1. Bit Replacement Audio Watermarking
Using Stereo Signals
Term Project
Muhammad Umer Kakli
Muhammad Naeem Tayyab
This Paper was presented in International Conference on
New Trends in Information & Service Science 2009
By Wei Cao, Yixin Yan, Shengming Li

2. Digital Watermarking involves embedding secret data
as an additional information to the digital signal such
as Audio, Image or Video Signal.
 Copyright Protection
 Video Fingerprinting
 Broadcast Monitoring

4. For Digital Audio Watermarking, the approaches can be
Blind or Non-blind.
For Non-blind watermarking techniques, original signals
are required for watermark extraction while in Blind
technique original signals are not required to extract
watermark.

5.  Removal Attacks: Trying to remove the watermark
signal without attempting to break the security of
watermark algorithm.
 Geometric Attacks: Don’t remove the watermark itself,
but intended to distort the watermark detector
synchronization with the embedded information.
 Cryptographic Attacks: Aim at cracking the security
methods in watermarking schemes and thus finding
a way to remove the embedded watermark
information
 Protocol Attacks: Aim at attacking the entire concept
of watermarking application.

6. Figure 1: Classification of Watermark Attacks

7.  LSB Encoding
 Echo Hiding
 Spread Spectrum Watermarking
 Watermarking the phase of the host signal

8.  Watermark added to the Least Significant Bit
(LSB)
 Easy to embed and retrieve
 High bit rate

9. De-merits of LSB Encoding Scheme
 Robustness of this method is very low
 Watermark bits can be easily detected and changed
 Addition of High level noise, re-sampling and MP3
Compression may completely destroy the watermark
information.
Proposed Scheme
 Instead of using LSB of mono-signal samples, it is
perceived that by using higher bits of Stereo-Signal
samples, it can be more robust.

10.  The proposed Audio Watermarking scheme improves
mono-signal watermarking method by introducing
watermarking threshold and making use of Stereo
Signals.
Mono Signals Vs. Stereo Signals
 Mono means audio signals are mixed together and
routed through a single channel while In Stereo two or
more individual channels are using. left channel's o/p is
connected to the left speakers and right channel's o/p is
connected to the right speakers. It gives the effect of
direction and the depth of sound.

11.  Watermarking threshold is calculated using the following
equation
Watermarking Threshold = 1/2n-(w+4)
n=Number of bits per sample
w=Watermark embed bit layer
 Samples whose values are higher than threshold are
considered as “Non-Silent” samples and they are used
to embed watermark bits. Samples whose values are
lower than threshold are considered as “Silent”
samples and they are not used for watermark
embedding.

12. After threshold calculation , let each “non-silent” sample
value of original stereo signal be represented in 16 bit
binary format.
a16,a15,a14,………….a3,a2,a1
where a1 is the bit in the 1st bit layer, a2 is the bit in the
2nd bit layer and so on.
A watermark bit stream is first generated and ith bit layers
of “non-silent” samples are used as watermark
embedded bit layer.

13. If the first bit of the stream is bit “1”, and ai of the first “non-silent”
sample of left-channel signal is also bit “1”, no action is taken. If ai is
not “1”, watermark embedding process of left-channel signal is
performed according to the following procedures,

14. If the first bit of the stream is bit “0”, and ai of the first “non-silent”
sample of right-channel signal is also bit “0”, no action is taken. If ai
is not “0”, watermark embedding process of right channel signal is
performed according to the following procedures,

15. Effect of Watermarking on Bit Layers
Figure 2: Percentage of Changed bits as embedding in the 5th bit layer

16. Signal to Noise Ration (SNR)
Signal to Noise Embedded Bit Layer of Watermark Signals
Ratios
(dB) 1st 2nd 3rd 4th 5th
Stereo Signals 72.09 70.15 67.38 59.36 46.25
Mono Signals 66.52 61.49 56.81 44.37 28.53
SNR of Watermarked Stereo and Mono Signals

17. Robustness under Attack of Addition of Noise
Figure 3: Watermark Extraction Rate Under addition of White noise (for 20dB & 50dB)

18. Robustness under MP3 Compression Attack
Embedded Bit Layer of Watermark
Extraction Rate (%) for MP3
Compression
Signals
1st 2nd 3rd 4th 5th
Extraction Rate (%) 49.80 49.82 49.83 50.06 50.11
Watermark Extraction Rate Under Attack of MP3 Compression

19. Robustness under Re-Sampling Attack
Watermark Embedded Bit Layer of Watermark Signals
Extraction Rate
(%) 1st 2nd 3rd 4th 5th
14.70 48.99 49.54 49.83 49.71 50.11
Re-sampling
Frequency
22.05 49.61 49.73 50.17 50.66 50.70
(KHz)
66.15 49.17 49.84 49.96 50.30 50.43
88.20 50.03 53.29 55.64 58.52 60.79
Watermark Extraction Rate Under Attack of Re-sampling

20. The noise level due to embedding watermark is
significantly reduced. This schemes is robust against
addition of white noise, even if the noise level is high.
The proposed scheme is not robust against MP3
Compression and re-sampling unless frequency is
integer multiple of original sampling frequency.