary/censoring agency to make a distinction between Skype video
calls and obfuscated bridge connections based on statistical analysis. As the bridges are not listed,
it is arduous for adversaries to find out their IP addresses for censorship. However, the bridges were870
detected in some instances based on their unconditional acceptance of incoming connections or their
location. Moreover, like Skype, SkypeMorph also supports UDP only. The data flow of the SkypeMorph
protocol is shown in Figure-17.
f. CensorSpoofer: Exploiting the asymmetric nature of web traffic researchers in [115] presented the
PoC of a censorship-resistant web browsing framework called “CensorSpoofer.” The proposed frame-875
work makes use of IP address spoofing to send data from the proxy to a user without revealing its
actual origin. Though a spoofed channel allows communication in a single direction only; however, the
asymmetric nature of web-browsing traffic can be exploited through a low-bandwidth indirect channel,
e.g., Email or steganographic instant messages to communicate requests from the user to the proxy.
Since the VoIP protocol maintains close synchronization without any endpoint and does not reveal its880
contents to the censor, to prevent detection, CensorSpoofer imitates an encrypted VoIP session to
tunnel the downstream data.
The techniques mentioned above, including StegoTorus, SkypeMorph, and CensorSpoofer, prevent
detection by using the approach of unobservability-by-imitation [116]. Therefore, these approaches are
also called “Parrot Circumvention” techniques. They try to imitate every observable aspect of the target885
protocol. However, this approach has numerous flaws [117] and the sessions of parrot circumvention
38

techniques can be identified even by weaker censors. Accordingly, a parrot can be distinguished from
its corresponding original protocol based on its reaction to errors and changing network conditions.
Generally, a protocol standard clearly defines how a particular error will be handled. However, in the
case of a parrot, it is very difficult to mimic error handling. Hence, differences in error handling can890
be mapped to fingerprint implementations of genuine protocols as well as associated parrots. In the
same way, a response to changing network conditions must be imitated by a parrot as close to the
target protocol as possible. In addition, if a side protocol is used, e.g., to renegotiate a signal codec, it
must be mimicked, too.
Consequently, researchers in [116] experimentally proved that the above-mentioned parrot circumven-895
tion techniques are observable through several passive and active attacks. Similarly, researchers in
[118] demonstrate that ToR pluggable transports can be detected with high probability through flow
analysis using various features that best describe the behavior of pluggable transports. The features
may include the number of bytes sent, each transmission/session duration, and the maximum packet
size.900
g. Meek and Adaptive Meek: To improve obfuscation and reduce indistinguishability, researchers in
[119] proposed Meek, a pluggable Transport to disguise ToR users' traffic as traffic to a cloud service
platform. However, Meek was vulnerable to various side-channel attacks. Hence, Meeks obfuscation
was observable. Subsequently, authors in [120] presented a traffic morphing based “Adaptive Meek”
that imitates ToR-based Meek traffic as the target traffic with minimum possible overheads to resist905
statistical traffic analysis attacks. The highlight of this approach is its focus on side-channel information
while replicating target traffic.
2.7. Blockchain-based Covert Communication
Public blockchain's inherent characteristics such as immutability, decentralization, and anonymity have
been leveraged by numerous researchers to make a network interference and tamper-resistant, and also use910
such networks for covert communications [121, 122, 123, 124, 125, 126, 127, 128, 129]. Correspondingly,
in the subsequent section, various blockchain-based covert communication techniques are illustrated.
a. Monero-based Secure Covert Channel: According to researchers in [123], public blockchains have
not been designed or developed to support covert communications. Therefore, some of the existing
covert communication techniques that leverage blockchain technology have the issues of ensuring915
anonymity at the transaction and network layer, insufficient robustness, and anti-tamper mechanism.
Therefore, Monero encrypts the receiving and sending addresses and transaction amount to ensure
the anonymity of communication and achieve a certain degree of anti-censorship and anti-tracing
capabilities. In addition, the Monero-based technique provides good transmission efficiency and bit
error rate (BER).920
b. Ethereum Whisper Protocol-based Covert Channel: To improve resistance against interference
and tampering attacks on covert communications, researchers proposed a unique approach based on
Ethereum blockchain's Whisper protocol [124]. The scheme uses the payload part of the protocol to
store secret data and employs the padding field to record the index value of the stored data.
c. Kleptography-based Covert Communication on Public Blockchain: Making use of Kleptogra-925
phy, [125] presented a technique for covert transmission of secret messages over public blockchains.
Where Kleptography [130] has been used in the past to attack public-key cryptosystems such as RSA,
39

Digital Signature Algorithm (DSA), Elliptic Curve Cryptography, and Diffie-Hellman key exchange. The
kleptographic algorithm enables an attacker to extract victim's private key. Moreover, the output of the
Kleptographic algorithm cannot be distinguished from the output of the original encryption algorithm.930
Correspondingly, Kleptography is an appropriate approach for the covert transmission of messages in
blockchain systems. However, to successfully realize this technique, there is a requirement to estab-
lish a communication channel external to the blockchain for negotiating data transmission algorithms
and associated key data. Once this is done, the sender generates a public-private key pair and then
encrypts the secret message using his public key. He then creates a special blockchain transaction935
such that the signature information in the transaction script conforms to the requirements of the Klepto-
graphic algorithm. Subsequently, to decrypt the secret message in the special transaction, the receiver
can compute the sender's private key from the signature data. However, the average data transaction
cost (max 80 bytes) as of 28 May 2022 is 1.13 US dollars.
d. Covert Communication based on Bitcoin Transaction: To prevent data loss in traditional covert940
communication techniques, researchers in [126] proposed a Bitcoin Transaction based covert ap-
proach. The scheme utilizes address interactions and transaction amounts to embed secret infor-
mation. Basically, the sender of the covert message determines the transaction sequence between
Bitcoin addresses by employing the index matrix of transaction addresses. In contrast to previous tech-
niques, the proposed method achieves high embedding capacity in a single transaction. Moreover, to945
prevent detection and hide secret information in the transaction amount and bitcoin address interac-
tion relationship, the proposed scheme conforms to the statistical distribution of transaction amount in
the public chain. On the other end, the receiver employs crawler technology to obtain the transaction
information of the corresponding address as per negotiated address sequence. Later, the transaction
information is arranged according to the timestamp, and the hidden information is extracted by parsing950
the Bitcoin transaction matrix.
e. Chain-based Covert Data Embedding: Authors in [129] proposed a covert communication technique
based on public keys. The public keys are used to produce Bitcoin addresses that transmit the informa-
tion. The proposed scheme can embed up to 8 bits in each public key. Although this scheme prevents
issues related to conventional covert techniques, such as data loss and susceptibility to data compres-955
sion and geometric attacks but it has low embedding and low communication efficiency. Moreover, it
uses a large number of public addresses.
f. Time-based Covert Channels: Researchers in [131] presented a covert communication technique
by utilizing the transaction time interval. However, according to [132] time-based approaches can be
observed by analyzing time interval and payload length.960
g. Storage Type Covert Communication Techniques: In addition to above approaches there are other
storage type covert communication techniques [133, 134, 135, 136]. These schemes use payload or
non-payload parts of the protocol such as digital signatures, recipient address, transaction payment
address, or some extended data parts, e.g., OP RETURN field in Bitcoin blockchain and Ethereum
smart contract event logs.965
For instance, in [133] the communicating parties use the payment addresses to convey a hidden
message to each other. The secret data embedding capacity of the proposed scheme depends on
the number of payment transactions published per block. Authors claim that the proposed technique
satisfies the definition of provable security as it is very hard to distinguish random payment transactions
40

from the ones bearing the covert payload. As a Proof of Concept (POC), a simplified version of970
blockchain [137] without any consideration for the consensus protocol is used. Moreover, for privacy
reasons, the user accounts are not reused after making a payment transaction. And the random
addresses are generated by taking a hash of the public key. E.g. to communicate stealthily, Alice
generates a number of public and private key pairs. Later, she generates the payment addresses by
taking the hash of the public keys. Now to send a message, Alice arranges the payment addresses975
so that the LSB of the addresses forms the secret message. Later, Bob can read the message from
the blockchain by reviewing the payment transactions submitted by Alice and the LSBs of the payee
addresses. For simplicity of reading by Bob, Alice submits one payment transaction per block. Hence,
one bit of the secret message is transmitted per block.
For security, Alice and Bob share a secret key to encrypt and decrypt the message and a random980
message start indicator (most likely the message start block number) that is used to locate the mes-
sage from the published blocks. The researchers claim that the proposed scheme is secure against
a chosen hidden text attack in which an adversary tries to distinguish between a random payment
transaction and the payment transactions bearing the secret message. In addition, the researchers
also claim that the proposed method is reliable and runs in expected polynomial time.985
However, based on the fact that public blockchains' transactions are publicly available hence, trans-
action layer covert communication approaches are vulnerable to insufficient anonymity. Which can
ultimately lead to the detection of covert communication [138]. Nonetheless, the use of OP RETURN
field and digital signature in Bitcoin transactions as covert data carriers infers low latency and high
resistance to detection [134].990
2.8. Other Unique Covert Techniques
a. Covert Underwater Acoustic Communication: Another interesting domain in covert communication
is Underwater Acoustic Communication (UAC), which has attracted extensive research in recent years.
The traditional Covert Underwater Acoustic Communication (CUAC) techniques hide the existence of
communication through transmission control and signal modulation. In this context, SS technology995
combined with acoustic carrier communication is used to achieve low SNR. However, artificially mod-
ulated signals can easily be identified in the marine environment. The second method of CUAC is to
generate communication signals by utilizing the acoustic signals inherent in the marine environment.
Correspondingly, researchers in [139] introduced a CUAC scheme in which communication signals are
disguised as ship-radiated noise and chaos signals to transmit secret messages. Hence, the commu-1000
nication signal cannot be distinguished from the normal ship-radiated noise by the adversary, which
is quite different from the traditional SS approaches. On the transmitter side, the correlation of the
chaos signal is utilized to embed secret messages, and later, corresponding pilot and synchronization
signals are added to the communication signal. Moreover, by analyzing the ship-radiated noise, the
continuous spectrum, line spectrum, and envelope function are extracted, and the ship-radiated noise1005
is reconstructed to generate the covert communication signal.
Whereas, on the receiver end, the pilot signal is used to detect the underwater acoustic channel and to
realize Doppler shift compensation and multipath equalization. Due to the characteristics of chaos sig-
nals, the proposed communication system has a low BER in the complex underwater acoustic channel
and is secure against multipath and Doppler channel interference. The proposed scheme has strong1010
transmission accuracy and security guarantees because the communication signal of the scheme is
41

