See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/363172933

Digital Watermarking System for Copyright Protection and Authentication of

Images Using Cryptographic Techniques

Article in Applied Sciences · August 2022

DOI: 10.3390/app12178724

CITATIONS READS

22 435

4 authors:

Sanivarapu Prasanth Vaidya Rajesh N V P S Kandala

BVRIT Hyderabad College of Engineering for Women VIT-AP University
29 PUBLICATIONS 393 CITATIONS 24 PUBLICATIONS 590 CITATIONS

SEE PROFILE SEE PROFILE

Khalid Hosny Mostafa M. Fouda

Zagazig University Idaho State University
197 PUBLICATIONS 5,506 CITATIONS 285 PUBLICATIONS 4,753 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Khalid Hosny on 01 September 2022.

The user has requested enhancement of the downloaded file.

 applied
sciences
Article
Digital Watermarking System for Copyright Protection and
Authentication of Images Using Cryptographic Techniques
Prasanth Vaidya Sanivarapu 1 , Kandala N. V. P. S. Rajesh 2 , Khalid M. Hosny 3 and Mostafa M. Fouda 4, *

1 Department of CSE, Aditya Engineering College, Surampalem 533437, India

2 School of Electronics Engineering, VIT-AP University, Vijayawada 522237, India
3 Faculty of Computers and Informatics, Zagazig University, Zagazig 44519, Egypt
4 Department of Electrical and Computer Engineering, Idaho State University, Pocatello, ID 83209, USA
* Correspondence: [email protected]; Tel.: +1-(208)-282-7768

Abstract: Digital images are transferred with ease through the network. Many users are using the
images without the knowledge of the owners. Therefore, a novel watermarking scheme is proposed
to ensure copyright protection and authentication of images using cryptography techniques. Here,
a quick response (QR) image is generated for a watermark image that contains public and private
keys prepared using a cryptosystem. Later, this QR image is scrambled using a chaotic logistic map.
The public and private keys are used to cipher and decipher the data. Next, the scrambled QR
watermark is embedded into a color image using a single-level discrete wavelet transform followed
by singular value decomposition using the key value. Finally, the inverse process is applied to extract
the watermark. The proposed method is validated using various image processing attacks. The
results are then compared with state-of-the-art watermarking schemes. The experimental results
show that the scheme provides good results in terms of robustness and imperceptibility.
Citation: Sanivarapu, P.V.;
Rajesh, K.N.V.P.S.; Hosny, K.M.; Keywords: digital watermarking; invisible watermark; QR code; RSA; singular value decomposition;
Fouda, M.M. Digital Watermarking discrete wavelet transform
System for Copyright Protection and
Authentication of Images Using
Cryptographic Techniques. Appl. Sci.
2022, 12, 8724. https://doi.org/ 1. Introduction
10.3390/app12178724
Recently, with the development of long-range informal communication on the web,
Academic Editors: David Megías, the capacity and dissemination of interactive media content have become extremely simple.
Minoru Kuribayashi and Wojciech On the other hand, this simplicity has led to the need for copyright protection, blocking
Mazurczyk information theft, and data genuineness [1,2].
Received: 5 August 2022
To handle the above issues, digital watermarking has emerged as an appropriate
Accepted: 27 August 2022
solution. Digital watermarking is a way of embedding a watermark into a significant
Published: 31 August 2022
image/media. A watermark acts as copyright data, shielding advanced information from
illicit replication and conveyance [3,4]. A watermark is a sort of marker clandestinely
Publisher’s Note: MDPI stays neutral
inserted in a signal (audio, video, or image information). A watermark embedded into
with regard to jurisdictional claims in
media may or may not relate to it. Watermarks are utilized to check the realness or
published maps and institutional affil-
uprightness of the watermarked signal [5,6].
iations.
Watermarking is a strategy that is broadly utilized and ceaselessly created by utiliz-
ing different strategies and executions [7,8]. In the proposed method, discrete wavelet
transform (DWT) and singular value decomposition (SVD) techniques are combined to
Copyright: © 2022 by the authors.
accomplish the vigor and imperceptibility of the watermark. The scheme is generally
Licensee MDPI, Basel, Switzerland. achievable for clients and has an oddity edge over the other existing digital watermarking
This article is an open access article methods [9]. The idea of embedding the watermark information is to prevent intruders or
distributed under the terms and other members from claiming to be the rightful owner of the data [10,11].
conditions of the Creative Commons The literature review is provided in Section 2. The methods used in the proposed
Attribution (CC BY) license (https:// scheme are provided in Section 3. Section 4 provides the process of embedding and
creativecommons.org/licenses/by/ extraction of the propounded method. Section 5 presents the experimental results with
4.0/). various images, attacks, and metrics. Finally, the conclusion is provided in Section 6.

