Digital Image Watermarking using Machine

Learning Techniques: A Technical Review

Swati Sharma*, Amit Kumar Singh ** and Pardeep Kumar
* ** *** Department of Computer Science & Engineering, Jaypee University of Information Technology Waknaghat, Solan,
Himachal Pradesh-India
*[email protected] **[email protected] ***[email protected]

Abstract: Recently, digital information can be easily manipulated, copied, distributed and stored, which has resulted in the
demand for safe ownership of the information. The watermarking provides a very good solution to the problem of copyright
protection and content authentication. This paper discusses basic concepts of digital watermarking and its characteristics,
important attacks on watermark system, general watermark embedding and extraction process, important transform and
machine learning techniques and some reported transform and spatial domain based watermarking method using machine
learning. Finally, the current state-of-the-art in the field is also discussed. This paper will be more important for researchers
to implement effective watermarking method using machine learning technique.

Keywords: Image Watermarking, Discrete wavelet transform (DWT), Discrete Cosine Transform (DCT), Support Vector
Machine (SVM), Singular Value Decomposition (SVD), Fuzzy Support Vector Machine (FSVM), Neural Network (NN).

Introduction
The digital media such as image, audio and video is an important way of communication in the world and being increasingly
used for delivery of multimedia content, thus it is easy to manipulate, store, distribute or reproduce the data using different
networking sites. This shows no difference in the quality between an original image and its copy image. However,
unrestricted copying and malicious tampering cause huge financial losses and problems for intellectual property rights [1-2].
Therefore, information hiding has become an important research area and watermarking is used as a data hiding technique for
the protection of digital images. Digital watermarking systems have been proposed to provide content protection,
authentication and copyright protection, protection against unauthorized copying and distribution [3-4]. In this technique, it
consists of protecting the illegal insertion of robust and imperceptible brand in a host image. The watermarking algorithms
must be imperceptible to the naked eye, robust against attacks, blind which means the original image is not necessary for the
detection and extraction of the brand. Important characteristics of digital watermarks are robustness, imperceptibility,
capacity, security and computational cost [5-7]. However, robustness, imperceptibility and capacity are tradeoff between each
other. So, there is a strong requirement to balance these characteristics. The image watermarking techniques is divided into
two domain methods: 1) Spatial domain methods (least significant bit substitution, spread spectrum etc.) are more simple
high capacity but are not robust against common signal processing attacks. 2) Transform domain methods (DWT, DCT and
SVD etc.) are more robust against common signal processing attacks but the computational complexity is higher than spatial
domain methods [8-10].
Machine learning is a technique to determine predict from desired observations or past behavior. It is a method that improves
the detection rate of watermarks after being attacked and contains numerous methods for different classification and patterns
for recognition of problem. It also provides increasing level of automation in the knowledge engineering process by replacing
time consuming activity with automatic techniques used for improvement of accuracy and efficiency [9-11].
Generally, four types of attacks on digital watermarking systems as discussed below [5]:
a. Active attacks: In active attacks hacker tries to remove watermark. They are aimed at distortion of watermark before
recognition. This shows the problem in copyright protection, copy control etc.
b. Passive attacks: In passive attacks, hackers just tried to figure out that there is watermark and identify it. In this
attack no damage or removal is done.
c. Forgery attacks: In forgery attacks hackers embed new valid watermark rather than removing one. In this hackers
can easily manipulates the data and makes corrupted image genuine.
d. Collusion attacks: In collusion attacks hackers has the same intension as for the active ones but used slightly
different approach. The hacker uses instances of same data to construct the new copy without watermark.
84 Advances in Engineering and Technology

General Watermarking System: Embedding and Extraction Process

In general, watermarking system consists of two processes, embedding and extraction. The embedding process is consisting
of encoding [8]. It is used to produce the watermarked image. The watermark embedding process takes a cover image (C),
watermark image (WI) and secret key (K) then it goes to embedded function gives watermarked image (WE). The extraction
process is consisting of recovery process. It is used to recover the corrupted image, which may or may not be the
watermarked image. The extraction process takes cover image and watermarked with secret key to recover the watermark
from the possibly distorted image [9-10]. The watermark embedding and extraction process is given in Figure 1(a) and
Figure 1(b) respectively. The watermark embedding process can be written as:
WE = F(C, WI, K)
Also, the watermark extraction process can be written as:
Watermark (W) = F (W or C, WE, K)

Figure 1: Watermark (a) Embedding process and (b) Extraction process [9]

Depending on the application requirements, the watermarking system can be blind, semi-blind or non- blind [9]. The
robustness of the semi-blind watermarking system is poor than the other two system.