the reconstructed mimic ship-radiated noise. Based on actual sea experiments, the researchers were
able to achieve a communication distance of 10 km at a communication rate of 40 bps with BER of 2.5
x 103.
b. NICScatter: Backscatter as a Covert Channel in Mobile Devices: Authors in [140] have discovered1015
that the wireless Network Interface Card (NIC) of a mobile device can be used by malware to covertly
convey information to an attacker. In this context, the NICScatter transmitter malware controls the
impedance of the NIC of a laptop or a mobile device to reflect (backscatter) surrounding RF signals
in order to communicate with the attacker covertly. The attacker can use two types of receivers to
capture the backscatter RF signals; Received Signal Strength Indicator (RSSI) and CSI. Hence, the1020
subject technique can affect many common wireless devices such as laptops and Unmanned Aerial
Vehicles (UAVs). One is based on RSSI/CSI measurement ability, which covers many common wire-
less devices, like laptops and UAVs [30].
Currently, the malware can pass sensitive information about the user to the attacker's mobile device
(Linux and Android OS) without requiring any special privileges. The covert channel can transmit at a1025
rate of 1.6 bps at a distance of 2 meters at a clear LoS or at a distance of 73cm through the walls.
c. Covert Communications through Network Configuration Messages: Network covert channels
can be defined as a way of transmitting hidden information by utilizing communication protocol fea-
tures whose main functionality is misused. Hence, various network protocols have been exploited by
researchers to covertly exchange secret information. These protocols range from TCP/IP, to HTTP,1030
DNS, and DHCP. Accordingly, researchers in [141] developed a POC of establishing three forms of
covert communications by taking advantage of xid,Sname and file, and options fields in DHCP
packet header. The secret message embedding capacity is relational to the ability to stay undetected.
Hence, there is no best covert channel under all circumstances, and usually, those who operate well in
certain scenarios cease to be useful in others. Therefore, based on three properties vital properties;1035
detectability, capacity, and reliability, the selection of a suitable covert technique may vary.
Correspondingly, in this scenario, the xid method provides a minimal 8 bytes per hour capacity. This
technique is thus recommended for the transmission of small amounts of highly sensitive data, e.g.,
an elliptic curve cryptography key. On the other hand, the Sname and f ile method provides an em-
bedding capacity of 760 bytes per hour. Whereas the options based covert channel provides a higher1040
capacity of 1020 bytes per hour. But it is recommended for loosely supervised networks.
As far as security/detectability of the DHCP-based covert channels is concerned, the authors claim that
the proposed covert communication could not be detected by a regular IDS such as Snort. However,
if active wardens and DHCP-specific IDS rules are designed, there is a high probability of detecting
the presence of subject covert channels. Nonetheless, non-detectability and embedding capacity can1045
be further improved if a compression technique, e.g., Huffman coding [142], is first used on the secret
information before it is hidden.
d. Practical Covert Channels for WiFi Systems: Wireless covert channels have the potential to ex-
filtrate information with high bandwidth by circumventing conventional access control mechanisms.
Moreover, these channels can operate under adverse conditions and can also tolerate a high amount1050
of signal variations. The main flexibility here is for the physical receivers that are not addressed within
wireless frames. Instead, they can simply eavesdrop on the transmission. Accordingly, researchers in
[143] analyze the possibilities to establish covert channels in WiFi systems by exploiting physical layer
42

Figure 18: AITSteg Framework.
characteristics.
Why physical layer? It is because the upper layers allow information embedding with only a few vari-1055
ations, e.g., using reserved bits or changing transmission timings. Nonetheless, a firewall can easily
detect such discrepancies [144]. On the other hand, physical wireless transmissions comprise sym-
bols containing high noise and random signal variations. Eavesdropping on the raw data in the air
results in a very large amount of data. Hence, it is nearly impossible to reveal any hidden information.
Correspondingly, traditional WiFi receivers are designed to reconstruct the signal despite variations.1060
Therefore, their performance does not degrade when additional information is embedded. Due to the
wireless broadcast nature, frames can contain oblivious sender and receiver addresses without being
detected by other network participants.
Correspondingly, researchers in [143] were able to establish four practical covert channels for the
physical layer of IEEE 802.11 a/g. The details of each of the covert channels are:1065
• First covert channel was created by utilizing the Short Training Field (STF) in combination with
Phase Shift Keying (PSK). This technique introduces phase shift to STF and is immune to reactive
jamming. Moreover, it has minimal influence on WiFi BER. At a transmission rate of 1 PSK symbol
per frame, the maximum achievable covert throughput is 375 kbps for 64-PSK.
• The second channel utilized the Carrier Frequency Offset (CFO) with Frequency Shift Keying1070
(FSK). To avoid detection, the lowest working ∆CF O should be chosen, e.g., 5 kHz. Subsequently,
by encoding 1 bit per 4 µs OFDM symbol, the achievable covert throughput is 250 kbit/s.
• Third covert channel was established using IEEE 802.11 a/g with additional subcarriers con-
forming to the IEEE 802.11 n spectrum mask (Camouflage Subcarriers). Correspondingly, this
method uses four additional subcarriers from IEEE 802.11 n, and it has no influence on WiFi1075
BER. Moreover, at the transmission rate of 4 QAM symbols per OFDM symbol, the maximum
covert throughput is 4.5 Mbps for 54 Mbps WiFi frames.
• The fourth covert channel was formed by replacing parts of the OFDM CP (CP Replacement).
Although this technique does not have much influence on WiFi BER in line-of-sight channels, but
it is affected by multipath effects. Moreover, at a transmission rate of 12 QAM symbols per OFDM1080
symbol, the maximum achievable covert rate is 6.75 Mbps for 1/2 CP with CPCP.
e. An Innovative Text Steganography Technique via Social Media: To preserve the confidentiality of
text messages sent via Short Message Service (SMS) or social media, researchers in [145] proposed
43

“AITSteg”. The proposed scheme provides end-to-end security for sharing of confidential informa-1085
tion/secret messages (SM) between end-users. Accordingly, as shown in Figure-18, the AITSteg
generates a hidden message (HM) containing the confidential information and then encodes it using
a symmetric key. Afterward, it embeds the generated HM in front of a cover message (CM) using four
zero-width Unicode characters (ZWC), including; zero-width non-joiner, POP directional, Left-to-Right
override, and Left-to-Right mark.1090
The hidden message embedding capacity of the proposed framework varies as per the social media
application that is being used for message transmission between end-users. E.g., in the case of a
simple SMS, a CM of length 1024 (UTF-8) characters can conceal 170 characters of HM. Similarly,
CM of 640 (UTF-8) and 16207 (UTF-8) characters can conceal HM of 106 and 2701 characters in
Facebook and WeChat, respectively. Likewise, the researchers used fifteen different social media ap-1095
plications to test their embedding capacity and the ability of AITSteg's HM to remain invisible. The
proposed scheme is claimed to be secure against MITM, message disclosure, and manipulation by
readers attacks.
3. Discussion/Gap Analysis
Covert communication techniques of numerous categories have been illustrated in the previous sec-1100
tions. However, some of these techniques strain system deployment by consuming significant resources
(e.g., bandwidth and energy). At the same time, a few of these techniques are not secure or have very
low secret message embedding capacity. Nonetheless, it is challenging to find a perfect steganography
method that fulfills all the security and performance requirements. Hence, there has to be a tradeoff be-
tween security, performance efficiency, and embedding capacity depending upon the requirements of the1105
communication environment.
Correspondingly, the overarching goal of covert wireless communication is the establishment of “shadow net-
works.” These networks comprise; relays that generate, transmit, receive and consume data, and jammers
that generate AN and hide the existence of communication/transmission from the wardens. Accordingly,
various SS techniques such as Direct Sequence Spread Spectrum (DSSS), Frequency Hopping Spread1110
Spectrum (FHSS), and their combination are being used for physical layer (wireless communications) se-
curity. However, numerous constraints and limits of covert wireless communication have been observed.
Accordingly, in one of the studies on secrecy and undetectability in a Multiple-Input Multiple-Output (MIMO)
setting [146], it has been identified that covert communication systems are constrained by average power.
Similarly, SS systems allow communication where it is prohibited because spreading the signal power over1115
a large time-frequency space substantially reduces an adversary's SNR. This impairs the attacker's ability to
discriminate between the noise and the information-carrying signal corrupted by noise. But here a question
arises: how small the power has to be for the communication to be undetectable, and how much hidden
data can be transmitted reliably. In this context, [147] investigated a scenario of AWGN channel, in which
a sequence of independent and identically distributed zero-mean Gaussian random variables with variance1120
σ2is used to corrupt a signaling sequence. Hence, for n number of channels d between two communicat-
ing parties, say, Alice and Bob, when Alice is not transmitting, the adversary (Willie) observes AWGN with
total power over n channel observations on average. Since Willie observes Alice's signal power when she
transmits in addition to the noise power, to prevent Willie from getting suspicious, the total power that Alice
can emit over n channel uses is limited to (σ2
ω√n). Hence Alice can transmit no more than (σ2
ω√n)/σ2
b)1125
44

