Upload
others
View
11
Download
0
Embed Size (px)
Citation preview
SECRET DATA TRANSMISSION USING STEGANOGRAPHY
A thesis submitted in partial fulfillment of the
requirements for the Degree of Doctor of Philosophy
(PhD) in Electronic Engineering
By
Muhammad Zeeshan Muzaffar
0909-PDEE-006
Department of Electronic Engineering
School of Engineering & Applied Sciences Isra University, Islamabad Campus
August 2017
Copyright © 2017 by Muhammad Zeeshan
Muzaffar
All rights reserved. No part of the material protected by this copyright notice
may be reproduced or utilized in any form or by any means, electronic or
mechanical, including photocopying, recording or by any information storage
and retrieval system, without the permission from the author.
ii
SECRET DATA TRANSMISSION USING STEGANOGRAPHY
By
Muhammad Zeeshan Muzaffar
(0909-PDEE-006)
Names of Supervisor
Signature: _________________ Dr. Ijaz Mansoor Qureshi Professor, Department of Electrical Engineering, Air University, Islamabad.
CERTIFICATE
It is certified that the research work contained in this thesis has been carried
out under the supervision of Prof. Dr. Ijaz Mansoor Qureshi, at Isra
University, Islamabad Campus is original. It is fully adequate, in scope and
quality, as a thesis for the degree of Doctor of Philosophy.
Signature: ____________________ Supervisor Prof. Dr. Ijaz Mansoor Qureshi Professor, Department of Electrical Engineering, Air University, Islamabad.
Signature: ____________________ External Examiner Prof. Dr. Abdul Jalil Professor, Department of Electronic Engineering International Islamic University Islamabad.
Signature: ____________________ External Examiner Dr. Ihsan ul Haq Principal ICT, Faculty of Engineering & Technology International Islamic University Islamabad.
iv
DEDICATED TO
PROPHET MUHAMMAD (P. B. U. H)
THE GREATEST SOCIAL REFORMER
&
MY WORTHY PARENTS
v
ACKNOWLEDGEMENT
I am thankful to pay my heartiest praises to the Almighty, the
beneficent and compassionate Allah, who blessed me to accomplish the
dream of my beloved parents. I am quite amiably obliged to my supervisor
Dr. I. M. Qureshi, for his massive patronization, extremely fervent guidance
and transcendent vision to ensue innovative concepts entailed to
Steganography. He has always provoked me to discover new heavens of
research. He has always demonstrated pristine paternal love and care for
me as his own child. Indeed, his profuse mentoring escorted me throughout
my life even other than studies.
I pay tribute to highly worthy and exalted teachers Dr. Aamir Saleem
Chaudhry, Dr. Aqdas Naveed Malik, Dr. Tanveer Ahmad Cheema, Dr. Abdul
Jalil, Mr. Rashid Bodla, and all other teachers who have taught and trained
me from nursery to this dignified level. Mr. Rashid Bodla is one of my
teacher who help me to explore the teaching capabilities of myself by giving
a special confidence of speaking in class.
Even All expressions and feelings are inferior and insufficient to level
their profuse love, encouragement and guidance, and also, cannot translate
all the immaculate sentiments; I abode in my heart for parents and friends.
I cannot afford to overlook to tribute all the remaining family
members, including my respectable brothers, and my beloved wife who,
remained balmy, caring, and complaisant and still are in all the arduous
moments are source of love and encouragement. Particularly, I am pleased
to appreciate my selfless and committed friends: Dr. Muhammad Shakeel,
vi
Dr. Atta ur Rehman and Dr. Adnan Aziz who constantly assisted me in
study sessions and course work. In the saga to sacrifices, Ghias Malik, is
appreciable, a true fellow.
vii
ABSTRACT
Cryptography, watermarking and steganography are among the
rapidly emerging techniques pertaining to sustain, authenticate, and exhume
the hidden data, especially when it is transmitted over a public
network/channel. In cryptography, the message is executed and encrypted
artistically that the intended message becomes incomprehensible. Whereas,
water marking technique conceals the data in some cover file quite tactfully
that the data engrossed in the host signal/ entity becomes imperceptible
which is likely to be authenticated later on. Likewise, steganography
approach veils the data in the cover signal indistinctly to deflect the
interception of undesired user. The exchange of encompassed message by
cryptographic system may create suspense for the intruder. But, contrary to
it, in steganography the user is less attracted to the hidden information.
Therefore, the harmonization of both of these technologies can produce
invisible higher level of message protection. In this dissertation, the problem
of imperceptibility, data rate and robustness is formulated and different
approaches are proposed and investigated to solve it.
