Professional Documents
Culture Documents
Hiding
K. Ramani #1, E. V. Prasad @2, S. Varadarajan *3, A. Subramanyam&4
#
CSED, SVN Engineering College, Tirupathi, India
1
ramanidileep@yahoo.com
@
Principal, JNTU College of Engineering, Kakinada, India
2
drevprasad@yahoo.com
*
EED, SVU College of Engineering, Tirupathi, India
3
varadasouri@gmail.com
&
CSED, AITS, Rajampet, AP, India
4
smarige@gmail.com
Abstract— Wavelet techniques can be successfully applied in detection of modifications to the data. By embedding
various signal and image processing applications, namely in watermark in commercial advertisements, an automated
image de-noising, segmentation, classification, motion monitoring system can verify whether the advertisements are
estimation and copy right protection. The proliferation of broadcasted as contracted or not. Indexing of video mail,
digitized media due to the rapid growth of networked
multimedia systems has created an urgent need for copyright
indexing of movies and news items, where markers and
enforcement technologies that can protect copyright ownership comments can be inserted that can be used by search engine.
of multimedia objects. Digital image watermarking is one such In the case of Medical Safety it is for embedding the date and
technology that has been developed to protect digital images patient’s name in medical images which could be useful for
from illegal manipulations. In this paper, a robust and safety measurement.
imperceptible watermarking scheme for copy right protection is Invisible watermarking requires a reasonable robustness
proposed. The method is based on decomposing an image using against attacks such as compression, as well as little or no
the Discrete Wavelet Transform, and then embedding locations degradation in subjective and objective image quality.
are generated from the low frequency sub-band by using secrete Watermarking in the frequency domain is more robust than
sort to improve the embedding intensity. From the experimental
results, the proposed scheme provides not only good quality,
watermarking in the spatial domain [1], because the
but also good robustness against external attacks, such as watermark information can spread out to the entire image [2].
rotation, compression, cropping, noise and scaling. In the viewpoint of frequency, the high frequency area should
be avoided for robustness.
Keywords—Discrete Wavelet Transform, Watermarking, In order to have efficient watermarking, one should
Copyright Protection consider three main criteria: robustness against attacks such as
compression, security and imperceptibility which is how
I. INTRODUCTION much the watermarked image is similar to the original one [3].
With rapid growth of multimedia technology, digital Early watermarking schemes that were introduced in the
watermarking for multimedia has become one of the widely spatial domain, where the watermark is added by modifying
used copyright protection methods. Digital watermarking is pixel values of the host image. Examples of such techniques
the process of conveying information by imperceptibly are given in [3], [4] and [5].
embedding it into the digital media. The purpose of Given its suitability to model the Human Visual
embedding such information depends on the application and System(HVS) behaviour, the Discrete Wave Transform(DWT)
the needs of owner/user of the digital media. has gained interest among watermarking researchers, as it is
Main applications of watermarking include Copyright witnessed by the number of algorithms following this
Protection, Fingerprinting, Copy protection, Image approach that have been proposed over the last few years.
authentication, Broadcast monitoring, Indexing and Medical There are number of steganography techniques based on least
safety areas. The objective of Copyright Protection is to significant bits (LSB) for embedding message, however these
embed information about the source/owner of the digital are highly insecure, because it can be easily detectable that the
media. Fingerprinting objective is to convey information data is hidden in cover object [6]. This problem can be
about the recipient of digital media in order to identify every avoidable by hiding data using Bit-Plane Complexity
single distributed copy of the media. Watermarking can be Segmentation (BPCS) steganography, in which image is
used to control data copying the digital media in case the decomposed into bit planes of sub band wavelet coefficients
watermark embedded in the media indicates that media is obtained by using Integer Wavelet Transform (IWT). At lower
copy protected. The objective of Image authentication is to levels of wavelet decomposition changes made in the cover
check the authenticity of the digital media. This requires the image can not identified by human naked eye. BPCS takes the
59
IV. PERFORMANCE EVALUATION
CA4 CH4 CA4 0 The performance of the proposed method is evaluated with
CA3 CA3' an 8-bit gray scale image of size 512X512 as the host image
CV4 CD4 0 0 and binary image with 64X64 pixels as the watermark image.