covert bits to Bob in n channel uses.
Correspondingly, jammers facilitate covert communication in covert networks by confusing the eavesdrop-
per as jammers operate independently from the relay transmission. Hence, the adversary cannot detect
covert communication by listening to the jammers. Thereby, jammers are known to have a parasitic effect
on Wardens' SNR. However, there is a question about the scaling behavior of such a network in the case1130
of multipath unicast communication in large wireless networks. To answer this question, [148] suggests
that variable jamming power and multipath fading should be incorporated into the jammer-assisted covert
communication model, as it may enable covert communication at a positive rate.
Likewise, covert transmission with mmWave is highly vulnerable to severe penetration losses and poor
diffraction of non-LoS links. Hence, deploying an IRS with LoS links for both sender and receiver can ad-1135
dress the limitations of mmWave covert communication. Accordingly, an IRS can be utilized in two ways to
enhance transmission stealthiness. Firstly, an IRS can reflect the desirable signals (e.g., information trans-
mission) in phase with the ones at the intended receiver. Secondly, an IRS can reflect the unwanted signals
in the opposite phase with the ones at the unintended receiver. As a result, increased transmission rate at
Bob and lower probability of detection at the adversary could be achieved simultaneously. However, there is1140
a tradeoff in configuring the EM responses of the IRS elements to achieve the above two objectives simul-
taneously. Additionally, in a wireless-powered covert communication system where the receiver is equipped
with RF energy harvesting capability, the sender can also perform wireless power transfer or simultaneous
wireless information and power transfer.
Over the period, the increase in image steganography has necessitated the importance of a comprehensive1145
analysis of these techniques. Subsequently, [35] evaluated most of the latest image steganography tech-
niques based on visual quality, embedding payload, robustness, undetectability/security, and computational
complexity. The researchers observed that embedding secret data by any steganographic method alters
the visual quality of the cover image. Although the human eye may not easily detect such a modification,
there is a requirement of standard measuring techniques that may estimate the visual modification levels to1150
decide whether the steganographic method is perceptual transparent or not. Similarly, embedding capacity
(number of secret bits hidden per pixel) is also an important factor in the selection of an ideal steganography
method. However, it has been seen that there is no one technique that is rugged, with good visual quality
and high embedding capacity. Hence, there is a tradeoff between embedding capacity, security, and visual
quality depending upon the covert communication application and information sensitivity.1155
Apropos above, security/undetectability is also an important factor while evaluating image steganography
techniques. Generally, image-based steganographic techniques are vulnerable to various types of ste-
ganalysis detection attacks. Most of these attacks include: visual steganalysis, statistical steganalysis,
histogram-based analysis [149], RS (Regular and Singular) steganalysis [150], Chi-square analysis [151],
Bit plane analysis, and non-structural analysis [152].1160
Evaluating the security and performance efficiency of compressed and raw domain video steganographic
techniques, it is proffered that DCT/DQCT/DWT coefficients and CAVLC/CABAC entropy-based techniques
have high embedding capacity with low computational complexity. However, they infer a high impact on
the video quality. On the other hand, the inter-frame prediction scheme has low embedding capacity, low
computational complexity, and low impact on video quality. Correspondingly, some of the raw domain video1165
steganography techniques have shown promising embedding capacity, however, with weak security. For in-
stance, a BPCS technique [88] has displayed an embedding capacity ratio of up to 45%. But it is not robust
45

against signal processing, noises, and compression attacks. Similarly, another BPCS technique [90] proven
to be rugged against 3D–SPIHT and Motion-JPEG2000 compression have a maximum embedding capacity
ratio of 28%. Hence, it can be deduced that there always has to be a compromise between embedding1170
capacity, video quality, and security.
Coming towards audio steganography, robustness and security are not the main attributes of temporal do-
main steganographic methods. However, the conventional LSB technique and its variants facilitate data
hiding. Moreover, some of the LSB variants [153],[154],[155] can tolerate noise addition at low levels but
at the cost of very low embedding capacity. On the other hand, in comparison to time-domain methods,1175
data hiding in frequency (transform) domain gives better results in terms of SNR [95]. Hence, it can be
deduced that audio steganography techniques in the transform domain benefit from the frequency masking
effect. Furthermore, to avoid stego signal degradation, most transform domain steganography techniques
use a perceptual model to determine the acceptable amount of embedded data. In addition, these tech-
niques have succeeded in realizing the security and the robustness of secret data against simple audio1180
signal manipulations such as amplification, filtration, or re-sampling. Moreover, embedded data can survive
noisy transmission environment or data compression induced by some of the encoding techniques such as:
ACELP, G.729, etc.
Compared to temporal and transform domain methods, robustness and security of embedded data are the
main characteristics of in-encoder approaches. Similarly, some of the coded domain methods have shown1185
considerably high embedding capacity compared to the underlying codecs' rate. Since hidden data are not
affected by the encoding process, data-extraction correctness is fulfilled in tandem-free operation. How-
ever, the integrity of embedded data in in-encoder audio steganography can be compromised if a voice
encoder/decoder exists in the network. In addition, hidden data can also be transformed if a voice en-
hancement algorithm such as echo or noise reduction is deployed in the network. Since bit-stream is more1190
sensitive to alterations than the original audio signal, the embedding capacity should be kept small to avoid
hidden data perceptibility. Hence, coded domain techniques are well suited for real-time applications.
Accordingly, [16] examined various audio steganography techniques (discussed in this study) to evaluate the
behavior of the respective algorithms when music and speech-audio signals are used separately to convey
secret messages. The results indicated that data embedded in music is less detectable than speech audio1195
signals. In this regard, reference steganalysis differentiates between the cover and stego signals based on
features extracted from high frequencies (lower in energy). Hence, the signal discontinuities are intensified
due to the noise generated by data embedding. As the number of low-energy frequency components in
music audio signals is smaller than that in speech audio signals, the detection rate is expected to be lower.
As far as text steganography is concerned, there are many variants of the two categories discussed in this1200
paper with their own advantages and disadvantages, e.g., [156],[157],[158],[159]. Some of such techniques
have utilized the format of the texts by just manipulating the different symbols of different languages. In
this regard, a lot of work is done in Arabic texts because this language contains lots of points that can be
manipulated very easily. Similarly, Indian languages and linguistic-based methods have also been explored.
However, most of these techniques are not secure against attacks involving detection and removal of embed-1205
ded data [25]. Therefore, there is a need to define a minimum standard (benchmarking) to evaluate these
techniques. Accordingly, a standard definition of robustness is also required to help design and development
of secure embedding methods. Nonetheless, data hidden in the text has a variety of applications, including
copyright verification, authentication, and annotation. In addition, most of the existing text steganography
46

