DWT Based Audio Watermarking Schemes : A Comparative Study

International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
DOI: 10.5121/ijci.2016.5420 173
DWT BASED AUDIO WATERMARKING SCHEMES:A
COMPARATIVE STUDY
N.V.Lalitha1
Ch. Srinivasa Rao2
and P.V.Y.JayaSree3
1
Department of Electronics and Communication Engineering, GMR Institute and
Technology, Rajam, A.P, India.
2
Department of Electronics and Communication Engineering, JNTU-K University,
Vizianagaram, A.P, India.
3
Department of Electronics and Communication Engineering, GIT, GITAM University,
Visakhapatnam, A.P, India.
ABSTRACT
The main problem encountered during multimedia transmission is its protection against illegal distribution
and copying. One of the possible solutions for this is digital watermarking. Digital audio watermarking is
the technique of embedding watermark content to the audio signal to protect the owner copyrights. In this
paper, we used three wavelet transforms i.e. Discrete Wavelet Transform (DWT), Double Density DWT
(DDDWT) and Dual Tree DWT (DTDWT) for audio watermarking and the performance analysis of each
transform is presented. The key idea of the basic algorithm is to segment the audio signal into two parts,
one is for synchronization code insertion and other one is for watermark embedding. Initially, binary
watermark image is scrambled using chaotic technique to provide secrecy. By using QuantizationIndex
Modulation (QIM), this method works as a blind technique. The comparative analysis of the three methods
is made by conducting robustness and imperceptibility tests are conducted on five benchmark audio
signals.
KEYWORDS
Discrete Wavelet Transform (DWT), Double Density DWT (DDDWT) and Dual Tree DWT (DTDWT),
Quantization Index Modulation (QIM)
1. INTRODUCTION
The swift growth in multimedia technology and the usage of internet, the major problem facing
by the owners is unauthorized copying, transmission and distribution of multimedia content.The
most common solutionfor protection of copyright is digital watermarking [1, 2]. Watermarking is
the process, in which watermark content is embedded into the digital content. Digital content may
be audio, image or video. Developing audio watermarking algorithms are not that much easy
[3,4] compared to image and video watermarking,. Firstly, Human Auditory System (HAS) is
much sensitive than Human Visual System (HVS). Therefore, even small changes in audio are
also recognized by the human ear. Secondly, video files are large compared to audio files in terms
of size. Hence, data hidden in audio files is quietly large compared with the image or video and
this high payload tends to degrade the audio quality. Therefore, trade-off exists between
robustness and imperceptibility.

174
Recently, several audio watermarking algorithms are developed. Most of the algorithms are based
on either time domain [5,6] or transform domain [7,8,9,10,11]. Watermarking in time domain is
easier to implement and needs less computational resources thanwatermarking in transform
domain [3,8] but, it is less robust against common signal processing attacks when compared to
transform domain watermarking. Generally, Fast Fourier Transform (FFT)[11], Discrete Cosine
Transform (DCT) [9], and Discrete Wavelet Transform (DWT)[10] are explored for transform
domain audio watermarking.
Still, there is a need for robust and high secured audio watermarking algorithms. In this paper, the
chaotic Gaussian map is used to encrypt the watermark image. The Logistic chaotic sequence is
used to develop synchronization code. Then, the watermark is embedded in
DWT/DDDWT/DTDWT coefficients of audio signal using QIM.
2. METHODS
2.1. Discrete Wavelet Transform (DWT)
The analysis filters (a1 and a2) decomposes the input signal x(n) into two sub-bands i.e., low-pass
frequency band (c(n)) and high frequency band (d(n)) and each of which is then down-sampled
by 2. The two sub-bands (c(n) and d(n)) are up-sampled by 2 and the synthesis filters (s1 and s2)
combines the two sub-bands to acquire a single signal y(n)[12] shown in Figure 1.
Figure 1. DWT decompose and combined process.
2.2. Double Density DWT (DDDWT)
Double –Density DWT [12] makes use of two distinct wavelets and a single scaling function. The
analysis filters decomposes the x(n) signal into three bands, and every sub-band is down-sampled
by 2. The filter bank for analysis consists of one low-pass filter (a1) and two high pass filters (a2
and a3). The synthesis filter bank consists of one low-pass filter (s1) and two high pass filters (s2
and s3). These3 sub-band coefficients pass through the system are up-sampled by two,
synthesized and then combined to develop the signal y(n) shown in Figure 2.
Figure 2. DDDWT decompose and combined process.
2
y(n)
s2
s122
2a2
a1 c(n)
d(n)
x(n)
d1(n) y(n)
c(n) s12
x(n)
2a1
2a2
d2(n)2
s22
s32

