A Video Dataset for an Efficient Camcording Attack Evaluation
Asma Kerbiche
1
, Saoussen Ben Jabra
1
, Ezzeddine Zagrouba
1
and Vincent Charvillat
2
1
Lab. LIMTIC, High Institute of Computer Science, University of Tunis El Manar,
2 Rue Abou Rayhane Bayrouni, 2080 Ariana, Tunisia
2
Lab IRIT, Team of Research VORTEX, ENSEEIHT, INP Toulouse, University of Toulouse, France
Keywords:
Dataset, Video Watermarking, Camcording Attack, Robustness Evaluation, Camcording Simulators.
Abstract:
Any video watermarking scheme dedicated to copyright protection should be robust against several attacks and
especially against malicious and dangerous attacks such as camcording. Indeed, this attack has become a real
problem for cinematographic production companies. However, until now the researchers don’t evaluate the
robustness of their video watermarking approaches against this attack or they consider it as a combination of
some usual attacks. To resolve this problem, several studies proposed camcording simulators which encourage
and help researchers in video watermarking domain to include the camcording in the robustness evaluation.
In this paper, a dataset of camcorder videos dedicated to an efficient robustness evaluation of watermarking
schemes is proposed which can help researches on camcording simulators’ creation. In this dataset, videos
are captured in realistic scenarios in the cinema and are recorded using five capture devices and from four
positions. In more, the proposed dataset contains marked versions of the proposed videos using three different
video watermarking techniques. This allows researchers comparing their approaches with these techniques.
Experimental results show that the robustness evaluation based on the proposed dataset is more efficient than
simulators based evaluation thanks to the diversity of the used capturing devices and the real conditions of
videos recording.
1 INTRODUCTION
With the evolution of hacking techniques and the
performance of smartphones that are, nowadays,
equipped with high quality cameras and powerful pro-
cessors allowing users to conveniently record, edit,
and share videos, movies’ hacking risks in screen-
ing rooms and movie theaters became a very dan-
gerous problem. Hence, video watermarking algo-
rithms that can resist to this kind of attack, called
”camcording”, and can help producers to protect their
copyrights should be developed. Unfortunately, un-
til now, researchers either still ignore this attack dur-
ing the watermarking evaluation, or they consider it
as a combination of usual attacks (rotation, compres-
sion, cropping ...) and this doesn’t reflect in any case
this attack. In order to facilitate and encourage re-
searchers to evaluate the robustness of their proposed
approaches against this attack, several researches are
developed to propose efficient camcording simula-
tors. Despite of their performance, these simulators
are insufficient to obtain a real comcorded video. In
this paper, we present the efficient methodology used
to collect a dataset of Camcorded videos captured in
realistic scenarios in a cinema using different captur-
ing devices placed in various positions. In addition,
some videos of this dataset are marked by three video
watermarking schemes in order to evaluate their ro-
bustness against camcording attack. The proposed
dataset can be used in order to help researchers to de-
velop efficient camcording simulations or to compare
their video watermarking schemes with existing tech-
niques used in the proposed dataset.
This paper is organized as follows: the first sec-
tion presents a survey of camcording attack where
we present the existing watermarking methods ded-
icated to this attack, the usual robustness evaluation
method and the developed simulators for camcording
attack. In the second section we explain the captur-
ing methodology and we give the characteristics of
the proposed dataset. In the section 3, the evaluation
of the proposed dataset is provided by comparing the
classical robustness evaluation and the proposed one.
In section 4, we enumerate some potentials applica-
tions of the proposed dataset. Finally, a conclusion
and some perspectives are drawn.
150
Kerbiche, A., Jabra, S., Zagrouba, E. and Charvillat, V.
A Video Dataset for an Efficient Camcording Attack Evaluation.
DOI: 10.5220/0006617201500158
In Proceedings of the 13th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2018) - Volume 4: VISAPP, pages
150-158
ISBN: 978-989-758-290-5
Copyright © 2018 by SCITEPRESS – Science and Technology Publications, Lda. All rights reserved
2 SURVEY OF CAMCORDING
ATTACK
Robustness is the most critical and important con-
straint that each watermarking algorithm must satisfy.