techniques are used only for text data. This can further be extended for other forms, such as tabular data1210
and logos. Also, there is a need to devise mechanisms to increase the embedding capacity of the existing
methods.
3.1. Open Research Challenges
Some of the open research challenges identified in the literature are:
a. Lightweight Encryption-based Secure Steganographic Techniques: For enhanced security of hid-1215
den messages, it is suggested that a steganographic technique integrating cryptography and error
correction code should be developed. The encryption and encoding of the secret data prior to the
embedding process will provide an additional security level and robustness against attackers during
the transmission. However, conventional encryption algorithms may be expensive. Therefore, there is
a need to incorporate lightweight cryptography to augment the security and performance efficiency of1220
the covert communication [38],[63].
b. Hybrid Techniques: Most of the existing steganalysis techniques target a particular steganographic
method. Hence, to increase the security of covert data and to add an element of confusion against
steganalysis techniques, a hybrid of multiple steganographic techniques should be developed duly
considering the strengths and weaknesses of existing methods. Hybrid steganography is likely to1225
provide a good line of defense against usual steganalysis [160].
c. Location-Sensitive Embedding: Recently, location-sensitive embedding techniques, also known as
adaptive steganography, have been developed to: improve payload, minimize distortion, and dynam-
ically secure secret data during the steganographic process. However, still, there is some time until
location-sensitive steganography gets matured with respect to modern steganalysis [41].1230
d. Intelligent Steganographic Techniques: Currently, no steganographic method satisfies all the se-
curity and performance requirements of an ideal steganographic technique, including imperceptibility,
ruggedness to steganalysis, embedding capacity, and computation complexity. Most of the existing
techniques may compromise security while increasing the embedding capacity or vice-versa. If a
method is considered secure with a high covert rate, it may have high computation complexity. Hence,1235
there is a need to design and develop an intelligent covert communication technique that should adjust
the level of security, embedding rate, and complexity concerning the underlying application environ-
ment.
e. Estimation of IRS Channels: In IRS-based covert channels, the wave manipulation is dependant on
the CSI to augment covert communication. However, the system performance is heavily dependent on1240
the availability and accuracy of CSI. Whereas instantaneous CSI of the reflection channels is difficult to
acquire due to an IRS's nearly passive operation. Therefore, a ML-based approach may be explored
that allows channel estimation without explicit feedback/detection [24].
f. IRS-Based Information/Communication Theoretic Models: An IRS-enhanced covert channel is
expected to provide enhanced embedding capacity due to its signal intensification and cancellation1245
capabilities. Hence, the conventional covert channels' embedding capacity needs to be revisited by
taking into account channel programmability. Furthermore, scaling laws of IRS-enhanced covert chan-
nel capacity need to be derived for a fundamental understanding of achievable performance limits [24].
47

g. Impact of Multiple IRSs: IRSs are anticipated to be deployed on the superficies of environmental
objects located with perplexing spatial patterns. Therefore, it is a common scenario that multiple IRSs1250
jointly shape the propagation environment. The aggregated impact of the operation of ambient IRSs
on IRS-enhanced covert communication is worth investigating by considering their spatial distribution.
h. Improved Video Steganography: To increase the ruggedness of video steganographic methods
against various attacks, it is recommended that video steganographic techniques incorporate selective
portions of the video (ROI (Regions of Interest)) to embed secret data instead of the entire video.1255
E.g., secret messages can be hidden into that portions of the video which contain human faces,
human bodies, cars, or any other moving object. It is believed that it would be very challenging for
an attacker to detect messages/data concealed into the ROI, which is changed frame-to-frame [21].
Moreover, since transform domain techniques are more robust against signal processing operations
and compression processes. There is a need to explore a video steganography technique that utilizes1260
transformation coefficients of the ROI rather than the actual pixel domain [21].
i. Augmenting ToR Pluggable Transports to Avoid Detection: ToR pluggable transports use various
obfuscation techniques to hide ToR traffic. However, flow analysis of multiple features such as the
packet size, the number of bytes sent, and the maximum packet size help an adversary in profiling the
pluggable transports. Hence, there is a need to develop sophisticated pluggable transports that mimic1265
a host protocol's all the features that an adversary may use to make a distinction between the two.
j. Lightweight Blockchain-based Steganography for Embedded Devices With an increase in the
use of embedded/micro IoT devices, hundreds of vendors have joined the IoT device manufacturing
industry. However, due to the heterogeneous nature of IoT devices and the use of different protocols
(mostly proprietary technologies), these devices have issues related to compatibility and failure to1270
establish secure communication. Hence, there is a need to design a lightweight blockchain-based
steganography technique for micro IoT devices.
4. Conclusion
In this work, a comparative study of the current state-of-the-art steganography techniques and ap-
proaches is presented. The classic methods range from wireless communication to image, video, audio,1275
and text-based steganography. In addition, we also reviewed some of the latest trends, including blockchain,
pluggable transport protocols, underwater acoustic communication, MODBUS protocol, and backscatter-
based covert communication. The survey is supported by an in-depth discussion on the strengths and
weaknesses of each type of steganography method. Finally, the discussion is summed up with a gist of
open research challenges.1280
The key takeaway from the discussion is that an ideal covert communication technique should provide a
higher embedding payload, LPD, and resistance against most types of steganalysis. However, in reality,
an ideal steganography technique does not exist. All of the methods discussed in this article have unique
strengths and limitations that depend on the adopted algorithm and the type of their applications. There-
fore, the significance of an embedding algorithm depends on the given application and the level of security1285
required.
48

References
[1] F.-X. Standaert, Introduction to side-channel attacks, in: I. M. Verbauwhede (Ed.), Secure Integrated Circuits and
Systems, Springer US, Boston, MA, 2010, pp. 27–42. doi:10.1007/978-0-387- 71829-3_2.
URL https://doi.org/10.1007/978-0-387-71829- 3 21290
[2] Y. Zhou, D. Feng, Side-channel attacks: Ten years after its publication and the impacts on cryptographic module
security testing., IACR Cryptol. ePrint Arch. 2005 (2005) 388.
[3] M. Rostami, F. Koushanfar, R. Karri, A primer on hardware security: Models, methods, and metrics, Proceedings
of the IEEE 102 (8) (2014) 1283–1295.
[4] Y. Yarom, K. Falkner, Flush+ reload: a high resolution, low noise, l3 cache side-channel attack, in: 23rd1295
{USENIX}Security Symposium ({USENIX}Security 14), 2014, pp. 719–732.
[5] P. C. Kocher, Timing attacks on implementations of diffie-hellman, rsa, dss, and other systems, in: Annual
International Cryptology Conference, Springer, 1996, pp. 104–113.
[6] P. Kocher, J. Jaffe, B. Jun, Differential power analysis, in: Annual international cryptology conference, Springer,
1999, pp. 388–397.1300
[7] D. Boneh, R. A. DeMillo, R. J. Lipton, On the importance of checking cryptographic protocols for faults, in:
International conference on the theory and applications of cryptographic techniques, Springer, 1997, pp. 37–51.
[8] A. Shamir, E. Tromer, Acoustic cryptanalysis: on nosy people and noisy machines, Online at http://people. csail.
mit. edu/tromer/acoustic (2004).
[9] C. C. Tiu, A new frequency-based side channel attack for embedded systems, Ph.D. thesis, University of1305
Waterloo (2005).
[10] N. Provos, P. Honeyman, Hide and seek: An introduction to steganography, IEEE security & privacy 1 (3) (2003)
32–44.
[11] G. J. Simmons, The prisoners’ problem and the subliminal channel, in: Advances in Cryptology, Springer, 1984,
pp. 51–67.1310
[12] N. F. Maxemchuk, Electronic document distribution, AT&T Technical Journal 73 (5) (1994) 73–80.
doi:10.1002/j.1538-7305.1994.tb00612.x.
[13] G. Simmons, The history of subliminal channels, IEEE Journal on Selected Areas in Communications 16 (4)
(1998) 452–462. doi:10.1109/49.668969.
[14] Simmons, G.J., How to insure that data acquired to verify treaty compliance are trustworthy, Proceedings of the1315
IEEE 76 (5) (1988) 621–627. doi:10.1109/5.4446.
[15] D. L. SCHILLING, in: Meteor burst communications Theory and practice, John Wiley & Sons, New York, 1993, p.
480. [link].
URL https://ci.nii.ac.jp/naid/10000050398/en/
[16] F. Djebbar, B. Ayad, K. A. Meraim, H. Hamam, Comparative study of digital audio steganography techniques,1320
EURASIP Journal on Audio, Speech, and Music Processing 2012 (1) (2012) 1–16.
[17] F. Djebbar, B. Ayad, H. Hamam, K. Abed-Meraim, A view on latest audio steganography techniques, in:
International Conference on Innovations in Information Technology, IEEE, Abu Dhabi, UAE, 2011, pp. 409–414.
doi:10.1109/INNOVATIONS.2011.5893859.
49

