International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

Edge based Robust Image Watermarking Scheme for

Copyright Protection of Images

M.Blessy Sharon S.J.Saritha

Computer Science and Engineering Computer Science and Engineering
JNTUCEP, Pulivendula, A.P, India JNTUCEP, Pulivendula, A.P, India
[email protected] [email protected]

Abstract— Digital image watermarking is one of the dominant knows it. This technique gives security to images. Some
means to catch out the unauthorized use of copyrighted images. It mathematical computations are compulsory to extract the
inserts secret information (watermark) into the host image that invisible watermark image. Invisible digital image watermarks
helps in finding the rightful ownership of image. Watermarking is are more secure than visible image watermarks [7, 8]. According
one of the solutions to prevent unlicensed use of digital media. This to some applications, the image watermarking techniques are
solution can help the protection of ownership rights by embedding categorized into three types: semi fragile, fragile and robust
the copyright information in the intended media. There are various approaches. The semi fragile watermarking is used to breakdown
algorithms to embedding a watermark image into digital image under every modification that a user-quantified threshold to be
(host or original). The proposed work is skilled with insertion of surpass, but fragile image watermarking is designed to know any
watermark images into host or original image in the edge positions. illegal changes on digital images. Robust Image watermarking is
Ultimately, produce high invisibility while embedding into reference primarily used for copyright protection of images under rightful
image and extraction of watermark might be without any ownership [1, 2, 4].Security involves encryption method[21]. In
distortions. This type of insertion makes easy and free from the false
our work, we mainly focus on the robust image watermarking .
In technology, spatial and transform domain watermarking are
positive errors. For finding the quality of the extracted watermark
two main categories of Digital Image Watermarking. Spatial
image and efficiency of algorithm, we used a well-known quality
domain watermarking is altering pixels of image directly in the
measure functions for images, such as PSNR (peak signal to noise given watermark. The advantages of this are easy
ration) and Jaccard function. The results of proposed work give implementation and low complexity.Cross-domain
high invisibility of the inserted watermark image, with very less transformation requires an extra computation for this method.
distortions of the extracted watermark image. Transform domain methods are widespread because of their
Keywords— Watermarking, Edge detection, Big Data, Stenography ability to abusing human visual perception and signal
characteristics in digital image watermarking. Commonly used
transforms comprise discrete cosine-transform (DCT), singular
I. INTRODUCTION value decomposition (SVD), discrete Fourier transforms (DFT),
and discrete wavelet transforms (DWT).
Now-a-days sharing of images is very easy due to technology.
one image worth thousands of words. Every day 250 billions of
images are upload and 350 million of images are downloaded
from facebook. In all these uploading and downloading users II. LITERATURE SURVEY
need security for their images. There are some advanced X.-y. Wang et al. (2016) proposed a scheme depends on local
technologies that make the privacy and security of our personal PHT (polar harmonic transform) to robust image watermarking
or confidential images. To change the ownership of images, for digital images. Advantages of this approach are: (1) the
many advanced techniques are existed in Image processing. So, probability density gradient was used for extracting uniform
providing of protection and security of images became feature points for an image by the enhanced speed up robust
compulsory. Digital Image Watermarking is one of the effective feature detector, (2) the local feature points for affine invariant
tool, that protects the confidentiality of Images. This method is are built adaptively as said by the difference of native probability
used to insert images in original image, so that there will not be density, and (3) to insert a watermark image into another digital
any visible changes in the original image. Digital watermarking image, used an effective and new two dimensional (2D)
is characterized as visible and invisible [4, 5, 9]. This type of transform and PHT (Polar Harmonic Transform). Experiments
visible image watermarking can be visible to normal eye without are conceded for testing out on a digital image with set of
any techniques but this digital image watermarking can be ruined hundred images are gathered from Internet, and the initial results
quickly. Scheme to provide security for images for rightful showed that, in the method he proposed for watermarking is
ownership but in invisible digital image watermarking technique, robust and imperceptible to regular image processing techniques
watermark is not visible to normal eye, only the rightful person such as JPEG compression, noise adding and filtering. In this,

