Content uploaded by Houria Kelkoul
Author content
All content in this area was uploaded by Houria Kelkoul on Dec 25, 2018
Content may be subject to copyright.
A Robust Combined Audio and Video Watermark
Algorithm Against Cinema Piracy
Houria Kelkoul∗, Youssef Zaz∗, Hicham Tribak∗and Gerald Schaefer†
Faculty of Science, University of Abdelmalek Essaadi, Tetouan, Morocco
Department of Computer Science, Loughborough University, Loughborough, U.K.
Abstract—The current trend in Moroccan digital cinema is to
make feature films available online. In order to counter illegal
distribution of media files in Morocco, this paper suggests an
effective algorithm that combines processing of both audio and
video streams. We extract an audio fingerprint pattern which
is embedded in video frames as a watermark using the discrete
wavelet transform. The two data streams are then synchronized
using a time code, and the final multimedia protected content
is exported in JPEG2000 format, which is the format used
for Digital Cinema Package (DCP) in Morocco. Experimental
results show that the proposed algorithm provides a high level
of robustness, while the use of watermarks in a combination
of audio and video streams allows to enhance the security of
multimedia content.
Index Terms—Multimedia processing; secrurity protection;
digital watermarking; audio fingerprinting; discrete wavelet
transform.
I. INTRODUCTION
There has been a significant emergence of Moroccan cinema
over the past two decades and Moroccan cinema has become
one of the most important national cinemas in Africa and
the Arab world. While only a handful of films per year were
produced in the 1970s, up to 32 feature films are now released
annually1.
This paper is devoted to the protection of digital cinema
products and to enable secure sharing of digital movies.
Various approaches have been proposed to address the problem
of securing digital cinema content, which comprise water-
marking, and fingerprinting techniques. Watermarking solu-
tions embed a secret copyright message into the multimedia
content, while fingerprinting methods identify, extract, and
compress distinctive constituents of video content and allow
identification of unknown videos by matching it to videos with
similar content [1].
In this paper, we demonstrate a solution to digital cinema
content protection that jointly exploits the audio and video
channels and is based on a combination of the two principal
techniques of fingerprinting and watermarking. The proposed
approach is compared to our previous approach from [12] and
the algorithm from [13] to demonstrate its usefulness.
The remainder of the paper is organised as follows: Sec-
tion II describes the problem of digital cinema copyright pro-
tection and highlights some related work. Section III presents
our proposed algorithm, while Section IV gives and discusses
1http://www.ccm.ma/bilan-cinematographique
the obtained experimental results. Section V concludes the
paper.
II. DI GI TAL CINEMA PROT EC TI ON
The majority of approaches involving watermarking and
fingerprinting technologies have been developed in separate
and independent ways since they have evolved in different
research areas [2]. For instance, an audio fingerprint is a
compact content-based signature that summarises information
of an audio recording [3]. The IFPI (International Federation
of the Phonographic Industry) and the RIAA (Recording
Industry Association of America) have instigated a study [4]
on audio fingerprinting technologies destined to evaluate nu-
merous fingerprinting systems. From this, it can be concluded
that a prototypical fingerprinting scheme in digital cinema field
should achieve several requirements including the ability to
correctly identify the identity of a cinematographic audio prod-
uct regardless of alterations. More detailed requirements can
help to distinguish between different fingerprinting schemes
and approaches [5], [6]. As an example, ‘cinemetrics’ [7] mea-
sures and visualises movie data to expose the features of films
and to create a pattern fingerprint for them. Characteristics
such as audio data is extracted, analysed and transformed into
a fingerprint matrix so that a movie can be identified after use
or alteration.
On the other hand, a variety of watermarking techniques
applied to the field of digital cinema have being proposed [8].
Watermarking is commonly defined as the procedure of em-
bedding a digital secret copyright message into a digital
signal. Watermarking methods can be divided into spatial and
frequency domain approaches [9]. Spatial watermarking algo-
rithms modify a selected subsets of frames using techniques
such as LSB or SSM modulation. Frequency (or transform)
Fig. 1. Illustration of video fingerprinting.
domain techniques are typically more effective compared to
spatial methods as they treat directly the band signal of the
data. Transform domains comonly employed include DCT,
DWT, and DFT. Figs. 1 and 2 illustrate the workings of
fingerprinting and watermarking approaches respectively.
