APPLICATION OF DIGITAL CELLULAR RADIO FOR MOBILE LOCATION ESTIMATION


IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

ROBUST AND IMPERCEPTIBLE WATERMARKING ON 

MEDICAL IMAGES USING COEFFICIENT PAIR 

MODIFICATION 

 LEDYA NOVAMIZANTI1,3*, ANDRIYAN BAYU SUKSMONO2,4, DONNY 

DANUDIRDJO2,4, AND GELAR BUDIMAN3,4 

 
1
Electrical Engineering and Informatics, School of Electrical Engineering and Informatics, 

Institut Teknologi Bandung, Bandung 40132, Indonesia  
2
School of Electrical Engineering and Informatics, Institut Teknologi Bandung, Bandung 40132, 

Indonesia  
3
School of Electrical Engineering, Telkom University, Bandung 40257, Indonesia 

4
Research Collaboration Center for Quantum Technology 2.0, Bandung 40132, Indonesia 

*
Corresponding author: ledyaldn@telkomuniversity.ac.id 

(Received: 7th September 2022; Accepted: 3rd October 2022; Published on-line: 4th January 2023)  

ABSTRACT:  Sensitive data including medical images and electronic patient records (EPR) 

have potential value in the era of big data and telemedicine applications. Distribution of 

medical images and EPR over public networks requires a high level of privacy and security. 

Robust and imperceptible watermarking techniques are needed to provide copyright 

preservation for medical images and protect patient information security. This paper improves 

the technique of Coltuc et al. by modifying the discrete cosine transform (DCT) coefficient 

pairs in the watermark embedding formula. Our proposed formula ensures that the difference 

between the two coefficients is at least 𝜁. If the difference between the two coefficients is less 
than 𝜁, then the new pixels are modified so that the difference is equal to 𝜁. The proposed 
method was evaluated on a variety of medical images, including X-ray, CT, US, MRI, and 

Colonoscopy, and compared to numerous robust watermarking techniques of the recent time. 

The experimental results demonstrate that the suggested method outperforms contemporary 

robust watermarking techniques in terms of imperceptibility, robustness, and security. The 

peak signal noise ratio (PSNR) for all modalities of watermarked medical images exceeds 54 

dB, and the average PSNR is approximately 56 dB. The proposed method is outstanding 

compared to Coltuc's method due to a 93% and 14% increase in bit error rate (BER) and 

normalized correlation (NC), respectively. Our work is superior to various state-of-the-art 

robust watermarking techniques, allowing it to be employed effectively in medical 

applications. 

ABSTRAK: Data sensitif termasuk imej perubatan dan rekod pesakit elektronik (EPR) 

mempunyai potensi nilai dalam era aplikasi data besar dan teleperubatan. Pengedaran imej 

perubatan dan EPR melalui rangkaian awam memerlukan tahap privasi dan keselamatan yang 

tinggi. Teknik penanda air yang mantap dan tidak dapat dilihat diperlukan untuk menyediakan 

pemeliharaan hak cipta untuk imej perubatan dan melindungi keselamatan maklumat pesakit. 

Kertas kerja ini menambah baik teknik Coltuc et al. dengan mengubah suai pasangan pekali 

transformasi kosinus diskret (DCT) dalam formula pembenaman tera air. Formula yang 

dicadangkan kami memastikan bahawa perbezaan antara dua pekali adalah sekurang-

kurangnya 𝜁. Jika perbezaan antara dua pekali kurang daripada 𝜁, maka piksel baharu diubah 
suai supaya perbezaannya sama dengan 𝜁. Kaedah yang dicadangkan telah dinilai pada 
pelbagai imej perubatan, termasuk X-ray, CT, US, MRI, dan Kolonoskopi, dan dibandingkan 

dengan banyak teknik penanda air yang mantap pada masa terkini. Keputusan eksperimen 

menunjukkan bahawa kaedah yang dicadangkan mengatasi teknik penanda air teguh 

88



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

kontemporari dari segi ketidakjelasan, keteguhan dan keselamatan. Nilai PSNR untuk semua 

modalitas imej perubatan bertanda air melebihi 54 dB, dan nilai purata PSNR adalah lebih 

kurang 56 dB. Kaedah yang dicadangkan adalah cemerlang daripada kaedah Coltuc kerana 

masing-masing peningkatan 93% dan 14% dalam BER dan NC. Kerja kami lebih unggul 

daripada pelbagai teknik penanda air teguh terkini, membolehkan ia digunakan dengan 

berkesan dalam aplikasi perubatan. 

KEYWORDS:  DCT; imperceptible; medical image; robust; watermarking 

1. INTRODUCTION 

Information and communication technology (ICT) is essential to all sectors of society, 

including government, banking, education, health, agriculture, and transportation. ICT 

integrates computer systems, digital data, communication devices, and the internet into one 

unified system. An important use of ICT for effective quality health care is the electronic 

storage of medical data. The information in the electronic medical data storage, including: 

patient identities, diagnosis reports, doctor's consultations, hospital information where the 

image was created, and other helpful information, can be used anywhere and anytime [1-3]. 

Medical data is susceptible and has great potential value in the era of big data because it is the 

most important basis for diagnosis in determining diagnostic methods and results [4,5]. 

Through public networks like the internet, medical professionals frequently exchange digitally 

created medical images of patients created using various modalities. Radiologists and doctors 

in the same field share these images for clinical interpretation [6].  

Medical professionals use telemedicine applications to remotely diagnose, evaluate, and 

treat patients via electronic communications. The program facilitates the transmission of 

medical images between two healthcare providers for improved diagnosis. However, medical 

image and EPR delivery over public networks demand a high level of security [7]. Modified, 

manipulated, or distorted medical data can cause incorrect diagnoses and severe health issues 

for everyone. In addition, the destruction and theft of medical images and identities can give 

rise to various legal and ethical concerns, including image retention and fraud, piracy, and 

illegal handling [1]. In contrast to financial institutions, which have robust data protection 

methods through two-factor authentication, medical records in healthcare systems are poorly 

protected. Medical facilities must secure data from illegal actions such as hacking and virus 

intrusion, staff negligence, and the theft of medical records on purpose [3]. The transfer of data 

in an intelligent healthcare system while preserving privacy, integrity, authority, and security 

is a complex topic that requires additional consideration [4]. 