Despite the importance of this constraint, the majority
of researchers evaluate the robustness of their algo-
rithms against attacks that no longer represent a real
risk to the video stream and ignore new techniques
of piracy. In fact, a video watermarking algorithm
can resist to usual attacks such as noise or geomet-
ric transformations while it may be inefficient against
more dangerous attacks such as camcording.
This last one is a malicious attack that consists
of capturing a video projected on a screen or in a
movie theater using a given capture device like cam-
era, smartphone, and camcorder in order to broadcast
it in an illegal way after having applied some trans-
formations to destroy the mark that has been inserted.
This attack has become a major problem for film’s in-
dustry in these recent. Hence, the robustness of video
watermarking algorithms against this attack has be-
come a necessity.
The Motion Picture Association affirms that the
annual loss caused by pirated movies is 6.1 billion
dollars and that over 90% of the pirated new release
titles are illegal recordings made by camcorder piracy
(of America, 2005), (Association., ) which is explic-
itly banned in many countries by law.In the United
Sates, for example, the law of Family Entertainment
and Copyright Act, prohibits the uses of recording de-
vices in theatres. The law also imposes a strict penalty
on anyone who pre-release works not just films. Also
in Japan, to avoid the increasing of the loss of box-
office revenues, an anti-camcorder law, which pro-
hibits recording movies even for private uses, has
been enforced and then permitted by the previous
copyright law that encourages the movie industry to
prevent any person from making illegal recordings.
2.1 Existing Video Watermarking
Techniques Dedicated to
Camcording Attack
To avoid the camcorder piracy, several watermark-
ing techniques have been proposed in order to iden-
tify the illegally Camcording position (B. Chupeau
and Lefbvre, 2008). The research and development
unit of Kodak proposed a robust video watermarking
method which permits to identify the cinema where
the projection took place as well as the time and date
of broadcast (Chandramouli R., 2001). Gosavi et
al. (Gosavi and Mali., 2017) proposed a video wa-
termarking technique which aims to detect the cam-
corder piracy based on DCT transform where the po-
sition of pirate is estimated by comparing original
video frames with watermarked ones. Nakashima et
al. (Yuta Nakashima and Babaguchi., 2009) proposed
a deterrent to camcorder piracy, by developing a sys-
tem for estimating the recording position from which
a camcorder recording is made. The system is based
on spread-spectrum audio watermarking for the mul-
tichannel movie soundtrack. It utilizes a stochastic
model of the detection strength, which is calculated
in the watermark detection process. Experimental re-
sults show that the system estimates recording posi-
tions in an actual theater with a mean estimation error
of 0.44m. The results show that the method does not
significantly spoil the subjective acoustic quality of
the soundtrack. Lee et al. (Min-Jeong Lee and Lee.,
2010) proposed also a video watermarking based on
spread spectrum way that satisfies the requirements
for protecting digital cinema and it enables the de-
tector to estimate the position where the camcorder
recording was made. The proposed position estimat-
ing model can detect the seat in a theater with a mean
absolute error of 33.84, 9.53, 50.38 cm.
2.2 Camcording Evaluation for Video
Watermarking Methods
Only a few researchers have tried to evaluate the ro-
bustness of their approaches against the Camcording
attack. In order to make this evaluation, some con-
sider the Camcording as a combination of usual at-
tacks and simply test robustness against this combi-
nation. However, this combination can’t provide, in
any way, the real case of the application of Cam-
cording. On the other hand, some researchers test
the robustness against Camcording by capturing their
watermarked video projected on standard screen like
LCD.
Do et al. (Hoseok Do, 2008) propose a blind dig-
ital video watermarking scheme, robust to camcorder
recording and to a variety of common video pro-
cessing and geometric distortions. They test the ro-
bustness against Camcording by capturing the water-
marked videos without specifying the camcorder and
the projection’s screen models. They tested three sce-
narios: recorder, recorder with rotation and recorder
with cropping but they didn’t specify the parameters
of each attack. Choi et al. (Dooseop Choi, 2010) pro-
pose a new blind MPEG-2 video watermarking algo-
rithm robust to camcorder recording and other attacks.