Fig. 2. Illustration of video watermarking.
In our previous work [12], we proposed an algorithm against
cinema movie piracy that employs a text copyright message
as a watermark. The watermark is embedded using spread
spectrum and DWT methods, whereas the approach in [13]
is based on DWT and SVD.
III. PROP OS ED D CP A LGORITHM
In this paper we present an algorithm exploiting audio
and video bands with the aim of enhancing the copyright
multimedia security process for digital cinema producers and
users.
Our algorithm proceeds as follows:
•Extracting the audio fingerprint:
1) The audio band signal is filtered with a band pass
filter with threshold frequencies of 600Hz and 3kHz.
2) The audio signal is converted using a third level
DWT to decompose the signal into units that are
localised both in time and frequency.
3) The filtered signal is divided into frames, and the
zero-crossing rate (ZCR) identified by
ZC Rm=1
2X
m
|sign(x(n))−sign(x(n−1))|(1)
where x(n)represent the value of the wavelet
coefficients in every frame.
4) The number of samples between two successive zero
crossings is counted.
5) The length of identical audio frames is compared to
a given threshold. Audio frames which are shorter
than the minimal threshold value are skipped, and
only the remaining ones used to generate a binary
output as
0if the value of a prior frame matches
the sign of the current audio frame
1if the values of the sequential frames
are different
6) The subsequent output presents the row fingerprint-
ing code which identifies the audio file.
•Watermark embedding:
1) The video content is sampled into frames.
2) Frames are converted using a third level DWT with
the aim of embedding the watermark.
3) As illustrated in Fig. 3, the watermark is embedded
into the LL blocks by
Y(i) = S0(i)∂∗w0(i)if w0(i) = 1
S0(i)∂∗w0(i)if w0(i) = −1(2)
where S0represents the low frequencies coefficient
LL and αis the watermark coefficient.
Fig. 3. DWT decomposition and watermark embedding.
Embedding the watermark in low frequency in-
formation obtained by wavelet transform ensures
robustness of the proposed algorithm against at-
tacks that have low pass characteristics like resizing
and several geometric distortions. In addition, low
frequencies are affected less by video compression
operations.
4) Audio and video synchronisation using the time
code and delivery of protected video.
Watermark extraction and original video identification can
then be performed by
1) IDWT: the main of low frequency coefficients LL
between original and watermarked video frames are
calculated as
w(s) = 1if mean(new)−mean(original)>0
−1if mean(new)−mean(original)60
(3)
2) Watermark bits (fingerprint) are extracted.
3) Audio frame identification and comparison using the
same operations mentioned above.
IV. EXP ER IM EN TAL RESULTS
An input stereo wave signal is extracted from a recording
predestined to a Moroccan film then converted to mono audio
data. The .wav file (Fig. 4) of 11 seconds in length is sampled
at 44.1 kHz and quantised to 16 bits per sample (i.e. CD
quality).
A low pass filter is used with the aim to remove noise
and to enable better transmission over band limited networks.
Fig. 4. Input wave file.
Fig. 5. Filtered wave signal.
The analysed band range is 600Hz to 3khz to remove high
frequency components. The result is shown in Fig. 5.
Next, the audio signal is converted using a third level DWT
(Fig. 6). The audio signal is thus decomposed into 3 layers of
wavelet frequencies.
Fig. 6. Third level wavelet transform.
To reduce the data size, the filtered signal is divided into
frames of size equal to video frames for the watermarking pro-
cess. The audio partitions are analysed and the zero crossing
rate identified as illustrated in Fig. 7.
Frames are extracted from the same MJPEG2000 video
sequence from which we extracted the audio band. This frames
are compressed in JPEG2000. The DWT transform is applied
to each frame separately and the watermark is embedded in
the low frequencies components as shown in Fig. 8.
To evaluate the performanc of our algorithm we calculate
PSNR and BER [10], [11] in relation to some common
transformations (rotation and resizing and cropping attacks)
and in comparison with the techniques of [12] and [13]. The
results are shown in Table I.