A. Performance Evaluation Metrics:
Step 3: The position of attaching the secrete information is Robustness is a measure of the immunity of the watermark
calculated with the formula given in Eq. (2) against attempts to remove or degrade it, intentionally or
unintentionally, by different types of digital image processing
[15], like image compression, guassian noise and Image
′
CH 3 " = CH 3 + (k1 * w(i, j ) ⊕ k 2 * s(i, j ))..............(2) cropping. The Gaussian noise is a watermark degrade attack,
where k1 and k2 are embedding intensities. The watermark Joint Photographic Experts Group(JPEG) compression is a
image is disturbed by a random seed and embeds it into watermark removal attack and cropping is a watermark
CH3 '. dispositioning geometrical attack. The similarity between the
original watermark and the watermark extracted from the
Step 4: An inverse wavelet transform is used to get the attacked image using the correlation factor, ρ is given by
watermarked image Eq.(5).
B. Extraction Algorithm
N N N
The proposed method is a non-blind watermarking method, ρ = ∑ϖ i ω i ′ / ∑ϖ i ∑ϖ i
′ ………………..(5)
hence the original host image is needed for extraction. The i =1 i =1 i =1
watermark is extracted as follows: where N is the number of pixels in watermark, ω and ω' are
the original and watermarks respectively. The correlation
Step 1: Both of the original host image and the watermarked factor ρ may take values between 0 to1. In general, a
image are decomposed by 3-level wavelet transform. correlation coefficient of about 0.75 or above is considered
acceptable.
Step 2: Getting secret sort s(i,j) from the original image.
Then, sub-bands CH3' and CH3 are generated by V. RESULTS
Eq. (1). The experimental results of the proposed scheme are as
follows:
Step 3: The embedded watermark image is retrieved by the The original host image, original watermark image,
following equation: watermarked image and extracted watermark image are shown
′
w′(i, j ) = {[CH 3 " (i, j ) − CH 3 (i, j )] / k1 ⊕ [k 2 * s (i, j )]′} ……(3) in Fig. 3.
60
watermark is recognizable. Beyond 300 it is some what
difficult to identify the extracted watermark.
(a) (b)
(a) (b)
(c) (d)
Fig. 3: (a) Original image (b) Original watermark (c)Watermarked image (c) (d)
(d)Extracted watermark Fig. 6. Extraction from rotation at (b) 50 and (d) 300
Compared to method [7] the PSNR between original image D. Salt and Pepper Noise:
and watermarked image is better in proposed method at level Salt and pepper noise is added to the watermarked image
3. with different noise densities of 0.2, 0.4, 0.6, 0.8, 0.9 and 1.0.
A. .Compression: Even for 1.0 noise density the extracted watermark is
recognizable.
The performance of the compressed images evaluated for
20%, 30%, 40%, 50%, 60% and 70% qualities on the
watermarked image sequentially. The results of extracted are
shown in Fig. 4. All extracted watermarks are recognizable.
Table 1 shows the results of different qualities of compression (a) (b)
rates. Even for 20% compression also the watermark is Fig. 7.Extraction of watermark from salt & pepper noise at noise densities
extractable. For more than 85 % the extracted watermark is (a) 0.2 and (b) 0.4
almost similar to watermark extracted without any
compression. E. Cropping:
The watermarked image is cropped to different block sizes,
150, 200, 250, 300, 350 and 400. Even when watermarked
image cropped to a block size of 150, still extracted
watermark is recognizable.