978-1-5386-1887-5/17/$31.00 ©2017 IEEE

3565
 International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

experimental results have exposed that the proposed A. Limitations

watermarking approach has combined those stimulating Images can be extracted from face book in online mode only but
distortions of geometric, like column or row removal, aspect not in offline. So net connection is compulsory. In this operation
ratio changing, a group of attack and projective transformations. logo image or original image must be 4 times smaller than
Also, watermarking can protect from some of the general image reference. If not, some of the logo image will skip while
processing techniques. This approach are associated to its lesser encryption. While decryption process we must know the size of
image watermark capacity, it is one of the drawback. And, it took the logo image. If any attacks will occur in the Embedded Image,
higher time-complexity for digital image feature point detector it is very difficult to get back the watermarked image or original
[1], because it uses the probability density gradient. image.
Y. AL-Nabhani et al. (2016) proposed an enhanced technique IV. PROPOSED WORK
for making watermarked digital images with high invisibility. To In the proposed work, it mainly contains three levels:
combine a binary watermark image in selected coefficient blocks
decomposing original image, embedding and extraction process.
of a digital image, they have developed Discrete Wavelet
Binary watermark image approach is used in embedding process.
Transform (DWT) with a Haar filter. Here to extract the
This technique is very very easy and also very difficult to guess
watermarked image without using the source image, a PNN
the key to extract the logo.
(Probabilistic Neural Network) approach is used. They also
evaluated the performance of the algorithm and the clarity of the A. Process of Watermark embedding
extracted watermark images, using digital image quality The watermark embedding process is shown in Fig.1. This
measurement function, such as NCC (normalized cross process contains the following levels: identifying the edge pixel,
correlation) and PSNR. Authors are conducted different conversion of edge pixels to binary, embedding the logo image,
experiments to confirm the robustness of the watermarked reconstruct image and test watermarked image.
image. All Tests intricate the use of several common attacks In proposed method, edge detection algorithm is used to insert
likes median filter, rotation, JPEG compression, Gaussian noise the original image into reference image. An edge in an image is a
and cropping against the watermark. The watermark images were local change in the image intensity, commonly associated with a
effectively extracted in all tests done on embedded images even discontinuity in the first derivative of the image intensity. There
attacks are happen, but the qualities of the extracted watermark are many edge detection algorithms present [13, 14, 15, 16].
images are varied depending on the type of attack on the Edges of the image can be identified by using edge detection
embedded watermarked images [2]. algorithms. The coordinates of an edge point can be the numeral
I.A. Ansari et al (2016). Proposed an approach with a decent row and column indices of the pixel. The edge coordinates may
capacity to secure digital image watermarking, for inserting be the original image coordinate system, but the edge detection
watermark image, Discrete Wavelet Transform followed by filter meanwhile filtering may scale or translate image
singular components are used on the digital images. This type of coordinates from the coordinate system of the image produced.
insertion makes the watermarked image without any false An edge portion may be abstracted as a small line segment about
positive error. Here multiple scaling factors are used for insertion a pixel size. The term edge is commonly used for either edge
of watermark. These scaling factor values are identified with fragments or edge points. An edge detector will give edge set,
Artificial Bee Colony (ABC) in order to get a decent trade off and it is partitioned into two subsets: correct edges and false
among imperceptibility and robustness. Arnold transform based edges. The false edges are, the edges that do not correspond to
encryption is also used to prevent the reliable access of edges in the image, so we called it as false positives edges [12].
watermark still after the triumphant extraction [3].

III. EXISTING METHOD

LSB (Least Significant Bit) is image watermarking approach,
which work as follows. In LSB method convert original image
and reference image into gray image each pixel is 8 bits in a
range from 0 to 255. When we change the least significant bit as
its value only one pixel value will influence. In image pixel one
bit variance will not change image visualization, with in image
we can hide the data using this property. In existing method
authors have to consider two LSB bits, as they will change the
pixel value at maximum value by "3" to store extra two bits of
data. In the LSB watermarking method least significant bit of the
reference image is replaced with original image data bits [17, 18,
and 20]. Fig. 1. The process of embedding logo image into Original Image