Appl. Sci. 2022, 12, 8724. https://doi.org/10.3390/app12178724 https://www.mdpi.com/journal/applsci

Appl. Sci. 2022, 12, 8724 2 of 13

2. Literature Survey
Concealing information in other media is extremely old, as depicted on account of
steganography. The term advanced watermarking first appeared in 1993 when
Tirkel et al. [12] introduced two methods to conceal information in pictures.
In the recent past, telehealth systems have increasingly improved watermarking
approaches [13]. These approaches provide authentication and security and give optimal
bandwidth utilization, another essential criterion of the telehealth communication system.
In [13], the authors proposed a watermarking scheme (WS) for telehealth applications.
This method embedded a signature watermark image and patient report of 80 characters
in length using lifting wavelet transform (LWT) and discrete cosine transform (DCT)
schemes. This work decomposed the host image into subbands using LWT, and DCT
further transformed the significant subbands. Simultaneously, the patient report and
the signature watermark were encrypted and embedded into the final DCT-transformed
subbands. The reverse process was applied to extract the watermarked information.
A blend of watermarking, cryptography, and error-correcting code for electronic pa-
tient records (EPRs) is proposed in the method [14]. The watermarking image (WI) and
EPR are embedded in this method after performing DWT and turbo encoding, respectively.
Later, an inverse DWT is applied to this embedded data and processed to obtain an en-
crypted watermark using pallier encryption. Simultaneously, the cover image is encrypted
using a pallier cryptosystem to obtain an encrypted cover image. Finally, these encrypted
images (cover and watermark) are again embedded and sent via a communication channel.
The reverse process is held at the destination to obtain the desired information.
A crypto watermarking scheme for telemedical applications is proposed in [15]. In [16],
a blind WS is framed to hide EPR data in the retinal image for telehealth applications. The
retinal image is decomposed into subbands, and the lower subband (LL) is subjected to
SVD; the EPR watermark is placed into this band. At the initial step, a bit-plane extraction
is performed on the host image, and the watermark image is integrated into the host image
using a chaotic mapping scheme. Further, the watermarked image (WI) is encrypted using
the fractional Hartley transform to obtain the crypto-watermarked image.
Another work in [17] is proposed for the detection of tampered medical images using
a crypto WS. However, it is a non-blind watermarking scheme, where the EPR is embedded
into the radiological images for authentication and security. The approach is based on DCT
and compressive sensing (CS), where CS is used to encrypt the watermark data and DCT is
applied to a host image. In [18], the authors attempted to embed patients’ two biometrics,
fingerprints and face, using a two-stage watermarking approach. Initially, the fingerprint
was encrypted using minutiae extraction and encoding into the original face image and a
key. Later, the watermarked image was further encrypted and embedded into the original
fingerprint to obtain the second level of the WI. Finally, the watermark was embedded by
combining the DCT subbands and the encrypted watermark image.
All the literature above is related to image watermarking with different transform
domains. They follow different types of watermarking approaches based on extraction,
and in embedding the watermark, different transformation techniques methods are utilized
(most commonly, DWT, DCT, and LWT). Previous watermarking schemes are weak in
terms of the security of watermark data, which inspired us to add crypto techniques to
protect the watermark. The proposed watermarking approach can overcome authentication
problems by embedding the QR code watermark. The proposed scheme converts the text
information to a scrambled QR image using a chaotic logistic map. The public and private
keys are used to cipher and decipher the data. As the images are vulnerable to attacks, the
proposed method overcomes this by combining DWT and SVD using adaptive embedding
factor values for images. Four subbands in the proposed scheme, LL, LH, HL, and HH,
are obtained after one level LWT. LL is selected based on its efficient properties. The LL
subband is again decomposed using QR, and then embedding of the scrambled QR code
watermark. The motivation behind this combination is to enhance the imperceptibility
and robustness. The robustness improvements are provided by applying DWT coefficients.
Appl. Sci. 2022, 12, 8724 3 of 13

The watermark is embedded by modifying the coefficients of DWT using secret keys. The
inverse process is utilized at the receiver end to retrieve the watermark data.

3. Preliminaries
This section gives the details of the techniques used in the propounded scheme. The
propounded scheme consists of two modules: embedding and extraction. The depiction of
these two processes is shown in sub sections. As the transform domain is more robust than
the spatial domain, the DWT algorithm is utilized in the embedding process as it has better
reconstruction without losing information. SVD is utilized in combination with DWT to
overcome noise and compression attacks. Moreover, a cryptographic algorithm (RSA) is
used to verify the algorithm’s robustness against the channel’s vulnerability.

3.1. RSA Algorithm

RSA is named after Rivest, Shamir, and Adleman, the three creators of the RSA
calculation [19]. It is an asymmetric cryptographic algorithm that contains two keys: public
and private. A public key is used to encrypt the data at the transmitter, and for decryption,
both these keys are used at the receiver. The primary rationale behind the RSA algorithm is
that it generates these two keys by factorizing a given large integer. Since a public key has
two numbers, one of them is a product of two prime numbers.
RSA keys are commonly 1024 or 2048 bits in length. If the key’s size increases, the
encryption’s strength increases exponentially. Similarly, using the same prime numbers, a
private key is also generated. Therefore, the algorithm’s robustness lies in foolproofing the
Appl. Sci. 2022, 12, x FOR PEER REVIEW 4 of 14
large number factorizing. RSA Algorithm is shown in Figure 1 respectively. The generated
public and private keys using RSA algorithm are provided in Table 1.

Figure RSAalgorithm.
1.RSA
Figure 1. algorithm.

Table
Table 1. Generationofofpublic
1. Generation public and
and private
private keyskeys
usingusing
RSA. RSA.

