Content uploaded by Khaled Elleithy
Author content
All content in this area was uploaded by Khaled Elleithy on Apr 03, 2015
Content may be subject to copyright.
78•1-4799•6776•6115/$31.00 ©2015 IEEE
A High Payload
V
DWT Domain
Ramadhan J. Mstafa,
I
EEE Studen
Department of Computer Science and
E
University of Bridgeport
Bridgeport, CT 06604, US
A
rmstafa@my.bridgeport
.
ed
u
Abstract—Video steganography has becom
e
due to the significant growth of video data over
performance of any steganography al
g
orithm
factors: embedding efficiency and embeddin
g
paper, a high embedding payload of vid
e
algorithm has been
p
ro
p
osed based on the
improve the security of the algorithm, a secr
e
encoded by ,, coding. Then, it is e
m
discrete wavelet transform (DWT) coefficients
As the DWT middle and hi
g
h fre
q
uenc
y
re
g
io
n
to be less sensitive data, the secret messa
g
e is e
m
the middle and high fre
q
uenc
y
DWT coefficie
n
algorithm is tested under two types of videos
and fast motion objects. The results of the
p
r
are compared to both the Least Si
g
nificant
B
algorithms. The results demonstrate better
p
e
r
proposed algorithm than for the others. The
h
proposed algorithm is approximately 28%, whi
a high embedding payload with a minimal
t
quality. The robustness of the proposed al
g
o
under various attacks. The results were consist
e
Keywords
—
Video Steganography; BCH
Embedding Efficiency; Embedding Payload
I. INTRODUCTION
Steganography is a process that involves
information (message) inside other carrier
protect the message from unauthorized users
.
(stego objects) will be seen by the Huma
n
(HVS) as one piece of data because the HVS
w
recognize the small change that occurs in
Message and cover data could be any type of
as text, audio, image, and video [2]. Fig.
general block diagram of steganography
development of steganalysis tools we
a
steganography schemes and rendering the
m
researchers have to develop secure steganog
r
that are protected from
b
oth attackers
a
detectors. Any successful steganography
V
ideo Steganography A
l
Based on BCH Codes (
t Membe
r
E
ngineering
A
u
Khaled M. Elleithy,
IE
Department of Computer
S
University of
Bridgeport, CT
elleithy@bri
d
e
a popular topic
the Internet. The
depends on two
p
a
y
load. In this
e
o ste
g
ano
g
ra
p
h
y
BCH coding. To
e
t messa
g
e is first
m
bedded into the
of video frames.
n
s are considered
m
bedded onl
y
into
n
ts. The proposed
that contain slow
o
p
osed al
g
orithm
B
it (LSB) and [1]
r
formance for the
h
idin
g
ratio of the
i
ch is evaluated as
t
radeoff of visual
rithm was tested
e
nt.
Codes; DWT;
hiding important
(cover) data to
.
The mixed data
n
Visual System
w
ill not be able to
the cover data.
data format such
1 illustrates the
concepts. The
a
kens unsecure
m
useless. Hence,
r
aphy algorithms
a
nd steganalysis
system should
consider two main important facto
r
embedding efficiency [3].
First, the embedding payload i
s
secret information that is going t
o
cover data. The algorithm has a hi
g
has a large capacity for the secre
t
efficiency includes the stego vis
robustness against attackers. Seco
n
rate and good quality of the c
o
embedding efficiency [4]. The ste
contains a high embedding effic
i
suspicion of finding hidden data a
n
detect through steganalysis tools.
H
the cover data after the embedding
p
the attention of attackers [5]. Th
e
directly affected by the security of
[6]. In traditional steganographic s
c
and embedding efficiency are oppo
s
of the secret message will decrease
that then weakens the embeddin
g
should be considered. The decidi
steganography algorithm and the
improve steganographic schemes,
m
matrix encoding and block code p
r
BCH, and Reed-Solomon codes [7
]
paper will provide “a state of the a
r
the frequency domain that uses
addition, this original, highly em
b
produces a reasonable tradeoff b
e
payload, and robustness.