[18] M. Hussain, M. Hussain, A survey of image steganography techniques (2013).1325
[19] J. Kour, D. Verma, Steganography techniques–a review paper, International Journal of Emerging Research in
Management & Technology 3 (5) (2014) 132–135.
[20] H. Kaur, J. Rani, A survey on different techniques of steganography, in: MATEC Web of Conferences, Vol. 57,
EDP Sciences, 2016, p. 02003.
[21] R. J. Mstafa, K. M. Elleithy, Compressed and raw video steganography techniques: a comprehensive survey and1330
analysis, Multimedia Tools and Applications 76 (20) (2017) 21749–21786.
[22] S. Mishra, V. K. Yadav, M. C. Trivedi, T. Shrimali, Audio steganography techniques: A survey, in: S. K. Bhatia,
K. K. Mishra, S. Tiwari, V. K. Singh (Eds.), Advances in Computer and Computational Sciences, Springer,
Singapore, 2018, pp. 581–589.
[23] M. Forouzesh, P. Azmi, A. Kuhestani, P. L. Yeoh, Covert communication and secure transmission over untrusted1335
relaying networks in the presence of multiple wardens, IEEE Transactions on Communications 68 (6) (2020)
3737–3749. doi:10.1109/TCOMM.2020.2978206.
[24] X. Lu, E. Hossain, T. Shafique, S. Feng, H. Jiang, D. Niyato, Intelligent reflecting surface (irs)-enabled covert
communications in wireless networks, IEEE Network (2020).
[25] R. Kumar, H. Singh, Recent Trends in Text Steganography with Experimental Study, Springer International1340
Publishing, Cham, 2020, pp. 849–872. doi:10.1007/978-3-030- 22277-2_34.
URL https://doi.org/10.1007/978-3-030-22277- 2 34
[26] T.-X. Zheng, H.-M. Wang, D. W. K. Ng, J. Yuan, Multi-antenna covert communications in random wireless
networks, IEEE Transactions on Wireless Communications 18 (3) (2019) 1974–1987.
[27] E. Bj ¨
ornson, J. Hoydis, M. Kountouris, M. Debbah, Massive mimo systems with non-ideal hardware: Energy1345
efficiency, estimation, and capacity limits, IEEE Transactions on Information Theory 60 (11) (2014) 7112–7139.
doi:10.1109/TIT.2014.2354403.
[28] R. Soltani, D. Goeckel, D. Towsley, B. A. Bash, S. Guha, Covert wireless communication with artificial noise
generation, IEEE Transactions on Wireless Communications 17 (11) (2018) 7252–7267.
[29] R. Negi, S. Goel, Secret communication using artificial noise, in: 62nd IEEE Vehicular Technology Conference,1350
Vol. 3, Dallas, TX, USA, 2005, pp. 1906–1910. doi:10.1109/VETECF.2005.1558439.
[30] S. Kumar, A. Singh, M. Kumar, Covert communication integrates into wavelet packet transform ofdm system over
rayleigh fading channel, Wireless Networks 26 (1) (2020) 81–89.
[31] S. W. Kim, H. Q. Ta, Covert communication by exploiting node multiplicity and channel variations, in: IEEE
International Conference on Communications (ICC), IEEE, 2020, pp. 1–6.1355
[32] L. Tao, W. Yang, S. Yan, D. Wu, X. Guan, D. Chen, Covert communication in downlink noma systems with
random transmit power, IEEE Wireless Communications Letters (2020).
[33] H. ZivariFard, M. R. Bloch, A. Nosratinia, Keyless covert communication via channel state information, arXiv
preprint arXiv:2003.03308 (2020).
[34] N. Hou, Y. Zheng, Cloaklora: A covert channel over lora phy, in: 28th International Conference on Network1360
Protocols (ICNP), IEEE, Madrid, Spain, 2020, pp. 1–11. doi:10.1109/ICNP49622.2020.9259364.
50

[35] M. Hussain, A. W. A. Wahab, Y. I. B. Idris, A. T. Ho, K.-H. Jung, Image steganography in spatial domain: A
survey, Signal Processing: Image Communication 65 (2018) 46–66.
doi:https://doi.org/10.1016/j.image.2018.03.012.
URL https://www.sciencedirect.com/science/article/pii/S092359651830256X1365
[36] C.-K. Chan, L. Cheng, Hiding data in images by simple lsb substitution, Pattern Recognition 37 (3) (2004)
469–474. doi:https://doi.org/10.1016/j.patcog.2003.08.007.
URL https://www.sciencedirect.com/science/article/pii/S003132030300284X
[37] R. Amirtharajan, J. B. Balaguru Rayappan, An intelligent chaotic embedding approach to enhance stego-image
quality, Information Sciences 193 (2012) 115–124. doi:https://doi.org/10.1016/j.ins.2012.01.010.1370
URL https://www.sciencedirect.com/science/article/pii/S0020025512000291
[38] K. Muhammad, J. Ahmad, N. U. Rehman, Z. Jan, M. Sajjad, Cisska-lsb: color image steganography using stego
key-directed adaptive lsb substitution method, Multimedia Tools and Applications 76 (6) (2017) 8597–8626.
[39] K. Muhammad, M. Sajjad, S. W. Baik, Dual-level security based cyclic18 steganographic method and its
application for secure transmission of keyframes during wireless capsule endoscopy, Journal of medical systems1375
40 (5) (2016) 114.
[40] T. Pevn`
y, T. Filler, P. Bas, Using high-dimensional image models to perform highly undetectable steganography,
in: International Workshop on Information Hiding, Springer, 2010, pp. 161–177.
[41] H.-D. Yuan, Secret sharing with multi-cover adaptive steganography, Information Sciences 254 (2014) 197–212.
doi:https://doi.org/10.1016/j.ins.2013.08.012.1380
URL https://www.sciencedirect.com/science/article/pii/S0020025513005720
[42] D.-C. Wu, W.-H. Tsai, A steganographic method for images by pixel-value differencing, Pattern Recognition
Letters 24 (9) (2003) 1613–1626. doi:https://doi.org/10.1016/S0167-8655(02)00402-6.
URL https://www.sciencedirect.com/science/article/pii/S0167865502004026
[43] C. Balasubramanian, S. Selvakumar, S. Geetha, High payload image steganography with reduced distortion1385
using octonary pixel pairing scheme, Multimedia tools and applications 73 (3) (2014) 2223–2245.
[44] X. Zhang, S. Wang, Efficient steganographic embedding by exploiting modification direction, IEEE
Communications Letters 10 (11) (2006) 781–783. doi:10.1109/LCOMM.2006.060863.
[45] A. Ali M, S. Ghazali, R. Amjad, A. Mohammed Hazim, S. Tanzila, Data hiding based on improved exploiting
modification direction method and huffman coding, Journal of Intelligent Systems 23 (4) (2014) 451–459.1390
doi:doi:10.1515/jisys-2014-0007.
URL https://doi.org/10.1515/jisys-2014-0007
[46] A. Masoud, I. Subariah, Adaptive steganography scheme using more surrounding pixels, in: International
Conference On Computer Design and Applications, Vol. 1, IEEE, Qinhuangdao, China, 2010, pp.
V1–225–V1–229. doi:10.1109/ICCDA.2010.5541442.1395
[47] T. Mingwei, S. Wen, C. Xiaoliang, H. Jie, An image information hiding using adaptation and radix, Optik 126 (23)
(2015) 4136–4141. doi:https://doi.org/10.1016/j.ijleo.2015.07.200.
URL https://www.sciencedirect.com/science/article/pii/S0030402615007536
[48] W. Hong, Adaptive image data hiding in edges using patched reference table and pair-wise embedding
technique, Information Sciences 221 (2013) 473–489. doi:https://doi.org/10.1016/j.ins.2012.09.013.1400
URL https://www.sciencedirect.com/science/article/pii/S0020025512006068
51

[49] J. Chen, A pvd-based data hiding method with histogram preserving using pixel pair matching, Signal
Processing: Image Communication 29 (3) (2014) 375–384.
doi:https://doi.org/10.1016/j.image.2014.01.003.
URL https://www.sciencedirect.com/science/article/pii/S09235965140001741405
[50] V. Potdar, E. Chang, Grey level modification steganography for secret communication, in: 2nd IEEE International
Conference on Industrial Informatics, 2004. INDIN ’04. 2004, Berlin, Germany, 2004, pp. 223–228.
doi:10.1109/INDIN.2004.1417333.
[51] K. Muhammad, J. Ahmad, H. Farman, Z. Jan, M. Sajjad, S. W. Baik, A secure method for color image
steganography using gray-level modification and multi-level encryption, KSII Transactions on Internet and1410
Information Systems (TIIS) 9 (5) (2015) 1938–1962.
[52] Y.-H. Yu, C.-C. Chang, Y.-C. Hu, Hiding secret data in images via predictive coding, Pattern Recognition 38 (5)
(2005) 691–705. doi:https://doi.org/10.1016/j.patcog.2004.11.006.
URL https://www.sciencedirect.com/science/article/pii/S0031320304003930
[53] W. Hong, T.-S. Chen, C.-W. Shiu, Reversible data hiding for high quality images using modification of prediction1415
errors, Journal of Systems and Software 82 (11) (2009) 1833–1842, sI: TAIC PART 2007 and MUTATION 2007.
doi:https://doi.org/10.1016/j.jss.2009.05.051.
URL https://www.sciencedirect.com/science/article/pii/S0164121209001204
[54] H.-C. Wu, H.-C. Wang, C.-S. Tsai, C.-M. Wang, Reversible image steganographic scheme via predictive coding,
Displays 31 (1) (2010) 35–43. doi:https://doi.org/10.1016/j.displa.2009.10.002.1420
URL https://www.sciencedirect.com/science/article/pii/S0141938209000596
[55] I. F. Jafar, K. A. Darabkh, R. T. Al-Zubi, R. A. Al Na’mneh, Efficient reversible data hiding using multiple
predictors, The Computer Journal 59 (3) (2016) 423–438. doi:10.1093/comjnl/bxv067.
[56] E. Pietka, Image standardization in pacs, in: I. N. BANKMAN (Ed.), Handbook of Medical Imaging, Biomedical
Engineering, Academic Press, San Diego, 2000, pp. 783–801.1425
doi:https://doi.org/10.1016/B978-012077790-7/50056-4.
URL https://www.sciencedirect.com/science/article/pii/B9780120777907500564
[57] Z. Pan, S. Hu, X. Ma, L. Wang, Reversible data hiding based on local histogram shifting with multilayer
embedding, Journal of Visual Communication and Image Representation 31 (2015) 64–74.
doi:https://doi.org/10.1016/j.jvcir.2015.05.005.1430
URL https://www.sciencedirect.com/science/article/pii/S1047320315000838
[58] W. Luo, F. Huang, J. Huang, Edge adaptive image steganography based on lsb matching revisited, IEEE
Transactions on Information Forensics and Security 5 (2) (2010) 201–214. doi:10.1109/TIFS.2010.2041812.
[59] K.-H. Jung, K.-Y. Yoo, Data hiding using edge detector for scalable images, Multimedia tools and applications
71 (3) (2014) 1455–1468.1435
[60] S. Sun, A novel edge based image steganography with 2k correction and huffman encoding, Information
Processing Letters 116 (2) (2016) 93–99. doi:https://doi.org/10.1016/j.ipl.2015.09.016.
URL https://www.sciencedirect.com/science/article/pii/S0020019015001738
[61] R.-Z. Wang, Y.-S. Chen, High-payload image steganography using two-way block matching, IEEE Signal
Processing Letters 13 (3) (2006) 161–164. doi:10.1109/LSP.2005.862603.1440
[62] M. A. Al-Husainy, Image steganography by mapping pixels to letters, Journal of Computer science 5 (1) (2009)
33.
52