Digital watermarking is a common approach for concealing digital information by 

masking signals [8] due to its additional qualities of robustness and imperceptibility. Digital 

watermarking technology is becoming crucial for various fields since it offers a variety of 

compact solutions for numerous approaches and applications, such as cloud computing, 

electronic health, and the Internet of Things [2,11]. Watermarking of digital images is allowed 

in two domains, i.e., the spatial domain and the frequency domain. In the spatial domain of 

watermarking technology, personal information is integrated directly into the cover image by 

modifying the pixel values [3,9]. In frequency domain watermarking, the cover image is first 

translated into the frequency domain, and then the hidden data is embedded by modifying the 

frequency coefficient value [3,9]. Watermarking is also categorized based on the method of 

watermark extraction. Non-blind techniques require both the host and the watermark image to 

extract the watermark. A semi-blind watermark requires a secret key and a watermarked image 

to extract the watermark. If only the secret key is required to extract the watermark, the 

89



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

approach is known as blind watermarking [10]. Based on the application, watermarking 

techniques can be classified as either robust or fragile [1,5]. 

The robustness of confidential information, and how the encoded information can 

withstand attacks conducted by attackers during data transmission, is the primary goal of robust 

watermarking [9]. Robust watermarking is therefore widely used to preserve digital data's 

copyright. However, fragile watermarking [1] is used to identify the invader-damaged portions 

of the watermarked image, i.e., those regions of the image that were damaged during 

transmission. The integrity of digital data material is therefore verified using fragile 

watermarking. Processing time for spatial domain-based watermarking is lower than for 

methods that embed data in the transformation domain. However, compared to transformation-

based strategies, spatial domain-based approaches are less resistant to attacks [7]. Two key 

requirements for creating effective watermarking methods are imperceptibility and robustness 

[6]. 

Compared to watermarking in the spatial domain, watermarking in the frequency domain 

offers additional benefits. That is why many researchers use frequency domain watermarking 

techniques. DCT is one of the widespread transformations and is often chosen in frequency 

domain watermarking techniques [6] because of its lower computational cost, high robustness 

and high compression [11]. In [12,13], the watermark was embedded by Coltuc and Chassery 

using the DCT domain watermarking method. The image is divided into 8×8 blocks. Next, 

DCT for each block is determined. One bit of information entered into each block depends on 

the difference between the two coefficients of the mid-band frequency. For example, bit "1" is 

encoded if the difference is negative, and a bit "0" otherwise. If the relative size of each 

coefficient does not match the bit to be encoded, the coefficients are swapped. The interchange 

of these coefficients does not considerably alter the watermarked image because the mid-

frequency DCT coefficients have almost identical magnitudes. However, this method can cause 

errors due to the rounding case of DCT. As a result, not all watermarks can be extracted 

properly. 

Coltuc’s method is improved by modifying the DCT coefficient pairs in the watermark 

embedding formula. Our proposed formula ensures that the difference between the two 

coefficients is at least 𝜁. If the difference between the two coefficients is less than 𝜁, then the 
new pixels are modified so that the difference is 𝜁. The new watermarking system on medical 
images based on the modification of DCT coefficient pairs is proposed to achieve the following 

objectives: (1) to achieve good imperceptibility from watermarked images; (2) to provide 

intellectual property protection for medical images; (3) to preserve the safety of patient data 

[6]. 

2. RELATED WORKS 

The discrete cosine transform (DCT), which has energy compaction features, is frequently 

employed in watermarking approaches. DCT can convert an image into a frequency band, 

which can be inverted to transform it back into the original image. Following the DCT process, 

the image pixels are turned into coefficients. Direct current (DC) is represented by the first 

coefficient DCT in the base array function's upper left corner, while the remaining coefficients 

represent alternating current (AC). Since the AC is so crucial to an image, any modifications 

to the AC can significantly affect the image. On the other hand, slight adjustments to the image 

have little impact on the AC. 

90



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

Thanki et al. [14] developed a watermarking technique for medical images based on fast 

discrete curvelet transform (FDCuT) and DCT. The block-wise DCT is applied to the medical 

image's high-frequency curvelet coefficient. A white Gaussian noise (WGN) order is added to 

the original image's mid-band frequency coefficient in compliance with the watermark bit to 

create a watermarked medical image. Images from X-rays, ultrasounds, MRIs, and computed 

tomography (CT) scans were utilized to evaluate how the suggested watermarking technique 

worked. The study results show that watermarks are more invisible in all types of medical 

images with PSNR above 45 dB. However, two uncorrelated WGN sequences generated during 

the embedding process are required to recover the watermark. Therefore, only binary 

watermarks may be embedded using this technique; text-based EPR data cannot be. Another 

drawback is that the resulting watermark image always contains noise. 

Novamizanti et al. [15] introduced the singular value decomposition (SVD) method to 

medical image watermarking techniques based on FDCuT and DCT. Watermark embedding 

by exchanging the singular value of the watermark and the host image result from the FDCuT 

and DCT transformations. The proposed algorithm has good imperceptibility with interval 

PSNR values 53 dB to 54 dB. However, this watermarking scheme is semi-blind, where it is 

necessary to output SVD results from the original watermark image at the extraction stage. 

This method is not resistant to blurring and geometry attacks. 

Lei et al. [16] reported a robust image watermarking algorithm based on DCT domain and 

QIM. The host signal's medium and low-frequency DCT coefficient is divided into two regions. 

Then the watermark bit is embedded by quantizing the ratio of the two halves. According to 

experimental findings, this method is only superior against attacks of the common type. This 

algorithm has limited capacity, where a 512×512 image host can only accommodate a 128-bit 

watermark. Then, the optimal parameter for each image is obtained as PSNR of 41 dB. 

Zhu et al. [17] presented a robust image steganography algorithm for JPEG compression. 

Firstly, a candidate DCT coefficient unaffected by JPEG compression is chosen. Second, the 

proposed distortion function gives each potential DCT coefficient a cost. The error correction 

code and the Syndrome Trellis Code help the method become even more robust and invisible. 

This algorithm's embedding capacity remains restricted to payloads between 0.01 and 0.1 

bpnzAC. For 512×512 image hosts, the proposed algorithm only produces PSNR values with 

the interval 42 dB to 47 dB. 

Rachmawanto et al. [18] designed an image watermarking algorithm using block-based 

DCT, where the DC coefficient is chosen to keep the watermark resistance included. The 

Beaufort cipher is used for the encryption process and the watermark distribution when 

inserted. The goal is to improve watermark security and imperceptibility aspects. Generally, 

the spread spectrum watermarking technique uses PN Sequence for watermark deployment. 

However, the proposed method is non-blind, where the cover image is still required for the 

watermark extraction process. 

An image watermarking technique based on DCT was researched by Byun et al. [19]. First, 

the DCT coefficient for the specified location is determined. Next, the variation value is 

calculated to adjust the coefficient under the embedding bit and quantization step. Finally, the 

watermark bit is inserted directly into the pixel value without using a full-frame DCT. 