The remainder of this paper
Section 2 presents some of the
discusses discrete wavelet transfor
m
principles of BCH codes. Section
and extracting phases of the
algorithm. Section 6 illustrates an
d
results. Section 7 contains the concl
u
Fig. 1 General Block Diagram of Steganography Algorithms
l
gorith
m
in
15, 11)
E
EE Senior Member
S
cience and Engi
n
eering
Bridgepor
t
06604, USA
d
geport.edu
r
s: embedding payload and
s
defined as the amount of
o
be embedded inside the
g
h embedding payload if it
t
message. The embedding
ual quality, security, and
n
d, both a low modification
o
ver data lead to a high
ganography algorithm that
i
ency will reduce attacker
n
d will be quite difficult to
H
owever, any distortion to
p
rocess occurs will increase
e
embedding efficiency is
the steganographic scheme
c
hemes, embedding payload
s
ite. Increasing the capacity
the quality of stego videos
g
efficiency. Both factors
ng factors depend on the
user requirements [4]. To
m
any of the algorithms use
r
inciples such as Hamming,
]
. The contributions of this
rt
” embedding algorithm in
e
rror correcting codes. In
b
edding payload algorithm
e
tween visual qualit
y
, data
is organized as follows:
related works. Section 3
m
. Section 4 explains some
5 presents the embedding
proposed steganography
d
explains the experimental
u
sions.
II. RELATED WORKS
In 2012, Zhang et al. proposed an efficien
t
BCH code for steganography. The embed
d
secret message into a block of cover data.
process is completed by changing various c
o
input block in order to make the syndrome
efficient embedder improves both stora
g
computational time compared with ot
h
According to the system complexity, Z
h
improves the system complexity from expone
n
In 2013, Liu et al. proposed a robust stega
n
using H.264/AVC compressed video str
distortion drift in the intra-frame. The preve
n
frame distortion drift can be achieved using
the intra-frame prediction. Some blocks wi
l
cover data for embedding the secret informat
i
will depend on the prediction of the intra
-
neighboring blocks to avert the distortion tha
t
the adjacent blocks. To improve system
robustness, Liu et al. applied BCH code to the
the embedding process. Then, the enco
d
embedded into the 4x4 DCT block of quan
t
with only a luminance component of the int
r
luminance (brightness) component to the HV
S
on its higher sensitivity than that of the color
In 2014, Diop et al. proposed an adapti
v
algorithm using a linear error correcting code,
low-density parity-check (LDPC) code. The a
l
how to minimize the effect of secret messa
g
the LDPC code. For that purpose, Diop et al.
d
the LDPC code is a better encoding algorit
h
codes [11].
Previously mentioned algorithms lack t
h
withstand hacker attacks. With the emb
flexibility exists to increase the capacity of
a
This paper proposes a high embedding
steganography algorithm in the DWT domai
n
codes.
III. DISCRETE WAVELET TRANSFOR
M
DWT is a well-known method that transfe
r
the time domain to the frequency domain
separates high, middle, and low frequencies a
n
from one another, while other methods, suc
h
the various frequencies into estimated regio
n
of the 2D-DWT image decomposition is ap
p
video frame. It splits the frame into fou
r
(approximation), LH (horizontal), HL (ve
r
(diagonal), using both a low pass filter Lo_D
z
filter Hi_Dz for the decomposition
p
roce
s
frequency sub-band, which is an approximati
o
frame reduced to a quarter of its size. The
L
sub-bands are middle and high frequenc
i
detailed information about any image. In th
e
image decomposition, the 2D-DWT is applie
band, producing four new sub-bands [13, 14]
algorithm, the BCH encoder was applied to
Then, the LH, HL, and HH coefficients we
r
data to embed the encoded secret message. Fi
g
2D-DWT.
t
embedder using
d
er conceals the
The embedding
o
efficients in the
values null. The
g
e capacity and
t
her algorithms.
h
ang’s algorithm
n
tial to linea
r
[8].
n
ography scheme
r
eam without a
n
tion of the intra-
the directions of
l
l be selected as
i
o
n
. This process
-
frame modes of
t
propagates from
efficiency and
e
message prior to
d
ed message is
t
ized coefficients
r
a-frame [9]. The
S
is chosen based
component [10].
v
e steganography
referred to as the
l
gorithm explai
n
s
g
e insertion using
d
emonstrated that
h
m than all other
h
e robustness to
edding payload,
a
secret message.
