S.A.kalaiselvan- robust video data hiding at forbidden zone

National Conference on Advances in Computer Science and Information Technology ISBN:978-81-924031-8-2
1
9 7 8 8 1 9 2 4 0 3 1 8 2
Robust Video Data Hiding at Forbidden Zone
Using Stegnography1
Kalaiselvan SA,2
Karthick V
1
Assistant professor/CSE,Arignar Anna Institute of Science & Technology,Sriperumbudur-
2
II Year M.Tech(IT),Bharath University,Chennai -
1
kalaiselvan_be@ymail.com, 2
vkarthick86@gmail.com
Abstract—Video data hiding is still an important research topic
due to the design complexities involved. We propose a new
video data hiding method that makes use of erasure correction
capability of repeat accumulate codes and superiority of
forbidden zone data hiding. Selective embedding is utilized in
the proposed method to determine host signal samples suitable
for data hiding. This method also contains a temporal
synchronization scheme in order to withstand frame drop and
insert attacks. The proposed framework is tested by typical
broadcast material against MPEG- 2, H.264 compression,
frame-rate conversion attacks, as well as other well-known
video data hiding methods. The decoding error values are
reported for typical system parameters. The simulation results
indicate that the framework can be successfully utilized in video
data hiding applications.
Index Terms—Data hiding, digital watermarking, forbidden
zone data hiding, quantization index modulation, repeat
accumulate codes, selective embedding..
I.INTRODUCTION
DATA HIDING is the process of embedding information
into a host medium. In general, visual and arual media are
preferred due to their wide presence and the tolerance of
human perceptual systems involved. Although the general
structure of data hiding process does not depend on the host
media type, the methods vary depending on the nature of
such media. For instance, image and video data hiding share
many common points; however video data hiding
necessitates more complex designs [6], [7] as a result of the
additional temporal dimension. Therefore, video data hiding
continues to constitutean active research area. Data hiding in
video sequences is performed in two major ways: bitstream-
level and data-level. In bitstream-level, the redundancies
within the current compression standards are exploited.
Typically, encoders have various options during encoding
and this freedom of selection is suitable for manipulation
with the aim of data hiding. However, these methods highly
rely on the structure of the bitstream; hence, they quite
ragile, in the sense that in many cases they cannot survive
any format conversion or transcoding, even without any
significant loss of perceptual quality. As a result, this type of
data hiding methods is generally proposed for fragile
applications, such as authentication. On the other hand,
datalevel methods are more robust to attacks. Therefore, they
are suitable for a broader range of applications. Despite their
fragility, the bitstream-based methods are still attractive for
data hiding applications. For instance, in [1], the redundancy
in block size selection of H.264 encoding is exploited for
hiding data. In another approach [17], the quantization
parameter and discrete cosine transform (DCT) coefficients
are altered in the bitstream-level. However, most of the
video data hiding methods utilize uncompressed video data.
Sarkar et al. [2] proposed a high volume transform domain
data hiding in MPEG-2 videos. They applied quantization
index modulation (QIM) to low frequency DCT coefficients
and adapted the quantization parameter based on MPEG-2
parameters. Furthermore, they varied the embedding rate
depending on the type of the frame. As a result, insertions
and erasures occur at the decoder, which causes de-
synchronization. They utilized repeat accumulate (RA) codes
in order to withstand erasures. Since they adapted the
parameters according to type of frame, each frame is
processed separately. RA codes are already applied in image
data hiding. In [3], adaptive block selection results in de-
synchronization and they utilized RA codes to handle
erasures. Insertions and erasures can be also handled by
convolutional codes as in [4]. The authors used
convolutional codes at embedder. However, the burden is
placed on the decoder. Multiple parallel Viterbi decoders are
used to correct de-synchronization errors. However, it is
observed [4] that such a scheme is successful when the
number of selected host signal samples is much less than the
total number of host signal samples. In [5], 3-D DWT
domain is used to hide data. They use LL subband
coefficients and do not perform any adaptive selection.
Therefore, they do not use error correction codes robust to
erasures. Instead, they use BCH code to increase error
correction capability. The authors performed 3-D
interleaving in order to get rid of local burst of errors.
Additionally, they proposed a temporal synchronization