They test the robustness of the proposed algorithm
against Camcording by making several recordings of
each video using a digital camcorder Sony, HDR-SR1
A Video Dataset for an Efficient Camcording Attack Evaluation
151
on a tripod 2.5m away from a 24-inch LCD screen
Dell, 2405FPW. Two scenarios are used for testing:
Recording 1: The recorded videos are resized to orig-
inal size and recompressed by Xvid, 1000 Kbit per
second. Recording 2: The recorded videos are re-
sized to original size and recompressed by Xvid, 500
Kbit per second. Asikuzzaman et al. (Md. Asikuz-
zaman and Pickering, 2014) propose a blind video
watermarking algorithm where the watermark is em-
bedded into both chrominance channels using a dual-
tree complex wavelet transform. This algorithm is
robust to downscaling in arbitrary resolution, aspect
ratio change, compression, and Camcording. The ro-
bustness against Camcording was tested by display-
ing watermarked video sequences at the rate of 25
fps and 30 fps on a 24-inch Samsung monitor and
recorded the content with an iPhone 4S. Li et al.
(Li Li, 2015) proposes an H.264/AVC HDTV water-
marking method that is robust to camcorder record-
ing, transcoding, recoding, and other geometric at-
tacks. They test the robustness against camcorder at-
tack by recording the watermarked video using a cam-
corder Sony HXR-MC1500C on a tripod 2 m away
from a 24 in. LCD monitor.
2.3 Review of Camcording Simulators
Most of robustness tests against Camcording attack
are carried out on a single screen model, using a sin-
gle capture model and applying, generally, only one
scenario of equipment’s disposition. In fact, bench-
marking of the camcorder path is far from being a
frequent practice today due to the heavy logistical ob-
stacles associated with this evaluation process. To
solve this problem, some researchers are now focus-
ing on the study of the impacts and distortions caused
by the Camcording with the aim of designing pre-
cise simulators for this attack. In addition, the re-
sults of this study and analysis could be reused to im-
prove video watermarking techniques. This process
has been adopted in some previous works to model the
printing and scanning process of still-image water-
marking and the acoustic path transmission but Cam-
cording is relatively not enough studied compared
to this works focusing only on spatial deformations.
Owing to the interaction between several devices, the
displayed content of the camcorder changes video in
some various ways, including temporal transforma-
tions, geometric distortions, variable and non-uniform
luminance transformations, alteration colors satura-
tion... It is necessary to understand the different phe-
nomena involved to design effective and precise sim-
ulators that imitate these effects.
Ben Zid et al. (Cherif Ben Zid and Doerr, 2013)
have study the luminance transforms due to the Cam-
cording process and investigate three different alter-
ations which are the spatial non-uniformity, the steady
state luminance response, and the transient luminance
response. To do this, they performed several con-
trolled experiments where they simulated different
configurations of the Camcording process. They used
two alternative displays and one camcorder device
which are a 24 ”LCD monitor Dell U241014, a home
theater video projector, Christie HD5Kc15 and a Sony
HDR-CX200ETM camcorder. They then excited the
system with several visual stimuli and looked at the
recorded answers to infer the underlying mechanisms
that take place as well as their characteristics. This
study can be improved because it is focused on only
three distortions that video content undergoes along
Camcording process and uses only two displays and
one capturing device.
Hajj-Ahmad et al. (Adi Hajj-Ahmad and Wu,
2017) have lead an investigation of the lumi-
nance flicker that is naturally present in camcorded
recordings due to the interplay between liquid-
crystaldisplay (LCD) screen and camcorder. To do
this, they have break down the acquisition pipeline
into three stages which are the emission of a back-
light signal by the screen, the integration of the light
emitted by the screen with a sensor of the camcorder,
and the sequential sampling of the different rows of
a video frame. They initially model the flicker signal
and demonstrate that its parameters are related to such
internal characteristics of the capture devices as the
back-light frequency of the LCD screen and the read-
out time of the camcorder. Then, they introduce an
estimation strategy to recuperate these hidden param-
eters directly from camcorded recordings and demon-
strate that such forensic cues could provide intelli-
gence on the pirate devices. They additionally dis-
cuss on how to recuperate the shape of the low power
flicker signal and demonstrate that it could be used to
infer which back-light technology employed in the pi-
rate LCD screen. The authors set out the prospects to
better understand the applicability of flicker forensics
which will involve large scale validation experiments
with a wide diversity of devices, hence the utility of
our proposed dataset.