payload video
n
based on BCH
M
(DWT)
r
s the signal from
[12]. The DWT
nd its boundaries
h
as DCT, group
n
s. The first level
p
lied to the cover
r
sub-bands, LL
r
tical), and HH
z
and a high pass
s
s. LL is a low
o
n of the original
L
H, HL, and HH
i
es that contain
e
second level of
e
d to the LL sub-
. In the proposed
the hidden data.
r
e used as cover
g
. 2 illustrates the
Fig. 2 First Level of t
h
The results demonstrate the fi
r
Fig. 3 shows the second level of the
Fig. 3 Second Level of the 2D-
D
To achieve a complete re
c
following wavelet equations must b
e
__ _
_
_,_
In the above equations, Lo_Dz
wavelet filter bank of the decompo
s
Lo_Rz and Hi_Rz signify the
w
reconstruction process. The foll
o
transfer functions of the Haar wavel
_
1
_1
_
1
_
1
IV. BCH C
O
Bose, Chaudhuri, and Hocque
n
encoder. It is one of the most po
w
methods, which can be used for det
e
in a block of data. The BCH c
Hamming code because BCH can
c
binary BCH (n, k, t) can correct err
o
codewords of the length n
,
,
length k
,
,
,…,
.
E
messages can both
b
e interpret
e
h
e 2D-DWT
r
st level of decomposition.
decomposition process.
D
WT Decompositio
n
c
onstruction process, the
e
satisfied:
_2
(1)
_
(2)
and Hi_Dz represent the
s
ition process. Furthermore,
w
avelet filter ban
k
of the
o
wing equations are the
et transform filters:
(3)
(4)
(5)
(6)
O
DES
n
ghe
m
invented the BCH
w
erful random cyclic code
e
cting and correcting errors
ode is different from the
orrect more than one bit. A
or
s of a maximum t bits for
,
,…,
and message
E
ncoded codewords and
e
d as polynomials, where
, and
. When m and t are any positive integers where
3and 2, there will be a binary BCH code
with the following properties:
¾ Block codeword length 21
¾ Message length
k
¾ Maximum correctable error bits
t
¾ Minimum distance 21
¾ Parity check bits
The BCH codes inventors decided that the generator
polynomial g(x) will be the polynomial of the lowest degree in
the Galois field GF (2), with ,,,…, as roots on the
condition that is a primitive of GF2 . When Mx is a
minimal polynomial of where 12, then the least
common multiple (LCM) of 2t minimal polynomials will be
the generator polynomial g(x). The g(x) function and the
parity-check matrix H of the BCH code [15] are described as
follows:
1 …
1 …
1 …
. . . . .
. . . . .
. . . . .
1 …
(7)
,,,…,
(8)
…
(9)
In this paper, the BCH code (15, 11, 1) is used with the
following parameters:
1) is a primitive element of the GF2 , where m=4 and
n2 1.
2) The primitive polynomial is 1 .
3) There are three minimal polynomials of ,,and
[16]:
¾ Mx1xx (10)
¾ Mx1xxxx (11)
¾ Mx1xx (12)
4) The single error correction is used and the generator
polynomial will be gxMx1xx.
5) The minimum distance of the applied BCH code (15,
11) is more than 2.
V. THE PROPOSED STEGANOGRAPHY ALGORITHM
The proposed algorithm uses uncompressed video
sequences based on the frames as still images. At the
beginning, the video stream is separated into frames; each
frame is converted to YCbCr color space. The reason for
converting to YCbCr color space is to remove the correlation
between the red, green, and blue colors. A luminance (Y)
component is the brightness data, which is more sensitive to
the human eye than the color (chrominance) components.
Consequently, the color parts can be subsampled into the video
sequences, and some insignificant data can be discarded.
A. Data Embedding Phase
The process of embedding the secret message consists of
two phases: first encoding the message using the BCH code
(Step 1 to 5) and then embedding the encoded message into the
cover videos (Step 6 to 13). This process can be completed by
the following steps:
Step1:
Input the secret message (text file).
Step2:
Change the bits positions of the whole secret
message by key1.
Step3:
Convert the whole secret message to a one
dimensional array (1-D).