2
9 7 8 8 1 9 2 4 0 3 1 8 2
technique to cope with temporal attacks, such as frame drop,
insert, and repeat. In this paper, we propose a new block-
based selective embedding type data hiding framework that
encapsulates forbidden zone data hiding (FZDH) [8] and RA
codes in accor- dance with an additional temporal
synchronization mechanism. FZDH is a practical data hiding
method, which is shown to be superior to the conventional
QIM [9]. RA codes are already used in image [3] and video
[2] data hiding due to their robustness against erasures. This
robustness allows handling de-synchronization between
embedder and decoder that occurs as a result of the
differences in the selected coefficients. In order to
incorporate frame synchronization markers, we partition the
blocks into two groups. One group is used for frame marker
embedding and the other is used for message bits. By means
of simple rules applied to the frame markers, we introduce
certain level of robustness against frame drop, repeat and
insert attacks. We utilize systematic RA codes to encode
message bits and frame marker bits. Each bit is associated
with a block residing in a group of frames. Random
interleaving is performed spatio-temporally; hence,
dependency on local characteristics is reduced. Host signal
coefficients used for data hiding are selected at four stages.
First, frame selection is performed. Frames with sufficient
number of blocks are selected. Next, only some
predetermined low frequency DCT coefficients are permitted
to hide data. Then the average energy of the block is
expected to be greater than a predetermined threshold. In the
final stage, the energy of each coefficient is compared
against another threshold. The unselected blocks are labeled
as erasures and they are not processed. For each selected
block, there exists variable number of coefficients. These
coefficients are used to embed and decode single message bit
by employing multidimensional form of FZDH that uses
cubic lattice as its base quantizer. We describe the utilized
data hiding method in Section II. Then the proposed video
data hiding method is presented in Section III. Experiment
results are given in Section IV, which is followed by the
concluding remarks. II. Forbidden Zone Data Hiding
Forbidden zone data hiding (FZDH) is introduced in [8]. The
method depends on the forbidden zone (FZ) concept, which
is defined as the host signal range where no alteration is
allowed during data hiding process. FZDH makes use of FZ
to adjust the robustness-invisibility tradeoff. Let s (bold
denoting a vector) be the host signal in RN and mC{0, 1} be
the data to be hidden. Then the marked signal x is obtained
as given in x = s, s ∈ FZm Mm(s), s ∈ AZm (1) where FZm,
allowed zone (AZm) pair defines the host signal zones where
alteration is allowed or not and Mm(.) is a mapping from RN
to a suitable partition of RN. The requirement on these zones
and partitions is simply based on the constraint that they
should be mutually exclusive for different m. The key point
of FZDH is the determination of the zonesand the partitions.
There could be infinite ways to achieve this; however, a
practical design can be performed by using quantizers. Such
a simple parametric form is given in (2),
Fig. 1. Sample embedding function of FZDH in 1-D. ci is a reconstruction
point of the quantizer.
where the mapping function is defined as
2) Here r is the control parameter, qm(.) is a quantizer
indexed by m, and e is defined as the difference vector
between the host signal and its quantized version em The
mapping function in (2) states that the host signal is
modified by adding an additional term, which is a scaled
version of the quantization difference. In 1-D, this additional
term is scalar, whereas in N-D host signal is moved along
the quantization difference vector and toward the
reconstruction point of the quantizer. Hence, embedding
distortion is reduced and became smaller than the
quantization error. FZm and AZm are defined using the control
parameter and the difference vector FZm = {s| _em_ ≤ r}, AZm
= {s| _em_ > r}. (4) In order to fulfill the requirement of
mutual exclusion, the reconstruction points of the quantizers
that are indexed by different m should be non-overlapping,
which can be achieved by using a base quantizer and shifting
its reconstruction points depending on m, similar to Dither
Modulation [9]. A typical embedding function that uses a
uniform quantizer is shown in Fig. 1. During data extraction
step, the generic minimum distance
decoder is utilized to decode the hidden data mˆ = argmin
md(y, ym) (5) where y is the received signal, ym is equal to its
FZDH embedding operation applied version as in (1), and
d(.,.) is a suitable distance metric. The decoder and