The most complete simulator that is already avail-
able as open source tool to researchers is the Cam-
Mark developed by Schaber et al. (P. Schaber, 2014).
This tool simulates a re-acquisition of a video from
a camcorder to support watermarking development
by enabling automated test cases for such camcorder
copy attacks (fig 1). The authors are thus trying to
simulate the typical artifacts of a camcorder captur-
ing: geometric modifications (aspect ratio changes,
VISAPP 2018 - International Conference on Computer Vision Theory and Applications
152
cropping, perspective and lens distortion), tempo-
ral modifications (unsynchronized frame rates and
the resulting frame blending), sub-sampling (rescal-
ing, filtering, Bayer color array filter), and histogram
changes (AGC, AWB), camera movement (e.g., a
hand-held camera) and background insertion. To do
this, they developed separate models for the various
effects that appear at various stages of the Camcord-
ing process and apply their model in a natural order
that resembles the actual physical order. In fact, there
are effects that are caused from display like crop-
ping, AR changes and visible parts of display, from
optics they consider the perspective distortion, lens
distortion and movement, from shutter there is frame
blending due to frame rate changes, from sensor there
is Resolution changes and bayer pattern interpolation
and finally the processing induces automatic gain con-
trol and automatic white balance. The results ob-
tained by CamMark are comparable to actual cam-
corder copies but cannot yet represent a real case of
camcording process. The authors set out the prospects
to consider the characteristics of different displaying
devices in future releases of CamMark tool. More-
over, this tool is not easy to use. In fact, it is based on
command line and the graphical user interface was a
part of the original first version and was removed in
version 1.1 and we are still waiting for a new Cam-
Mark interface which will offer a web-based interface
to an entirely server-side processing.
Figure 1: CamMark simulation.
3 PROPOSED DATASET FOR
CAMCORDING VIDEOS
In this paper, we propose a video dataset dedicated to
evaluate the robustness of watermarking approaches
against camcording attack. This dataset can help re-
searchers to build their own camcording simulators by
studying the distortion caused by camcorded videos.
In more, it gives the researchers the possibility to
compare their approaches of video watermarking with
the robustness of tested techniques. In fact, the pro-
posed dataset contains different videos (marked and
not marked) recorded from different positions in a
movie theater using different capturing devices.
3.1 Chosen Benchmark Videos
The first step in the dataset building is the choice of
test videos. Indeed, we have chosen five color free
right video (fig2) which present the most sensitive
scenes for a video watermarking. The first video is the
sequence Stephan which is a benchmark for video wa-
termarking algorithms evaluation and researches on
video processing. The second video is a sequence
of the famous cartoon HD ”Big Buck Bunny”. The
third one is a sequence of a HD cartoon ”Sintel”
which presents complex scenes with a fast luminance
changing. The fourth video is a video clip ”Varoshse-
quence” which presents a rock concert. The last one
is an Italian advertising video ”Segugio”. The charac-
teristics of these videos are presented in Table 1.
Table 1: Video tests characteristics.
Videos Resolution Size Duration
Stephan SD
346*280
7.29 Mo 0 :12 sec
Big Buck
Bunny
SD
854*480
6.08 Mo 0 :20 sec
Sintel HD
1280*720
10 Mo 0:16 sec
Varosh Se-
quence
HD
1280*720
9.94 Mo 0:16 sec
Segugio HD
1280*720
9.63 Mo 0:16 sec
Figure 2: Chosen videos.
