International Journal of Scientific & Engineering Research Volume 2, Issue 6, June-2011 1
ISSN 2229-5518
Improved Audio Watermarking
Using DWT-SVD
N.V.Lalitha, G.Suresh, Dr.V.Sailaja
—————————— • ——————————
HE rapid development of the Internet and the digital information revolution cased significant changes in
the global society, ranging from the influence on the
world economy to the way people nowadays communi- cate [1]. Digitizing of multimedia data has enabled relia- ble, faster and efficient storage, transfer and processing of digital data. It also leads to the consequence of illegal production and redistribution of digital media. Digital watermarking is identified as a partial solution to related problems which allow content creator to embed hidden data such as author or copyright information into the multimedia data [2]. In cryptographic techniques signifi- cant information is encrypted so that only the key holder has access to that information. Once the information is decrypted the security is lost. Information hiding is unlike cryptography, message is embedded into digital media, which can be distributed and used normally. Information hiding doesn’t limit the use of digital data. Within past
few years several algorithms for embedding and extrac-
magic triangle. In order to satisfy the requirements of magic triangle, watermarks are seen embedded in Fourier domain [3], time domain [4], sub-band domain [6], wave- let domain [7] and by echo hiding [5].
The DCT and DWT transforms have been extensively used in many digital signal processing applications. SVD is a useful tool of linear algebra with several applications in image compression, watermarking and other areas of signal processing. A few years ago, SVD is explored for image watermarking applications [9, 10]. The brief intro- duction of these three techniques are presented in this section .
is a technique for converting a signal into elementary fre- quency components [11]. The most common DCT defini- tion of a 1-D sequence of length N is
N =1
tion of watermark in audio sequence have been published
C (u) = a (u)
f ( x) cos J7t (2 x - 1)u l,
(1)
[3-7]. Almost all audio watermarking algorithms work by
x =0
L 2N J
exploiting the perceptual property of Human Auditory
System (HAS). The simplest visualization of the require-
For u=0,1,2,…,N-1. Similarly, the inverse transformation is defined as
ments of information hiding in digital audio is possible
N =1
J7t (2 x - 1)u l
via a magic triangle [8]. Inaudibility, robustness to attacks
f ( x) = a (u)C (u) cos ,
(2)
and the watermark data rate are in the corners of the
u =0
L 2N J
for x=0,1,2,….N-1. In both equations (1) and (2) a (u) is
defined as
————————————————
G.Suresh, Assoc.Prof, ECE Dept., Krishna’s Pragati Institute of
Technology, Rajahmundry, A.P.INDIA. PH-09885837385.
E-mail: sureshg_ece@yahoo.co.in
Dr. V.Sailaja ,Prof, ECE Dept., GIET, Rajahmundry,A.P. INDIA,
E-mail:lalithanarla.ece@gmail.com
1
a (u) = N
2
\ N
for u = 0
for u -: 0
(3)
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 2
ISSN 2229-5518
It is clear from (1) that for u=0, C(u=0)=
1 N -1
N x =0
f ( x) .
sampling by 2.
Thus the first transform coefficient is the average value of the sample sequence. In literature, this value is referred to as the DC Coefficient. All other transform coefficients are called the AC Coefficients [12].
In particular, a DCT is a Fourier-related transform similar to the Discrete Fourier Transform (DFT), but using only real numbers. DCTs are equivalent to DFTs of rough- ly twice the length, operating on real data with even symmetry (since the Fourier transform of a real and even function is real and even), where in some variants the input and/or output data are shifted by half a sample.
has received a tremendous amount of interest in many important signal processing applications including audio and image watermarking [13, 14 and 15]. With the DWT, the audio signal can be transformed into frequency do- main ranging from low frequency to high frequency. Be- sides, the high frequency spectrum is less sensitive to human ear. That is the reason why the high frequency component is usually discarded in the compression process. Therefore, information to be hidden can be em- bedded into the low frequency component to against the compression attack.
The DWT is defined by the following equation
explored for watermarking [17]. The SVD for square ma- tices was discovered independently by Beltrami in 1873 and Jordan in 1874, and extended to rectangular matrices by Eckart and Young in the 1930s. It was not used as a computational tool until the 1960s because of the need for sophisticated numerical techniques. In later years, Gene Golub demonstrated its usefulness and feasibility as a tool in a variety of applications [18]. SVD is one of the most useful tools of linear algebra with several applica- tions in image compression and other signal processing fields.
Watermark embedding procedure
W ( j, k ) =
x(k )2- j /2 (2- j n - k )
j k
The D represents the Details sub-band, and A represents
the approximation sub-band.
Where (t) is a time function with finite energy and fast decay called the mother wavelet. The DWT analysis can be performed using a fast, pyramidal algorithm related to multirate filterbanks [16].