According to the findings, the suggested watermarking technology offers less computing 

complexity and ensures robustness. A color host image with a size of 512×512 can only 

accommodate a watermark of 1024 bits, indicating that the embedding capacity of this 

technique is still constrained. The proposed method's average PSNR is about 42 dB. 

91



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

Ko et al. [20] reported a watermarking method based on the correlation of DCT 

coefficients between blocks. The coefficient variance in the two blocks of DCT is calculated 

and changed depending on the watermark bits to adjust this difference to a specified range. The 

level of the DCT coefficient's modification is based on the DC coefficient and the median of 

the alternating current (AC) coefficient, which is ordered in zigzag order. The experimental 

findings demonstrate the robustness of the suggested technique against several single and 

combined attacks. For the Lena watermarked image, only a PSNR of 41.6 dB is obtained. In 

addition, the resistance is weak against types of noise attacks, including BER is 17.06 for salt-

and-pepper noise and 9.98 for Gaussian noise. 

A robust watermarking technique based on DCT, speeded up robust features (SURF), and 

perceptual hashing was proposed by Nawaz et al. [21]. The watermark image is preprocessed 

to enhance its security using affine transformation with feature matrix and chaotic encryption 

technology. SURF feature points are utilized to choose the DCT region and extract scalable 

watermarks after geometric rectification during the extraction process. Studies reveal that the 

algorithm efficiently defends against geometric and conventional attacks and can effectively 

maintain the security of images with an NC value higher than 90%. However, this paper does 

not report the PSNR results from the watermarked images without attack. This technique 

utilizes a correlation function to confirm the similarity of the derived watermarks. Hence 

scrambled data from the original watermark is still required at the watermark extraction stage. 

Kumar et al. [22] presented a secure watermarking framework on medical images by 

applying IntDCT and SVD. The differential evolution technique was used to determine suitable 

scaling factors by randomizing the watermark with the step space fill curve. In addition to the 

watermarked image, the 𝑈𝑊,𝑆,and 𝑉𝑊 values during insertion are also required for the 
watermark extraction process. The experimental findings demonstrate that the designed 

framework is robust to geometric attacks. 

Fares et al. [23] introduced watermarking approaches based on DCT and Schur 

decomposition to protect medical images. Integration is performed at the intermediate 

frequencies. According to the findings, the designed algorithm produces acceptable 

imperceptibility with PSNR values at intervals 47 dB to 49 dB. However, capacity is limited, 

where a 1024×1024 image host can only accommodate a 1024-bit watermark. The suggested 

method is robust under some conventional attacks but, it is not resistant to geometric attacks, 

such as cropping and scaling.  

3. PRELIMINARY 

In this section, the DCT technique and the method of Coltuc are presented and used in the 

proposed watermarking method. The energy compaction and redundancy removal properties 

of DCT make this transformation popular in image processing, especially image watermarking. 

However, DCT-based watermarking is prone to rounding errors. When this DCT is applied to 

the Coltuc’s method, which exchanges two DCT coefficients to embed a watermark into the 

host image, the watermarking method is challenging in developing watermarking techniques, 

especially in medical images. 

3.1  Discrete Cosine Transform 

As one of the prominent transformation techniques, DCT works by changing the image 

from the spatial to the frequency domain [13]. This technique has been used extensively in 

image processing. Some of the outstanding properties of DCT in image processing applications 

[24,25] include: 

92



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

1. Decorrelation can eliminate redundancy between neighboring pixels. 

2. DCT offers better energy compaction for correlated images. 

3. Separability and symmetry: Transformation matrices can be computed offline, thus 
providing computational efficiency. 

4. Orthogonality: This property makes some reduction in pre-computing complexity. 

Generally, the image is segmented into 8×8 pixel blocks. Then, each of those blocks is 

subjected to a 2D-DCT. Mathematically, the 2D-DCT transformation and the 2D-DCT inverse 

for a block size of 8×8 can be seen in Eq. (1) and (2), respectively, as follows: 

𝐹(𝑢,𝑣) = 
1

4
𝐶(𝑢)𝐶(𝑣)∑ ∑ 𝑓(𝑥,𝑦) 7𝑦=0

7
𝑥=0 𝑐𝑜𝑠(

(2𝑥+1)𝑢𝜋

16
)𝑐𝑜𝑠(

(2𝑦+1)𝑣𝜋

16
) (1) 

𝑓(𝑥,𝑦) =
1

4
𝐶(𝑢)𝐶(𝑣)∑ ∑ 𝐹(𝑢,𝑣)7𝑣=0

7
𝑢=0 𝑐𝑜𝑠(

(2𝑥+1)𝑢𝜋

16
)𝑐𝑜𝑠(

(2𝑦+1)𝑣𝜋

16
) (2) 

with 𝐶(𝜔) =
1

√2
 for 𝜔 = 0,and 𝐶(𝜔) = 1,for otherwise ; 𝐹(𝑢,𝑣) is the DCT coefficient; 

𝑓(𝑥,𝑦) is the pixel value in the spatial domain.   

There are two ways to apply DCT to images: block-wise and without block-wise. DCT 

separates the image into three frequency sub-bands, i.e., low (LF), middle/ mid (MF), and high 

(HF), in a block-wise approach. Figure 1 shows the position of the DCT coefficient with blocks 

of size (8×8) pixels. The 8×8 pixel block transformation yields 64 DCT coefficients. The 

position of the upper left corner of the DCT coefficient, 𝐹(0,0), is the DC component, and the 
other 63 coefficients are the AC component. The DC component represents the average color 

of the entire frequency-transformed region. Thus, embedding data in DC components can cause 

visual artifacts that are visible to the normal human eye. Embedding data is done in the MF 

coefficient region, because this region carries relatively less important image information than 

the HF and LF regions, and is less significant in perception. Thus, data embedding in the MF 

region does not significantly affect image quality or communication performance [14,26]. 

 

Fig. 1: The 2D discrete cosine transform (DCT) coefficients with blocks of size (8×8) pixels. 

 3.2  Coltuc’s Method 

Coltuc and Chassery [12,13] applied the DCT domain in the robust watermarking stage. 