[63] A. A. Abdulla, H. Sellahewa, S. A. Jassim, Stego quality enhancement by message size reduction and fibonacci
bit-plane mapping, in: International Conference on Research in Security Standardisation, Springer, 2014, pp.
151–166.1445
[64] A. Alan Anwar, S. Harin, J. Sabah A., Steganography based on pixel intensity value decomposition, in: Mobile
Multimedia/Image Processing, Security, and Applications, Vol. 9120, International Society for Optics and
Photonics, SPIE, 2014, pp. 19 – 27.
URL https://doi.org/10.1117/12.2050518
[65] G. Swain, S. K. Lenka, A better rgb channel based image steganography technique, in: International Conference1450
on Computing and Communication Systems, Springer, Vellore, India, 2011, pp. 470–478.
[66] V. Thanikaiselvan, S. Subashanthini, R. Amirtharajan, Pvd based steganography on scrambled rgb cover images
with pixel indicator, Journal of Artificial Intelligence 7 (2) (2014) 54.
[67] M. Khan, S. Muhammad, M. Irfan, R. Seungmin, B. Sung Wook, Image steganography using uncorrelated color
space and its application for security of visual contents in online social networks, Future Generation Computer1455
Systems 86 (2018) 951–960. doi:https://doi.org/10.1016/j.future.2016.11.029.
URL https://www.sciencedirect.com/science/article/pii/S0167739X16306768
[68] K. Muhammad, M. Sajjad, I. Mehmood, S. Rho, S. W. Baik, A novel magic lsb substitution method (m-lsb-sm)
using multi-level encryption and achromatic component of an image, Multimedia Tools and Applications 75 (22)
(2016) 14867–14893.1460
[69] N. N. El-Emam, M. Al-Diabat, A novel algorithm for colour image steganography using a new intelligent
technique based on three phases, Applied Soft Computing 37 (2015) 830–846.
doi:https://doi.org/10.1016/j.asoc.2015.08.057.
URL https://www.sciencedirect.com/science/article/pii/S1568494615005700
[70] S¸ . Do ˘
gan, A new data hiding method based on chaos embedded genetic algorithm for color image, Artificial1465
Intelligence Review 46 (1) (2016) 129–143.
[71] D.-C. Wu, Y.-M. Wu, Covert communication via the qr code image by a data hiding technique based on module
shape adjustments, IEEE Open Journal of the Computer Society 1 (2020) 12–34.
[72] Qr code security: What are qr codes and are they safe to use?, 2021, Last accessed on 20 Jul 2021.
URL https://www.kaspersky.com.au/resource-center/definitions/what-is-a-qr-code-how-to-scan1470
[73] C. Boncelet, Chapter 7 - image noise models, in: A. Bovik (Ed.), The Essential Guide to Image Processing,
Academic Press, Boston, 2009, pp. 143–167. doi:https://doi.org/10.1016/B978-0-12- 374457-9.00007- X.
URL https://www.sciencedirect.com/science/article/pii/B978012374457900007X
[74] B. Liu, F. Liu, C. Yang, Y. Sun, Secure steganography in compressed video bitstreams, in: 3rd International
Conference on Availability, Reliability and Security, IEEE, Barcelona, Spain, 2008, pp. 1382–1387.1475
doi:10.1109/ARES.2008.140.
[75] H. Zhu, R. Wang, D. Xu, X. Zhou, Information hiding algorithm for h.264 based on the prediction difference of
intra 4x4, in: 3rd International Congress on Image and Signal Processing, Vol. 1, IEEE, Yantai, China, 2010, pp.
487–490. doi:10.1109/CISP.2010.5646792.
[76] S. K. Kapotas, A. N. Skodras, A new data hiding scheme for scene change detection in h.264 encoded video1480
sequences, in: International Conference on Multimedia and Expo, IEEE, Hannover, Germany, 2008, pp.
277–280. doi:10.1109/ICME.2008.4607425.
53

[77] C. Xu, X. Ping, T. Zhang, Steganography in compressed video stream, in: First International Conference on
Innovative Computing, Information and Control - Volume I (ICICIC’06), Vol. 1, IEEE, Beijing, China, 2006, pp.
269–272. doi:10.1109/ICICIC.2006.158.1485
[78] F. Pan, L. Xiang, X.-Y. Yang, Y. Guo, Video steganography using motion vector and linear block codes, in:
International Conference on Software Engineering and Service Sciences, IEEE, Beijing, China, 2010, pp.
592–595. doi:10.1109/ICSESS.2010.5552283.
[79] J. Huang, Y. Q. Shi, Reliable information bit hiding, IEEE Transactions on Circuits and Systems for Video
Technology 12 (10) (2002) 916–920. doi:10.1109/TCSVT.2002.804897.1490
[80] M. Steinebach, F. Petitcolas, F. Raynal, J. Dittmann, C. Fontaine, S. Seibel, N. Fates, L. Ferri, Stirmark
benchmark: audio watermarking attacks, in: International Conference on Information Technology: Coding and
Computing, IEEE, Las Vegas, NV, USA, 2001, pp. 49–54. doi:10.1109/ITCC.2001.918764.
[81] Y. Liu, Z. Li, X. Ma, J. Liu, A robust data hiding algorithm for h.264/avc video streams without intra-frame
distortion drift, in: Proceedings of the 2012 Second International Conference on Electric Information and Control1495
Engineering - Volume 01, ICEICE ’12, IEEE Computer Society, USA, 2012, p. 182–186.
[82] B. G. Mobasseri, M. P. Marcinak, Watermarking of mpeg-2 video in compressed domain using vlc mapping, in:
7th Workshop on Multimedia and Security, MM&Sec ’05, Association for Computing Machinery, New York, USA,
2005, p. 91–94. doi:10.1145/1073170.1073187.
URL https://doi.org/10.1145/1073170.10731871500
[83] R. Wang, L. Hu, D. Xu, A watermarking algorithm based on the cabac entropy coding for h. 264avc, Journal of
Computer Information System 7 (6) (2011) 2132–2141.
[84] L. Der-Chyuan, H. Chen-Hao, Lsb steganographic method based on reversible histogram transformation function
for resisting statistical steganalysis, Information Sciences 188 (2012) 346–358.
doi:https://doi.org/10.1016/j.ins.2011.06.003.1505
URL https://www.sciencedirect.com/science/article/pii/S0020025511002775
[85] R. Zhang, V. Sachnev, M. B. Botnan, H. J. Kim, J. Heo, An efficient embedder for bch coding for steganography,
IEEE Transactions on Information Theory 58 (12) (2012) 7272–7279. doi:10.1109/TIT.2012.2217072.
[86] S. K. Moon, R. D. Raut, Analysis of secured video steganography using computer forensics technique for
enhance data security, in: 2nd International Conference on Image Information Processing (ICIIP-2013), IEEE,1510
Shimla, India, 2013, pp. 660–665. doi:10.1109/ICIIP.2013.6707677.
[87] E. Kawaguchi, R. O. Eason, Principles and applications of BPCS steganography, in: A. G. Tescher, B. Vasudev,
V. M. B. Jr., B. Derryberry (Eds.), Multimedia Systems and Applications, Vol. 3528, International Society for
Optics and Photonics, SPIE, 1999, pp. 464 – 473.
URL https://doi.org/10.1117/12.3374361515
[88] S. Sun, A new information hiding method based on improved bpcs steganography, Adv. MultiMedia 2015 (Jan.
2015). doi:10.1155/2015/698492.
URL https://doi.org/10.1155/2015/698492
[89] S. Jeremiah, N. Hideki, N. S. Mahdad, K. Eiji, Bpcs steganography using ezw lossy compressed images, Pattern
Recognition Letters 23 (13) (2002) 1579–1587. doi:https://doi.org/10.1016/S0167-8655(02)00122-8.1520
URL https://www.sciencedirect.com/science/article/pii/S0167865502001228
[90] H. Noda, T. Furuta, M. Niimi, E. Kawaguchi, Application of bpcs steganography to wavelet compressed video, in:
International Conference on Image Processing, (ICIP ’04)., Vol. 4, IEEE, Singapore, 2004, pp. 2147–2150 Vol. 4.
doi:10.1109/ICIP.2004.1421520.
54