175
2.3. Dual Tree DWT (DTDWT)
The dual tree DWT of a signal x(n) is a parallel combination of two DWTs [13]. Therefore, it is
2-times expensive than DWT. The filters are chosen in a way that the upper DWT can be inferred
as real part of the wavelet and lower DWT can be inferred as imaginary part of wavelet [14] and
is shown in Figure 3.
Figure 3. DTDWT decompose and combined process.
3. SYNCHRONIZATION CODE GENERATION AND INSERTION
The synchronization code [7,8,9] is used to resist the de-synchronization attacks.
Desynchronization attack means the watermark cannot be recognized from the watermarked
audio because of lack of synchronization. Desynchronization attacks are cropping, shifting and
MP3 compression, they will change the audio signal length, which leads to unsuccessful
extraction of the watermark.To overcome this problem, exact location of the watermark should be
identified before the extraction process. For synchronization code generation, the logistic chaotic
sequence is used, that is defined as:
= (1 − ) (1)
Where is the initial value that is from 0 to 1, is the real parameter.
Synchronization code is generated using eq(1) based on the following condition.
2
2h2
h1
2
2h2
h1
x(n)
g1 2
g2 2
g1 2
g2 2
g1 2
g2 2
2
2h2
h1

176
=
1, > 1/2
0, ℎ
(2)
The host audio A is divided into two parts and . Synchronization code that is generated
from the eq(2) is hosted into the first part of audio signal with length LS is embedded as
follows:
( ) = "
# $ %
&'( )
(
) ∗ +, = 0
( , (
&'( )
(
) ∗ +) +
(
.
, = 1
(3)
where + is the embedding strength.
Embedded and attacked watermarked audio signal is also split into two parts. From first part of
watermarked signal synchronization code will be detected with following condition.
=
0, +/4 ≤ 1 $( ( ), +) < 3+/4
1, ℎ
(4)
4. WATERMARK EMBEDDING AND EXTRACTION
4.1. Pre-processing of a Watermark
To improve the security and robustness, watermark image must be pre-processed by using chaotic
scrambling technique. Gaussian map [11] is one of the chaotic encryption methods. Gaussian
map chaotic encryption technique is defined as:
4 = (56(78)9)
+ : (5)
Where z1 is the initial value that ranges from 0 to 1. ; and : are the real parameters.
< =
1, 4 > =ℎ
0, ℎ
(6)
Where=ℎ is the predefined threshold. Two dimensional binary watermark is converted into a
vector of size M X M. This is encrypted by < using following condition:
> = ?@A( , < ) (7)
4.2. Watermark Concealing Procedure
The watermark concealing procedure is given in Figure 4 . In this procedure, total audio signal is
segmented into two parts. The synchronization code is insert in audio signal first part to
overcome the de-synchronization attacks. The audio signal second part is used to host the pre-
processed watermark image.