Firstly, the cover image 𝑋 is segmented into 8×8 blocks. Next, DCT is applied to each block. 
The two MF coefficients �̂�𝑎 and �̂�𝑏 in each DCT block are chosen to insert one watermark bit 
𝑤ϵ {0,1}, as follows [27]: 

93



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

(�̂�𝑎
′ , �̂�𝑏

′  ) =

{
 

 
(�̂�𝑎, �̂�𝑏), 𝑖𝑓 �̂�𝑎 − �̂�𝑏 > 0 and 𝑤 = 1 
(�̂�𝑏, �̂�𝑎), 𝑖𝑓 �̂�𝑎 − �̂�𝑏 ≤ 0 and 𝑤 = 1 
(�̂�𝑎, �̂�𝑏), 𝑖𝑓 �̂�𝑎 − �̂�𝑏 ≤ 0 and 𝑤 = 0 
(�̂�𝑏, �̂�𝑎), 𝑖𝑓 �̂�𝑎 − �̂�𝑏 > 0 and 𝑤 = 0 

 (3) 

with 𝑎 and 𝑏 being the two AC channel indexes. The index is the coordinates of the pixels. 
Thus, the AC channel index represents the coordinates of the DCT coefficients located in the 

AC component. For example, 𝑎=(2,3) and 𝑏=(3,2) are the selected AC channel indexes, so 
the two MF coefficients �̂�(2,3) and �̂�(3,2) are used to insert a watermark bit, as shown in Fig. 1. 

By replacing �̂�𝑎
′  and �̂�𝑏

′  for the appropriate coefficients, the DCT coefficients for 𝑌 are 
obtained. The watermarked image 𝑌 is then created by applying the inverse DCT. For the 
decoding process, 𝑌 is segmented into 8×8 blocks. Next, each block is converted using the 
DCT technique. The embedded watermark for a coefficient pair of �̂�𝑎 and �̂�𝑏 is recovered as: 
if �̂�𝑎 − �̂�𝑏 > 0, then the bit watermark is 1, and 0 otherwise. 

One bit of information is entered into each block depending on the difference between the 

two coefficients of MF. The new coefficient value does not change if the difference is positive 

and the bit is "1", or the difference is negative, and the bit is "0". However, for other conditions, 

the positions of the two MF coefficients are swapped. The exchange of these coefficients does 

not change the watermarked image significantly because the MF of DCT coefficients generally 

have almost the same magnitude. However, this condition is prone to extraction errors due to 

the rounding factor of DCT, as shown in Fig. 2(a). As a result, not all watermarks can be 

extracted properly. One solution is to select two MF coefficients �̂�𝑎 and �̂�𝑏 to embed a robust 
watermark. However, this method becomes impractical.  

4.   PROPOSED METHOD: WATERMARKING IMAGES VIA 

MODIFYING DCT COEFFICIENT PAIR 

In this section, the modification of the insertion formula in Coltuc’s method is described 

in subsection 4.1. The watermarking embedding step is presented in subsection 4.2, and the 

watermarking extraction step is presented in subsection 4.3. The watermarking framework for 

medical images is generally shown in Fig. 3. First, a watermark in the form of an EPR is hidden 

into a medical image, resulting in a 𝑌 watermarked image, and then sent through a 
communication channel. During transmission, the watermarked image undergoes various 

attacks, resulting in a medical image that is attacked as �̂�. The attacks can be in the form of 
noise and intentional and unintentional modifications. In the extraction process, only 𝑌 (for 
noise-free channel) or 𝑌 (for noise channel) is needed to restore the original watermark. Thus, 
it can preserve medical images with copyright and maintain patient data security. 

This paper enhances the robust watermarking method Coltuc et al. by modifying the DCT 

coefficient pairs in the watermark embedding formula. The proposed formula ensures that the 

difference between the two coefficients is at least as much as 𝜁. If the difference between the 
two coefficients is less than 𝜁, then the new pixels are modified so that the difference is equal 
to 𝜁. This new watermarking system on medical images based on modifying the DCT 
coefficient pair is suggested to obtain acceptable imperceptibility of watermarked images, 

impart data confidentiality of medical images, and preserve patient information privacy. 

94



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

 

Fig. 2: Illustration of watermark embedding formula (a) Coltuc’s method, (b) Proposed. 

 

Fig. 3: The framework of medical image watermarking. 

4.1 Modification of the Watermark Embedding Formula in Coltuc’s Method 

The value of the MF coefficient of the DCT generally has the same magnitude. After 

rounding operations to produce watermarked images and entering the decoding process, not all 

watermarks can be extracted correctly. We modified Eq. (3) so that the watermarked host image 

has high imperceptibility, protects the copyright (watermark) embedded in medical images, 

and maintains the privacy of patient information. 

(�̂�𝑎
′ , �̂�𝑏

′  ) =

{
 
 
 
 

 
 
 
 
(�̂�𝑎, �̂�𝑏),             𝑖𝑓 �̂�𝑎 − �̂�𝑏 > 𝜁 and 𝑤 = 1    

(�̅� +
𝜁

2
, �̅� −

𝜁

2
)  𝑖𝑓 |�̂�𝑎 − �̂�𝑏| ≤ 𝜁 and 𝑤 = 1 

(�̅� +
𝜁

2
, �̅� −

𝜁

2
)  𝑖𝑓 �̂�𝑎 − �̂�𝑏 < −𝜁 and 𝑤 = 1  

(�̂�𝑎, �̂�𝑏),              𝑖𝑓 �̂�𝑎 − �̂�𝑏 < −𝜁 and 𝑤 = 0 

(�̅� −
𝜁

2
, �̅� +

𝜁

2
), 𝑖𝑓 |�̂�𝑎 − �̂�𝑏| ≤ 𝜁 and 𝑤 = 0

(�̅� −
𝜁

2
, �̅� +

𝜁

2
), 𝑖𝑓 �̂�𝑎 − �̂�𝑏 > 𝜁 and 𝑤 = 0   

 

 (4) 

 

The proposed formula ensures that the difference between the two MF coefficients, namely 

�̂�𝑎 and �̂�𝑏, is at least as much as 𝜁. If the distance between the two MF coefficients is less than 
𝜁, then the new pixel is modified so that the difference in distance is as significant as 𝜁. 

95



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

Modification of the insertion formula of Coltuc’s method through the DCT coefficient pair can 

be seen in Eq. (4) and then simplified to Eq. (5). Figure 2(b) presents the illustration of the 

proposed formula 

(�̂�𝑎
′ , �̂�𝑏

′  ) =

{
(�̂�𝑎, �̂�𝑏),                                             if (�̂�𝑎 − �̂�𝑏 < −𝜁 and 𝑤 = 0) or (�̂�𝑎 − �̂�𝑏 > 𝜁 and 𝑤 = 1)    

(
�̂�𝑎+�̂�𝑏+(2𝑤−1)𝜁

2
,
�̂�𝑎+�̂�𝑏−(2𝑤−1)𝜁

2
), otherwise                                                                                          

 (5) 

with �̅� is the average of the two MF coefficients of the DCT, namely �̂�𝑎 dan �̂�𝑏. 

4.2 Embedding Stage 

The input of the watermarking embedding stage is a medical image 𝑋 as the host, and the 
output is a watermarked host image 𝑌. The sizes of 𝑋 and 𝑌 are M × M, while the size of 𝑊 is 
N × N. Fig. 4(a) presents the watermarking embedding stage, and the detailed embedding steps 

are described as follows. 