[91] K. Patel, K. K. Rora, K. Singh, S. Verma, Lazy wavelet transform based steganography in video, in: International1525
Conference on Communication Systems and Network Technologies, IEEE, Gwalior, India, 2013, pp. 497–500.
doi:10.1109/CSNT.2013.109.
[92] H. Kekre, A. Athawale, S. Rao, U. Athawale, Information hiding in audio signals, International Journal of
Computer Applications 7 (9) (2010) 14–19.
[93] D. C. Kar, C. J. Mulkey, A multi-threshold based audio steganography scheme, Journal of Information Security1530
and Applications 23 (2015) 54–67. doi:https://doi.org/10.1016/j.jisa.2015.02.001.
URL https://www.sciencedirect.com/science/article/pii/S2214212615000034
[94] P. Singh, A comparative study of audio steganography techniques, J Int Res J Eng Technol 3 (4) (2016) 581–585.
[95] W. Bender, D. Gruhl, N. Morimoto, A. Lu, Techniques for data hiding, IBM systems journal 35 (3.4) (1996)
313–336.1535
[96] S. Shirali-Shahreza, M. Manzuri-Shalmani, High capacity error free wavelet domain speech steganography, in:
International Conference on Acoustics, Speech and Signal Processing, IEEE, Las Vegas, NV, USA, 2008, pp.
1729–1732. doi:10.1109/ICASSP.2008.4517963.
[97] K. Gopalan, S. Wenndt, A. Noga, D. Haddad, S. Adams, Covert speech communication via cover speech by tone
insertion, in: Aerospace Conference Proceedings (Cat. No.03TH8652), Vol. 4, IEEE, Big Sky, MT, USA, 2003,1540
pp. 4 1647–4 1653. doi:10.1109/AERO.2003.1235095.
[98] F. Djebbar, B. Ayad, K. Abed-Meraim, H. Hamam, Unified phase and magnitude speech spectra data hiding
algorithm, Security and Communication Networks 6 (8) (2013) 961–971.
[99] D. Guerchi, H. Harmain, T. Rabie, E. Mohamed, Speech secrecy: An fft-based approach, International Journal of
Mathematics and Computer Science 3 (2) (2008) 1–19.1545
[100] A. H. Ali, L. E. George, A. Zaidan, M. R. Mokhtar, High capacity, transparent and secure audio steganography
model based on fractal coding and chaotic map in temporal domain, Multimedia Tools and Applications 77 (23)
(2018) 31487–31516.
[101] A. Nishimura, Data hiding for audio signals that are robust with respect to air transmission and a speech codec,
in: International Conference on Intelligent Information Hiding and Multimedia Signal Processing, IEEE, Harbin,1550
China, 2008, pp. 601–604. doi:10.1109/IIH- MSP.2008.333.
[102] K. Hofbauer, G. Kubin, High-rate data embedding in unvoiced speech, in: 9th International Conference on
Spoken Language Processing, ISCA, Pittsburgh, PA, USA, 2006, pp. 1–4.
[103] G. S. Kang, T. M. Moran, D. A. Heide, Hiding information under speech, Tech. rep., NAVAL RESEARCH LAB
WASHINGTON DC TRANSMISSION TECHNOLOGY BRANCH (2005).1555
[104] B. Geiser, P. Vary, High rate data hiding in acelp speech codecs, in: International Conference on Acoustics,
Speech and Signal Processing, IEEE, Las Vegas, NV, USA, 2008, pp. 4005–4008.
doi:10.1109/ICASSP.2008.4518532.
[105] N. Aoki, A technique of lossless steganography for g.711 telephony speech, in: International Conference on
Intelligent Information Hiding and Multimedia Signal Processing, IEEE, Harbin, China, 2008, pp. 608–611.1560
doi:10.1109/IIH-MSP.2008.122.
[106] Aoki, Naofumi, A semi-lossless steganography technique for g.711 telephony speech, in: 6th International
Conference on Intelligent Information Hiding and Multimedia Signal Processing, IEEE, Darmstdt, Germany,
2010, pp. 534–537. doi:10.1109/IIHMSP.2010.136.
55

[107] Y. F. Huang, S. Tang, J. Yuan, Steganography in inactive frames of voip streams encoded by source codec, IEEE1565
Transactions on Information Forensics and Security 6 (2) (2011) 296–306. doi:10.1109/TIFS.2011.2108649.
[108] J. Peng, S. Tang, Covert communication over voip streaming media with dynamic key distribution and
authentication, IEEE Transactions on Industrial Electronics (2020).
[109] N. Sashaa, A. Saeed, J. Daryl, Covert communication using modbus protocol in iot devices, in: 7th International
Conference of Control Systems, and Robotics (CDSR’20), ASET, Niagara Falls, Ca, 2020, pp. 1–8.1570
doi:10.11159/cdsr20.127.
[110] Pluggable transports for circumvention, 2022, Last accessed on 25 Apr 2022. (2012).
URL https://gitweb.torproject.org/torspec.git/blob/HEAD:/proposals/180-pluggable-transport.txt
[111] Z. Weinberg, J. Wang, V. Yegneswaran, L. Briesemeister, S. Cheung, F. Wang, D. Boneh, Stegotorus: A
camouflage proxy for the tor anonymity system, in: ACM Conference on Computer and Communications1575
Security, CCS ’12, Association for Computing Machinery, New York, NY, USA, 2012, p. 109–120.
doi:10.1145/2382196.2382211.
URL https://doi.org/10.1145/2382196.2382211
[112] J. Appelbaum, N. Mathewson, Pluggable transports for circumvention, 2022, Last accessed on 20 Apr 2022.
(2010).1580
URL https://gitweb.torproject.org/torspec.git/blob/HEAD:/proposals/180-pluggable-transport.txt.
[113] M. Leech, M. Ganis, Y. Lee, R. Kuris, D. Koblas, L. Jones, Socks protocol version 5–rfc 1928, Internet
Engineering Task Force (1996).
URL http://tools.ietf.org/html/rfc1928
[114] H. Mohajeri Moghaddam, B. Li, M. Derakhshani, I. Goldberg, Skypemorph: Protocol obfuscation for tor bridges,1585
in: ACM Conference on Computer and Communications Security, CCS ’12, Association for Computing
Machinery, New York, NY, USA, 2012, p. 97–108. doi:10.1145/2382196.2382210.
URL https://doi.org/10.1145/2382196.2382210
[115] Q. Wang, X. Gong, G. T. Nguyen, A. Houmansadr, N. Borisov, Censorspoofer: Asymmetric communication using
ip spoofing for censorship-resistant web browsing, in: ACM Conference on Computer and Communications1590
Security, CCS ’12, Association for Computing Machinery, New York, NY, USA, 2012, p. 121–132.
doi:10.1145/2382196.2382212.
URL https://doi.org/10.1145/2382196.2382212
[116] C. Brubaker, A. Houmansadr, V. Shmatikov, Cloudtransport: Using cloud storage for censorship-resistant
networking, in: International Symposium on Privacy Enhancing Technologies Symposium, Springer, 2014, pp.1595
1–20.
[117] A. Houmansadr, C. Brubaker, V. Shmatikov, The parrot is dead: Observing unobservable network
communications, in: IEEE Symposium on Security and Privacy, 2013, pp. 65–79. doi:10.1109/SP.2013.14.
[118] K. Shahbar, A. N. Zincir-Heywood, An analysis of tor pluggable transports under adversarial conditions, in: IEEE
Symposium Series on Computational Intelligence (SSCI), 2017, pp. 1–7. doi:10.1109/SSCI.2017.8280829.1600
[119] D. Fifield, C. Lan, R. Hynes, P. Wegmann, V. Paxson, Blocking-resistant communication through domain
fronting., Proc. Priv. Enhancing Technol. 2015 (2) (2015) 46–64.
[120] H. Xie, L. Wang, S. Yin, H. Zhao, H. Shentu, Adaptive meek technology for anti-traffic analysis, in: International
Conference on Networking and Network Applications (NaNA), 2020, pp. 102–107.
doi:10.1109/NaNA51271.2020.00025.1605
56