As a multirate filterbank the DWT can be viewed as a
constant Q filterbank with octave spacing between the centers of the neighboring higher frequency subband. In the Pyramidal algorithm the signal is analyzed at differ- ent frequency bands with different resolution by decom- posing the signal into a coarse approximation and detail information. The coarse approximation is then further decomposed using the same wavelet decomposition step. This is achieved by successive highpass and lowpass fil- tering of the time domain signal and is defined by the following equations:
Step 3: Apply SVD to the DCT performed approximation sub-band A. SVD decomposes the DCT coefficients into three matrices namely, U, S, VT. Where U is Unary matrix, S is Singular matrix.
Where S = singular matrix of original audio signal
Sw = singular matrix of watermark audio signal
Sem = singular matrix of watermarked audio signal
: Produce the final watermarked audio signal as
yhigh [k ] =
n
ylow [k ] =
n
x[n]g[2k - n]
x[n]h[2k - n]
follows:
� Apply the inverse SVD operation using the U and
VT matrices, which were unchanged, and the S matrix,
which has been modified according to Equation (3).
Where
yhigh [k ]
, ylow [k ]
are the outputs of the high-
� Apply the inverse DCT operation to obtain each
pass (g) and lowpass (h) filters, respectively after sub-
watermarked audio frame. The overall watermarked au-
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 3
ISSN 2229-5518
dio signal is obtained by summing all Watermarked frames.
VT matrices, which were unchanged, and the S matrix, which has been modified according to Equation (4).
� Apply the inverse DCT operation to obtain each
watermarked audio frame.
Orignal Audio signal
Fram- ing
DC T
SV D
Watermark Audio Signal
DC T
SV D
X
Watermarked Audio
Signal Framing DCT
SVD
+ Sw=(Swa-S)/k
Inverse
SVD
Inverse SVD
IDC T
IDC T
Is any frames
Is water- mark de- tected
Step 1: Perform steps 2 and 3 of the embedding proce- dure until the S matrix is obtained for all frames of the watermarked audio signal.
Where Sex = singular matrix of extracted watermark audio signal.
� Apply the inverse SVD operation using the U and
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 4
ISSN 2229-5518
Step 3: Apply SVD to the DWT performed approximation sub-band A. SVD Decomposes the DWT coefficients into three matrices namely, U, S, VT. Where U is Unary matrix, S is Singular matrix.
Step 1: Perform steps 2 and 3 of the embedding proce- dure until the S matrix is obtained for all frames of the watermarked audio signal.
ex em
Where S = singular matrix of original audio signal
Sw = singular matrix of watermark audio signal
Sem = singular matrix of watermarked audio signal
� Apply the inverse SVD operation using the U and
VT matrices, which were unchanged, and the S matrix, which has been modified according to Equation (3).
� Apply the inverse DWT operation to obtain each
watermarked audio frame. The overall watermarked au- dio signal is obtained by summing all Watermarked
frames.
Where Sex = singular matrix of extracted watermark audio signal.
� Apply the inverse SVD operation using the U and
VT matrices, which were unchanged, and the S matrix, which has been modified according to Equation (4).
� Apply the inverse DWT operation to obtain each
watermarked audio frame.
Watermarked Audio Signal
Orignal Audio signal
Watermark Audio
Signal
Framing
Framing
DW T
SVD
DWT
SV D
X
DWT SVD
Sw=(Swa-S)/k
+ Inverse SVD
Inverse
SVD
IDWT
IDWT
Is any frames
Is wa- termark detected
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 5
ISSN 2229-5518
Four classes of audio signals like speech, pop music, rock music and instrumental were used to study the per- formance of the DCT-SVD and DWT-SVD algorithms. These classes were chosen because each class has different spectral properties.
the perceptual quality of the embedded watermark audio within the original audio signal. To measure impercepti- bility, we use Signal-to-Noise Ration (SNR) as an objective measure, and a listening test as a subjective measure.
For subjective quality evaluation, a listening test was performed with five listeners to estimate the Mean Opi- nion Score (MOS) grade of the watermarked signals for four different signals. Each listener was presented with the pairs of original signal and the watermarked signal and asked to report whether any difference could be de- tected between the two signals. The five people listed to each pair for 15 times, and they gave a grade for the pair. The average grade for each pair from all listeners corres- ponds to the final grade for the pair. MOS evaluation cri- terion and MOS for the two techniques are listed in the Tables1 and 2 respectively.
