Transcript 下載/瀏覽Download

A High-Capacity Steganography
Scheme for
JPEG2000 Baseline System
Liang Zhang, Haili Wang, and Renbiao Wu,
Senior Member, IEEE
IEEE TRANSACTIONS ON IMAGE PROCESSING, VOL. 18, NO. 8, AUGUST 2009
Received September 09, 2008; revised April 01, 2009.
First published April 24, 2009; current version published July 10, 2009.
1
Adviser: Chih-Hung Lin
Speaker:Po-Kai Shen
Date : 98/11/24
Outline
1.
2.
3.
4.
5.
6.
7.
8.
Author
Introduction
Jpeg2000 baseline coding system
Steganography based on twice bit-plane
encoding
Redundancy evaluation
Synchronization information and scrambling
measure
Simulation
2
Conclusion
Author(1)
Liang Zhang was born in 1970. He received
the Ph.D. degree in electronic information
engineering from Tianjin University, Tianjin,
China, in 2003.
He is an Associate Professor. He is currently
with the Tianjin Key Lab of Advanced Signal
Processing in Civil Aviation University of China.
His current research interests include image
processing, information hiding, and intelligent
visual surveillance.
3
Author(2)
 Haili Wang was born in 1983.
She is now a postgraduate specializing in
signal processing.
Her current research interests include image
processing and information hiding.
4
Author(3)
 Renbiao Wu (M’95–SM’01) received the
B.Sc. And M.Sc. degrees from Northwestern
Polytechnic University, Xian, China, in 1988
and 1991, respectively, and the Ph.D. degree
from Xidian University, Xian,in 1994, all in
electrical engineering.
From May 1994 to February 1996, he was a
Postdoctoral Fellow at the College of Marine
Engineering, Northwestern Polytechnic
University, where he was promoted to Associate
Professor in December 1995.
From March 1996 to February 1997, he was
a Visiting Scholar at the Center for
Transportation Research, Virginia Polytechnic
Institute and State University,Blacksburg.
5
Author(3)
From March 1997 to December 1998, he was a
Visiting Scholar at the Department of Electrical and
Computer Engineering, University of
Florida,Gainesville.
Since January 1999, he has been with the Tianjin
Key Lab for Advanced Signal Processing, Civil
Aviation University of China, Tianjin, China, where
he is currently a Chaired Professor and director of
the lab.
6
Author(3)
From August 2004 to January 2005, he was a
Distinguished Research Scholar in the Department of
Electrical and Electronic Engineering, Imperial
College London, London, U.K. Dr. Wu was the
recipient of the National Outstanding Young
Investigator Award of China in 2003.
His research interests include space-time adaptive
processing, adaptive arrays, feature extraction and
image formation, spectral estimation and their
applications to radar and wireless communication
systems.
7
Introduction
1) Modern information hiding technology is an
important branch of information security.
2) Steganography
three competing aspects :
 Capacity
 Security
 Robustness
8
Introduction
3)
Section III shows procedures of JPEG2000 baseline
system and points out the problem due to bitstream
truncation.
4)
Section IV describes the principle of twice bit-plane
encoding and illustrates the operation procedures.
5)
Section V gives a detailed description on redundancy
evaluation, and explains how embedding points and their
intensity are adjusted.
9
Introduction
6)
Section VI, we define measures for synchronization and
security.
7)
Section VII shows the simulation results.
8)
Section VIII draws a conclusion.
10
Jpeg2000 baseline coding system

JPEG2000 uses uniform scalar quantizers with enlarged
“deadzones.”

Truncating the embedded bitstream associated with any
given codeblock has the effect of quantizing the wavelet
coefficients in that codeblock more coarsely.

That is to say, there still exists a lossy procedure after
entropy encoding.
11
Steganography based on twice
bit-plane encoding
12
Steganography based on twice
bit-plane encoding
1)
There are three sub-steps involved in the determination of
embedding points and embedding intensity for a code block.
2)
Scrambled synchronization information and secret messages
are embedded into the selected embedding points from the
lowest embed-allowed bit-plane to higher ones.
3)
Secondary bit-plane encoding is operated after information
13
embedding.
Steganography based on twice
bit-plane encoding

Ensured at the cost of increased computational complexity
and slightly changed compression ratio.
14
Redundancy evaluation
(1)

Where
xi
is the quantized wavelet coefficient with the bits
lowerthan the highest no-zero bit are replaced by zeros.

The parameter
coefficient
 i is the quantization step of the wavelet
xi . The parameter 
assumes a value between
0 and 1. According to , a typical value of
of the first step is denoted as
yi
 is 0.7. The result
15
Redundancy evaluation
(2)

In the second step, the neighborhood masking effect is
exploited to process the wavelet coefficients as the following:(2)

The neighborhood contains wavelet coefficients within a
window of N by N, centered at the current position.