Inputs andand
Inputs Outputs of RSA
Outputs of RSA Algorithm
Algorithm
X(o) (0–1) U (3.56–4) Prime Numbers Public Key Private Key Encrypted Message Decrypted Message
X(o) (0–1) U (3.56–4) Prime Numbers Public Key Private Key Encrypted Message Decrypted Message
0.2 3.6 (3, 5) (1, 15) (1, 15) 31514649 (0.2, 3.6)
0.2
0.4 3.73.6 (5, (3,
7) 5) (1, 15) (27, 55) (1, 15) 2716334111620
(23, 35) 31514649 (0.4, 3.7) (0.2, 3.6)
0.4
0.5 3.63.7 (5, 7)
(5, 13) (23, 35) (43, 65)(27, 55) 226273451624
(19, 65) 2716334111620 (0.5, 3.6) (0.4, 3.7)
0.6
0.5 3.83.6 (5, (5,
11)13) (11, 55)
(19, 65) (11, 55)(43, 65) 3746544451461
226273451624 (0.6, 3.8) (0.5, 3.6)
0.8 3.9 (7, 13) (35, 91) (35, 91) 55249602528 (0.8, 3.9)
0.6 3.8 (5, 11) (11, 55) (11, 55) 3746544451461 (0.6, 3.8)
0.8 3.9 3.2. (7, 13) Wavelet Transform
Discrete (35, 91) (DWT) (35, 91) 55249602528 (0.8, 3.9)
The primary benefit of wavelet analysis over Fourier analysis is its capability to cap-
ture time and frequency localization. Daubechies and Mallat introduced the DWT in the
The RSA algorithm employs the following procedure to generate public and private keys:
late 1980s [20,21]. DWT subdivides a signal into a set of mutually orthogonal wavelet basis
•functions
Pick [22].
two big
The prime
2D-DWT numbers, r, andused
is commonly s. for image processing applications [23].
•The DWT = r × s by multiplying
Find tdecomposition on images these
yields values,
information
where about low-frequency
t is referred LL modulus
to as the sub- for
bands; horizontal and
encryption and vertical
decoding.edge details as LH and HL subbands, respectively; and di-
•agonalUseedge features k
a number asless
HH than
subbands. This tprocess
t so that is called
is roughly primeone-level decomposition,
to (r − 1) × (s − 1), which
which can be extended
indicates that thetoonly
various levels
factor in to extract further
common higher-level
between k and (r feature
− 1) ×information
(s − 1) is 1. Select
from images [23]. The 2D-DWT subband decomposition process for the first three levels
“k” so that 1 < k < ϕ (t), k is prime to ϕ (t), and gcd (e,d(t)) = 1.
is shown in Figure 2. The further mathematical details of DWT can be found in [24,25].
The three levels of DWT decompositions are shown in Figure 2 Respectively.
 Appl. Sci. 2022, 12, x FOR PEER REVIEW 4 of 14

Appl. Sci. 2022, 12, 8724 4 of 13

• The public key is <e, t> for t = r × s. The public key <e, t> encrypts a plaintext
message m. The mathematical methodology is employed to obtain ciphertext C from
the original message: C = mk mod t.
• The following formula is employed to calculate the d and set the private key in a way
Figure that
1. RSAD k mod {(r − 1) × (s − 1)} = 1.
algorithm.
• <d, t> is the private key. The private key <d, t> is used to decipher the ciphertext
Table 1. Generation of public and private keys using RSA.
message c. The below formula is used to generate plain text m out from ciphertext c:
m= cd mod
Inputs and Outputs
t. of RSA Algorithm
X(o) (0–1) U (3.56–4) Prime Numbers
• Different inputs of x(o), u, prime numbers, andDecrypted
Public Key Private Key Encrypted Message Message
the generated public and private keys
0.2 3.6 (3, 5) (1, 15) (1, 15) 31514649 (0.2, 3.6)
0.4 3.7 (5, 7) with an encrypted
(23, 35) message
(27, 55) and decrypted
2716334111620 messages are
(0.4, 3.7) shown in Table 1.
0.5 3.6 (5, 13) (19, 65) (43, 65) 226273451624 (0.5, 3.6)
0.6 3.8 3.2. Discrete Wavelet
(5, 11) (11, 55) Transform
(11, 55) (DWT)
3746544451461 (0.6, 3.8)
0.8 3.9 (7, 13)The primary
(35, 91)
benefit(35,
of91) 55249602528
wavelet analysis over Fourier(0.8, 3.9)
analysis is its capability to capture
time and frequency localization. Daubechies and Mallat introduced the DWT in the late
3.2. Discrete Wavelet Transform (DWT)
1980s The[20,21]. DWTofsubdivides
primary benefit wavelet analysisa over
signal into
Fourier a setisofitsmutually
analysis capability to orthogonal
cap- wavelet basis
functions
ture time and[22]. Thelocalization.
frequency 2D-DWTDaubechies
is commonlyand Mallat used for image
introduced the DWTprocessing
in the applications [23].
late 1980s
The DWT [20,21]. DWT subdivideson
decomposition a signal into ayields
images set of mutually orthogonal
information wavelet
about basis
low-frequency LL subbands;
functions [22]. The 2D-DWT is commonly used for image processing applications [23].
horizontal and vertical edge details as LH and HL subbands,
The DWT decomposition on images yields information about low-frequency LL sub-
respectively; and diagonal
edge features as HH subbands. This process is called one-level
bands; horizontal and vertical edge details as LH and HL subbands, respectively; and di- decomposition, which
agonal
can beedge features as
extended toHH subbands.
various Thisto
levels process
extractis called one-level
further decomposition,
higher-level feature information from
which can be extended to various levels to extract further higher-level feature information
images [23]. The 2D-DWT subband decomposition process for the first three levels is shown
from images [23]. The 2D-DWT subband decomposition process for the first three levels
in Figure
is shown 2. The
in Figure further
2. The mathematical
further details
mathematical details of can
of DWT DWT can be
be found found in [24,25]. The three
in [24,25].
levels
The threeoflevels
DWT decompositions
of DWT decompositions areareshown
shown in Figure
in Figure 2 Respectively.
2 Respectively.