Subsequently, in addition of the original versions,
we have marked these videos by using three differ-
ent video watermarking algorithms: multi-frequency
insertion (DWT-SVD-DCT) and feature region based
approach proposed by Kerbiche et al. (A. Kerbiche,
A Video Dataset for an Efficient Camcording Attack Evaluation
153
2012), the algorithm proposed by Agilandeeswari et
al. (L. Agilandeeswari, 2013) and based on the
two transforms DWT-SVD, and spatial algorithm pro-
posed by Datta et al. (S. Datta, 2014) which is
based on the LSB method. We choose to test three
robust and efficient video watermarking algorithms
which embed signature in different domain of inser-
tion (multi-frequency and spatial) in order to study
their robustness against a realistic case of camcording
attack. In addition, these camcorded marked videos
can be used by researches to compare their propos-
als with these videos. Moreover, the usefulness of
these camcorded watermarking versions, was to eval-
uate the proposed dataset and compare it with existing
camcording methods, and this by testing their robust-
ness.
3.2 Used Devices
The materials and capture devices that we used to
record the chosen videos are: 3 tripods, laser meter,
two smartphones (Samsung galaxy note 2 and iPhone
4s), a webcam Logitech V-u0028, a Panasonic camera
HDC-tm900 and a Canon camera EOS 450D. Each
capturing device has at least 16 GB of storage ca-
pacity and the videos are recorded with frame rate
ranging from 20 to 30 frames per second. The char-
acteristics of these capture devices are illustrated in
Table 2.
Table 2: Capture devices characteristics.
Smartphones
Processor Camera Resolution
Galaxy
Note 2
1.6 Ghz 8 Mpx 1920×1080
Quad core
2Go Ram
Iphone 4s 1 Ghz 8 Mpx 1920×1080
Apple A5
Cameras
Capturing system Resolution
Panasonic
HDC-
tm900
3MOS (2 × 2.53 Mpx) 1920×1080
Canon
EOS
450D
CMOS (12.2 Mpx) 1280×720
Webcam
Logitech
V-u0028
5 Mpx 800×600
3.3 Capturing Scenarios
Once videos are ready we displace to the Utopia cin-
ema in Toulouse which was at our disposal for 2
hours. This room is 9.78m in width, 21.7m in length,
2.84m height at the back and 3.67m in the middle. It
is composed of 224 seats (14 × 16) and the size of the
projection screen is 8.16 meters (fig3).
Figure 3: Utopia cinema.
Turning to the capture step, to camcorder our
videos we have placed the capturing devices in sev-
eral places (fig4): in front, in the left, in the right and
in the projection room given that the employees of
the cinema affirmed that the majority of camcordred
videos are captured from the projection room. For this
reason we used two different scenarios:
Concerning the first one, we exposed our captur-
ing devices as follows (fig5):
1. The Panasonic camera to the right of the screen to
12m of the projection screen, 50cm from the wall
and 1.50m height.
2. The Canon camera to the left of the screen, to
16.6m of the projection screen, 1m from the wall
and 1.20m height.
3. Samsung Galaxy Note 2 in front of the screen
to 10.5m of the projection screen, 4.5m from the
wall and 1.50m height.
4. IPhone 4s in the left of the screen to 6.34 m of
the projection screen, 1m from the wall and 1.80m
height.
VISAPP 2018 - International Conference on Computer Vision Theory and Applications
154
Figure 4: Videos’ camcordering.
5. Logitech Webcam in the projection room in front
of the screen at 19m, 4.5m from the wall and 2.5m
height.
Figure 5: Capturing’s Scenario 1.
In the second scenario, we exposed our capturing
devices as follows (fig6):
1. The Panasonic camera in the projection room in
front of the screen at 19m, 4.5m from the wall and
2.5m height.
2. Samsung Galaxy Note 2 in the right of the screen
to 13.5m of the projection screen, 1.5m from the
wall and 0.95m height.
3. IPhone 4s in the right of the screen to 13m of the
projection screen, 1.3m from the wall and 0.95m
height.
4. Logitech Webcam in front of the screen at 15m,
4.5m from the wall and 1.10m height.