3
9 7 8 8 1 9 2 4 0 3 1 8 2
embedder should be synchronized in terms of the zones,
partitions, and system parameters. For soft decoding,
channel observation probabilities (om) are computed using
the distances (6)
Fig. 2. Embedder flowchart of the proposed video data hiding framework
for a single frame.
III. Proposed Video Data Hiding Framework
We propose a block based adaptive video data
hiding method that incorporates FZDH, which is shown to
be superior to QIM and competitive with DC-QIM [8], and
erasure handling through RA Codes. We utilize selective
embedding to determine which host signal coefficients will
be used in data hiding as in [3]. Unlike the method in [3], we
employ block selection (entropy selection scheme [3]) and
coefficient selection (selectively embedding in coefficients
scheme [3]) together. The de-synchronization due to block
selection is handled via RA Codes as in [2] and [3]. The de-
synchronization due to coefficient selection is handled by
using multi-dimensional form of FZDH in varying
dimensions. In [2], the frames are processed independently.
It is observed that [10] intra and inter frames do not yield
significant differences. Therefore in order to overcome local
bursts of error, we utilize 3-D interleaving similar to [5],
which does not utilize selective embedding, but uses the
whole LL subband of discrete wavelet transform.
Furthermore, as in [5], we equip the method with frame
synchronization markers in order to handle frame drop,
insert, or repeat attacks. Hence, it can be stated the original
contribution of this paper is to devise a complete video data
hiding method that is resistant to de-synchronization due to
selective embedding and robust to temporal attacks, while
making use of the superiority of FZDH.
A. Framework
The embedding operation for a single frame is
shown in Fig. 2. Y-channel is utilized for data embedding. In
the first step, frame selection is performed and the selected
frames are processed block-wise. For each block, only a
single bit is hidden. After obtaining 8 × 8 DCT of the block,
energy check is performed on the coefficients that are
predefined in a mask. Selected coefficients of variable length
are used to hide data bit m. m is a member of message bits or
frame synchronization markers. Message sequence of each
group is obtained by using RA codes for T consecutive
frames. Each block is assigned to one of these groups at the
beginning. After the inverse transform host frame is
obtained.
Fig. 3. Decoder flowchart of the proposed video data hiding
framework for
a single frame.
Decoder is the dual of the embedder, with the
exception that frame selection is not performed. Fig. 3 shows
the flowchart for a single frame. Marked frames are detected
by using frame synchronization markers. Decoder employs
the same system parameters and determines the marked
signal values that will be fed to data extraction step. Non-
selected blocks are handled as erasures. Erasures and
decoded message data probabilities (om) are passed to RA
decoder for T consecutive frames as awhole and then the
hidden data is decoded. B.
Selective Embedding
Host signal samples, which will be used in data
hiding, are determined adaptively. The selection is
performed at four stages: frame selection, frequency band
determination, block selection, and coefficient selection.
1) Frame selection: selected number of blocks in the whole
frame is counted. If the ratio of selected blocks to all blocks
is above a certain value (T0) the frame is processed.
Otherwise, this frame is skipped.
2) Frequency band: only certain DCT coefficients are
utilized. Middle frequency band of DCT coefficients shown
in Fig. 4 is utilized similar to [2].