Step4:
Encode the message by using the BCH (15, 11)
encoder.
Step5:
XOR the encoded data, which consists of 15 bits
(11 bits of message + 4 bits of parity), with the 15
bits of random value using key2.
Step6:
Input the cover video stream.
Step7:
Convert the video sequence into a number of
frames.
Step8:
Split each frame into the YUV color space.
Step9:
Apply the two dimensional DWT separately to
each Y, U, and V frame component.
Step10:
Embed the message into the middle and high
frequency coefficients (LH, HL, and HH) of each
of the Y, U, and V components.
_,,, 0
|_,,|, 0 (13)
_,,, 0
|_,,|, 0
(14)
_,,, 0
|_,,|, 0 (15)
Where ,, are the Y, U, and V
coefficients, and
S
is the encoded secret
message, 000,…,111.
E
is the embedding
process.
Step11:
Apply the inverse two dimensional DWT on the
frame components.
Step12:
Rebuild the stego frames from the YUV stego
components.
Step13:
Output the stego videos, which are reconstructed
from all embedded frames.
Two keys were used in the proposed steganography
algorithm; each key was predefined by the sender and receiver
in both the embedding and the extracting processes. The first
key (key1) is used to randomly change the position of all bits in
the secret message to make the message unreadable and chaotic
before encoding by the BCH. The second key (key2) is used
after the encoding process; the encoded message is divided into
15-bit groups, and each group is XORed with the 15-bit
numbers (the 15-bit numbers were randomly generated). One
of the strengths of the proposed algorithm is the usage of the
two keys, which improve the security and robustness of the
proposed algorithm. The block diagram of data embedding
process for the proposed steganography algorithm is shown in
Fig. 4.
Fig. 4 Block Diagram of the Data Embeddin
g
B. Data Extracting Phase
This section introduces the process o
encoded message from the stego videos
decoding the encoded message using the B
C
process can be completed
b
y the following ste
p
Step1:
Input the stego videos.
Step2:
Convert the ste
g
o video se
q
uenc
e
of frames.
Step3:
Divide each frame into the YUV c
o
Step4:
Apply the 2D-DWT se
p
aratel
y
t
o
V component.
Step5:
Extract the encoded messa
g
e fro
m
hi
g
h fre
q
uenc
y
coefficients (LH,
each Y, U, and V component.
,,
_
,,
|
,,
|
,,
_
,,
|
,,
|
,,
_
,,
|
,,
|
g
Process
o
f retrieving the
first, and then
C
H decoder. This
p
s:
e
s into a number
o
lor s
p
ace.
o
each Y, U, and
m
the middle and
HL, and HH) of
0
0
(16)
0
0
(17)
0
0
(18)
Where
,
,
coefficients, and
is t
h
EX
is the extractin
g
p
r
o
Step6:
Segment the entire en
c
groups.
Step7:
XOR each
g
rou
p
w
numbers that were
g
e
n
the sender side (key
2
).
Step8:
Decode the messa
g
e
decoder.
Step9:
Produce an arra
y
from
t
Step10:
Re
p
osition the messa
g
order using key
1
Step11:
Out
p
ut the secre
t
The block diagram of the dat
a
proposed steganography algorithm i
Fig. 5 Block Diagram of the D
a
are the distorted YUV
h
e retrieved secret message.
o
cess.
c
oded messa
g
e into 15-bits
w
ith the random 15-bits
n
erated b
y
the same ke
y
at
b
y
the BCH (15, 11)
t
he resul
t
ed groups.
g
e a
g
ain to the original bit
t
messa
g
e as a text file.
a
extracting process of the
s shown in Fig. 5.
a
ta Extracting Process
VI. EXPERIMENTAL RESULTS AND DI
S
In this section, the performance of the pr
o
is evaluated through several experiments.
T
environment utilizes several variables: the co
v
a dataset consisting of seven video seque
n
Interchange Format (CIF) type; also, the fo
4:2:0. In addition, the resolution of each vide
o
and all videos are equal in length with 150 fr
a
file is used as a secret message. The wor
k
using MATLAB to test the proposed algorith
m
results are implemented using both fast a
n
videos.