Existing Transform based Watermarking Techniques Using Machine Learning method

Review of image watermarking using noticeable machine learning techniques is carried out and presented below.
Vatsa et al. [12] proposed biometric based image watermarking algorithm where face image is embedded in the fingerprint.
The embedding watermarking method based on Discrete Wavelet Transform (DWT) and Support Vector Machine (SVM).
The experimental result is shown that the method is robust and the face image is resilient to geometric and frequency attacks.
The integration of SVM improved the face recognition by 10%. Yen et al. [13] proposed a digital watermarking technique
using spatial domain based on support vector machine. The watermarking technique uses only 128bits in training SVMs. To
embed the watermark bits, the proposed scheme modifies blue channels of the central and surrounding pixels at same time.
Watermarks are embedded in spatial domain and extracted directly from a watermarked image without the requirement of
original image. The experimental result shows that the proposed scheme provides high PSNR of a watermarked images and
low extraction error rate.
Jianzhen et al.[14] proposed a RST( Rotation, Scaling and translation) invariant watermarking technique utilizing SVM and
image moments for synchronization. In watermarking technique to estimate RST transform parameters SVM is utilized to
learn the image geometric pattern represented by six combined low order image moments. The experimental result shows that
scheme can resist JPEG compression, noise and geometric attacks. Lei Li et al.[15], proposed a image watermarking scheme
using spatial domain based on Fussy Support Vector Machine (FSVM). In the embedding process, the 8 * 8 block of the
cover image is divided into sub-block of the texture features as input vectors using support vector machine. The image sub-
 Digital Image Watermarking using Machine Learning Techniques: A Technical Review 85

block is divided into a weak texture and a strong texture. The strong texture information is embedded into the cover image.
The method has been shown that the robustness of the FSVM based method is better than SVM based method against
important attacks. Jain et al.[16], proposed a watermarking algorithm based on support vector machine using color image. In
the embedding process, the watermark is embedded into the discrete wavelet domain of the original image and extracted by
training support vector machine. In addition, the method is using momentum coefficient to reduce the error and increase the
rate of the learning. The experimental results have been shown that the method is imperceptible against signal processing
attacks. However, the value of PSNR is below than 27 dB for most of the attacks.
Ramamurthy et al.[17] proposed a robust digital image watermarking scheme using neural network and fuzzy logic approach.
The approach shows comparison to embed watermark into host image using quantization in DWT domain based on BPNN
(Back Propagation Neural Network) and DFIS (Dynamic Fuzzy Inference System). The experimental result shows the
watermarking technique is robust and imperceptible to the attacks. B.Jagadeesh et al.[18], proposed a robust and blind Image
watermarking algorithm for copyright protection of images. The embedding watermarking method is used by DWT based on
the support vector machine. The experimental result shows that method is secured and robust for various attacks. However,
the value of NC and PSNR is less than 0.9711 and 35dB for most of the attacks. Vafaei et al. [19] proposed a robust blind
watermarking method. The watermarking method uses the Neural Networks in Discrete Wavelet Transform domain. The
neural networking technique is used to maximize the strength of watermark image. The experimental result shows that
method is robust and imperceptible to various attacks. B.Jagadeesh et al.[20] proposed a novel image watermarking method
in discrete wavelet transform domain using support vector machine. The embedding watermarking method is used to extract
the watermark from the watermarked image even after different image processing attacks. The experimental result shows that
the given algorithm is secure and robust to different image attacks. Yahya et al. [21] proposed a model for information
security using stego SVM classification. The embedding technique uses LSB in image steganography that hides data behind a
cover-image in a spatial and discrete cosine transform (DCT) domain. The technique proposed a new model that utilizes
Human Visual System (HVS) and embedding technique through shifted LSB called Stego SVM- Shifted LSB in DCT
domain to preserve the imperceptibility and increase the robustness of stego-images.
Zhang et al.[22] proposed a technique of image watermarking capacity using neural network. The watermarking technique is
used for hiding the information in the form of images and watermarking is used as a form of communication. The
experimental result shows that the attraction basin of associative memory decided watermarking capacity. Shi et al.[23]
proposed a new color watermark embedding technique with circulation, based on non-overlapping SVD for hiding important
information in images. In the watermarking method cover image is decomposed into small watermarks and then watermark is
embedded into one single block with circulation. The experimental result shows that scheme is robust against different image
attacks. Thai et al.[24] proposed a technique for image classification using support vector machine and artificial neural
network. In the technique, image is divided into sub images and each sub image is classified into the responsive class by
ANN then SVM compiled all the classified result of ANN. The Experimental result shows the feasibility of the technique.
From the above, it is clear that image watermarking techniques in transform domain have been found to give high robustness,
imperceptibility, capacity and security. The digital watermarks are potentially useful in various applications including
ownership assertion, fingerprinting, copy prevention or control, telemedicine, e-commerce, e-governance, media forensics,
and artificial intelligence and healthcare etc. Digital cinema is also considered as a practical application, where the
information can be embedded as a watermark in every frame. Digital image watermarking is the most effective solution in
these areas and its use to protect the information is increasingly exponentially day by day.
The transform based watermarking schemes can be further explored for video and audio watermarking.The state of art in
transform domain based image watermarking using machine learning language as available in the literature is given below.