3566
 International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

In Edge detection, gradient is a parameter to the degree of While calculating an approximation in the gradient, it's far
alternate in a function, and an photograph to be considered to be precarious that the x and y partial derivatives be computed at
an array of samples of positive non-stop characteristic of exactly the equal role in area. However, the use of the above
photograph intensity. With the aid of the usage of a discrete approximations, ex is sincerely the approximation to the
approximation to the gradient we analogy, appropriate gradient at the interpolated point [i, j +1/2] and Ey at [i +1/2, j]
modifications within the grey values in an image may be for this reason, 2 x 2 first differences, as opposed to 2 x 1 and 1
detected. The two-dimensional equivalent of the primary spin-off x 2 mask, are time and again used for the x and y partial
gradient vector is derivatives.
†ˆ C. Process of Extracting Watermarking
୶ †š Process of extraction of watermarking logo image from reference
ሾˆሺšǡ ›ሻሿ ൌ ቆ ቇ ൌ ൮ †ˆ ൲ (1) image, we require original host image. The algorithm for the
୷
†› process of extracting watermarking image is as follows

There are two houses associated with the gradient: • Algorithm2: Extract Watermark Image from Reference
(1) The vector E[j(x, y)] factors in the manner of the maximum Image
rate of boom of the feature j(x, y), and
1. while(i<=(length(mat1))){ //mat1
(2) The value of the gradient, given via
is the embedded image
(2) 2. k=1
ሾˆሺšǡ ›ሻሿ ൌ  ට୶ଶ ൅ ୷ଶ
3. y1<-c(0,0,0,0,0,0,0,0)
4. while(k<=4&&i<=length(mat1)){
Equation (2) is the most rate of increase of (x, y) in step with unit
distance inside the direction. It is process, to approximate the 5. if(mat7[i]!=0){
gradient via entire values: 6. if(k==1){
ሾˆሺšǡ ›ሻሿ ൎ  ȁ୶ ȁ ൅ ห୷ ห (3) 7. z1=i
8. x1=pbin(as.integer(mat1[z1]))
Or 9. y1[2]<-x1[8]
(4)
ሾˆሺšǡ ›ሻሿ  ൎ ƒšሺȁ୶ ȁǡ ห୷ หሻ 10. y1[1]<-x1[7]
11. }else if(k==2){
By analyzing the vector, we define the direction of the gradient: 12. z2=i
୷ (5) 13. x2=pbin(as.integer(mat1[z2]))
Ƚሺšǡ ›ሻ ൌ  –ƒିଵ ൬ ൰
୶ 14. y1[4]<-x2[8]
15. y1[3]<-x2[7]
B. Numerical Approximation 16. }else if(k==3){
For images, the derivatives in equation (1) are approximated via 17. z3=i
variations. The best gradient approximation is
18. x3=pbin(as.integer(mat1[z3]))
‫ܧ‬௫ ؆ ݂ሾ݅Ǥ ݆ ൅ ͳሿ െ ݂ሾ݅ǡ ݆ሿ (6) 19. y1[6]<-x3[8]
(7) 20. y1[5]<-x3[7]
‫ܧ‬௬  ؆ ݂ሾ݅ǡ ݆ሿ െ ݂ሾ݅ ൅ ͳǡ ݆ሿ
21. }
Here j to the x direction and i to the negative y course. These 22. else if(k==4){
will be carried out with simple convolution masks as exposed 23. z4=i
beneath
24. x4=pbin(as.integer(mat1[z4]))
‫ܧ‬௫ ൌ  ሺെͳͳሻ (8)
25. y1[8]<-x4[8]
(9) 26. y1[7]<-x4[7]
ͳ
‫ܧ‬௬ ൌ  ൬ ൰ 27. }
െͳ

3567
 International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

28. k=k+1 ‫ܬ‬ଵଵ (13)

‫ ݀ݎܽܿܿܽܬ‬ൌ
29. } ‫ܬ‬଴ଵ ൅ ‫ܬ‬ଵ଴ ൅ ‫ܬ‬ଵଵ
30. i=i+1
The J11 means the total number of attributes where in two data
31. if(i>=length(mat1)) objects have a 1. The J10 and J01 mean the total number of
32. break(); attributes where one data object has a 1 and the other has a 0.
The all matching attributes are then divided by the all non-
33. } similar attributes plus the similar ones. A perfect matching score
34. if(p<=(m1[1]*m1[2])){ would then be a 1.
35. mat12[p]<-pdec(y1) //mat12 is A. Result of Existing Method
the extracted image The Fig. 3 shows that, three original images are watermarked
36. p=p+1 with a logo images and also the resultant Steg-image. The
result also shows that the resultant of the extracted image from
37. } the steg-image.
38. }
This is the process of extraction .