A. Visual Quality
The main challenge of using video stegan
o
as much data as possible without degrading
t
of the stego video. PSNR is an objective qua
l
used to calculate the difference between the
stego video frames. It can be obtained
b
y fol
l
[17]:
10
∑∑,
,
Where O and S denote the original and s
t
components, respectively, and m and n
resolutions. In Fig. 6, the PSNR of the Y
calculated for all seven videos. The results o
f
the
Akiyo
,
Container
,
Bus
, and
Foreman
v
stable, while in the
Soccer
and
Tennis
vide
o
frequently changing. The reason for the v
a
quality is because the sporting videos cont
a
objects than the others. Overall, the
Akiyo
v
i
visual quality.
Fig. 6 PSNR Comparisons for the Y-Components of
A
Figs. 7 and 8 show the PSNR of the U-c
o
V-component, respectively, for all seven vi
d
figure, the demonstrated results of the PSNR
-
S
CUSSION
o
posed algorithm
T
he experimental
v
e
r
data comprise
n
ces of Common
rmat of YUV is
o
is 352 288,
a
mes. A large text
k
is implemented
m
efficiency. The
n
d slo
w
motion
o
graphy is to hide
t
he visual quality
l
it
y
measurement
original and the
lowing equations
19
20
t
ego YUV frame
are the video
Y
-components are
f
the PSNR-Y for
v
ideos are more
o
s, the quality is
a
rying the visual
a
in fas
t
er motion
i
deo has the best
A
ll Seven Videos
o
mponent and the
d
eos. In the first
-
U for the
Akiyo
and
Container
videos are changing
other, as compared to other five v
i
Coastguard
video has the highest d
B
second figure, the PSNR-V for a
calculated; the
Coastguard
and
S
o
quality. In Fig. 9, the PSNR co
m
each video is shown. The compari
s
the objective quality for each of t
h
and
Foreman
videos ranged betwe
e
all contain slower motion object
s
Coastguard
,
Soccer
, and
Tennis
vi
d
time (ranges between 35-47 dB).
T
these videos contain faster motion
o
visual quality.
Fig. 7 PSNR Comparisons for the U-Co
m
Fig. 8 PSNR Comparisons for the V-Co
m
Table I shows the average of th
e
V component for all video seque
n
each part is measured by separatel
y
frames per video. The averages are
v
the type of videos and the speed of
t
slightly from one frame to
i
deos. The PSNR-U of the
B
s among the group. In the
l
l video streams has been
o
cce
r
videos have a better
m
parison for 150 frames of
s
on shows that the result of
h
e
Akiyo
,
Container
,
Bus
,
e
n 40-42 dBs; these videos
s
while the PSNR of the
d
eos change frequently over
T
he changes occur because
o
bjects that lead to unstable
m
ponent of All Seven Videos
m
ponent of All Seven Videos
e
PSNR for each Y, U, and
n
ces. The visual quality of
y
averaging each of the 50
v
arious and depend on both
t
he motion object.
Fig. 9 PSNR Comparisons for 150 Frames of All
TABLE I. T
HE
A
VERAGE
PSNR
FOR
E
ACH
Y,
U,
A
N
FOR
A
LL
S
EVEN
V
IDEOS
Video
Se
q
uences
Frame
Number
PSNR
Y
PSN
R
U
Akiyo
1-50 44.799 36.60
4
51-100 44.787 36.68
1
101-150 44.905 36.62
1
Coastguard
1-50 41.126 46.16
6
51-100 41.064 45.05
4
101-150 40.648 44.35
8
Container
1-50 39.423 42.13
2
51-100 39.365 42.17
1
101-150 39.442 42.31
2
Bus
1-50 37.661 41.18
7
51-100 36.902 40.91
6
101-150 37.830 41.73
7
Soccer
1-50 42.429 42.94
7
51-100 40.471 42.35
7
101-150 50.163 43.40
7
Foreman
1-50 41.206 42.48
9
51-100 41.374 41.98
2
101-150 41.370 41.90
9
Tennis
1-50 39.952 42.19
4
51-100 38.278 36.11
4
101-150 34.538 37.55
0
B. Embedding Payload
According to [18], the proposed algori
embedding payload. The obtained hiding ra
t
reasonable tradeoff is noticed between th
e
embedded message in each video (6.12 Mbyte
Seven Videos
N
D
V
C
OMPONENT
R
PSNR
V
4
43.622
1
43.663
1
43.692
6
46.990
4
46.126
8
46.177
2
40.653
1
40.752
2
40.934
7
41.818
6
42.563
7
43.417
7
44.887
7
45.349
7
46.079
9
43.025
2
42.532
9
42.565
4
37.921
4
34.966
0
37.283
i
thm has a high
t
io is 28.12%. A
e
amount of the
s) and the quality
(35.58 - 45.68 dBs). The hiding r
a
equation 21. A number of exper
i
compare the embedding capacity
and the embedding capacity of both
Table II shows the compa
r
algorithms, according to the amo
u
frame.