1. Split 𝑋 into 8×8 non-overlapping blocks.  

2. For each block, apply DCT using Eq. (1) so that it produces �̂�. 

3. In each DCT block, the two MF coefficients �̂�𝑎 and �̂�𝑏 are decisive in embedding one-
bit watermark 𝑤ϵ {0,1}. The robust watermark embedding formula is expressed in Eq. 
(5). 

4. The DCT coefficient of 𝑌 is done as follows: 

�̂�𝑖  = {
�̂�𝑖
′, 𝑖𝑓 𝑖 𝜖 {𝑎,𝑏}

�̂�𝑖, otherwise
 (6) 

5. Process the inverse DCT using Eq. (2), and a watermarked image is obtained. 

 
(a) 

 

 
 

(b) 

Fig. 4: The proposed watermarking method using DCT coefficient pair (a) embedding stage (b) 

extraction stage. 

4.3  Extraction Stage 

The input of the watermarking extraction stage is the 𝑌 watermarked host image, and the 
output is the extracted watermark 𝑊∗. While, 𝑌 and 𝑊∗ sizes are M × M, and N × N, 
respectively. Figure 4(b) presents the watermarking extraction stage, and the extraction steps 

in detail are described as follows. 

1. Split 𝑌 into 8×8 non-overlapping blocks.  

96



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

2. Apply DCT to each block using Eq. (1) so that it produces �̂�. 

3. In each DCT block, the two MF coefficients �̂�𝑎 and �̂�𝑏 become determinants in 
extracting the one-bit watermark 𝑤ϵ {0,1}. The robust watermark extraction formula is 
stated as follows. 

𝑊∗ = {
1, if �̂�𝑎 − �̂�𝑏 > 0
0, otherwise       

 (7) 

At the transmission is without noise, 𝑊∗ equals 𝑊. 

5.   RESULTS AND DISCUSSION 

This section evaluates the effectiveness of the suggested robust and imperceptible 

watermarking technique. Subjective visual observation and objective quantitative analysis are 

used to evaluate the proposed method. Numerous attacks with varying parameters were 

performed to assess robustness further. Finally, the proposed method's imperceptibility and 

robustness are compared to state-of-the-art works. 

This experiment uses the grayscale format and image sizes 512×512 of the host medical 

image. A total of five medical images from different modality types were used as test data, 

including MRI, X-Ray, Computed Tomography (CT), ultrasound (US), and colonoscopy. 

Medical images were retrieved from the MedPixTM medical image database [26] and CVC-

ClinicDB database [28]. The host medical image modality used in the experiment is shown in 

Fig. 5(a-e). Meanwhile, the watermark is a random binary image whose size depends on the 

block size discussed in the previous section. For an 8×8 block, the amount of watermarks 

embedded into the original host medical image is (512×512)/(8×8)=4096 bits. In other words, 

blocks measuring 8×8 yield a watermark of size 64×64. 

The efficiency of the proposed method is thoroughly evaluated on two watermarking 

criteria, including imperceptibility and robustness. The performance metric for imperceptibility 

criteria uses peak signal noise ratio (PSNR) [9]. The watermarked host image must be invisible 

to humans to guarantee information security. Therefore, imperceptibility/ invisibility criteria 

are essential metrics in watermarking techniques. The PSNR measures how visually similar the 

watermarked image and the original host are. Both images should look the same, so there is no 

significant difference between the two. The similarity between the two images increases with 

increasing PSNR values and vice versa. Generally, the watermarked image is acceptable if the 

PSNR is more than 37 dB and the watermark is not discernible to the human visual system 

[29]. Here, 𝑋 is the original host image, 𝑌 is the watermarked host image, and M×M is the host 
image size. Mathematically, PSNR (in dB) is formulated as follows [11,15]. 

𝑃𝑆𝑁𝑅 = 10 ×  𝑙𝑜𝑔
2552× M × M

∑ ∑ (𝑿𝒊,𝒋−𝒀𝒊,𝒋)
2M

𝑗=1
M
𝑖=1

              (8) 

The bit error rate (BER) and normalized correlation (NC) are the performance metric to 

evaluate a watermarking technique's robustness criteria. Technically, BER measures the 

erroneous bit rate after the watermark extraction process. If the BER value is close to 0, then a 

few messages are lost, and vice versa. Ideally, the value of BER should be 0. If EB is the 

amount of incorrectly decoded bits, and TB is the number of watermark bits, then BER can be 

calculated [14], [27] as EB TB⁄ . 

97



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

 

Fig. 5: Five modalities of original medical image and a watermark (a) X-ray , (b) CT, (c) US, (d) 

MRI, (e) Colonoscopy, (f) watermark. 

Principally, the NC measures the similarity between the original and extracted watermark 

images. The NC can be described mathematically by Eq. (9). If the NC value is near 1, then 

the watermarking method is very robust. 

𝑁𝐶 =
∑ ∑ 𝑊𝑖,𝑗 𝑊𝑖,𝑗

∗N
𝑖−1

N
𝑖=1

∑ ∑ 𝑊𝑖,𝑗
2N

𝑖−1
N
𝑖=1

              (9) 

where 𝑊 and 𝑊∗ are the original watermark and extracted watermark images, respectively. 
N×N is the watermark image's size [14]. After applying the embedding and extraction 

processes, the proposed scheme's imperceptibility and robustness tests are carried out for varied 

medical images. 

5.1  𝜁 Parameter Evaluation 

The proposed watermark embedding formula ensures that the difference between the two 

MF coefficients, namely �̂�𝑎 and �̂�𝑏, is at least 𝜁. If the distance between the two MF coefficients 
is less than 𝜁, then the new pixels are modified so that the distance difference is 𝜁. We changed 
the value of the 𝜁 parameter in the watermark embedding formula to get the PSNR and expected 
robustness level. Experiments were carried out on five types of medical images, i.e., X-ray, 

CT, US, MRI, and Colonoscopy. Figure 6 presents the effect of the 𝜁 parameter on PSNR and 
BER. Figure 6(a) shows the relationship between 𝜁 and BER.  

Higher 𝜁 results in lower BER. The criterion BER is 0 when 𝜁 ≥ 3. It means that the 
extracted watermark can be recovered very well. BER of 0 means there are no bit errors in 

watermark extraction. While Fig. 6(b) shows the relationship between BER and PSNR over 𝜁, 
which corresponds to Fig. 6(a). The higher the 𝜁 value, the higher the resistance, but the lower 
the PSNR. When the resistance condition is ideal, namely BER is 0, the PSNR value ranges 

around 48 dB to 59 dB for the range of 𝜁 is 3 to 10. Thus, 𝜁 determines the imperceptibility and 
robustness of the watermark embedding into the host medical image. 