(a) (b)
Fig.4: Extraction from Compression at (a) 20% and (b) 70%
B. Scaling:
The watermarked image is scaled by 20%, 40%, 60%, 80%,
100% and 150%. In Fig. 5, the extracted watermarks are given,
which shows that even the watermarked image is scaled by
different values still extracted watermark is recognizable.
(a) (b)
(a) (b)
Fig. 5. Extraction from Resizing at (a) 20% and (b) 150%
C. Rotation:
The watermarked image is scaled by 10, 50, 100 , 150 , 200 (c) (d)
and 300. In Fig. 6, the extracted watermarks are shown for 50 Fig.8 Watermark Extraction from cropped image of (b) 20% and (d) 30%
and 300 rotation of watermarked image, still the extracted watermarked image.
61
The PSNR and correlation values for extracted watermark 2) Scaling: Fig.11 and Fig.12 show the results of
are shown in Table 1, Table 2, Table 3, Table 4 & Table 5 for extracting watermarks from scaled watermarked images.
compression, scaling, rotation, salt pepper noise and cropping When watermark image is scaled to 20% and 40% DCT and
respectively. DWT techniques failed to extract the watermark. The
proposed method is better results for all scaling values.
Table 1: PSNR and Correlation factor of the compressed Images
3) Rotation: Fig 13 and 14 show the results of extracted
compression 20% 30% 40% 50% 60% 70% watermarks from rotated watermarked images. DCT method
rate
is having better PSNR and correlation factor values for
PSNR(dB) different angles of rotation compared to DWT and DWT with
51.3 51.45 51.57 51.59 51.61 52.1 secrete sorting methods. But the proposed method is having
Correlation 0.903 0.9085 0.912 0.9233 0.9269 0.9464
better results than DWT method.
factor
4) Cropping: As shown Fig.15 and Fig. 16, the proposed
Table 2: PSNR and Correlation Factor of the scaled Images method suffers much less from cropping attack compared
Scaling 20% 40% 60% 80% 100% 150%
DCT and DWT methods. But, DWT is having better
performance than DCT for this attack.
PSNR 41.04 43.51 44.17 45.18 48.92 49.98
(dB) 5) Salt and Pepper Noise: The robustness against salt and
Correlati- 0.4324 0.4625 0.5062 0.6904 0.8007 0.8123 pepper noise is tested for all three methods. Again the
on
factor proposed method is having good correlation value for
different noise densities. However the difference in PSNR
Table 3: PSNR and Correlation Factor of the rotated Images values is less compared to other two methods at 0.6, 0.8 and
Rotation 10 50 100 150 200 300 1.0 noise density values.
angle
(degrees)
PSNR 49.73 45.11 42.34 41.21 38.83 34.60
(dB)
Correlation 0. 81 0. 77 0. 71 0.70 0.69 0. 68
factor
Table 4: PSNR and Correlation Factor of the Salt & Pepper Noised images
Noise 0.2 0.4 0.6 0.8 0.9 1.0
density
62
Fig. 14 Correlation factor of different watermarking methods based on
Fig. 11 Correlation factor of different watermarking methods based on rotation attack
scaling attacks
63
REFERENCES
[1] W. Bender, D. Grahul, N. Morimoto, and A. Lu, “Techniques for Data
Hiding”, IBM Systems Journal, vol. 35, no. 3-4, 1996, IBM, USA, pp.
313-336.
[2] I. J. Cox, J. Killian, T. Legthton, and T. Shamoon, “Secure Spread
Spectrum for Multimedia,” IEEE Trans. On Image Processing, vol. 6,
no.12, 1997, pp. 1673-1687.
[3] R. G. Schyndel, A. Z. Tirkel, N. R. Mee, and C. F. Osborne, “A digital
watermark,” IEEE Int. Conf. Image Processing. Vol.2, pp. 86-90,Texas,
U.S.A, November, 1994.
[4] R. B. Wolfgang, and E. J. Delp, “A watermark for digital images,”
School of Electrical Engineering, Purdue University, USA, Tech. Rep.,
1995.