Figure 2. Three
Figure levelslevels
2. Three of DWTof
decompositions.
DWT decompositions.
3.3. Singular Value Decomposition (SVD)
3.3. Singular Value Decomposition (SVD)
SVD is a factorization of a given matrix with various applications in image processing
[26]. TheSVD is a values
singular factorization
(SVs) of an of a given
image matrix
have excellent with various
soundness, applications
i.e., (i) they will not in image process-
change
ing evenThe
[26]. aftersingular
modifying values
the original image,
(SVs) ofand
an (ii) SVs address
image have inborn
excellent arithmetical
soundness, i.e., (i) they will
image properties. Here, an image can be addressed as a framework of positive scalar qual-
not
ities.
change even after modifying the original image, and (ii) SVs address inborn arith-
Appl. Sci. 2022, 12, x FOR PEER REVIEW 5 of 14
metical image properties. Here, an image can be addressed as a framework of positive
scalar qualities.
The SVD of a matrix M can be factorized into three matrices: U, S, and V product,
The SVD of a matrix M can be factorized into three matrices: U, S, and V product,
where UU
where and
and V orthonormal
V are are orthonormalmatrices.matrices. S is matrix
S is a diagonal a diagonal matrix
with positive withinpositive values in
values
descending
descending order,
order, as shown
as shown in Figurein3.Figure 3. The
The equation formequation
of SVD forform of SVD
the matrix for the matrix M is
M is M
=MU=× SU× × × VSTis, where
VT,Swhere utilized Stois
embed the watermark
utilized to embedinformation.
the watermark information.

Figure3. 3.
Figure Singular
Singular valuevalue
decomposition of a 4 × 4 of
decomposition a4×4
matrix. matrix.

4. Proposed Method
This section discusses the details of the proposed digital watermarking processes.
The proposed system embeds the QR code watermark in the transform domain using
DWT and SVD using the cryptography technique (RSA). The embedding process is pro-
vided in Section 4.1, and the reverse process of embedding is implemented in watermark
Appl. Sci. 2022, 12, 8724 5 of 13

4. Proposed Method
This section discusses the details of the proposed digital watermarking processes. The
proposed system embeds the QR code watermark in the transform domain using DWT
and SVD using the cryptography technique (RSA). The embedding process is provided in
Section 4.1, and the reverse process of embedding is implemented in watermark extraction,
which is detailed in Section 4.2.

4.1.
Appl. Sci. 2022, 12, x FOR PEER REVIEW Embedding 6 of 14
This section discusses the process of watermark embedding using the cryptographic
algorithm stepwise. The watermark embedding architecture is shown in Figure 4.

Theproposed
Figure4.4.The
Figure proposedscheme:
scheme:embedding
embeddingprocess.
process.

Step 1: The watermark data consisting of name and country is created as a QR code.
4.2.
StepExtraction
2: Generate a public key and an encrypted message by inputting private key values
into This section
the RSA goes details the extraction procedure. In semi-blind watermark extrac-
algorithm.
tion,
Steppartial
3: The data of significant
QR code datausing
is scrambled is required. The
a Chaotic extraction
Logistic Mapprocedure
(CLM) to is provided
provide in
water-
Figure 5. security.
mark data
Import the
Step1:4:Examine
Step thesignificant
image thatimage where the watermark has to be concealed.
is watermarked.
Step 5: Convert the imported color image
Step 2: The color watermarked image is converted into red,togreen, and blue
red, green, andcomponents.
blue layers.
Consider
Step3:6:As
Step the blue is
the watermark layer and apply
embedded one-level
in the Haar wavelet
blue component, thedecomposition to obtain
same is considered for
the four
extraction.subbands (LL, LH, HL, and HH).
Step 4: The blue component is applied with one-level DWT with a Haar wavelet.
Step 5: The LL subband is considered, and SVD is applied to generate a singular value
matrix.
Step 6: A scrambled QR watermark is extracted based on the key values and partial data
 Appl. Sci. 2022, 12, 8724 6 of 13

Step 7: The LL subband and scrambled QR image are selected and decomposed using SVD
decomposition.
Step 8: Both images’ singular values are considered and combined with a key value to
generate watermarked singular values.
Step 9: A watermarked LL subband is created using an invertible SVD.
Step 10: By fusing the watermarked LL subband and the other subbands to create the
watermarked blue layer, an inverse DWT is implemented for one level.
Step 11: A watermarked color image is created by combining a blue layer with the red,
green, and other layers.
Step 12: The watermarked image with the public key and key value are communicated to
the receiver.