4
9 7 8 8 1 9 2 4 0 3 1 8 2
3) Block selection: energy of the coefficients in the mask is
computed. If the energy of the block is above a certain
Fig. 4. Sample coefficient mask denoting the selected frequency band.
Fig. 5. Typical block partitioning for message bits and frame
synchronization markers.
value (T1) then the block is processed. Otherwise, it is
skipped.
4) Coefficient selection: energy of each coefficient is
compared to another threshold T2. If the energy is above T ,
then it is used during data embedding together with other
selected coefficients in the same block.
C. Block Partitioning
Two disjoint data sets are embedded: message bits
(m1) and frame synchronization markers (m2). The block
locations of m2 are determined randomly depending on a
random key. The rest of the blocks are reserved for m1. The
same partitioning is used for all frames. A typical
partitioning is shown in Fig. 5. m2 is embedded frame by
frame. On the other hand, m1 is dispersed to T consecutive
frames. Both of them are obtained as the outcomes of the RA
encoder.
D. Erasure Handling
Due to adaptive block selection, de-synchronization
occurs between embedder and decoder. As a result of attacks
or even embedding operation decoder may not perfectly
determine the selected blocks at the embedder. In order to
overcome this problem, error correction codes resilient to
erasures, such as RA codes are used in image [3] and video
[2] data hiding in previous efforts. RA code is a low
complexity turbo-like code [11]. It is composed of repetition
code, interleaver, and a convolutional encoder. The source
bits (u) are repeated R times and randomly permuted
depending on a key. The interleaved sequence is passed
through a convolutional encoder with a transfer function 1/(1
+ D), where D represents a first-order delay.
Fig. 6. RA encoder (u denotes source bits and u + v denote encoded bits).
In systematic RA code, input is placed at the
beginning of the output as shown in Fig. 6. In this paper, we
utilize systematic RA codes to obtain m1 as u1+v1 and m2 as
u2+v2. Here, u1 denotes the uncoded message bits and u2 is
the uncoded frame synchronization marker bits. RA code is
decoded using sum-product algorithm. We utilize the
message passing algorithm given in [12].
E. Frame Synchronization Markers
Each frame within a group of T consecutive frames
is assigned a local frame index starting from 0 to T − 1.
These markers are used to determine the frame drops, inserts
and repeats, as well as the end of the group of frames at
which point all necessary message bits are available for RA
decoder. Frame indices are represented by K2 bits. After RA
encoder RK2 bits are obtained. Hence, RK2 blocks are
reserved for frame markers. K2 >> log2T, so that a small
portion of 2K2 codewords is valid. Therefore, we can detect
the valid frames with higher probability. Using the
sequential frame index information, the robustness increases.
Furthermore, RA code spreads the output codewords of the
adjacent frame indices; hence, errors are less likely to occur
when decoding adjacent frame indices. Once one reserves
RK2 blocks for frame markers, T (N − RK2) blocks remain for
message bits. Then, the actual number of message bits (K1)
becomes equal to _T (N − RK2)/R_, where _·_ denotes floor
operation. The remaining blocks at the end of last frame are
left untouched.
F. Soft Decoding
At the decoder, a data structure of length RK1 is
kept for channel observation probability values, om. The
structure is initialized with erasures (om = 0.5 for m = 0 and
m = 1). At each frame, frame synchronization markers are
decoded first. Message decoding is performed once the end
of the group of frames is detected. Two frame index values
are stored: current and previous indices. Let fcur and fpre

5
9 7 8 8 1 9 2 4 0 3 1 8 2
denote the current and previous frame indices, respectively.
Then the following rules are used to decode u1. 1) If fcur > T,
then skip this frame. (This case corresponds to unmarked
frame.) 2) If fcur = fpre, then skip this frame. (This case
corresponds to frame repeat.)3) Otherwise, process the
current frame. Put om values in the corresponding place of
the data structure. Nonselected blocks are left as erasures.
If fcur< fpre, then the end of the group of frames is reached.
Decode the message bits and obtain u1. Initialize data
structure.
IV. Experiments
We perform experiments in three stages. First, we
compare QIM and FZDH by means of their raw decoding
error performances without any error correction. Second, we
observe the performance of the proposed framework against
various common video processing attacks. Third, we
compare the proposed video data hiding framework against
JAWS [14], [15] and the method in [2] by using MPEG-2
compression attack.
A. FZDH Versus QIM
We utilize MPEG-2 DVB-S videos from five
different TV channels. The total duration of the host video
set is equal to 60 min (approximately 90 000 frames). The
resolution of the videos is 720 by 576. Initial bitrates of the
videos range from 6 Mb/s to 9 Mb/s. The marked videos are
re-encoded at various bitrates and decoding errors are
computed. The raw channel performance is measured by
hiding the same data bit (i.e., constant m) to the whole video.
Additionally, frame selection is not active. Hence de-
synchronization due to selective embedding is not effective.
QIM and FZDH are compared at the same embedding
distortion and data hiding rate. Two different embedding
distortion values are utilized: 48 dB and 51 dB average
SNR. Embedding distortion is computed as the average
PSNR between host and marked frames. The data hiding
parameters that yield these values are tabulated in Table I.
We should note that different pairs of (_, r) may yield the
same embedding distortion. We make use of typical values
determined manually. T1 is selected as 2000 and T2 is set to
1000. A typical host and marked frame pair for FZDH (at 48
dB) is shown in Figs. 7 and 8. Comparison results against
MPEG-2 compression attack are shown in Fig. 9 for Intra
and Inter frames, respectively, for 48 dB and 51 dB cases.
We observe that FZDH is superior to QIM, especially at low
compression bitrates and small mbedding distortion values.
These conditions correspond to low WNR values; hence, the
results comply with the reported results for AWGN in [8].
Furthermore, we observe the Intra and Inter frames do not
yield significant differences.
B. Common Video Processing Attacks
At the second stage, we apply error correction and
assess the performance of FZDH against some common
video processing attacks. We utilize a typical TV broadcast
material of 10 min We prefer a smaller duration, which is
still accurate to draw conclusions, due to the computational
burden of RA decoding. The format of the test video is
MPEG-2 at 9 Mb/s and its resolution is 720 by 576.
Fig. 7. Typical host frame.
Fig. 8. Corresponding marked frame using FZDH with = 40 and r = 4.