5.2 Imperceptibility Analysis 

For integrity and ensuring information security, watermarked host medical images must 

be imperceptible/ invisible to humans. Figure 7 shows a watermarked host X-ray image when 

not under attack and the extracted watermark. The experiment was carried out when 𝜁 = 3. The 
PSNR, BER, and NC results of each corresponding image are also shown in Fig. 7. The 

watermarked image is acceptable because the PSNR is more than 37 dB, and the watermark is 

not discernible to the human visual system [30]. 

98



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

(a) (b) 

Fig. 6: Evaluation of 𝜁 (a) The BER of extracted watermark 
(b) The trade-off between PSNR and BER.

Fig. 7: Imperceptibility and robustness performance. Rows represent the extracted watermarked 

image and watermark along with their corresponding PSNR, NC, and BER. Columns represent type 

medical images, i.e., X-ray, CT, US, MRI, and Colonoscopy. 

Based on Fig. 7, the imperceptibility criterion of the suggested method is that all PSNR of 

each watermarked medical image is greater than 54 dB. The average PSNR generated for the 

five watermarked medical image modalities is 56.3604 dB. Then, all extracted watermarks 

yielded BER is 0 and NC is 1 when the watermarked host image was not subjected to attack. It 

demonstrates that the method of watermarking is imperceptible. Furthermore, the suggested 

approach satisfies both the subjective and objective standards for watermarking invisibility. 

When the watermarked host image is not subjected to an assault, extracted watermarks have a 

BER of 0 and an NC of 1. Therefore, the process of watermarking has a high level of 

imperceptibility. Thus, the suggested method satisfies both the subjective and objective 

requirements for watermarking invisibility. 

5.3  Robustness Analysis 

99



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

Various watermarking attacks were applied to watermarked medical images to examine 

the robustness criteria of the proposed method. Table 1 represents different attacks with 

parameter specifications used in the experiment. These attacks include compression, filtering, 

noise, histogram equalization, sharpening, and motion blur. In particular, compression attacks 

include JPEG and JPEG2000. JPEG compression parameters with quality factor (QF) 80 and 

JPEG2000 compression with compression ratio (CR) = 4 and 8. Filter attacks consist of 

median, Gaussian, average, and lowpass filters. The parameters in the filtering attack use a 

filter with a window size of 3×3. Noise attacks include Gaussian, salt & pepper, and speckle. 

The parameters in the noise addition attack are set by the variance or noise density of 0.001. 

Parameters of motion blur attack with linear motion of the camera with pixels Len = 7, with 

angle degrees Theta = 4. At the same time, histogram equalization and sharpening use default 

parameters.  

Table 1: Various Attacks with Parameter Specifications in Experiments 

Attack Parameter Specifications 

JPEG compression (JC) QF = 80, 90 

JPEG2000 compression (J2k) CR = 8, 12 

Median filter (MF) 3×3 

Gaussian filter (GF) 3×3 

Average filter (AF) 3×3 

LPF 3×3 

Gaussian noise (GN) Mean = 0, Variance = 0.001 

Salt & peppers noise (SP) Noise density = 0.001 

Speckle noise (SN) Mean = 0, Variance = 0.001 

Histogram equalization (HE) - 

Sharpening (SH) - 

Motion blur (MB) Len = 7, Theta = 4 

 

In the extraction stage, if the extracted watermark can recover from damaged watermarked 

images, then the watermarking method is said to be robust and secure. Here, BER and NC 

performance metrics examine the robustness of the proposed method under various 

watermarking attacks. In addition, the result of the suggested scheme is compared with the 

existing techniques [15,17] for the same medical image dataset. The BER of extracted 

watermark on numerous watermarked medical images is shown in Fig. 8. Meanwhile, Figure 

9 presents the results of the extracted watermark from the attacked watermarked X-ray along 

with corresponding BER and NC. 

Various attacks are rendered on watermarked medical images. These medical images 

include X-ray, CT, US, MRI, and Colonoscopy. On the extraction side, the extracted watermark 

image is expected to be recovered from the damaged watermarked medical image. Based on 

Fig. 8, JPEG compresses images without significantly altering their appearance. Medical 

images with watermarks are compressed with JPEG using varying quality factors. Then, the 

proposed method is used to extract the watermarked image. The experiments indicated that the 

proposed method performs well for all the medical images we tested on JPEG compression 

with QF > 80 and JPEG2000 compression with CR < 12. The extracted watermark can still be 

interpreted visually if BER < 0.3 [30]. On X-ray, CT, and Colonoscopy images, watermarks 

can be extracted well even at JPEG compression with QF = 90 and JPEG2000 compression 

with CR = 12. Remarkably, watermarks can be best extracted, with BER is 0 for all 

100



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

compression parameters given in watermarked Colonoscopy image. Based on Fig. 9, 

watermarks can be extracted well with BER of 0 and NC of 1 for JPEG compression attacks 

with QF = 90 and JPEG2000 with CR = 8, and 12 given to the attacked watermarked X-ray 

images. 

 

Fig. 8: The BER of extracted watermarks when condition with attack. 

 

Fig. 9: The BERs and NCs of extracted watermarks from watermarked X-rays attacked.   

Four filters, including the median, Gaussian, average, and low pass, and three different 

noises, such as Gaussian, salt & pepper, and speckle, are applied to watermarked medical 

images. Afterward, the watermark image is extracted using the proposed method. The 

experiment results (Fig. 8 and Fig. 9) prove that the proposed method performs well for all 

medical images provided a filtering attack. Remarkably, watermarks can be extracted very well 

with BER 0 for watermarked Colonoscopy medical images subjected to average filter and LPF 

attacks. The experiment results also prove that the proposed scheme performs well for all 

medical images under the noise attack. Remarkably, watermarks can be successfully extracted 

with a BER of 0 for all watermarked medical images subjected to speckle noise attacks. 

101



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

A series of other attacks, such as histogram equalization, sharpening, and motion blur are 

also applied to watermarked medical images for endurance testing. Based on the experiment 

results (Fig. 8 and Fig. 9), the proposed scheme performs well for all medical images given the 

three attacks. Impressively, watermarks can be well extracted with a BER of 0 for all 

watermarked medical images given a sharpening attack. Watermarks can also be extracted with 

a BER of 0 for motion blur attacks rendered in watermarked MRI and Colonoscopy images. 