The parameter
i
is the total number of wavelet coefficients
in the neighborhood.
16
Redundancy evaluation
(2)

The parameter  assumes a value between 0 and 1,
together with
i
, is used to control the strength of
embedding intensity adjustment due to neighborhood
masking.

The symbol
x̂k denotes the neighboring wavelet coefficients
greater than or equal to 16, and all its bits lower than the
highest no-zero bit are set to be zeros.
17
Redundancy evaluation
(3)
(4)

In the third step, a weighting factor about brightness
sensitivity is used in the processing.

The symbol
I l denotes the subband at resolution
level l 0,1...k .and with orientation  LL, LH , HL, HH  .


I l i, j  denotes the wavelet coefficient located

at i, j  in subband I l .
The symbol
The levels of discrete wavelet decomposition is k.
18
Redundancy evaluation

The pixel value has a dynamic range of [ -128, 127]. The local
average brightness is normalized by dividing 128. Then the
result of the third step,
zi , is given by
(5)
Quantization redundancy is calculated by the following equation:
(6)

The redundancy of the wavelet coefficient
measured by
ri
.
xi
can be
19
Redundancy evaluation

Use the wavelet coefficients with
ri
not less than 2 to carry
message bits.

The rule of adjustment on embedding points and intensity is
as follows:
1)
If ri  2 , then this candidate embedding point should
be removed.
2)
n
n 1
If 2  ri  2 , then the embedding capacity of this
point is determined to be n bits.
20
Synchronization information and
scrambling measure
 The first part of the synchronization information is a 2-bit flag
that indicates whether a certain code block contains secret
message.
 The second part of the synchronization information is a 12-bit
fragment that indicates the length of the secret message
embedded in this code block.
21
Synchronization information and
scrambling measure
 The third part of the synchronization information is a 12-bit
fragment that indicates the length of the secret message
embedded in this code block.
22
Synchronization information and
scrambling measure
 A 64-bit secret key is used as a seed to generate a
s
sequence of pseudo random binary
numbers, which is used
i
to scramble the message bits.
 N is the total number of message bits.
 The symbol
mi denotes the i th
message bit, and
ni
the
i th binary number of the pseudo random sequence. The
operator ⊕ denotes binary addition. The scrambled
message bits, denoted as
si
selected wavelet coefficients.
, are to be embedded into
23
Simulation
Fig. 6.
(a) Original image used as cover media.
(b) the binary logo image used as secret message.
24
Simulation
 In the first step, the lowest embed-allowed bit-plane of each
code block is determined.
 In the second step, the wavelet coefficients with magnitudes
not less than a given threshold are chosen as candidate
embedding points.
 In the third step, the candidate embedding points are
adjusted image adaptively based on redundancy evaluation
to increase hiding capacity
 In the fourth step, we embed message bits into the selected
wavelet coefficients and finish encoding the stego-image
25
Simulation
 The threshold is set to 16

The parameters in those equations are set to be: N=5 ,
α=0.7 , β=0.2
26
 Evaluation results for wavelet coefficients A, B, and D are as
follows:
rA  1.72
rB  2.59
rC  13.4
Simulation
1
2
B: 2  2.59  2
D: 23  13.4  2 4
1)
If ri  2 , then this candidate embedding point should
be removed.
2)
n
n 1
If 2  ri  2 , then the embedding capacity of this
point is determined to be n bits.
27
Simulation
 Experiment shows that information hiding has caused slight
change on PSNR (Peak Signal to Noise Ratio) and the
28
actual compression ratio.
Simulation
 In order to test and measure the effectiveness on hiding
capacity enlargement, we simply bypass the redundancy
evaluation for comparison. Two methods are tested in the
experiments.
 Method 1: With redundancy evaluation.
 Method 2: Without redundancy evaluation.
29
Simulation
Fig. 10.
(a) Crown
(b) Baboon
TABLE I
HIDING CAPACITY OF THE THREE TEST IMAGES
30
（A compression ratio of 0.8 bits per pixel）
Simulation
Fig. 11. Hiding capacity of different compression ratios.
31
Simulation
32
Fig. 12. Four images in the database
Simulation
33
Fig. 13. ROC curves tested on different payloads.
Simulation
 The detector does work only if the message length greatly
exceeds the hiding capacity.
 The proposed steganography scheme can be considered
undetectable in the situation of lower payloads than hiding
capacity.
34
Conclusion
 The contributions of this work are mainly focused on
dealing with two problems: bitstream truncation and
redundancy measurement.
35
備註（1）
36