6
9 7 8 8 1 9 2 4 0 3 1 8 2
The system parameters are tuned manually. We utilize the
following values during the experiments: T0 = 0.05, T1 =
1000,T2 = 500, K2 = 10, T = 3. One should note that
threshold values are selected for this resolution and block
size of 8 by 8. Different dimensions might require some
other threshold values. Typical R values are used according
to the attack. Once these values are set, the embedding rate is
determined. For instance for R = 150, K1 is 99, i.e., 33 bits
are hidden per frame. First, we observe the effect of the
parameters on the number of selected block rate. Results for
a 4 s video segment are shown in Fig. 10. The number of the
selected blocks depends on the content and varies slowly
with time. The abrupt changes correspond to shot
boundaries. We observe that embedder and decoder select
different number of blocks. Interestingly, for low rates the
decoder can select higher number of blocks. However, due to
frame selection at embedder (with respect to T0), the
decoder can correctly determine the group of frame and
extract the hidden data as a result of the frame
synchronization markers. Second, we observe the decoding
error performance against compression attack. We utilize
typical bitrates for this resolution. We increase R to the point
where one obtains error-free
decoding. The results are tabulated in Table II for two
different embedding distortion values. 46.0 dB and 40.46 dB
embedding distortion values are obtained with the following
FZDH parameters { = 40, r = 4}, { = 80, r = 8},
respectively. The results indicate that we need repetition
number higher than the erasure rate. The reason for this
observation is due to the fact that decoding errors occur as a
result of compression as well as the erasures due to the block
selection. Furthermore, we observe that H.264 appears to be
a stronger attack compared to MPEG-2. Therefore, we need
higher repetition for error-free decoding.

7
9 7 8 8 1 9 2 4 0 3 1 8 2
Fig. 9. FZDH versus QIM. (a) Intra frames, 48 dB average
embedding distortion. (b) Inter frames, 48 dB average
embedding distortion. (c) Intra frames, 51 dB average
embedding distortion. (d) Inter frames, 51 dB average
embedding distortion.
Fig. 10. Typical selected block rates and decoding errors for
T0 = 0.05, T1 = 1000, K1 = 10, T = 3, R = 150, = 80, r = 8
and MPGE-2 4 Mb/s compression. Circle denotes decoding
error, square denotes selected block rate at embedder, dot
denotes selected block rate at decoder, and plus denotes the
frame locations where message is embedded. Third, we test
the performance of the method against another common
video processing: frame-rate conversion. The frame rate of
the original video is 25 f/s. We change this framerate to a
higher as well as a lower value and measure the decoding
error rate. We should note that frame-rate conversion could
be achieved in various ways, some of which could be quite
complex. However, we utilize an open source codec
(ffmpeg1), which performs frame-rate conversion by frame
drop/repeat and re-encoding. First, we present the selected
block rates at embedder and decoder in Fig. 11 by using the
same video segment in Fig. 10. The frame rate of the marked
video is changed to 30 f/s from 25 f/s. We observe that even
if the message locations are shifted, we can successfully
decode the message bits as a result of the frame
synchronization markers. The total decoding error results are
tabulated in Table III. We observe that different rates have
similar results. Frame insertions and drops do not differ as
long as they can be detected correctly by synchronization
markers. We should note that for lower f/s value of 23.98,
frame drop rate is quite small and at most one frame per
group of frames can be dropped. Additionally, we require
higher repetitions than compression attack. Finally, we test
the scaling performance. For this purpose, we downscale the
marked video, and then, upscale the attacked video to its
original size. Scaling operations are performed again using
ffmpeg library. The decoding error values for three different
dimensions are given in Table IV. Scaling test results
indicate that CIF resolution attack totally removes the hidden
data. On the other hand, we can obtain better results for
VGA and SVGA resolutions. However, error-free decoding