5.4 Comparison of the Proposed Method with State-of-the-Art Work 

In Fig. 10, the imperceptibility performance of the proposed method is compared with 

Coltuc's method [13] and the recently developed watermarking method [14,15] for medical 

image identity protection. The watermark payload under the same conditions, which is 1 bit for 

every 64 pixels of the host medical image. The PSNR value of the proposed method is 

compared with the related work. The methods are compared using 𝜁 = 3 and without 
watermarking attacks. The proposed scheme offers superior performance to the three existing 

schemes [13-15]. In the proposed method, the average PSNR value of watermarked medical 

images is 56.3604 dB. Meanwhile, the mean of PSNR for watermarked medical images from 

the [13-15] scheme are 50.5647 dB, 48.3150 dB, and 53.6184 dB, respectively. It means that 

the insertion of watermark bits does not influence the watermarked medical image of the 

proposed method. The proposed method produces high imperceptibility over the the state-of-

the-art works [13-15]. 

 

Fig. 10: Comparison under different medical images and methods.  

Specifically, Table 2 summarizes the robustness performance of the proposed method 

compared to Coltuc’s method [13] to evaluate how much contribution the proposed 

watermarking method makes. The host image used as test data is an X-ray. The suggested 

method performs well compared to the previous method against all types of watermarking 

attacks. In fact, the watermark can be extracted perfectly in a JPEG compression attack with 

QF = 90, JPEG2000 with CR = 8 and 12, speckle noise, and sharpening. The proposed method 

is outstanding from Coltuc’s method [13], with the increase in BER and NC being 93% and 

14%, respectively. 

Table 2: Comparison of BERs and NCs of the Proposed Method to the Coltuc’s Method [13] in 

Attacked Watermarked X-Ray 

Attack BER NC 

102



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

 

 

[13] Proposed [13] Proposed 

JPEG (QF=80) 0.2488 0.0056 0.8491 0.9888 

JPEG (QF=90) 0.2493 0 0.8523 1 

J2k (CR=8) 0.1177 0 0.9605 1 

J2k (CR=12) 0.1626 0 0.9337 1 

MF [3×3] 0.2842 0.0374 0.4059 0.9221 

GF [3×3] 0.1492 0.0405 0.6942 0.9172 

AF [3×3] 0.0615 0.0208 0.8693 0.9563 

LPF [3×3] 0.0625 0.0190 0.8673 0.9598 

GN (0.001) 0.2261 0.0042 0.8644 0.9950 

SP (0.001) 0.0173 0.0054 0.9877 0.9982 

SN (0.001) 0.2146 0 0.8652 1 

HE 0.1060 0.0068 0.9237 0.9863 

SH 0.0225 0 0.9843 1 

MB 0.0603 0.0022 0.9706 0.9956 

 

Table 3: NCs Comparison of the Proposed Method to the State-of-the-Art [14,15] in Attacked 

Watermarked X-Ray 

Attack Thanki [14] Novamizanti [15]  Proposed 

JPEG (QF=80) 0.9282 0.8769 0.9888 

JPEG (QF=90) 0.9806 0.8973 1 

GN (0.001) 0.6377 0.9970 0.9950 

SP (0.001) 0.7532 0.9473 0.9982 

HE 0.9708 0.8909 0.9863 

SH 0.9674 0.9890 1 

The proposed method is also compared with the recently developed watermarking methods 

[14,15] to maintain the privacy and protection of the EPR as a watermark. Host X-ray images 

were used as test data for all compared methods. Watermarking in both techniques [14,15] was 

performed in the FDCuT and DCT domains. The extraction process of both methods is semi-

blind. The difference is the insertion of the watermarking technique [14] using the correlation 

sequence method. While in [15], the watermarking embedding technique uses the SVD method. 

Compared with the state-of-the-art works [14,15], the proposed method is superior in terms of 

resistance to JPEG compression attacks, sharpening, histogram equalization, and sharpening. 

In addition, on a JPEG compression attack with QF = 90 and sharpening, the watermark can 

be extracted perfectly. The proposed method modifies the DCT coefficient pairs to embed a 

robust watermark. Thus, the average result demonstrates that the suggested method 

outperforms the present scheme regarding imperceptibility, robustness, and security. The 

proposed approach can be part of a robust reversible watermarking scheme [31] due to the blind 

extraction stage of the proposed watermarking method. 

3. CONCLUSION 

This paper proposes a modification of the DCT coefficient pair in the watermark 

embedding process. The proposed formula ensures that the difference between the two 

coefficients is at least as 𝜁. If the difference between the two coefficients is less than 𝜁, then 
the new pixels are modified so that the difference is equal to 𝜁. This simple watermarking 

103



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

technique based on modifying DCT coefficient pairs is offered to achieve high imperceptibility, 

provide intellectual property protection for medical images, and ensure patient data security. 

The suggested method was evaluated on different types of medical images, including X-ray, 

CT, US, MRI, and Colonoscopy, and compared to numerous robust watermarking techniques 

of the current work. The experimental results indicate that the suggested method outperforms 

various contemporary watermarking techniques regarding imperceptibility, robustness, and 

security. The PSNR value for all categories of watermarked medical images exceeds 54 dB, 

and the average PSNR value is approximately 56 dB. In terms of resistance to JPEG 

compression attacks, histogram equalization, and sharpening, the suggested method is superior 

to numerous state-of-the-art robust watermarking methods. Future research could apply the 

presented technology to a robust reversible watermarking strategy and medical image 

protection for telemedicine applications. 

ACKNOWLEDGEMENT 

This research is part of L. Novamizanti's dissertation research funded by the Doctoral 

Dissertation Research Grant from the Ministry of Research and Technology of the Republic of 

Indonesia, file numbers: 083/E5/PG.02.00.PT/2022 and 343/IT1.B07.1/SPP- LPPM/V/2022. 

The Indonesian Endowment Fund for Education (LPDP) Scholarship and Telkom University 

also partially funded this research. 

REFERENCES 

[1] Memon, NA. Alzahrani A. (2020) Prediction-based reversible watermarking of CT scan images

for content authentication and copyright protection. IEEE Access, 8: 75448-75462.

doi: 10.1109/ACCESS.2020.2989175.

[2] Sajedi H, Rahbar S. (2020) Information hiding methods for E –Healthcare. Smart Health, 15:

100104. doi: 10.1016/j.smhl.2019.100104.

[3] Ahmad GI, Singla J, Giri KJ. (2021). Security and Privacy of E-health Data. In Multimedia

Security, Springer Singapore, pp. 199-214.

[4] Lanxiang C, Wutong BAI, Zhiqiang YAO. (2020) A Secure and Privacy-Preserving Watermark

Based Medical Image Sharing Method. Chinese J. Electron., 29(5): 819-825.

doi: 10.1049/cje.2020.07.003.

[5] Magdy M, Hosny KM. (2022) Security of medical images for telemedicine : a systematic review.