3.4 Dataset Characteristics
The proposed dataset contains 180 camcordred wa-
termarked videos. Figure 7 presents some of those
Figure 6: Scenario 2.
videos: (a) Segugio video camcordred with Panasonic
in the right, (b) Sintel video camcordred with Pana-
sonic in the projection room, (c) Big Buck Bunny
video camcordred with Samsung Galaxy Note2 in the
right, (d) Stephan video camcordred in the right with
IPhone 4s and (e) VaroshSequence camcordred in the
front with webcam.
Camcordred videos by IPhone 4s and Canon cam-
era are .mov files which is a common multimedia con-
tainer file format developed by Apple and compati-
ble with both Macintosh and Windows platforms and
it commonly uses the MPEG-4 codec for compres-
sion. The video obtained by the Panasonic camera are
.MTS files which is a file extension for an AVCHD
(Advanced Video Coding High Definition) video clip
format for high-definition video. The MTS file format
supports 1080i (a high definition video format with
1080 horizontal scan lines, interlaced) and 720p (720
horizontal scan lines, progressive scan, rather than in-
terlaced) in a relatively small file size. AVCHD files
are based on the MPEG 4 codec. Finally, the video
obtained by the webcam and the Samsung Galaxy
note 2 are .mp4 files. Table3 presents the characteris-
tics of the camcordred videos.
Table 3: Camcordred videos’ characteristics.
Format Resolution
Panasonic .mts 1920 × 1080
Canon .mov 1280 × 720
IPhone 4s .mov 1920 × 1080
Galaxy note 2 .mp4 1920 × 1080
Webcam .mp4 800 × 600
The whole dataset will be soon available
on our website http://camcordingvideos.github.io/
dataset.html for download or you can drop an email
to the authors. The dataset website allows users to
search videos by watermarking algorithm (not water-
marked version, algo1, algo2, algo3) and capturing
device. The users may choose to download all result-
ing videos, or select a subset from them.
A Video Dataset for an Efficient Camcording Attack Evaluation
155
Figure 7: Camcordred videos.
4 DATASET EVALUATION
The proposed dataset was evaluated based on robust-
ness results obtained after the evaluation of the three
watermarking approaches which were applied on the
proposed benchmark videos. Indeed, in order to prove
the efficiency of the proposed dataset, we compare
the obtained results with those obtained after applying
other techniques of camcording which are: the Cam-
Mark simulator and the results obtained after cam-
cording with a Smartphone (Samsung note 2) using a
monitor screen Dell Professional 469-3134 19” LED
LCD. Figure 8 presents camcordred versions of the
watermarked video Stephan with these three cam-
cording techniques. In more, we compare these re-
sults with those cited in the papers corresponding to
the tested approaches (given in table 4) in order to
prove that usual evaluation based on classic attacks
only, can not reflect the real performance of video wa-
termarking algorithms.
The robustness was evaluated on five videos ob-
tained from the proposed dataset:
The Video Bug Buck Bunny camcordred in the
projection room with the Panasonic camera.
The video Stphan camcordred in the right of the
projection screen with the smartphone Samsung
Galaxy note2.
The video Sintel camcordred in the left of the
Figure 8: Different versions of camcorded Stephan video.
Table 4: Performances of the watermarking algorithms as
cited in their papers.
Algorithms (A. Ker-
biche,
2012)
(L. Ag-
ilan-
deeswari,
2013)
(S. Datta,
2014)
Robust- MPEG4 Good Good Poor
ness Collusion Good Poor Poor
Rotation Good Good Good
Cropping Good Good Good
Noise Good Good Poor
Invisibility Good Good Good
screen projector with IPhone 4s.
The video VaroshSequence camcordred in front of
the screen projector with the Logitech webcam.
The video Segugio camcordred in the left of the
screen projector with the Canon camera.
For the CamMark simulator we have used the
standard configuration presented in figure 9 and pro-
jected on the three simulated screens: SonyHDR-
TD20 Monitor BrightEnv, SonyHDR-TD20 Projec-
tion BrightEnv and SonyHDR-TD20 Projection Dark-
Env. Then, the robustness was tested on these videos:
The Video Bug Buck Bunny simulated with the
projector SonyHDR-TD20 Monitor BrightEnv.