Fig. 7. Input wave signal framing.
Fig. 8. Video watermaking using audio fingerprint.
The obtained results show that in comparison with previous
techniques, embedding the audio fingerprint as a watermark
enhances the robustness with respect to attacks that have
low pass characteristics like resizing, rotation and cropping.
Moreover, embedding watermark bits in low frequency regions
of frames intensifies the strength of the proposed algorithm
without any visible negative effect that could affect the image
characteristics and the requirement of imperceptivity.
V. CONCLUSIONS
The aim of the presented approach was to design a copyright
protection algorithm for multimedia content identification for
digital cinema content in Morocco. Our proposed algorithm
combines watermarking and fingerprinting techniques by ex-
tracting an audio fingerprint from video content for the purpose
of using it to identify copyright information in a reliable and
traceable fashion.
REFERENCES
[1] R. Moolman and W.C. Venter, Video fingerprinting using robust hashing
of scale invariant features of frames. Southern African Telecommunica-
tion Networks and Applications Conference, 2012.
[2] Z. Shahid, M. Chaumont and Wi. Puech, H.264/AVC video watermark-
ing for active fingerprinting based on Tardos code. Signal, Image and
Video Processing, 7(4) pp. 679–694, 2013.
[3] P. Cano, E. Batlle, T. Kalker and J. Haitsma, A review of audio
fingerprinting. Journal of VLSI Signal Processing 41, 271–284, 2005.
[4] RIAA/IFPI, Request for information on audio fingerprinting technolo-
gies. 2001.
[5] T. Kalker and J. Haitsma, Audio identification technology overview.
Journal of VLSI Signal Processing 41(3) pp. 271-284.
[6] T. Kalker, Applications and challenges for audio fingerprinting. 111th
AES Convention, 2001.
[7] F. Brodbeck , Cinemetrics. Royal Academy of Art, The Hague 2011.
TABLE I
EXP ERI ME NTAL RE SU LTS IN T ERM S OF PSNR A ND BER FO R TH E PRO POS ED A LGO RIT HM , [12] AND [13]
frame
PSNR BER
rotation cropping resizing rotation cropping resizing
new [12] [13] new [12] [13] new [12] [13] new [12] [13] new [12] [13] new [12] [13]
125 45.51 40,37 30.19 80.37 78.20 71.23 43.08 41.20 35.12 0.032 0.043 0.064 0.072 0.069 0.032 0.082 0.078 0.066
130 65.79 50.47 29.10 50.43 46.45 38.23 33,12 28.19 26.09 0.012 0.025 0.031 0.032 0.029 0.021 0.052 0.043 0.032
373 55.20 44.87 23.11 70.45 65.34 61.23 53.13 41.17 38.70 0.035 0.046 0.084 0.032 0.026 0.020 0.182 0.092 0.089
450 75.68 36.47 31.09 40.48 38.37 31.23 83.17 71.23 60.20 0.054 0.064 0.081 0.071 0.070 0.033 0.0162 0.010 0.009
[8] N.K. Dubey, A review of watermarking application in digital cinema
for piracy deterrence. 4th International Conference on Communication
Systems and Network Technologies, 2014.
[9] P. Singh and R.S. Chadra, A survey of digital watermarking techniques,
applications and attacks. International Journal of Engineering and Inno-
vative Technology, 2(9), 2013.
[10] M. Zhao and Y. Dang, Color image copyright protection digital water-
marking algorithm based on DWT and DCT. 4th International Confer-
ence on Wireless Communications, Networking and Mobile Computing,
2008.
[11] A. Hu and N. Chen, A blind watermarking algorithm for color image
based on wavelet transform and Fourier transform. 9th International
Conference for Young Computer Scientists, 2008.
[12] H. Kelkoul and Y. Zaz, Digital cinema watermarking State of art
and comparison. International Journal of Computer and Information
Engineering, 11(2), 2017.
[13] C. Sharma, G. Singh and G. Singh, Efficient video watermarking tech-
nique for quality loss of data. Indian Journal of Science and Technology,
9(47), 2016.