Appl. Sci. 2022, 12, x FOR PEER REVIEW

4.2. Extraction 7o
This section goes details the extraction procedure. In semi-blind watermark extraction,
partial data of significant data is required. The extraction procedure is provided in Figure 5.

Figure The
Figure5.5. proposed
The scheme:
proposed extraction
scheme: process.
extraction process.
Step 1: Examine the image that is watermarked.
5. Experimental
Step Results image is converted to red, green, and blue layers.
2: The color watermarked
Step 3:A As the watermark
sample is embedded
color images in the
with a size 512component,
of blue × 512 × 3 and
the same is considered
watermark QR code wit
size of 256 × 256 are considered for the evaluation of the proposed method, which
for extraction.
Step 4: The blue component is applied with one-level DWT with a Haar wavelet.
shown in Figures 6 and 7. Several attacks such as noise attacks and geometric attacks
applied to the sample images to test the method’s robustness, which is discussed in det

5.1. Evaluation Metrics

 Appl. Sci. 2022, 12, 8724 7 of 13

Step 5: The LL subband is considered, and SVD is applied to generate a singular value matrix.
Step 6: A scrambled QR watermark is extracted based on the key values and partial data
of the significant image.
Step 7: Inverse scrambling is applied to the extracted watermark using the CLM algorithm.
Step 8: The watermark is extracted with the decrypted message using the public key, which
contains the private key values to verify the watermarked data.

5. Experimental Results
A sample color images with a size of 512 × 512 × 3 and watermark QR code with a
Appl. Sci. 2022, 12, x FOR PEER REVIEW
size of 256 × 256 are considered for the evaluation of the proposed method, which is shown 8o
in Figures 6 and 7. Several attacks such as noise attacks and geometric attacks are applied
to the sample images to test the method’s robustness, which is discussed in detail.

Figure 6. Sample
Figure 6. Samplecolor
color image.
image.
Appl. Sci. 2022, 12, 8724 8 of 13
Figure 6. Sample color image.

Figure
Figure 7. Sample
7. Sample watermarked
watermarked images andimages
extracted and extracted
watermarks. watermarks.
5.1. Evaluation Metrics
The peak-signal-to-noise ratio (PSNR) and normalized correlation coefficient (NCC)
metrics evaluate how effectively the proposed watermarking system works. These are
discussed in detail in this section.

5.1.1. Peak-Signal-to-Noise Ratio

PSNR is an articulation of the proportion between a signal’s most extreme conceivable
worth and the watermarked signal. The PSNR is normally represented in the logarithmic
decibel scale. PSNR between the watermarked picture and cover picture is utilized to assess
intangibility and is effortlessly characterized through the mean square error (MSE).
MSE permits the analysis of the valid pixel upsides of a unique image to a corrupted
image. The higher the PSNR values, the higher the quality of the watermarked image:

M −1 N −1 2
∑m =0 ∑n=0 ( Im,n − W MIm,n )
MSE =
Im,n
 
max I
PSNR I,W MI = 20log10 √ (1)
MSE
The MAX ML refers to the extreme value of the image that is possible.

5.1.2. Normalized Correlation Coefficient

NCC is the measurement utilized to discover the relationship coefficient between the
original watermark and the extracted watermark. When NCC is close to 1, the extracted
watermark equals the original watermark. The NCC is defined as follows:

∑m n
x =1 ∑y=1 qr ( x, y ) × Eqr ( x, y )
NCCqr,eqr = q q  (2)
m 2 m 2
∑ x=1 qr ( x, y) ∑ x=1 Eqr ( x, y)
Appl. Sci. 2022, 12, 8724 9 of 13

The PSNR and NCC values without attacks and with attacks on watermarked images
are shown in Tables 2–4 respectively.

Table 2. PSNR and NCC values without attacks.

Images PSNR and NCC Values without Attacks

House 42.25, 1.00 Lake 41.68, 1.00
Tree 43.56, 1.00 Pepper 40.35, 1.00
Girl 42.85, 1.00 House 2 41.74, 1.00
Mandrill 43.12, 1.00 Einstein 43.36, 1.00
Lena 42.22, 1.00 Monalisa 43.81, 1.00
Jetplane 41.81, 1.00 Monarch 42.65, 1.00

Table 3. PSNR and NCC values with different attacks.

PSNR and NCC Values after Attacks

Images
Salt and Pepper Gaussian Mean Filtering JPEG Compression
House 36.11, 0.9789 39.01, 0.9898 33.12, 0.9901 37.12, 0.9938
Tree 36.05, 0.9796 38.87, 0.9885 33.25, 0.9905 37.28, 0.9942
Girl 35.56, 0.9898 38.86, 0.9896 33.18, 0.9908 37.25, 0.9935
Mandrill 35.66, 0.9899 38.96, 0.9901 33.22, 0.9906 37.24, 0.9932
Lena 35.28, 0.9887 38.15, 0.9914 33.15, 0.9910 37.16, 0.9949
Jetplane 35.47, 0.9815 38.42, 0.9912 33.16, 0.9903 37.21, 0.9936
Lake 35.22, 0.9829 38.36, 0.9908 33.11, 0.9907 37.27, 0.9937
Pepper 35.84, 0.9867 38.51, 0.9925 33.08, 0.9915 37.28, 0.9932
House 2 36.24, 0.9814 39.12, 0.9904 33.13, 0.9911 37.24, 0.9941
Einstein 36.05, 0.9885 39.21, 0.9916 33.17, 0.9916 37.28, 0.9934
Monalisa 35.69, 0.9829 39.11, 0.9892 33.22, 0.9914 37.22, 0.9914
Monarch 35.88, 0.9785 39.05, 0.9845 33.20, 0.9909 37.24, 0.9909

Table 4. PSNR and NCC values with geometric attacks on watermarked images.