The video Stephan SonyHDR-TD20 Projection
BrightEnv.
The video Sintel SonyHDR-TD20 Projection
DarkEnv.
The video VaroshSequence SonyHDR-TD20
Monitor BrightEnv.
The video Segugio SonyHDR-TD20 Projection
DarkEnv.
VISAPP 2018 - International Conference on Computer Vision Theory and Applications
156
Finally, for the third type of camcording, we have
camcorded the same videos and positions used for
the test of the proposed dataset using the smartphone
Samsung note 2 and a LCD screen monitor. The ob-
tained robustness results are shown in the table 5.
Figure 9: Configuration of the CamMark simulator.
Table 5: Comparative study of Camcording techniques
based on robustness of the tested watermarking algorithms.
Algorithms (A. Ker-
biche,
2012)
(L. Ag-
ilan-
deeswari,
2013)
(S. Datta,
2014)
Proposed
dataset
(a) X X -
(b) - - -
(c) X X -
(d) X X -
(e) X - -
CamMark (a) X X X
(b) X X -
(c) X X X
(d) - X -
(e) X - -
Camcording (a) X X X
on a LSD (b) X - -
screen (c) X X -
monitor (d) X X -
(e) - X -
According to this comparative study, we can no-
tice that despite the performance proved by the au-
thors of the tested watermarking algorithms and the
good robustness results obtained with usual attacks,
some algorithms have not resist to camcording at-
tack and we have failed to detect the presence of the
mark on several tested videos whatever the camcord-
ing techniques used. In fact, the algorithm proposed
by Datta et al. (S. Datta, 2014), which is based on
the insertion in spatial domain, doesn’t succeed to de-
tect the mark on any video while for the algorithm
proposed by Kerbiche et al. (A. Kerbiche, 2012) we
were able to detect the presence of the mark on the
most of selected videos.
Moreover, we can also notice that for each cam-
cording technique we have obtained different results.
In fact, we success to detect the inserted mark on
some videos camcordred by the CamMark simulator
and by using the standard capturing scenario (Cam-
cording on a LCD screen monitor) but we fail to de-
tect it on the dataset’s video. In fact, for the algorithm
proposed by Datta et al. (S. Datta, 2014) the detector
fails to extract the mark for the five dataset’s videos
but successes for the CamMark simulated video and
the camcorded video on a LCD screen. These results
are logical as the proposed dataset represents the most
realistic case of camcording which causes a variety of
distortions that are not yet well simulated. In addition,
the test of camcording using only a standard screen
monitor and using one capture device is insufficient
to evaluate the robustness against this attack.
5 POTENTIAL APPLICATIONS
Among applications that can use the proposed dataset
and benefit from the availability of such a dataset, we
can enumerate:
Evaluation protocols of video watermarking algo-
rithms: The proposed dataset can be used in or-
der to compare the efficiency of video watermark-
ing algorithms against the Camcording attacks.
In fact, the captured watermarked videos can be
used directly by video watermarking researches
to compare their algorithms directly with the wa-
termarked videos in the proposed dataset or after
applying combination of attacks on them.
Camcorder’s simulators development: This
dataset can greatly helps the studies on camcorder
simulators. In fact, these videos are captured with
several capture devices and with different dispo-
sition scenarios of these tools, which will allows
researchers in this field to study the effects and
impacts caused by camcorder process and try to
simulate them.
A Video Dataset for an Efficient Camcording Attack Evaluation
157
6 CONCLUSION
This paper presents a survey of Camcording attack
and proposes a camcordred videos dataset which con-
tains recorded videos from a movie theater with sev-
eral capture devices and from different shooting. This
dataset can be useful for the comparative study of
video watermarking algorithms. In fact, it will allow
researchers comparing their approaches with the tech-
niques tested in the proposed dataset. Moreover, ex-
perimental results show that the robustness evaluation
based on the proposed dataset is more realist and effi-
cient than the evaluation based on Camcording simu-
lators thanks to the diversity of the used capturing de-
vices and the real conditions of videos recording. For
this reason, the proposed dataset is useful, specially,
to improve the camcorder’s simulators development
by helping researches studding the impact of Cam-
cording and the caused distortion using several cap-
ture devices on videos. In future, we intend to carry
out an in-depth study of the distortions caused by the
camcording on the dataset’s videos.