177
Figure 4. Flowchart of watermark embedding process.
The concealing procedure is detailed as follows:
Step 1: Apply DWT/DDDWT/DTDWT on second part of audio signal.
Step 2: Wavelet coefficients are segmented into frames, and number of frames must be greater
than the watermark size.
Step 3: The pre-processed watermark is embedded into each frame using the following rule.
B′
( ) = "
# $ %
CD( )
E
) ∗ F, > = 0
( , (
CD( )
E
) ∗ F) +
E
.
, > = 1
(8)
where F is the embedding strength.
Step 4: Reconstruct the modified frames.
Step 5: Apply inverse wavelet transform on watermarked audio.
4.3. Extraction Algorithm
The process of extraction is the exact reverse process of concealing process and the algorithm is
given below:
Part A
Synchronization
code insertion
Embedding
Watermarked
Audio
Frame
Reconstruction
Inverse Wavelet
Transforma3
Binary
Watermark
Image
Pre-processing
Original
Audio
Part B
Synchronization
code generation
Segmented
into frames
DWT / DDDWT
/ DTDWT

178
Step1: Apply DWT/DDDWT/DTDWT on the second part of attacked watermarked audio signal.
Step2: Wavelet coefficients are segmented into frames.
Step3: Binary encrypted watermark vector is extracted from each frame by using following
equation.
G′
=
0, F/4 ≤ 1 $(B′′
( ), F) < 3F/4
1, ℎ
(9)
Step4: The decryption process is same as encryption to determine the binary watermark sequence.
Step5: Finally, convert the one dimensional extracted and decrypted binary sequence into two
dimensional watermark image of size M X M.
5. SIMULATION RESULTS
The experimental results give the comparative analysis of the three methods. The performance of
the three methods iscompared in terms of robustness, imperceptibility and payload. The
experiment is carried on 5 different types of 16-bit audio signals in the .WAV format with the
sampling rate 44.1 kHz. Each audio is of 10sec duration.
Binary image of 64 X 64 size is used as a watermark. For increasing the security of the
watermark, a Gaussian map chaotic encryption technique is used. Figure 5 illustrates Original and
encrypted watermark images.
Figure 5. Original watermark and its encrypted watermark images.
5.1. Imperceptibility Test
The audio signal quality should not be degraded upon embedding. The two approaches to perform
the perceptual audio quality evaluation [15]. i) Objective test by perceptual evaluation of audio
signal ii) Subjective listening test based on HAS.
i) Objective evaluation test:
To evaluate the objective quality, SNR metric is used. International Federation of the
Phonographic Industry (IFPI) quotes that watermarked audio should have SNR more than 20dB
[8]. SNR Vs Quantization step for three methods are shown in Figure 6.

179
Figure 6. SNR Vs Quantization step Table 1. SNR in dB for benchmark audio
Table 1 shows the SNR values and their average SNRs for different classes of benchmark audio
signals at Q=0.07 are above 20dB and hence meets IFPI requirement.
ii) Subjective Listening Test:
The SNR measure is not sufficient to measure imperceptibilty [8]. Therefore, subjective listening
test is also important to evaluate the imperceptibility. Subjective Difference Grade (SDG) is a
popular method to evaluate the watermarked audio quality [11]. Table 2 shows the SDG ranges,
which is from 5.0 to 1.0. This listening test is performed with ten listeners. Subjects are listened
original and watermarked audio signals and they report if any variation is identified between two
signals using SDG. The average SDG values are also called as Mean Opinion Score (MOS). The
MOS values for DWT,DDDWT and DTDWT is 4.5, 4.8 and 4.7 respectively at Q=0.07.
Table 2. SDG Ranges
Report by subject Quality Grade
Imperceptible Excellent 5
Perceptible, but not annoying Good 4
Slightly annoying Fair 3
Annoying Poor 2
Very annoying Bad 1
5.2. Robustness Test
Robustness of this scheme is evaluated with the below attacks on watermarked audio.
i) Resampling: The watermarked audio is resampled to 22.05 kHz, 11 kHz and 8
kHz and sampled back to 44.1 kHz.
ii) Re-quantization: Quantized down to 8-bit and re-quantized back to 16-bit.
iii) Noise: Added with random noise of 30dB signal.
iv) Low-pass Filtering: Cut-off frequency of 20 kHz is applied.
v) Echo addition: 10 ms and 1% decay of echo signal is added.
vi) MP3 Compression: 128 kbps and 256 kbps MPEG compression is applied to the
watermarked audio signal and then decoded back to the .WAV format.
DWT DDDWT DTDWT
Audio-1 31.1205 41.0349 27.7986
Audio-2 42.311 30.6061 27.2856
Audio-3 41.2256 27.0774 53.433
Audio-4 58.0209 41.3026 48.2897
Audio-5 29.8392 36.1878 36.0735
Average 40.5034 35.2417 38.5760
0.01 0.02 0.03 0.04 0.05 0.06 0.07 0.08 0.09 0.1 0.11
15
20
25
30
35
40
45
50
55
60
Quantization Step
SNR(dB)
SNR Vs Quantization Step for different methods
DWT
DDT
DT