PSNR and NCC Values after Attacks

Images
Cropping Rotation Scaling Translation
House 28.36, 0.9801 29.06, 0.9896 38.66, 0.9964 36.42, 0.9777
Tree 29.05, 0.9815 28.58, 0.9889 38.59, 0.9948 36.28, 0.9765
Girl 29.23, 0.9836 29.12, 0.9885 39.28, 0.9947 36.19, 0.9787
Mandrill 29.26, 0.9885 29.14, 0.9889 39.22, 0.9957 36.33, 0.9789
Lena 28.21, 0.9869 28.12, 0.9904 38.92, 0.9954 36.23, 0.9778
Jetplane 29.60, 0.9878 27.86, 0.9911 37.99, 0.9958 36.19, 0.9781
Lake 29.35, 0.9886 28.26, 0.9919 38.84, 0.9961 36.28, 0.9776
Pepper 29.48, 0.9891 27.87, 0.9921 37.94, 0.9955 36.22, 0.9785
House 2 29.54, 0.9912 28.56, 0.9902 38.65, 0.9961 36.18, 0.9788
Einstein 29.22, 0.9905 29.14, 0.9919 39.19, 0.9951 36.19, 0.9781
Monalisa 29.35, 0.9897 29.04, 0.9901 39.21, 0.9968 36.24, 0.9782
Monarch 29.32,0.9856 28.87, 0.9842 38.92, 0.9964 36.28, 0.9778
 shown in Figure 8.

5.2.1. Salt and Pepper Noise

Salt and pepper noise is added to an image by the option of both choo
Appl. Sci. 2022, 12, 8724 10 of 13
255-pixel worth) and irregular dullness (with 0-pixel esteem) all around
model is otherwise called information drop noise because it is measured
of
5.2.salt
Noiseand pepper crystals on the image. The sample salt and pepper
Attacks
Noiseand
images in images is an arbitrary variety
the watermark of brilliance
extracted from or it shading data in and
with PSNR the image.
NCCIt values
is corruption of a signal by the addition of data. Images containing multiplicative noise
Figure
consist of8more
with a density
noise data. Noiseofattacks
0.5. on sample images and extracted watermarks are
shown in Figure 8.

Figure 8. Noise attacks on sample images and extracted watermarks.

Figure 8. Noise attacks on sample images and extracted watermarks.
5.2.1. Salt and Pepper Noise
Salt and pepper noise is added to an image by the option of both choosing noise (with
255-pixel worth) and irregular dullness (with 0-pixel esteem) all around the image. This
model is otherwise called information drop noise because it is measured by the number of
salt and pepper crystals on the image. The sample salt and pepper noise attacks on images
and the watermark extracted from it with PSNR and NCC values are provided in Figure 8
with a density of 0.5.

5.2.2. Gaussian Noise Attack

Gaussian noise is a measurable commotion with a feasible density equivalent to an
ordinary state, otherwise called Gaussian dissemination. Gaussian noise values with a
variance of 0.01 are tested on various images.

5.2.3. Mean Filtering Attack

Mean filtering with a size of 3 × 3 is applied, producing an average value of their
neighbors, including themselves. Average or mean filtering also reduces the intensity
disparity between adjacent pixels.

5.2.4. JPEG Compression

This is an image processing standard designed by the Joint Photography Experts
Group. It uses DCT for compression, which is lossy image compression. Different compres-
sion ratios can be applied to the images based on the requirement. The proposed method is
tested with compression 90.
 Geometric attacks on sample images and extracted watermarks are sho
5.3.1. Cropping Attack

Appl. Sci. 2022, 12, 8724

A piece of the picture is trimmed, which is expected to influence
11 of 13
th
stalled in the picture. As the watermark is installed into the picture ut
SVD strategies, we can, in any case, recuperate the watermark picture,
5.3.
ness Geometric
remains Attacks
unblemished.
Geometric attacks on sample images and extracted watermarks are shown in Figure 9.

Figure
Figure 9. Geometric attacks attacks
9. Geometric on sampleon
images and extracted
sample images watermarks.
and extracted watermarks.
5.3.1. Cropping Attack
5.3.2. Rotation
A piece of the Attack
picture is trimmed, which is expected to influence the watermark
installed in the picture. As the watermark is installed into the picture utilizing DWT and
A rotation
SVD strategies, attack
we can, is an
in any case, entirely
recuperate thedistinguishable assault
watermark picture, and performed to
the genuineness
of the first picture. It ensures that none of the mathematical mutilations in
remains unblemished.
worthiness
5.3.2. of the
Rotation Attack watermark picture. The watermarked picture cannot ex
in the framework
A rotation attack is esteems.
an entirely distinguishable assault performed to change the files
of the first picture. It ensures that none of the mathematical mutilations influence the
trustworthiness of the watermark picture. The watermarked picture cannot experience
change in the framework esteems.