REFERENCES
A. Kerbiche, Saoussen Ben Jabra, E. Z. ((2012)). A ro-
bust video watermarking based on image mosacing
and multi-frequential embedding. In IEEE Interna-
tional Conference on Intelligent Computer Communi-
cation and Processing,.
Adi Hajj-Ahmad, Sverine Baudry, B. C. G. D. and Wu,
M. (January (2017)). Flicker forensics for camcorder
piracy. In IEEE Transactions on information forensics
and security, volume 12.
Association., M. P. Anti-piracy fact
sheet asia-pacific region. In
[Online]http://www.mpaa.org/AsiaPacificPiracyFact-
Sheet.pdf.
B. Chupeau, A. M. and Lefbvre, F. (Jan. (2008)). In-theater
piracy: Finding where the pirate was. In in Proc. SPIE
Security, Forensics, Steganography, and Watermark-
ing of Multimedia Contents X.
Chandramouli R., Memon N., R. M. (2001). Invisible wa-
termarking for digital cinema. In Digital watermark-
ing.
Cherif Ben Zid, Severine Baudry, B. C. and Doerr, G.
(March (2013)). A sneak peek into the camcorder
path. In Proc. SPIE 8665, Media Watermarking, Se-
curity, and Forensics.
Dooseop Choi, Hoseok Do, H. C. T. K. ((2010)). A
blind mpeg-2 video watermarking robust to camcorder
recording. In Journal of Signal Processing, volume
90, pp 13271332.
Gosavi, C. S. and Mali., S. N. (May (2017)). Secure, ro-
bust video watermarking to prevent camcorder piracy.
In Indian Journal of Science and Technology,, vol-
ume 10.
Hoseok Do, Dooseop Choit, H. C. T. K. ((2008)). Digital
video watermarking based on histogram and tempo-
ral modulation and robust to camcorder recording. In
IEEE International Symposium on on Signal Process-
ing and Information Technology ISSPIT.
L. Agilandeeswari, K. M. ((2013)). A robust video wa-
termarking algorithm for content authentication using
discrete wavelet transform (dwt) and singular value
decomposition (svd). In International Journal of Se-
curity and Its Applications,.
Li Li, Zihui Dong, J. L. J. D. Q. H. C.-C. C. T. W. ((2015)).
An h.264/avc hdtv watermarking algorithm robust to
camcorder recording. In Journal of Visual Communi-
cation and Image Representation, volume 26, pp 1-8.
Md. Asikuzzaman, Md. Jahangir Alam, A. J. L. and Pick-
ering, M. R. ((2014)). A blind and robust video wa-
termarking scheme using chrominance embedding. In
International Conference on Digital Image Comput-
ing: Techniques and Applications (DICTA).
Min-Jeong Lee, K.-S. K. and Lee., H.-K. (November
(2010)). Digital cinema watermarking for estimating
the position of the pirate. In IEEE Transactions on
multimedia, volume 12.
of America, M. P. A. (2005). Us piracy fact sheet. In [On-
line]. http://www.mpaa.org/USPiracyFactSheet pdf.
P. Schaber, S. Kopf, C. W. W. E. ((2014)). A camcorder
copy simulation as watermarking benchmark for digi-
tal video. In ACM Multimedia Systems Conference.
S. Datta, A. N. ((2014)). Data authentication using digi-
talwatermarking. In International Journal of Advance
Researchin Computer Science and Management Stud-
ies,.
Yuta Nakashima, R. T. and Babaguchi., N. (April (2009)).
Watermarked movie soundtrack finds the position of
the camcorder in a theater. In IEEE Transactions on
multimedia, volume 11.
VISAPP 2018 - International Conference on Computer Vision Theory and Applications
158