1. The method proposed in [12, 14, 16, 18, 20] based on the combination of DWT and SVM to achieve high
robustness, imperceptibility and good quality of watermarked image. In [12] biometric is used to embed the face
image with the fingerprint in order to achieve security also.
2. The method proposed in [13, 15] based on the combination of spatial domain with machine learning techniques
(FSVM and SVM) to lower the attacks on digital data. In the technique [15] shown that the robustness of the FSVM
based method is better than SVM based method against important attacks.
3. The method proposed in [17, 19] based on the combination of DWT with Neural Network uses to increase the
strength of watermark image. In the technique [19] host image was decomposed into wavelet bits embedded in the
sub-band coefficients.
4. The method proposed in [21] based on the combination of DCT and SVM uses LSB in image steganography that
hides data behind a cover-image. The technique is robust to various attacks.
86 Advances in Engineering and Technology

Table 1: Wavelet based watermarking using machine learning model

Results
Sn. Authors, Year Technique used Watermark Type, Size
(Maximum value)dB

Value of verification of face and

Vatsa et al. [12], Fingerprint,FaceImage (512*512)
1 DWT, SVM fingerprint ranges from 0-1 and
2005 / Extracted Face Image
shows accuracy is improved by 10%.
Color image of Lena, Baboon ,
Spatial domain
2. Yen et al.[13], 2006 Monkey , House (512*512)/ Binary PSNR= 45.536
method , SVM
image of Rose (64*64)
Jianzhen et al.[14], Grey scale image of lena (512*512)/
3. DWT, SVM PSNR= 39.12
2009 logo image
Spatial Domain, Grey scale image of Cameraman
4. Li et al.[15], 2010 PSNR=20.866,NC=0.986
SVM, FSVM (256*256)/ logo image
Lena, Baboon (256*256)/ logo color
5. Jain et al.[16],2011 DWT, SVM PSNR=43.499
image (32*32)
DWT , Neural
Ramamurthy et al. Office_4 cover image (512*512)/
6. Network, Fuzzy PSNR=51.3593,NC=0.9965
[17], 2012 Barbara grey scale image (64*64)
logic
B.Jagadeesh et Lena , Goldhill and pepper
7. DWT, SVM PSNR=45.15,NC=1
al.[18], 2013 (512*512)/ logo(64*64)
Grey scale image of Lena , Baboon,
Vafaei et al.[19], DWT, Neural
8. Airplane, Barbara(512*512)/logo PSNR=48.25, NC=0.99
2013 Network
watermark Binary image (8*8)
Grey scale images of Goldhill,
B.Jagadeesh et
9. DWT ,SVM Mandrill and Peppers (512*512) PSNR=45.94,NC=0.961
al.[20],2014
/logo image (64*64)
Yahya et al.[21], Lena, Baboon of size 1024 bits/ logo
10. DCT, SVM PSNR=49.86, NC=1.0
2015 image

Conclusion
Digital watermarking can be used in any area where there is need to protect multimedia data for the purpose of identification,
annotation and copyright with guaranteed security and authenticity. This paper has presented detailed review of
watermarking techniques based on spatial and frequency domain using machine learning techniques. The different
watermarking techniques are use to learn about the parameters based on the performance along with the state of art. The use
of machine learning method with watermarking is the blooming area that can further be explored.

Acknowledgement
The Author’s are sincerely thankful to the potential/ anonymous reviewer’s for their critical comments and suggestions to
improve the quality of the paper.