5.3.3. Scaling Attack

In an image scaling attack, the intruders try to manipulate the image without knowing
by downscaling and upscaling. The image is downsampled twice in the proposed system
to test the robustness.

5.3.4. Translation Attack

In an image, a translation attack shifts the image to a specific direction on the X and Y
axes. The proposed system shifts the image to two pixels in the X direction and two in the
Y direction.
Different types of attacks are applied to test the robustness of the method. In addition
to this, the method is also compared with related watermarking schemes [4,8,27,28], as
shown in Table 5. The comparison is tested with various attacks where the proposed
scheme has better robustness results than other methods.
Appl. Sci. 2022, 12, 8724 12 of 13

Table 5. Comparison of NCC values with related watermarking methods.

Image Attacks Schemes Lena Mandrill Peppers Airplane

Vaidya et al. [4] 1.00 1.00 1.00 1.00
Mun et al. [27] 0.9885 0.9886 0.9895 0.9879
No Attack Agoyi et al. [28] 0.8694 0.8837 0.9030 0.9025
Hosny et al. [6] 0.9995 0.9995 0.9995 0.9995
Proposed Method 1.00 1.00 1.00 1.00
Vaidya et al. [4] 0.8458 0.7156 0.9465 0.9325
Mun et al. [27] - - - -
Salt and
Agoyi et al. [28] - - - -
Pepper noise
Hosny et al. [6] 0.9916 0.9916 0.9916 0.9916
Proposed Method 0.9887 0.9899 0.9867 0.9815
Vaidya et al. [4] 0.8489 0.8053 0.9279 0.9114
Mun et al. [27] - - - -
Gaussian noise Agoyi et al. [28] 0.7955 0.7796 0.8083 0.7956
Hosny et al. [6] 0.9905 0.9905 0.9905 0.9905
Proposed Method 0.9914 0.9901 0.9925 0.9912
Vaidya et al. [4] 0.8944 0.8946 0.8942 0.8965
Mun et al. [27] 0.9921 0.9860 0.9948 0.9888
Cropping Agoyi et al. [28] - - - -
Hosny et al. [6] - - - -
Proposed Method 0.9869 0.9885 0.9891 0.9878
Vaidya et al. [4] 0.9827 0.9956 0.9980 0.9963
Mun et al. [27] 0.9837 0.9742 0.9768 0.9879
Scaling Agoyi et al. [28] 0.9263 0.9675 0.9644 0.9381
Hosny et al. [6] 0.9940 0.9940 0.9940 0.9940
Proposed Method 0.9954 0.9957 0.9955 0.9958
Vaidya et al. [4] 0.9053 0.8983 0.8819 0.8956
JPEG Mun et al. [27] 0.9457 0.6537 0.9194 0.9472
Compression Agoyi et al. [28] 0.8469 0.8469 0.8901 0.8854
(90) Hosny et al. [6] 0.9928 0.9928 0.9928 0.9928
Proposed Method 0.9949 0.9932 0.9932 0.9936

6. Conclusions
A novel watermarking scheme is proposed to ensure copyright protection and au-
thentication of images using cryptographic techniques. This scheme fuses the asymmetric
encryption RSA algorithm with spatial domain schemes called the DWT and SVD algorithm.
This hybrid algorithm ensures visual inspection, copyright protection, and authentication
robustness. Moreover, the proposed watermarking scheme was tested with various image
and signal processing attacks. The proposed method showed better results compared to
related watermarking techniques. However, we only tested our method on a few image
datasets. Therefore, in our subsequent works, we will attempt to add more diversified
data such as medical images and include more noise attacks to provide a generalized
watermarking method for the research field.

Author Contributions: Conceptualization, P.V.S., K.N.V.P.S.R., K.M.H. and M.M.F.; Methodol-

ogy, P.V.S., K.N.V.P.S.R., K.M.H. and M.M.F.; Software, P.V.S., K.N.V.P.S.R., K.M.H. and M.M.F.;
Writing—original draft, P.V.S., K.N.V.P.S.R. and M.M.F.; Writing—review & editing, K.M.H. All
authors have read and agreed to the published version of the manuscript.
Funding: This research received no external funding.
Institutional Review Board Statement: Not applicable.
Informed Consent Statement: Not applicable.
Data Availability Statement: Not applicable.
Appl. Sci. 2022, 12, 8724 13 of 13

Conflicts of Interest: The authors declare no conflict of interest.

