International Journal of Scientific & Engineering Research, Volume 1, Issue 2, November-2010 1
ISSN 2229-5518
A Survey of Frequency and Wavelet Domain
Digital Watermarking Techniques
Dhruv Arya
Abstract— Due to improvement in imaging technology and the ease with which digital content can be reproduced and manipulated there is a strong need for a digital copyright mechanism to be put in place. There is a need for authentication of the content as well as the owner. It has become easier for malicious parties to make scalable copies of copyrighted content with any compensation to the content owner. Digital Watermarking is being seen as a potential solution to this problem. Till date many different watermarking schemes have been proposed. This paper presents a comprehensive survey of the current techniques that have been developed and their effectiveness.
Index Terms— Image Retrieval, Transforms, Watermarking
—————————— • ——————————
The field of digital watermarking is rather new; indeed,
at this point many of its terms are not well defined. We define watermarking to be a process that embeds data, called a watermark into a multimedia object to help protect the owner ’s rights to that object.
A digitally watermarked image is obtained by invisibly hiding asignature information into the host im- age. The signature is recovered using an appropriate de- coding process. The challenge is to ensure that the wa- termarked image is perceptually indistinguishable from the original and that the signature be recoverable even when the watermarked image has been compressed or transformed by standard image processing operations.
This paper describes a various digital watermark
algorithms studying their strengths and weaknesses. Considering texture, luminance, corner and the edge in- formation in the image to generate a mask that makes the addition of the watermark less perceptible to the human eye. The operation of embedding and extraction of the watermark is done in both the spatial and the frequency domain thereby providing us information about the ro- bustness against common attacks including image com- pression and filtering. We use pseudo random sequences in embed the watermark. Weighted Peak Signal to Noise Ratio is used to evaluate the perceptual change between the original and the watermarked image.
It should be noted that he reason why digital watermark- ing is possible is that human vision system (HVS) is not perfect. Digital watermark utilizes the limitation of HVS
to make itself invisible, thus avoiding to degrade original
digital products, as well being hard to get identified or destroyed.
In order to be an effective watermarking system it should possess the following characterstics –
should unambiguously identify the owner. Furthermore,
the accuracy of owner identification should degrade gracefully in the face of attack.
IJSER © 2010 http://www.ijser.org
2 International Journal of Scientific & Engineering Research, Volume 1, Issue 2, November-2010
ISSN 2229-5518
10log [
( , )2 /
( ( , )
( , ))2 ]
SNR
10 m I i j
I i j
D i j
Fragile watermarks are ready to be destroyed by random
image processing methods. The change in watermark is easy to be detected, thus can provide information for im- age completeness. Robust watermarks are robust under most image processing methods and can be extracted from heavily attacked watermarked image. Thus it is pre- ferred in copyright protection.
n m n
When SNR approaches infinity, the original image and output image are totally the same.
Another similar one is Peak SNR (PSNR). For images with
255 gray levels, the PSNR is defined as:
10 m
The process of watermarking in the simplest form in-
PSNR 10log [
2552 /
n m n
(I (i, j) D(i, j))2 ]
volves two steps – Watermark Embedding and Water- mark Extraction.
Consider the function f() that denotes the embedding
Fig. 1. Process of Watermarking
function and I, the original image. W, denotes the water- mark to embed and I’, denotes the watermarked image.
I’ = f(I,W)
Common approach is as follows:
Select a group of coefficients from the original image
J = j1, j2… jn
Corresponding watermark sequence is
X = x1, x2… xn
Embed X into J according to certain model to obtain the adjusted sequence
J’ = J + X = j’1, j’2…j’n.
Put j’ back and take the place of J, then we get the water- marked image I’. Let e() denote the extraction function and I’ the image to be examined. Extract the watermark from I’. If the correlation function C(W, W’) satisfies
CWWT>= (T is the threshold value)
Then we consider there is a watermark W in I’.
Signal-to-noise ratio (SNR) is a common metric in signal processing industry. Suppose the original image is I(m,n) the output image is I’(m,n) then generally SNR is defined as:
Compared to spatial-domain watermark, watermark in frequency and wavelet domain is more robust and com- patible to popular image compression standards. Thus frequency and wavelet domain watermarking obtains much more attention. To embed a watermark, a frequen- cy or wavelet transformation is applied to the host data. Then, modifications are made to the transform coeffi- cients. Possible frequency image transformations include the Discrete Fourier Transform (DFT), Discrete Cosine Transform (DCT). Similarly in the wavelet domain we use the Discrete Wavelet Transform (DWT).
The first efficient watermarking scheme was introduced by Koch et al. In their method, the image is first divided into square blocks of size 8x8 for DCT computation. A pair of mid-frequency coefficients is chosen for modifica- tion from 12 predetermined pairs. Bors and Pitas devel- oped a method that modifies DCT coefficients satisfying a block site selection constraint. After dividing the image into blocks of size 8x8, certain blocks are selected based on a Gaussian network classifier decision. The middle range frequency DCT coefficients are then modified, us- ing either a linear DCT constraint or a circular DCT detec- tion region. A DCT domain watermarking technique based on the frequency masking of DCT blocks was in- troduced by Swanson. Cox developed the first frequency- domain watermarking scheme. After that a lot of water- marking algorithms in frequency domain have been pro- posed.