Table 1. MOS evaluation criterion
Score | Watermark imperceptibility |
5 | Imperceptibility |
4 | Perceptibility but not annoying |
3 | Slightly annoying |
2 | Annoying |
1 | Very annoying |
Table 2. MOS for the two techniques
n=0,1,2,3,……N
Signal to Noise Ratio (SNR) is a statistical difference me- tric which is used to measure the similitude between the undistorted original audio signal and the distorted wa- termarked audio signal.
Table 3. PSNR Evaluation
[2].
SNR in db is calculated using the following equation
An efficient audio watermarking algorithm in the fre- quency domain by embedding the inaudible audio water mark is presented here. It is verified that the DWT-SVD
SNR = 10 log10
N -1
n=0
x2 (n)
[ x(n) - xl (n)]2
technique is robust for most of the attacks rather than the
DCT-SVD. By means of combining the two transforms
DWT-DCT along with SVD, inaudibility and different
Where, N is the length of audio signal x(n) is the original signal
xl(n) is the watermarked signal
levels of robustness can also be achieved.
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 6
ISSN 2229-5518
[1] “ Algorithms for Audio Watermarking and Stegano
Graphy”, NEDELJKO, CVEJIC Department of Elec
Trical and Information Engineering, Information
Processing
Laboratory, University of OULU 2004.
[2] L.T.Bruton, J.D.Gordy, Performance Evaluation of Dig ital Audio Watermarking Algorithms, Proceedings of the 43rd IEEE Midwest Symposium, Michigan, Volume
1, pp.456-459, 8-11 August 2000.
[3] M.D.Swanson, B.Zju,A.H.Tewfik, Robust audio wa termarking using perceptual masking, Signal processing, vol.66,pp.337-355,1998.
[4] P.Bassia, I.Pitas, Robust audio watermarking in time domain, IEEE Trans. On Mutimedia, vol.3,No.2, pp.232-241, June 2001.
[5] W.Bender, D.Grul,N.Morimoto, A.Lu, Techniques for data hiding,IBM System Journal,vol.35, pp.313-336,
1996.
[6] Xin Li, Hong Heather Yu, Transparent and Robust Audio Data Hiding in Sub-band domain, Proc. Inter national Conference on Information Technology: cod ing and computing, Las Vegas, pp.74-79, March 27-
29,2000.
[7] M.F.Mansour,A.W.Tewfik, Audio watermarking by time scale modification, Proc International Conference
on Acoustic, Speech, and Signal processing, USA,
vol.3 pp.1353-1356, May 7-11, 2001.
[8] Nedeljko Cvejic, Tapio Seppnen, Watermark
Bit Rate in Diverse Signal Domains, Interna
tional Journal of Signal Processing, vol.1,
2004.
[9] Bao P, Ma X (2005) Image adaptive water marking using wavelet domain singular value decomposition. IEEE Trans Circuits Syst Vid eo Technol 15(1):96-102
[10] Yavuz E, Telatar Z (2007) Improved SVD- DWT based digital image watermarking against watermark am biguity. In:Proceedings of the ACM symposium on applied computing, pp 1051-1055
[11] Rao K and P.Yip, Discrete Cosine Transform algo ithms, advantages, applications. Academic Press, USA, 1990.
[12] The Discrete Cosine Transform(DCT): Theory and Application, Syed Ali Khayam, De partment of Elec trical & Computer Engineering, Michigan State Uni versity, March 10th 2003.
[13] Michael A., Geraon and Peter G. Graven, “A high-rate butied-data channel for audio CD”, Journal of the Audio Engineering Society, 43(1/2):3-22,January- February 1995.
[14] Shaoquan Wu, J.Huang, Y.Q. Shi, “Self-Synchronized
Audio Watermark in DWT Domain”,IEEE (ISCAS), v-
712-v-715,2004.
[15] N. Sriyingyong, and K. Attakitmongcol, “Wavelet- Based Audio Watermarking Using Adaptive Tabu Search” Wireless Pervasive Computing, 2006 1st In ternational Symposium on 16-18 an. 2006 Page(s):1-
5,2006.
IJSER © 2011 http://www.ijser.org
International Journal of Scientific & Engineering Research Volume 2, Issue 6, June -2011 7
ISSN 2229-5518
[16] S.G Mallat “A Theory for Multiresolution Signal De composition: The Wavelet Representation” IEEE Transactions on Pattern Analysis and Machine Intelli gence, Vol.11,1989,674-693.
[17] R. Liu and T.Tan, “A SVD-Based watermarking Scheme for Protecting Rightful Ownership”, IEEE Transactions on Multimedia, 4(1),March 2002,pp.121-
128.
[18] D.Kahaner, C. Moler and S.Nash, Numerical Methods and Software (New Jersey: Prentice-Hall, Inc, 1989).
[19] ITU-R, “Methods for objective measurements of per
ceived audio quality”, BS.1387.
IJSER © 2011 http://www.ijser.org