8
9 7 8 8 1 9 2 4 0 3 1 8 2
is not possible with the utilized system parameters. One
should increase the repetition rate, embedding distortion, or
number of frames in order to achieve error-free decoding.
C. Proposed Framework Against JAWS and Sarkar
We compare the proposed framework against the
canonical video watermarking methods JAWS and a more
recent quantization based method [2]. JAWS is a spatial
domain additive spread spectrum based watermarking
method [14]. It is utilized for DVD copyright protection [16]
and broadcast monitoring [15]. In AWS, luminance channel
of the frame in tiles. A pseudo-random watermark pattern is
generated. The size of this pattern matches the size of the tile
elements. The payload is increased by using shifted versions
of the base watermark. The amount of payload depends on
the number of shifts and possible shift locations determined
by a fixed grid. The superposition of the base watermark and
its shifted versions is added to each tile element of the
luminance channel spatially. The strength of the watermark
is adjusted by means of a global scaling parameter. At the
decoder, the received frame is folded by averaging tile
elements and the same base watermark is generated.
Correlation is performed to detect the locations of the peaks
in Fourier domain. The relative positioning of the peaks
gives the decoded message bits. We compare the proposed
framework and JAWS at the same data hiding rate and
embedding distortion. We utilize the same host video in
Section IV-B. For JAWS, we utilize a tile with four
elements, four shift locations inside a grid of 32 by 32. The
resultant data hiding bit rate is 30 bits per frame. This rate is
assured with R = 150 (33 bits per frame) in the proposed
framework. The global scaling parameter of JAWS is varied
to adjust the embedding distortion. The results of JAWS
against MPEG-2 compression attack are given in Table V.
When we compare Tables II and V, we observe that at the
same embedding distortion (40 dB PSNR), the proposed
framework is significantly superior to JAWS. The results
indicate that for high payload applications JAWS cannot
achieve an acceptable performance level. We compare the
proposed framework and adaptive quantization based
scheme of Sarkar et al. [2] by means of the results reported
in [2]. In this method, conventional QIM is applied in
uncompressed domain to selected low-frequency DCT
coefficients. The quantization parameter is adaptively
adjusted according to the type of the frame. The resulting
desynchronization due to coefficient selection is handled by
means of RA codes. The different utilization of I/P/B frames
results in the unexpected situation that the decoding error
decreases with decreasing embedding distortion. However,
we base our comparison with the best result obtained in [ 2].
We utilize the same host video as in [2]. Since the host video
is QVGA (320 by 240) size, we adjust method parameters
accordingly. First, we reduce the repetition number, R, to 4
and obtain embedding rate of 300 bits per frame, whereas the
best result is obtained for 294.2 bits per frame in [2].
Second, we utilize = 70, r = 10 with T = 3 and obtain an
embedding distortion of 37.06 dB PSNR, whereas the
corresponding value is 37.02 dB in [2]. In these conditions,
the proposed method yields 0.006 decoding error rate for
MPEG-2 compression attack at 4 Mb/s. On the other hand,
the performance of [2] is given in terms of “frame error
rate,” which corresponds to the frames with non-converging
RA decoder, and the best result reported in[2] is 0.008.
Assuming zero error for converging RA decoder, one can
claim that they have comparable performances. However, we
should note that at this level of embedding distortion visible
artifacts occur and hence for a more realistic comparison
embedding distortion should be decreased to an acceptable
level. V.
Conclusion
In this paper, we proposed a new video data hiding
framework that makes use of erasure correction capability of
RA codes and superiority of FZDH. The method is also
robust to frame manipulation attacks via frame
synchronization markers. First, we compared FZDH and
QIM as the data hiding method of the proposed framework.
We observed that FZDH is superior to QIM, especially for
low embedding distortion levels. The framework was tested
with MPEG-2, H.264 compression, scaling and frame-rate
conversion attacks. Typical system parameters are reported
for error-free decoding. The results indicate that the
framework can be successfully utilized in video data hiding
applications. For instance, Tardos fingerprinting [18], which
is a randomized construction of binary fingerprint codes that
are optimal against collusion attack, can be employed within
the proposed framework with the following settings. The
length of the Tardos fingerprint is AC2 0In 1 ε1 [19], where