180
vii) Additive Noise: Additive Gaussian Noise with 50 dB and 60 dB.
viii) Cropping: 1000 samples of the watermarked audio signal are made zero at
beginning, middle and ending parts.
ix) Signal Addition: Beginning samples are added with original audio samples.
x) Signal Subtraction: Watermarked audio signal beginning samples are subtracted
with original audio samples.
For comparison of original watermark and extracted watermark, Bit Error Rate (BER) and
Normalized Correlation (NC) are used.
BER =
KLMNOP QR OPPQP NSTU
KLMNOP QR TQTVW NSTU
(10)
XY =
∑ ∑ ([]5[^)(_]5_^)]
`∑ ∑ ([]5[^)9
] ∑ ∑ (_]5_^)9
]
(11)
Table 3 shows BER and NC for all mentioned signal processing attacks for three methods
at Q=0.07.
Table 3. BER and NC values for signal processing attacks.
Method DWT DDDWT DTDWT
Signal Processing Attack BER NC BER NC BER NC
Without attack 0 1 0 1 0.0002 0.9994
Resampling(22.05kHz) 0.0007 0.9982 0 1 0.1182 0.7316
Resampling(11kHz) 0.1741 0.6096 0.1528 0.6508 0.3726 0.2303
Resampling(8kHz) 0 1 0 1 0.0012 0.9971
Re-quantization 0 1 0 1 0.0447 0.8954
Noise 0 1 0 1 0.0059 0.9861
Filtering 0 1 0.0002 0.9994 0.0269 0.9363
Echo addition 0 1 0.0002 0.9994 0.0203 0.952
MP3 Compression (256) 0 1 0 1 0.0063 0.9848
MP3 Compression (128) 0.0004 0.9988 0.0012 0.9971 0.0354 0.9167
Additive Noise (50dB) 0 1 0 1 0.0591 0.863
Additive Noise (60) 0 1 0 1 0.0146 0.9651
Cropping (middle) 0 1 0 1 0.0002 0.9994
Cropping (end) 0 1 0 1 0.0002 0.9994
Cropping (front) 0.0022 0.9948 0.0022 0.9948 0.0024 0.9942
Signal Addition 0.002 0.9953 0.0022 0.9948 0.0022 0.9948
Signal Subtraction 0.002 0.9953 0.0022 0.9948 0.0024 0.9942
6. CONCLUSIONS
The performance of DWT based audio watermarking schemes viz., DWT, DDDWT and DTDWT
is analyzed. SNR is above 20 dB for all the three schemes. The watermarked signal is tested
against various signal processing attacks for different classes of audio signals and the
performance parameters BER and NC are obtained. The parameters shows that DDDWT