Fig. 3. Encrypted and Decrypted Output Images of Existing Method

Fig. 2. The process of Extraction of Logo Image from the Embedded Image The detailed comparisons of the three Images with Jaccard and
PSNR values are as follows:
V. RESULTS AND DISCUSSIONS
In proposed work, performance of watermarking is tested with
TABLE I. JACCARD AND PSNR VALUES OF EXISTING SYSTEM
robustness and invisibility. For comparison of the equality of
watermark embedded image with original image and extracted Image Jaccard PSNR
watermark logo image with original logo image here we are 1 21.68367 74.94915
using PSNR, Jaccard and Normalized cross correlation functions 2 7.699275 43.9036
[19]. PSNR function is used for objective image quality metrics. 3 17.66417 72.85485
Here PSNR function is used to test the robustness and invisibility
of a watermark image, which is measured as the ratio of the
number of properly matched local regions of watermarked image B. Results of proposed work
and number of original reasons of watermarked image. Suppose In this proposed work, three Fb profile pictures have taken
the size of Original image (S) and watermarked or logo image namely pic1, pic2, pic3 [11] and used in this proposed work. The
(W) is m X n and both can reach Smax is a maximum pixel size of the original images are variable pic1 and pic2 are of
value, The PSNR value can be determined as: dimensions 600 X 600, pic3 is with the dimensions of 800 X
݉ ൈ ݊ ൈ  ሺܵ௠௔௫ ሻଶ 600. A gray scaled logo image of size 87 X 85 is used as a
ܴܲܵܰ ൌ ͳͲ Ž‘‰ଵ଴ ቆ ቇ watermark . The original and watermark images are shown in
σ௠ ௡
௜ୀଵ σ௝ୀଵሺܵሺ݅ǡ ݆ሻ െ ܹሺ݅ǡ ݆ሻሻଶ
Fig 4. Fig 5 shows embedded process of the watermark logo
(12) image into three pictures.

When dealing with data objects that have binary attributes, it is

more effective to calculate similarity using a Jaccard
Coefficient. The equation to find the Jaccard coefficient is as
follows:

3568
 International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

TABLE II. THE VALUES OF THE JACCARD AND PSNR FUNCTIONS

S. Images Proposed Method LSB Method

No
Jaccard PSNR Jaccard PSNR
1 img1 0.472222 0.5984405 0.2457 0.34405
2 img2 0.68367 0.7225731 0.26367 0.63158
3 img3 0.841112 0.7654452 0.34122 0.54452

Fig. 4. Original and Watermark Images

Fig. 5. Embedded Process of a Watermark Logo Image into three Facebook Profile Pictures

C. False positive error checking

False positive error takes place when a incorrect or wrong
watermark is extracted from the embedded image. The
following image here is the extracted watermark image
without referring the source image using embedded image.
From Fig.6, we can easily observe that there is some noise in
the image, this is because if unlicensed person change the
source image by wrong image. So, this tells that this
approach is secured towards false positive error.
Fig. 6. Extracted watermarklogo ,when originalimage not provided