9
9 7 8 8 1 9 2 4 0 3 1 8 2
A is a function of false positive probability (ε1), false
negative probability, and maximum size of colluder
coalition, (Co). The minimum segment durations required
for Tardos fingerprinting in different operating conditions
are given in Table VI. We also compared the proposed
framework against the canonical watermarking method,
JAWS, and a more recent quantization based method [2].
The results indicate a significant superiority over JAWS and
a comparable performance with [2]. The experiments also
shed light on possible improvements on the proposed
method. First, the framework involves number of thresholds
(T0, T1, and T2), which are determined manually. The range
of these thresholds can be analyzed by using a training set.
Then some heuristics can be deduced for proper selection of
these threshold values. Additionally, incorporation of human
visual system based spatio-temporally adaptation of data
hiding method parameters as in [13] remains as a future
direction.
References
[1] S. K. Kapotas, E. E. Varsaki, and A. N. Skodras, “Data
hiding in H- 264 encoded video sequences,” in Proc. IEEE
9th Workshop Multimedia Signal Process., Oct. 2007, pp.
373–376.
[2] A. Sarkar, U. Madhow, S. Chandrasekaran, and B. S.
Manjunath, “Adaptive MPEG-2 video data hiding scheme,”
in Proc. 9th SPIE Security Steganography Watermarking
Multimedia Contents, 2007, pp. 373–376.
[3] K. Solanki, N. Jacobsen, U. Madhow, B. S. Manjunath,
and S. Chandrasekaran, “Robust image-adaptive data hiding
using erasure and error correction,” IEEE Trans. Image
Process., vol. 13, no. 12, pp. 1627–.
[4] M. Schlauweg, D. Profrock, and E. Muller, “Correction
of insertions and deletions in selective watermarking,” in
Proc. IEEE Int. Conf. SITIS, Nov.–Dec. 2008, pp. 277–284.
[5] H. Liu, J. Huang, and Y. Q. Shi, “DWT-based video data
hiding robust to MPEG compression and frame loss,” Int. J.
Image Graph., vol. 5, no. 1, pp. 111–134, Jan. 2005.
[6] M. Wu, H. Yu, and B. Liu, “Data hiding in image and
video: I. Fundamental issues and solutions,” IEEE Trans.
Image Process., vol. 12, no. 6, pp. 685–695, Jun. 2003.
[7] M. Wu, H. Yu, and B. Liu, “Data hiding in image and
video: II. Designs and applications,” IEEE Trans. Image
Process., vol. 12, no. 6, pp. 696– 705, Jun. 2003.
[8] E. Esen and A. A. Alatan, “Forbidden zone data hiding,”
in Proc. IEEE Int. Conf. Image Process., Oct. 2006, pp.
1393–1396. [9] B. Chen and G. W. Wornell, “Quantization
index modulation: A class of provably good methods for
digital watermarking and information embedding,” IEEE
Trans. Inform. Theory, vol. 47, no. 4, pp. 1423–1443, May
2001.
[10] E. Esen, Z. Do˘gan, T. K. Ates, and A. A. Alatan,
“Comparison of quantization index modulation and
forbidden zone data hiding for compressed domain video
data hiding,” in Proc. IEEE 17th Signal Process. Commun.
Applicat. Conf., Apr. 2009, pp. 404–407.
[11] D. Divsalar, H. Jin, and R. J. McEliece, “Coding
theorems for turbo-like codes,” in Proc. 36th Allerton Conf.
Commun. Control Comput., 1998, pp. 201–210. [12] M. M.
Mansour, “A turbo-decoding message-passing algorithm for
sparse parity-check matrix codes,” IEEE Trans. Signal
Process., vol. 54, no. 11, pp. 4376–4392, Nov. 2006

S.A.kalaiselvan- robust video data hiding at forbidden zone

More Related Content

What's hot (20)

Viewers also liked (8)

Similar to S.A.kalaiselvan- robust video data hiding at forbidden zone (20)

Recently uploaded (20)

S.A.kalaiselvan- robust video data hiding at forbidden zone