181
outperforms DTDWT for different values of quantization step. Also, DDDWT performance is
almost nearer to DWT scheme.
REFERENCES
[1] I.Cox, M.Miller, J.Bloom, J.Fridrich, T.Kalke, “Digital Watermarking and steganography”, Second
Edition, Morgan Kaufmann, Burlington, MA, 2007.
[2] M.Swanson, M.Kobayashi, A.Tewfik, “Multimedia data embedding and watermarking techniques”,
Proc.IEEE 86 (1998) 1064-1087.
[3] E.Ercelebi, L.Batakci, “Audio watermarking scheme based on embedding strategy in low frequency
components with a binary image”, Digital Signal Processing, Vol.19, pp 265-277, 2009.
[4] M.D.Swanson, B.Zhu, A.H.Tewfik, L.Boney, “Robust audio watermarking using perceptual
masking”, Signal processing, 66(3), pp.337-355, 1998.
[5] Lie W.N, Chang L.C, “Robust high quality time domain audio watermarking based on low frequency
amplitude modification”, IEEE Transactions on Multimedia, 8(1), pp.46-59, 2006.
[6] P.Basia, I.Pitas, N.Nikolaidis, “Robust audio watermarking in the time domain”, IEEE Transactions
on Multimedia, 3(2), pp. 232-241, 2001.
[7] X.Y.Wang, H.Zhao, “A novel synchronization invariant audio watermarking scheme based on DWT
and DCT”, IEEE Transactions on Signal Processing, 54(12) pp. 4835-4840,2006.
[8] Lei, I.Y.Soon, F.Zhou, Z.Li, H.Lei, “A robust audio watermarking scheme based on lifting wavelet
transform and singular value decomposition”, Signal Processing, Vo. 92, pp. 1985-2001, 2012.
[9] B.Y.Lei, I.Y.Soon, Z.Li, “Blind and robust audio watermarking scheme based on SVD-DCT”, Signal
Processing, Vo. 91, pp. 1973-1984, 2011.
[10] V.Bhat K, I. Sengupta, A. Das, "An adaptive audio watermarking based on the singular value
decomposition in the wavelet domain", Digital Signal Processing 20 (2010) 1547-1558.
[11] P.K.Dhar, T.Shimamura, “Audio watermarking in transform domain based on singular value
decomposition and Cartesian-Polar Transformation”, International Journal of Speech Technology,
Vol.17, pp.133-144, 2014.
[12] Ivan W.Selesnick,“The Double Density DWT”, http://eeweb.poly.edu/iselesni/double/double.pdf
[13] N.Kingsbury, “Image Processing with complex wavelets”, Phil.Trans. R. Soc. London A, 1997.
[14] N.G. Kingsbury, “The dual-tree complex wavelet transform: A new technique for shift invariance and
directional filters”, in proceedings of Eighth IEEE DSP Workshop, Salt Lake City, UT, Aug,9-12,
1998.
[15] Y.Q.Lin, W.H.Abdulla, “Perceptual Evaluation of Audio Watermarking using Objective Quality
Measures”, in proceedings of the IEEE International Conference on Acoustics, Speech and Signal
Processing, 2008, pp. 1745-1748.

182
AUTHORS
N. V. Lalitha is presently pursuing PhD at GIT, GITAM University. She obtained her
M.Tech from Jawaharlal Nehru Technological University, Kakinada and B.Tech from
Jawaharlal Nehru Technological University. Presently, she is working as Assistant professor
in the Department of Electronics and Communication Engineering at GMR Institute of
Technology, Rajam, Srikakulam District. Her research interests are Audio and Image
Processing. She is a Life Member of IETE.
Srinivasa Rao Ch is currently working as Professor in the Department of ECE, JNTUK
University College of Engineering, Vizianagaram, AP, India. He obtained his PhD in Digital
Image Processing area from University College of Engineering, JNTUK, Kakinada, AP,
India. He received his M. Tech degree from the same institute. He published 40 research
papers in international journals and conferences. His research interests are
DigitalSpeech/Image and Video Processing, Communication Engineering and Evolutionary
Algorithms. He is a Member of CSI. Dr Rao is a Fellow of IETE.
P. V. Y. Jayasree is currently working as Associate Professor in the Department of ECE,
GIT, GITAM University. She obtained her PhD from University College of Engineering,
JNTUK, Kakinada, AP, India. She received M.E. from Andhra University. She published
more than 50 research papers in international journals and conferences. Her research interests
are Signal Processing, EMI/EMC, RF & Microwaves.

DWT Based Audio Watermarking Schemes : A Comparative Study

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to DWT Based Audio Watermarking Schemes : A Comparative Study (20)

Recently uploaded (20)

DWT Based Audio Watermarking Schemes : A Comparative Study