[121] J. Tian, G. Xiong, Z. Li, G. Gou, A survey of key technologies for constructing network covert channel, Security
and Communication Networks 2020 (2020).
[122] S. Underwood, Blockchain beyond bitcoin, Communications of the ACM 59 (11) (2016) 15–17.
[123] L. Liu, L. Liu, B. Li, Y. Zhong, S. Liao, L. Zhang, Msccs: A monero-based security-enhanced covert
communication system, Computer Networks 205 (2022) 108759.1610
doi:https://doi.org/10.1016/j.comnet.2021.108759.
URL https://www.sciencedirect.com/science/article/pii/S1389128621006010
[124] L. Zhang, Z. Zhang, Z. Jin, Y. Su, Z. Wang, An approach of covert communication based on the ethereum
whisper protocol in blockchain, International Journal of Intelligent Systems 36 (2) (2021) 962–996.
[125] F. Gao, L. Zhu, K. Gai, C. Zhang, S. Liu, Achieving a covert channel over an open blockchain network, IEEE1615
Network 34 (2) (2020) 6–13. doi:10.1109/MNET.001.1900225.
[126] X. Luo, P. Zhang, M. Zhang, H. Li, Q. Cheng, A novel covert communication method based on bitcoin transaction,
IEEE Transactions on Industrial Informatics 18 (4) (2022) 2830–2839. doi:10.1109/TII.2021.3100480.
[127] W. She, L. Huo, Z. Tian, Y. Zhuang, C. Niu, W. Liu, A double steganography model combining blockchain and
interplanetary file system, Peer-to-Peer Networking and Applications 14 (5) (2021) 3029–3042.1620
[128] L. Zhang, Z. Zhang, W. Wang, Z. Jin, Y. Su, H. Chen, Research on a covert communication model realized by
using smart contracts in blockchain environment, IEEE Systems Journal (2021)
1–12doi:10.1109/JSYST.2021.3057333.
[129] H. Cao, H. Yin, F. Gao, Z. Zhang, B. Khoussainov, S. Xu, L. Zhu, Chain-based covert data embedding schemes
in blockchain, IEEE Internet of Things Journal (2020) 1–1doi:10.1109/JIOT.2020.3040389.1625
[130] A. Young, M. Yung, The dark side of “black-box” cryptography or: Should we trust capstone?, in: N. Koblitz (Ed.),
Advances in Cryptology — CRYPTO ’96, Springer Berlin Heidelberg, Berlin, Heidelberg, 1996, pp. 89–103.
[131] J. C. Wray, An analysis of covert timing channels, Journal of Computer Security 1 (3-4) (1992) 219–232.
[132] J. Han, C. Huang, F. Shi, J. Liu, Covert timing channel detection method based on time interval and payload
length analysis, Computers & Security 97 (2020) 101952.1630
[133] J. Partala, Provably secure covert communication on blockchain, Cryptography 2 (3) (2018).
doi:10.3390/cryptography2030018.
URL https://www.mdpi.com/2410-387X/2/3/18
[134] S. T. Ali, P. McCorry, P. H.-J. Lee, F. Hao, Zombiecoin 2.0: managing next-generation botnets using bitcoin,
International Journal of Information Security 17 (4) (2018) 411–422.1635
[135] L. Zhang, Z. Zhang, W. Wang, R. Waqas, C. Zhao, S. Kim, H. Chen, A covert communication method using
special bitcoin addresses generated by vanitygen, Computers, Materials & Continua 65 (1) (2020) 597–616.
[136] O. Zohar, Unblockable chains–a poc on using blockchain as infrastructure for malware operations, 2022, Last
accessed on 22 May 2022. (2020).
URL https://github.com/platdrag/UnblockableChains1640
[137] J. Chen, S. Micali, Algorand, arXiv preprint arXiv:1607.01341 (2016).
[138] A. Fionov, Exploring covert channels in bitcoin transactions, in: 2019 International Multi-Conference on
Engineering, Computer and Information Sciences (SIBIRCON), 2019, pp. 0059–0064.
doi:10.1109/SIBIRCON48586.2019.8957861.
57

[139] S. Huang, X. Hou, W. Liu, G. Liu, Y. Dai, W. Tian, Mimicking ship-radiated noise with chaos signal for covert1645
underwater acoustic communication, IEEE Access 8 (2020) 180341–180351.
doi:10.1109/ACCESS.2020.3027022.
[140] Z. Yang, Q. Huang, Q. Zhang, Nicscatter: Backscatter as a covert channel in mobile devices, in: 23rd Annual
International Conference on Mobile Computing and Networking, MobiCom ’17, Association for Computing
Machinery, Snowbird, Utah, USA, 2017, p. 356–367. doi:10.1145/3117811.3117814.1650
URL https://doi.org/10.1145/3117811.3117814
[141] R. Rios, J. A. Onieva, J. Lopez, Covert communications through network configuration messages, Computers &
Security 39 (2013) 34–46, 27th IFIP International Information Security Conference.
doi:https://doi.org/10.1016/j.cose.2013.03.004.
URL https://www.sciencedirect.com/science/article/pii/S01674048130004971655
[142] D. A. Huffman, A method for the construction of minimum-redundancy codes, Proceedings of the IRE 40 (9)
(1952) 1098–1101. doi:10.1109/JRPROC.1952.273898.
[143] J. Classen, M. Schulz, M. Hollick, Practical covert channels for wifi systems, in: Conference on Communications
and Network Security (CNS), IEEE, Florence, Italy, 2015, pp. 209–217. doi:10.1109/CNS.2015.7346830.
[144] S. Zander, G. Armitage, P. Branch, A survey of covert channels and countermeasures in computer network1660
protocols, IEEE Communications Surveys Tutorials 9 (3) (2007) 44–57. doi:10.1109/COMST.2007.4317620.
[145] M. Taleby Ahvanooey, Q. Li, J. Hou, H. Dana Mazraeh, J. Zhang, Aitsteg: An innovative text steganography
technique for hidden transmission of text message via social media, IEEE Access 6 (2018) 65981–65995.
doi:10.1109/ACCESS.2018.2866063.
[146] A. Hero, Secure space-time communication, IEEE Transactions on Information Theory 49 (12) (2003)1665
3235–3249. doi:10.1109/TIT.2003.820010.
[147] B. A. Bash, D. Goeckel, D. Towsley, S. Guha, Hiding information in noise: fundamental limits of covert wireless
communication, IEEE Communications Magazine 53 (12) (2015) 26–31. doi:10.1109/MCOM.2015.7355562.
[148] S. Lee, R. J. Baxley, M. A. Weitnauer, B. Walkenhorst, Achieving undetectable communication, IEEE Journal of
Selected Topics in Signal Processing 9 (7) (2015) 1195–1205. doi:10.1109/JSTSP.2015.2421477.1670
[149] A. Ker, Steganalysis of lsb matching in grayscale images, IEEE Signal Processing Letters 12 (6) (2005)
441–444. doi:10.1109/LSP.2005.847889.
[150] J. Fridrich, M. Goljan, R. Du, Reliable detection of lsb steganography in color and grayscale images, in:
Workshop on Multimedia and Security: New Challenges, MM&Sec ’01, Association for Computing Machinery,
New York, NY, USA, 2001, p. 27–30. doi:10.1145/1232454.1232466.1675
URL https://doi.org/10.1145/1232454.1232466
[151] A. Westfeld, A. Pfitzmann, Attacks on steganographic systems, in: International workshop on information hiding,
Springer, 1999, pp. 61–76.
[152] V. Sedighi, R. Cogranne, J. Fridrich, Content-adaptive steganography by minimizing statistical detectability, IEEE
Transactions on Information Forensics and Security 11 (2) (2016) 221–234. doi:10.1109/TIFS.2015.2486744.1680
[153] N. Cvejic, T. Seppanen, Increasing robustness of lsb audio steganography using a novel embedding method, in:
International Conference on Information Technology: Coding and Computing, (ITCC 2004), Vol. 2, IEEE, Las
Vegas, NV, USA, 2004, pp. 533–537 Vol.2. doi:10.1109/ITCC.2004.1286709.
58

[154] N. Cvejic, T. Sepp¨
anen, Increasing robustness of lsb audio steganography by reduced distortion lsb coding., J.
Univers. Comput. Sci. 11 (1) (2005) 56–65.1685
[155] M. A. Ahmed, M. L. M. Kiah, B. Zaidan, A. Zaidan, A novel embedding method to increase capacity and
robustness of low-bit encoding audio steganography technique using noise gate software logic algorithm,
Journal of Applied Sciences 10 (1) (2010) 59–64.
[156] R. Kumar, A. Malik, S. Singh, B. Kumar, S. Chand, A space based reversible high capacity text steganography
scheme using font type and style, in: International Conference on Computing, Communication and Automation1690
(ICCCA), IEEE, Greater Noida, India, 2016, pp. 1090–1094. doi:10.1109/CCAA.2016.7813878.
[157] S. Shi, Y. Qi, Y. Huang, An approach to text steganography based on search in internet, in: International
Computer Symposium (ICS), IEEE, Chiayi, Taiwan, 2016, pp. 227–232. doi:10.1109/ICS.2016.0052.
[158] M. Fateh, M. Rezvani, An email-based high capacity text steganography using repeating characters,
International Journal of Computers and Applications 43 (3) (2021) 226–232.1695
arXiv:https://doi.org/10.1080/1206212X.2018.1517713,doi:10.1080/1206212X.2018.1517713.
URL https://doi.org/10.1080/1206212X.2018.1517713
[159] A. Majumder, S. Changder, A novel approach for text steganography: Generating text summary using reflection
symmetry, Procedia Technology 10 (2013) 112–120.
[160] M. Hussain, A. W. Abdul Wahab, N. Javed, K.-H. Jung, Hybrid data hiding scheme using right-most digit1700
replacement and adaptive least significant bit for digital images, Symmetry 8 (6) (2016) 41.
59