3
TABLE II. C
APACITY
E
MBEDDING
C
O
M
A
LGORITHM WITH BOTH
[1]
A
N
Video
Resolution YUV
Proposed
Algorithm
(Bits/Frame)
176 X
144
Y
57024
U
14256
V
14256
352 X
288
Y
228096
U
57024
V
57024
The proposed algorithm has
capacity of [1] and the LSB algori
t
and 2.2 times, respectively, withou
t
Fig 10 shows the capacity embe
d
proposed algorithm.
Fig. 10 Comparison of the Proposed
A
In table III, there are five vid
motion objects (the
Soccer
,
Tennis
motion objects; the
Akiyo
and
Co
motion objects). Part (a) of the tabl
e
has the lowest PSNR and its origin
a
(b) of the table indicates the stego
PSNR and its original frame in eac
h
that the minimum and maximum P
S
slow motion objects are very clos
e
Akiyo
and
Container
videos. H
o
maximum PSNR of the videos that
h
different in dBs range such as
0
1
2
3
4
5
6
7
Proposed Alorithm Reference
Capacity Embedding (Mbytes)
a
tio can be calculated as in
i
ments were conducted to
of the proposed algorithm
the LSB algorithm and [1].
r
ison between the three
u
nt of secret data in each
100% 21
M
PARISON OF THE
P
ROPOSED
N
D THE
LSB
ALGORITHMS
[1]
(Bits/Frame)
LSB
Algorithm
(Bits/Frame)
4096 25344
Not used 6336
Not used 6336
8192 101376
Not used 25344
Not used 25344
improved the embedding
t
hm by approximately 41.7
t
visual quality degradation.
d
ding improvement of the
A
lgorithm with [1] and LSB
eos of both fast and slow
, and
Coastguard
have fast
Co
ntaine
r
videos have slow
e
shows the stego frame that
a
l frame in each video.
Part
frame that has the highest
h
video. It can be observed
S
NR of the videos that have
e
to one another, as in the
wever, the minimum and
h
ave fast motion objects are
the
Soccer
,
Tennis
, and
[1] LSB Algoritm
Proposed Alorithm
Reference [1]
LSB Algoritm
Coastguard
videos. Overall, the objective quality of both video
types is considerable.
C. Robustness
To evaluate the performance of the proposed algorithm for
correctly retrieving the secret message, two objective metrics
have been used: 1) the Similarity Function (SF) and 2) the Bit
Error Rate (BER). Both metrics are used to test whether the
extracted secret message has been corrupted during
communication. The SF and BER can be calculated as in the
following equations [19]:
∑∑,
,
∑∑,
∑∑
,
22
∑∑,
,
100% 23
where and
are the embedded and extracted secret
messages, respectively, and a and b are the dimensions of the
secret message array. The algorithm is tested under different
types of attacks (Gaussian noise with the zero mean and
variance=0.01 and 0.001, Salt & pepper noise with the
density=0.01 and 0.001, and median filtering). To achieve the
robustness of the algorithm, the higher SF and the lower BER
must be obtained. Table IV illustrates the performance of the
proposed algorithm under attacks while it retrieves the hidden
data with a high SF and a low BER.
D. Security Analysis
Despite of the high embedding payload, the security of the
proposed steganography algorithm has also improved. This is
mainly because two keys had been used before the embedding
process to produce the unreadable message to safeguard it
against attackers. Moreover, one of the strongest error
correcting methods has been applied on the secret message, the
BCH (15, 11) codes.