References
1. Vaidya, S.P. Multiple decompositions-based blind watermarking scheme for color images. In Proceedings of the IEEE International
Conference on Recent Trends in Image Processing and Pattern Recognition, Solapur, India, 21–22 December 2018; Springer:
Berlin/Heidelberg, Germany, 2018; pp. 132–143.
2. Hosny, K.M.; Darwish, M.M. Invariant image watermarking using accurate polar harmonic transforms. Comput. Electr. Eng. 2017,
62, 429–447. [CrossRef]
3. Sanivarapu, P.V.; Rajesh, K.N.V.P.S.; Reddy, N.V.R.; Reddy, N.C.S. Patient data hiding into ECG signal using watermarking in
transform domain. Phys. Eng. Sci. Med. 2020, 43, 213–226. [CrossRef]
4. Vaidya, S.P.; PVSSR, C.M. Adaptive digital watermarking for copyright protection of digital images in wavelet domain. Procedia
Comput. Sci. 2015, 58, 233–240. [CrossRef]
5. Vaidya, P.; PVSSR, C.M. A robust semi-blind watermarking for color images based on multiple decompositions. Multimed. Tools
Appl. 2017, 76, 25623–25656.
6. Hosny, K.M.; Darwish, M.M. Robust color image watermarking using invariant quaternion Legendre-Fourier moments. Multimed.
Tools Appl. 2018, 77, 24727–24750. [CrossRef]
7. Vaidya, P.; PVSSR, C.M. Adaptive, robust, and blind digital watermarking using Bhattacharyya distance and bit manipulation.
Multimed. Tools Appl. 2018, 77, 5609–5635.
8. Hosny, K.M.; Darwish, M.M.; Fouda, M.M. Robust color images watermarking using new fractional-order exponent moments.
IEEE Access 2021, 9, 47425–47435. [CrossRef]
9. Hosny, K.M.; Darwish, M.M.; Li, K.; Salah, A. Parallel multi-core CPU and GPU for fast and robust medical image watermarking.
IEEE Access 2018, 6, 77212–77225. [CrossRef]
10. Hosny, K.M.; Darwish, M.M. New geometrically invariant multiple zero watermarking algorithm for color medical images.
Biomed. Signal Process. Control 2021, 70, 103007. [CrossRef]
11. Hosny, K.M.; Darwish, M.M. Reversible color image watermarking using fractional-order polar harmonic transforms and a
chaotic sine map. Circuits Syst. Signal Process. 2021, 40, 6121–6145. [CrossRef]
12. van Schyndel, R.G.; Tirkel, A.Z.; Osborne, C.F. A digital watermark. In Proceedings of the IEEE 1st International Conference on
Image Processing, Austin, TX, USA, 13–16 November 1994; Volume 2, pp. 86–90.
13. Singh, A.K. Robust and distortion control dual watermarking in LWT domain using DCT and error correction code for color
medical image. Multimed. Tools Appl. 2019, 78, 30523–30533. [CrossRef]
14. Anand, A.; Singh, A.K. Joint watermarking-encryption-ECC for patient record security in wavelet domain. IEEE MultiMedia 2020,
27, 66–75. [CrossRef]
15. Kaur, G.; Agarwal, R.; Patidar, V. Crypto-watermarking of images for secure transmission overcloud. J. Inf. Optim. Sci. 2020, 41,
205–216.
16. Zermi, N.; Khaldi, A.; Kafi, R.; Kahlessenane, F.; Euschi, S. A DWT-SVD based robust digital watermarking for medical image
security. Forensic Sci. Int. 2021, 320, 110691. [CrossRef] [PubMed]
17. Borra, S.; Thanki, R. Crypto-watermarking scheme for tamper detection of medical images. Comput. Methods Biomech. Biomed.
Eng. Imaging Vis. 2020, 8, 345–355. [CrossRef]
18. Lebcir, M.; Awang, S.; Benziane, A. Robust blind watermarking approach against the compression for fingerprint image using
2D-DCT. Multimed. Tools Appl. 2022, 81, 20561–20583. [CrossRef]
19. Zhou, X.; Tang, X. Research and implementation of RSA algorithm for encryption and decryption. In Proceedings of the IEEE
2011 6th International Forum on Strategic Technology, Harbin, China, 22–24 August 2011; Volume 2, pp. 1118–1121.
20. Shensa, M.J. The discrete wavelet transform: Wedding the a trous and Mallat algorithms. IEEE Trans. Signal Process. 1992, 40,
2464–2482. [CrossRef]
21. Vaidya, S.P. A blind color image watermarking using brisk features and contourlet transform. In Proceedings of the International
Conference on Recent Trends in Image Processing and Pattern Recognition, Solapur, India, 21–22 December 2018; Springer:
Berlin/Heidelberg, Germany, 2018; pp. 203–215.
22. Vaidya, S.P.; PVSSR, C.M.; Santosh, K.C. Imperceptible watermark for a game-theoretic watermarking system. Int. J. Mach. Learn.
Cybern. 2019, 10, 1323–1339. [CrossRef]
23. Nason, G.P.; Silverman, B.W. The discrete wavelet transform in s. J. Comput. Graph. Stat. 1994, 3, 163–191.
24. Van Fleet, P.J. Discrete Wavelet Transformations: An Elementary Approach with Applications; John Wiley & Sons: Hoboken, NJ, USA, 2011.
25. Chang, C.; Girod, B. Direction-adaptive discrete wavelet transform for image compression. IEEE Trans. Image Process. 2007, 16,
1289–1302. [CrossRef]
26. Vaidya, S.P.; PVSSR, C.M. A robust and blind watermarking for color videos using redundant wavelet domain and SVD. In Smart
Computing Paradigms: New Progresses and Challenges; Springer: Berlin/Heidelberg, Germany, 2020; pp. 11–17.
27. Mun, S.-M.; Nam, S.-H.; Jang, H.-U.; Kim, D.; Lee, H.-K. A robust blind watermarking using convolutional neural network. arXiv
2017, arXiv:1704.03248.
28. Agoyi, M.; C¸elebi, E.; Anbarjafari, G. A watermarking algorithm based on chirp z-transform, discrete wavelet transform, and
singular value decomposition. Signal Image Video Process. 2015, 9, 735–745. [CrossRef]

View publication stats