Most frequency-domain algorithms make use of the
spread spectrum communication technique. By using a bandwidth larger than required to transmit the signal, we can keep the SNR at each frequency band small enough, even the total power transmitted is very large. When in- formation on several bands is lost, the transmitted signal
can still being recovered by the rest ones. The spread
IJSER © 2010 http://www.ijser.org
International Journal of Scientific & Engineering Research, Volume 1, Issue 2, November-2010 3
ISSN 2229-5518
spectrum watermarking schemes are the use of spread spectrum communication in digital watermarking. Simi- lar to that in communication, spread spectrum water- marking schemes embed watermarks in the whole host image. The watermark is distributed among the whole frequency band. To de-stroy the watermark, one has to add noise with sufficiently large amplitude, which will heavily degrade the quality of watermarked image and be considered as an unsuccessful attack.
One major reason why frequency domain wa-termarking schemes are attractive is their compatibility with existing image compression standards, in par-ticular, the JPEG standard. The compatibility ensures those schemes a good performance when the watermarked image is sub- ject to lossy compression, which is one of the most com- mon image processing methods today. In consequence, those schemes become particularly useful in practical ap-
The JPEG2000 standard had adopted a new technique, the wavelet transform. Although this standard has not been widely used yet, any new watermarking algorithm that intends to endure in the future should get along with it. The difference between different wavelet domain me- thods depends on the way the watermark is weighted. The reason for this is to reduce the presence of visual arti- facts.
The DWT (Discrete Wavelet Transform) separates an im-
age into four components, a lower resolution approxima- tion image (LL) a horizontal (HL), a vertical (LH) and a diagonal (HH) detail component. The process can then be repeated to computes multiple “scale” wavelet decompo- sitions.
One of the many advantages of the wavelet transform is that that it is believed to more accurately model aspects of the HVS (Human Visual System) as compared to the FFT or DCT. This allows us to use higher energy watermarks in regions that the HVS is known to be less sensitive to, such as the high resolution detail bands {LH,HL,HH). Embedding watermarks in these regions allow us to in- crease the robustness of our watermark, at little to no ad- ditional impact on image quality. One of the most straightforward techniques is to use a similar embedding technique to that used in the DCT, the embedding of a CDMA sequence in the detail bands according to the equ-
ation
I
u , v
I
u , v
Wi + U 
Wi 
xi , u, v E HL, LH
Wi , u, v E LL, HH
Fig.1 . DCT Domain Watermarking Technique using Mid Band Coeffi- cient Exchange with PSNR = 2.53. Going from Top Left to Right
A. Original Image B.Watermarked Image C. Watermark D. Extracted
Watermark
plications on the Internet.
A widely accepted point now is the frequency-domain watermark should be embedded into the mid-band of the transformed host image. Watermarks in high frequency band tend to have less influence on the quality of original image, while watermarks in low band will achieve a bet- ter robustness (since a large portion of high frequency components may be quantized to zero under JPEG com- pression, as shown in figure 4). And the mid-bind scheme is right a tradeoff between the imperceptibility and ro- bustness..
Fig. 2. Wavelet Domain Watermarking Technique – Going from Top Left to Right A. Original Image B. Watermarked Image C. Extracted Watermark D. Watermark
4 International Journal of Scientific & Engineering Research, Volume 1, Issue 2, November-2010
ISSN 2229-5518
Where Wi denotes the coefficient of the transformed im- age, xi the bit of the watermark to be embedded, and a scaling factor. To detect the watermark we generate the same pseudo-random sequence used in CDMA genera-
tion and determine its correlation with the two trans- formed detail bands. If the correlation exceeds some thre- shold T, the watermark is detected. This can be easily ex- tended to multiple bit messages by embedding multiple watermarks into the image. During detection, if the corre- lation exceeds T for a particular sequence a “1” is recov- ered; otherwise a zero. The recovery process then iterates through the entire PN sequence until all the bits of the watermark have been recovered (as shown in the results in Fig 2). Also as the embedding uses the values of the transformed value in embedded, the embedding process should be rather adaptive; storing the majority of the wa- termark in the larger coefficients.
In this paper, I have introduced some basic concepts in digital watermarking, including its foundation, proper- ties, requirements as well as the comparison between dig- ital watermark techniques.Schemes in frequency domain and wavelet domain are introduced with analysis of pros and cons, in terms of imperceptibility, robustness, im- plementation complexity etc., for each domain.
The last part of this paper presents some experimental
results, taking the typical frequency-domain and wavelet domain algorithms. The watermark is embedded into the regions of low interest of host image to achieve a good tradeoff between the imperceptibility and robustness of the watermarking system. The results show that these algorithms have a satisfactory performance under image cropping and JPEG lossy compression.
I would like to thank Prof. Akash Tayal of Indira Gandhi Institute of Technology, New Delhi and Nivedita Yadav Research student at IIT Delhi for their guidance, support and motivation without which the paper could not have been completed
[1] J. Cox, J. Kilian, F. T. Leighton, and T. Shamoon, "Secure spread spectrum watermarking for multimedia," IEEE Trans. Image Processing, vol. 6, pp. 1673-1687, Dec. 1997
[2] A. Bors and I. Pitas, “Image watermarking using DCT domain constraints.” in Proc. IEEE. Int. Conf. Image Processing, Lausanne, Switzerland, Sept. 1996, pp. 231-234
[3] R.C. Gonzalez, R.E. Woods, “Digital Image Processing”, Upper
Saddle River, New Jersey, Prentice Hall, Inc., 2002
[4] Perez-Gonzalez, F.; Hernandez, J.R.;” A tutorial on digital wa- termarking” Security Technology, 1999. Proceedings. IEEE 33rd Annual 1999 International Carnahan Conference on. Oct. 1999
Page(s):286 – 292
[5] Ingemar J. Cox, Matthew L. Miller, Jeffrey A. Bloom “Digital
Watermarking” Morgan Kaufmann Publishers ISBN: 1-55860-
714-5
IJSER © 2010 http://www.ijser.org