VII. CONCLUSION
In this paper, a high payload video steganography
algorithm in the DWT domain based on BCH codes has been
proposed. The steganography algorithm decomposes the video
into frames; then, it divides each frame into three components
(Y, U, and V). Before the embedding process, the secret
message is segmented and encoded using BCH (15, 11) codes
to increase the efficiency of the algorithm. The 2D-DWT has
been applied to each component; both the middle and high
frequency coefficients (LH, HL, and HH) are selected for
embedding the secret data. In addition, during the embedding
and extracting processes, this algorithm used two keys, which
improved the security of the system.
The proposed algorithm has a high embedding payload.
The amount of the secret data in each video is approximately
6.12 Mbytes and the HR is 28.12%. According to embedding
payload, the proposed algorithm outperformed both the LSB
and [1] algorithms. The visual quality of the stego videos is
also high: the PSNR ranged between 35.58 - 45.68 dBs with an
SF=1 and a BER=0. The efficiency of the proposed algorithm
is verified through a number of experiments in which a number
of cover videos with fast and slow motion objects were used.
Various attacks were also conducted to verify the efficiency of
the algorithm. The experimental results showed that the
proposed algorithm is robust against Gaussian and impulsive
noises. Furthermore, it was resistant against a median filtering
attack, contrary to [1] that is not robust enough against all
attacks and to the LSB algorithm which is susceptible to many
attacks, as well. For future work, we would like to improve the
embedding payload of the proposed algorithm with the respect
of the video quality by using other techniques that operate in
frequency domain. Also, we would like to conduct efficient
linear block codes to enhance the security of the algorithm.
REFERENCES
[1] M. A. Alavianmehr, et al., "A lossless data hiding scheme on video raw
data robust against H.264/AVC compression," in Computer and
Knowledge Engineering (ICCKE), 2012 2nd International eConference
on, 2012, pp. 194-198.
[2] H. Yuh-Ming and J. Pei-Wun, "Two improved data hiding schemes," in
Image and Signal Processing (CISP), 2011 4th International Congress
on, 2011, pp. 1784-1787.
[3] R. J. Mstafa and K. M. Elleithy, "A highly secure video steganography
using Hamming code (7, 4)," in Systems, Applications and Technology
Conference (LISAT), 2014 IEEE Long Island, 2014, pp. 1-6.
[4] C. Chin-Chen, et al., "A High Payload Steganographic Scheme Based
on (7, 4) Hamming Code for Digital Images," in Electronic Commerce
and Security, 2008 International Symposium on, 2008, pp. 16-21.
[5] L. Guangjie, et al., "An Adaptive Matrix Embedding for Image
Steganography," in Multimedia Information Networking and Security
(MINES), 2011 Third International Conference on, 2011, pp. 642-646.
[6] W. Jyun-Jie, et al., "An embedding strategy for large payload using
convolutional embedding codes," in ITS Telecommunications (ITST),
2012 12th International Conference on, 2012, pp. 365-369.
[7] R. Zhang, et al., "Fast BCH Syndrome Coding for Steganography," in
Information Hiding. vol. 5806, S. Katzenbeisser and A.-R. Sadeghi,
Eds., ed: Springer Berlin Heidelberg, 2009, pp. 48-58.
[8] Z. Rongyue, et al., "An Efficient Embedder for BCH Coding for
Steganography," Information Theory, IEEE Transactions on, vol. 58, pp.
7272-7279, 2012.
[9] Y. Liu, et al., "A Robust Data Hiding Algorithm for H. 264/AVC Video
Streams," Journal of Systems and Software, 2013.
[10] I. E. Richardson, H. 264 and MPEG-4 video compression: video coding
for next-generation multimedia: John Wiley & Sons, 2004.
[11] I. Diop, et al., "Adaptive steganography scheme based on LDPC codes,"
in 2014 16th International Conference on Advanced Communication
Technology (ICACT), 2014, pp. 162-166.
[12] E. Prasad, "High Secure Image Steganography based on Hopfield
Chaotic Neural Network and Wavelet Transforms," International
Journal of Computer Science & Network Security, vol. 13, 2013.
[13] G. Prabakaran and R. Bhavani, "A modified secure digital image
steganography based on Discrete Wavelet Transform," in Computing,
Electronics and Electrical Technologies (ICCEET), 2012 International
Conference on, 2012, pp. 1096-1100.