References
[1] Cheddad, J. Condell, K. Curran and P. McKevitt, “Digital Image Steganography : Survey and Analyses of Current Methods”, Signal
Processing, vol.90, no.3, 2010.
[2] Babak Mahdian and Stanislav Saic, “A bibliography on blind methods for identifying image forgery”, Signal Processing: Image
Communication 25, 2010.
[3] Huang-Chi Chen, Yu-Wen Chang and Rey-Chue Hwang, “A Watermarking Technique based on the Frequency Domain ”, Journal of
Multimedia,vol.7, no.1, 2012.
[4] Vaishali S. Jabade and Dr. Sachin R. Gengaje, “Literature Review of Wavelet Based Digital Image Watermarking Techniques”,
International Journal of Computer Applications, vol.31, no.1, 2011.
[5] Prabhishek Singh and R S Chadha, “A Survey of Digital Watermarking Techniques, Applications and Attacks”, International Journal
of Engineering and Innovative Technology, vol.2, no.9, 2013.
[6] Kaiser J.Giri, Mushtaq Ahmad Peer and P. Nagabhushan, “A Robust Color Image Watermarking Scheme Using Discrete Wavelet
Transform”, International Journal of Image Graphics,2015.
[7] Henri Bruno Razafindradina and Attoumani Mohamed Karim, “Blind and Robust Image Watermarking based on Wavelet and Edge
Insertion”, International Journal on Cryptography and Information Security, Vol.3, no.3, 2013
 Digital Image Watermarking using Machine Learning Techniques: A Technical Review 87

[8] Harshita rawat, Ashwani kumar and Satendra kumar, “Robust Digital Image Watermarking Scheme for Copyright Protection”,
International Journal of Computer Applications, vol.75, no.18, 2013.
[9] Amit Kumar Singh, Mayank Dave and Anand Mohan, “Wavelet Based Image Watermarking: Futuristic Concepts in Information
Security”, Proceedings of the National Academy of Sciences, India Section A: Physical Sciences, vol. 84, Issue 3, pp. 345-359, 2014.
[10] Simon Tong and Edward Chang, “support Vector Machine Active Learning for Image Retrieval”, ACM, 2001
[11] Erkan Yavuz and Ziya Telatar, “Improved SVD-DWT Based Digital Image Watermarking Against Watermark Ambiguity”, 2007.
[12] Mayank Vatsa, Richa Singh and Afzal Noore, “Improving biometric recognition accuracy and robustness using DWT and SVM
watermarking”, IEICE Electronics Express, vol.2, no.12, pp.362-367, 2005.
[13] Shwu-Huey Yen and Chia-Jen Wang, “SVM Based watermarking technique”, Tamkang Journal of Science and Engineering, vol. 9,
no.2, 2006.
[14] Wu Jianzhen, “A RST Invariant Watermarking Scheme Utilizing Support Vector Machine and Image Moments for Synchronization”,
IEEE International Conference on Information Assurance and Security, 2009.
[15] Lei Li, Wen-Yan Ding and Jin-Yan Li, “A Novel Robustness Image Watermarking Scheme Based on Fuzzy Support Vector
Machine”, IEEE Pattern Recognition and Intelligence System, 2010.
[16] Yogendra Kumar Jain and Saurabh Tiwari, “An Enhanced Digital Watermarking for Color Image using SVM”, International Journal
of Computer Science and Information Technology, vol.2, no.5, 2011.
[17] Nallagarla Ramamurthy and Dr.S.Varadarajan, “Robust Digital image watermarking scheme with Neural Network and fuzzy logic
approach”, International Journal of Emerging Technology and Advanced Engineering, vol.2, no.9, 2012.
[18] B.Jagadeesh, P.Rajesh Kumar and P.Chenna Reddy, “Robust Digital Image Watermarking Scheme in Discrete Wavelet Transform
Domain using Support Vector Machine”, International Journal of Computer Applications, vol.73, no.14, 2013.
[19] M. Vafaei, H. Mahdavi Nasab and H. Pourghassem, “A new blind Robust Watermarking method based on Neural Networks in
Wavelet Transform Domain”, World Applied Science Journal, vol.22, no.11, 2013.
[20] B.Jagadeesh, P.Rajesh Kumar and P.Chenna Reddy, “Digital Image Watermark Extraction in Discrete Wavelet Transform Domain
using Support Vector Machine”, ACEEE International Journal of Recent Trends in Engineering and Technology , vol.11, 2014.
[21] Saadiah Yahya, Hanizan Shaker Hussain and Fakariah Hani M. Ali, “DCT Domain Stega SVM- shifted LSB Model for highly
Imperceptible and robust cover image”, International Conference on Computing and Informatics, vol. 43, 2015.
[22] Fan Zhang and Hongbin Zhang, “Image watermarking capacity using Neural Network”, IEEE International Conference of Web
Intelligence, 2004.
[23] Zhenghao Shi and Lifeng He, “Application of neural network in medical Image Processing”, International Symposium on Networking
and Network Security, 2010.
[24] Le Hoang Thai, Tran Son Hai and Nguyen Thanh Thuy, “Image Classification using Support vector machine and Artificial neural
networks”, International Journal of Information Technology and Computer Science, vol..32, no.38, 2012.