[5] S. P. Maity, and M. K. Kundur, “Robust and blind spatial
watermarking in digital image,” Dept. of Electronics and Telecomm.,
India, Tech. Rep., 2001.
Fig.17 Correlation factor of different watermarking methods based on salt and [6] K. Ramani, Dr. E. V. Prasad, and Dr. S. Varadarajan, “Steganography
pepper noise attack Using BPCS to the Integer Wavelet Transformed Image”, International
Journal of Computer Science and Network Security, Vol.7 No 7, July
2007, pp. 293-301.
[7] Ramani, K, Prasad, E.V and Varadarajan, S, “DWT Based
Watermarking For Biometric Data”, Indian Journal of Information
Science and Technology, Vol. 3, November 2007, pp. 35-41.
[8] S. Joo, Y. Suh, J. Shin, H. Kikuchi, and S. J. Cho., “A new robust
watermark embedding into wavelet DC components,” ETRI Journal,
24, 2002, pp. 401-404.
[9] Ejima. M and A. Myazaki,2001. “On the evaluation of performance of
digital watermarking in the frequency domain”, in proc. Of the IEEE
Int. Conf. on Image Processing. 2:546-549.
[10] Voloshynovskiy. S., S. Pereira and T. Pun. 2001. “Attacks on Digital
watermarks: classification, Estimation-Based attacks and Benchmarks”,
Comm, Magazine.,39(8):118-126.
[11] N.F. Johnson, S.C. Katezenbeisser, “A Survey of Steganographic
Techniques” in Information Techniques for Steganography and Digital
Watermarking, S.C. Katzenbeisser et al., Eds. Northwood, MA: Artec
House, Dec. 1999, pp 43-75
Fig.18 PSNR of different watermarking methods based on salt and pepper [12] G. Langelaar, I. Setyawan, R.L. Lagendijk, “Watermarking Digital
noise attack Image and Video Data”, in IEEE Signal Processing Magazine, Vol 17,
pp 20-43, September 2000.
VI. CONCLUSION [13] Z. Yuehua, C. Guixian and D. Yunhai, “An image watermarking
The spatial domain watermarking scheme is vulnerable to algorithm based on discrete cosine transform block classifying,” ACM
Int. Conf. October, 2004, pp. 234-235, Shanghai, P. R. China, 2004.
image processing attacks, hence, the proposed scheme [14] J. L. Liu, D.C. Lou, M. C. Chang, & H. K. Tso, “A robust
modifies the original image in transform domain and embeds watermarking scheme using self-reference image,” Computer
a watermark in the difference values to overcome the weak Standards & Interfaces, 28, 2005, pp. 356-367.
robustness problem in spatial domain, and methods in [6] and [15] A. A. Reddy, and B. N. Chatterji, “ A new wavelet based logo-
watermarking scheme,” Pattern Recognition Letters, 26, 2005, pp.
[7]. 1019-1027
The time taken by the proposed scheme for embedding and [16] A. Piva, M. Barni, F. Bartolini, and V. Cappellinni, “DCT- watermark
extraction of watermarking is several times less than using recovering without restoring to uncorrupted original image,” in
Shoemaker’s method for embedding and extraction of JPEG International Conference on Image Processing, Vol .III, pp.520-
523,1997.
images and also it is more robust. The proposed method has
simple steps for watermarking and extraction and it is resistant
against many image processing attacks. But, the proposed
method is non-blind, which requires the original image for
extraction of the watermark. The proposed method is
compared with DCT [16] and DWT [7] methods. For JPEG
compression and rotation attacks the DCT is showing better
performance compared to proposed and DWT method. For
cropping, scaling and noise affects the proposed method is
more robust than other two methods.
Further research can be done in extending this work to
combined Discrete Cosine Transform-Discrete Wavelet
Transform, Dual Tree Complex Wavelet Transform and
Combined Discrete Cosine Transform-Dual Tree Complex
Wavelet Transform methods.
64