[14] O. S. Faragallah, "Efficient video watermarking based on singular value
decomposition in the discrete wavelet transform domain," AEU -
International Journal of Electronics and Communications, vol. 67, pp.
189-196, 2013.
[15] Y. Hoyoung, et al., "Area-Efficient Multimode Encoding Architecture
for Long BCH Codes," Circuits and Systems II: Express Briefs, IEEE
Transactions on, vol. 60, pp. 872-876, 2013.
[16] A. K. Panda, et al., "FPGA Implementation of Encoder for (15, k)
Binary BCH Code Using VHDL and Performance Comparison for
Multiple Error Correction Control," in Communication Systems and
Network Technologies (CSNT), 2012 International Conference on, 2012,
pp. 780-784.
[17] A. A. Judic e, et al., "An Image High Capacity Steganographic Methods
by Modified OPA Algorithm and Haar Wavelet Transform,"
I
nternational Journal of Computer Science & Net
w
2014.
[18] L. Tse-Hua and A. H. Tewfik, "A novel high-cap
a
system,"
I
mage Processing, IEEE Transactions o
n
2440, 2006.
TABLE III. M
INI
M
Minimum PSNR
Original Frames
S
Frame No. 74 in Akiyo Video Stego f
r
Frame 26
t
h
in Containe
r
Video
Stego f
r
Frame 2
nd
in Soccer Video Stego f
r
Frame No. 86 in Tennis Video Stego f
r
Frame No. 90 in Coastguard
Video Stego f
r
a)
TABLE IV. P
E
Type of
Attack
Akiyo
SF BER
% SF
No attacks 1 0
1
(Salt &
Pepper)
Density=
0.01 0.955 4.5 0.96
5
0.001 0.963 3.7 0.97
3
(Gaussian
white)
Variance=
0.01 0.923 7.7 0.93
3
0.001 0.909 9.1 0.91
9
Median filtering 0.986 1.4 0.98
7
w
ork Security, vol. 14,
a
city data-embedding
n
, vol. 15, pp. 2431-
[19] Y. He, et al., "A real-time dual water
m
video stream for Video-on-Demand
J
ournal of Electronics and Communic
a
M
UM AND
M
AXIMUM
PSNR
FOR EACH OF THE
F
IVE
V
IDEO
S
TREAMS
Maximum PS
N
S
tego frames Original Frames
r
ame PSNR 41.44 dB Frame No. 107 of Akiyo Video
S
r
ame PSNR 40.63 dB Frame 143
rd
in Container Video
S
r
ame PSNR 41.68 dB Frame 136
th
in Soccer Video
S
r
ame PSNR 34.99 dB Frame No. 39 in Tennis Video
S
r
ame PSNR 43.24 dB Frame No. 72 in Coastguard
Video
S
b)
E
RFORMANCE OF THE
P
ROPOSED
A
LGORITHM
U
NDER
A
TTACKS
Bus Coastguard Container Foreman
BER
%SF BER
%SF BER
%SF BER
% S
F
1
0 1 0 1 0 1 0
5
3.5 0.945 5.5 0.975 2.5 0.965 3.5 0.9
2
3
2.7 0.953 4.7 0.983 1.7 0.959 4.1 0.9
3
3
6.7 0.913 8.7 0.943 5.7 0.919 8.1 0.9
0
9
8.1 0.899 10.1 0.929 7.1 0.898 10.2 0.8
7
7
1.3 0.986 1.4 0.998 0.2 0.975 2.5 0.9
5
m
arking algorithm of H.264/AVC
service," AEU - International
a
tions, vol. 66, pp. 305-312, 2012.
NR
Stego frames
S
tego frame PSNR 41.95 dB
S
tego frame PSNR 40.99 dB
S
tego frame PSNR 47.43 dB
S
tego frame PSNR 41.08 dB
S
tego frame PSNR 47.00 dB
Soccer Tennis
F
BER
% SF BER
%
1 0 1 0
2
3 7.7 0.921 7.9
3
2 6.8 0.933 6.7
0
2 9.8 0.901 9.9
7
4 12.6 0.865 13.5
5
9 4.1 0.961 3.9