3569
 International Conference on Energy, Communication, Data Analytics and Soft Computing (ICECDS-2017)

steganalysis:Concepts and practice. pages 3549, 2003

VI. CONCLUSION AND FUTURE SCOPE [18] M Pavani, S Naganjaneyulu, C Nagaraju, A survey on LSB based
steganography methods; International Journal Of Engineering And
In the proposed edge based robust image watermarking Computer Science ISSN:2319-7242 Volume2 Issue 8 August, 2013 Page
technique , a logo image is inserted into edge positions of the No. 2464-2467
original embedded image. This technique gives high invisibility [19] Ismail Avcibas, Nasir Memon, and BlentSankur. Steganalysis using image
and also robustness for the logo image. The results of the edge quality metrics. IEEE transactions on Image Processing, 12:221229, 2001.
based robust image watermarking technique gives high [20] J. Sun, Z.B. Xu, and H.Y. Shum. Image super-resolution using gradient
invisibility of the embedded watermarked image, and also it profile prior. InComputer Vision and Pattern Recognition, 2008. CVPR
satisfied the extracted watermark image from the embedded '08. IEEE Conference on, pages18, 2008
image. The watermark images were extracted in all tests [21] BH krishna ,IRS reddy,S Kiran,RPK Reddy,Multiple text encryption,key
successfully, but in some attacks the quality may be various. entrenched,distributed cipher under pairing functions and transposition
ciphers
Because of the principal components insertion this scheme is
also free from the false positive errors.

References

[1] Wang Xiang-yang, Liu Yu-nan, Li Shuo, Yang Hong-ying, Niu Pan-pan,
Zhang Yan, Anew robust digital watermarking using local polar harmonic
transform, Computers andElectrical Engineering,
http://dx.doi.org/10.1016/j.compeleceng.2015.04.001(2015)
[2] Yahya AL-Nabhani, Hamid A. Jalab, Ainuddin Wahid, Ra_dahMd
Noor,Journal of King Saud University - Computer and Information
Sciences,http://dx.doi.org/10.1016/j.jksuci.2015.02.002 (2015)
[3] Irshad Ahmad Ansaria, Millie Panta, Chang WookAhn, ABC optimized
secured image watermarking scheme to find out the rightful ownership,
Journal of Optik, Elsevier, (2016)
[4] Xiang-yangWang, Yu-nan Liu, Meng-meng Han, Hong-ying Yang, Local
quaternion PHTbased robust color image watermarking algorithm, Journal
of Vis. Commun. Image R,
[5] Renato J. CintraTodor V. Cooklev, Robust image watermarking using
non-regularwavelets, Springer-Verlag London Limited (2008)
[6] YaqingNiu Matthew Kyan Sridhar Krishnan Qin Zhang, A combined just
noticeabledistortion modelguided image watermarking, Springer-Verlag
London Limited (2010)
[7] WangChun-peng,WangXing-yuan, XiaZhi-qiu, Geometrically invariant
image watermarking based on fast Radial Harmonic Fourier Moments,
Journal of Signal Processing:Image Communication (2016)
[8] Jun Lang, Zheng-guangZhang, Blind digital water marking method in the
fractionalFourier transform domain, Journal of Optics and Lasers in
Engineering, Elsevier (2013)
[9] Qingtang Su. Beijing Chen, Robust color image watermarking technique
in the spatialdomain, Journal of Soft Computing, Springer links (2017)
[10] Coltuc D, Chassery JM Very fast watermarking by reversible contrast
mapping. IEEESignal Process Lett 14(4):255258 (2007)
[11] D Praveen Kumar. C. Nagaraju, Theoretical Analysis and Security
Preserving Extraction of Public Images from Facebook, IEEE Digital
Xplore (2017)
[12] Ramesh Jain, RangacharKasturi, Brian G. Schunck Published by
McGraw-Hill, Inc.,
[13] GS Rao, C NagaRaju, LSS Reddy, EV Prasad; A Novel Fingerprints
IdentificationSystem Based on the Edge Detection; International Journal
of Computer Science andNetwork Security 8, 394-397
[14] C NagaRaju, S NagaMani, G rakesh Prasad, S Sunitha, Morphological
edge detectionalgorithm based on multi-structure elements of di_erent
directions, International Journalof Information and Communication
Technology Research 1 (1)
[15] C. NagarajuSharada Mani, G. SyamPrasad, O.Rama Devi, A Novel Edge
DetectionTechnique for Color Images, International Journal of Advanced
Engineering & Applications
[16] C. NagarajuD.VishnuVardhanBandaruBhavana,Edge Preserving Based on
WeightedGuided Image Filtering, International Journal of Science and
Research (IJSR) ISSN :2319-
7064;http://dx.doi.org/10.21275/v5i6.NOV164734
[17] R. Chandramouli, M. Kharrazi, and N. Memon. Image steganography and

3570