Multimed. Tools Appl., 81: 25101-25145.   doi: 10.1007/s11042-022-11956-7

[6] Alzahrani A, Memon NA. (2021) Blind and Robust Watermarking Scheme in Hybrid Domain for

Copyright Protection of Medical Images. IEEE Access, 9: 113714-113734.

doi: 10.1109/ACCESS.2021.3104985.

[7] Thakur S, Singh AK, Ghrera SP, Mohan A. (2020) Chaotic based secure watermarking approach

for medical images. Multimed. Tools Appl., 79: 4263-4276.   doi: 10.1007/s11042-018-6691-0

[8] Hassan B, Ahmed R, Li BO, Hassan O. (2019) An Imperceptible Medical Image Watermarking

Framework for Automated Diagnosis of Retinal Pathologies in an eHealth Arrangement. IEEE

Access, 7: 69758-69775.   doi: 10.1109/ACCESS.2019.2919381.

[9] Anand A, Singh AK. (2021) Watermarking techniques for medical data authentication : a survey.

Multimed. Tools Appl., 80: 30165-30197.   doi: 10.1007/s11042-020-08801-0

[10] Evsutin O, Dzhanashia K. (2022) Watermarking schemes for digital images : Robustness

overview. Signal Process. Image Commun., 100: 116523.   doi: 10.1016/j.image.2021.116523.

[11] Saini D, Ali M. (2020) A robust medical image watermarking framework based on SVD and DE

In Integer DCT domain (Workshop Paper), pp. 373-378.

doi: 10.1109/BigMM50055.2020.00064.

[12] Coltuc D, Chassery JM. (2007) Distortion-free robust watermarking: a case study. Secur.

Steganography, Watermarking Multimed. Contents IX, 6505: 65051N, doi: 10.1117/12.702445.

104



IIUM Engineering Journal, Vol. 24, No. 1, 2023 Novamizanti et al. 
https://doi.org/10.31436/iiumej.v24i1.2597 

[13] Coltuc D. (2007) Towards distortion-free robust image authentication. J. Phys. Conf. Ser., 77:

012005.   doi: 10.1088/1742-6596/77/1/012005.

[14] Thanki R, Borra S, Dwivedi V, Borisagar K. (2017) An efficient medical image watermarking

scheme based on FDCuT–DCT. Eng. Sci. Technol. an Int. J., 20(4): 1366-1379.

doi: 10.1016/j.jestch.2017.06.001.

[15] Novamizanti L, Wahidah I, Wardana NPDP. (2020) A Robust Medical Images Watermarking

Using FDCuT-DCT-SVD. Int. J. Intell. Eng. Syst., 13(6): 266-278.

doi: 10.22266/ijies2020.1231.24

[16] Lei M, Liu X, Wang M, Yang Y, Qu Z. (2018) Robust image watermarking based on quantization

index modulation in the DCT domain. J. Internet Technol., 19(2): 507-514.

doi: 10.3966/160792642018031902019.

[17] Zhu Z, Zheng N, Qiao T, Xu M. (2019) Robust steganography by modifying sign of DCT

coefficients. IEEE Access, 7: 168613-168628.   doi: 10.1109/ACCESS.2019.2953504.

[18] Rachmawanto EH, Setiadi DRIM, Sari CA, Rijati N. (2019) Imperceptible and secure image

watermarking using DCT and random spread technique. Telkomnika (Telecommunication

Comput. Electron. Control., 17(4): 1750-1757.   doi: 10.12928/TELKOMNIKA.v17i4.9227.

[19] Byun SW, Son HS, Lee SP. (2019) Fast and Robust Watermarking Method Based on DCT

Specific Location. IEEE Access, 7: 100706-100718.   doi: 10.1109/ACCESS.2019.2931039.

[20] Ko HJ, Huang CT, Horng G, WANG SJ. (2020) Robust and blind image watermarking in DCT

domain using inter-block coefficient correlation. Inf. Sci. (NY)., 517: 128-147.

doi: 10.1016/j.ins.2019.11.005.

[21] Nawaz SA, Li J, Bhatti UA, Mehmood A, Shoukat MU, Bhatti MA (2020) Advance hybrid

medical watermarking algorithm using speeded up robust features and discrete cosine transform.

PLoS One, 15(6): 1-21.   doi: 10.1371/journal.pone.0232902.

[22] Kumar S, Jha RK. (2019) FD-based detector for medical image watermarking. IET Image

Processing, 13(10): 1773-1782 2019. doi: 10.1049/iet-ipr.2018.5485.

[23] Fares K, Khaldi A, Redouane K, Salah E. (2021) DCT & DWT based watermarking scheme for

medical information security. Biomed. Signal Process. Control, 66: 102403.

doi: 10.1016/j.bspc.2020.102403.

[24] Khayam SA. (2003) The Discrete Cosine Transform (DCT): Theory and Application. In

Information Theory and Coding, Michigan State University, pp. 1-13.

[25] Hamidi M, Haziti ME, Cherifi H, Hassouni ME. (2018) Hybrid blind robust image watermarking

technique based on DFT-DCT and Arnold transform. Multimed. Tools Appl., 77(20): 27181-

27214.   doi: 10.1007/s11042-018-5913-9.

[26] Hernandez JR, Amado M, Perez-Gonzalez F. (2000) DCT-domain watermarking techniques for

still images: detector performance analysis and a new structure. IEEE Trans. Image Process.,9(1):

55-68.   doi: 10.1109/83.817598.

[27] Wang X, Li X, Pei Q. (2020) Independent Embedding Domain Based Two-Stage Robust

Reversible Watermarking. IEEE Trans. Circuits Syst. Video Technol., 30(8): 2406-2417.

doi: 10.1109/TCSVT.2019.2915116.

[28] Rahim T, Novamizanti L, Ramatryana INA, Shin SY. (2021) Compressed medical imaging based

on average sparsity model and reweighted analysis of multiple basis pursuit. Comput. Med.

Imaging Graph., 90: 101927.   doi: 10.1016/j.compmedimag.2021.101927.

[29] Guo Y, Li BZ., Goel N (2017) Optimised blind image watermarking method based on firefly

algorithm in DWT-QR transform domain. IET Image Process, 11(6): 406-415.

doi: 10.1049/iet-ipr.2016.0515

[30] Budiman G, Suksmono AB, Danudirdjo D. (2020) Compressive Sampling with Multiple Bit

Spread Spectrum-Based Data Hiding. Appl. Sci., 10(12): 1-21.   doi: 10.3390/app10124338.

[31] Novamizanti L, Suksmono AB, Danudirdjo D, Budiman G. (2022) Robust Reversible

Watermarking using Stationary Wavelet Transform and Multibit Spread Spectrum in Medical

Images. Int. J. Intell. Eng. Syst., 15(3): 343-354.   doi: 10.22266/ijies2022.0630.29.

105