In first approach, a novel technique to embed information into the
audio signals is proposed. In this regard, the set of all possible values of
amplitudes of audio signals are termed as “audio sample space”. An
algorithm is proposed to subdivide this sample space into subspaces and the
information was embedded into these subspaces. On the other hand, an
algorithm for decoding on receiver side is also proposed. The algorithm has
the capability to work on real time systems and provide sufficient security at
commercial level. The amount of imperceptibility achieved is, quite a distinct
viii
benefit concerning perceptual evaluation of speech quality (PESQ).
In second approach, an innovative steganography technique has
been engineered named as weighted pattern matching (WPM) which is
utilized to collaborate the lifting wavelet transform. The message bits are
insert in the indistinguishable places that are picked from the coefficients of
detail sub-bands by taking edge of the proposed WPM technique. WPM
captures the correlation between the message data block and detail
coefficients help us to configure the exact location that can contain the data
block invisibly. The ultimate results of the experiment exhibit that the WPM
technique enhances invisibility significantly in addition to lossless massage
retrieval.
In third approach, another sustained efficient, imperishable approach
brimmed of heavy payload, named compressive weighted pattern matching
(CWPM) has been invented and applied. CWPM technique has been born
from the combination of compressive sensing (CS) with WPM. Use of CS
provides the higher level of security and bigger payload by means of
compression and encryption. CWPM holds the position where data block can
be embedded on the basis of a weighted correlation.
Our engineered techniques have been compared with the well-known
steganography elaborated in literature review. The results ingenuity proves
that all designed models are far better firm and ingratiating.
ix
ABBREVIATIONS
Abbreviation Term
AWGN ---------------------------------------------------Additive White Gaussian Noise
BER------------------------------------------------------------------------------ Bit Error Rate
CS ------------------------------------------------------------------- Compressed Sensing
CSM ------------------------------------------------------------- Changing Slope Method
CWPM ------------------------------------- Compressive Weighted Pattern Matching
DFT ------------------------------------------------------------ Discrete Fourier Transform
DWT --------------------------------------------------------- Discrete Wavelet Transform
FFT ----------------------------------------------------------------- Fast Fourier Transform
IDFT ------------------------------------------------Inverse Discrete Fourier Transform
IDWT ---------------------------------------------- Inverse Discrete Wavelet Transform
IEEE ------------------------------ Institute of Electronic and Electrical Engineering
IFFT ----------------------------------------------------- Inverse Fast Fourier Transform
ILWT ------------------------------------------------- Inverse Lifting Wavelet Transform
LWT--------------------------------------------------------------Lifting Wavelet Transform
NC ------------------------------------------------------------------ Normalized Correlation
PESQ --------------------------------------- Perceptual Evaluation of Speech Quality
x
PSNR --------------------------------------------------------- Peak Signal to Noise Ratio
SNR ------------------------------------------------------------------- Signal to Noise Ratio
WPM --------------------------------------------------------- Weighted Pattern Matching
xi
TABLE OF CONTENTS
Page
CERTIFICATE --------------------------------------------------------------------------- iii
ACKNOWLEDGEMENTS ------------------------------------------------------------ v
ABSTRACT ------------------------------------------------------------------------------- vii
ABBREVIATIONS------------------------------------------------------------------------ ix
TABLE OF CONTENTS --------------------------------------------------------------- xi
LIST OF TABLES ----------------------------------------------------------------------- xiii
LIST OF FIGURES --------------------------------------------------------------------- xiv
CHAPTER I – INTRODUCTION ---------------------------------------------------- 01 1. Problem Statement ------------------------------------------------------------------ 03 2. Contribution of the Thesis --------------------------------------------------------- 05 3. Organization of the Thesis -------------------------------------------------------- 06 CHAPTER II – LITERATURE REVIEW ABOUT STEGANOGRAPHY --- 08 1. Introduction ---------------------------------------------------------------------------- 08 2. Types of Steganography ---------------------------------------------------------- 08
2.1 Audio Steganography --------------------------------------------------------- 09 2.2 Video Steganography --------------------------------------------------------- 11 2.3 Image Steganography --------------------------------------------------------- 14 2.4 Text Steganography ----------------------------------------------------------- 15
3. Summary ------------------------------------------------------------------------------- 17 CHAPTER III – INTRODUCTION TO COMPRESSIVE SENSING -------- 18 1. Introduction ---------------------------------------------------------------------------- 18 2. Compressed Sensing --------------------------------------------------------------- 19
2.1 The Sensing Problem --------------------------------------------------------- 20 2.2 Signal Representation and Sparsity -------------------------------------- 21 2.3 Incoherent Sampling ----------------------------------------------------------- 22 2.4 Under-Sampling and Sparse Signal Recovery ------------------------- 23
3. Summary ------------------------------------------------------------------------------- 24 CHAPTER IV – CHANGING SLOPE METHOD: A TIME DOMAIN APPROCH FOR AUDIO STEGANOGRAPHY --------------------------------- 25 1. Introduction ---------------------------------------------------------------------------- 25 2. System Model ------------------------------------------------------------------------- 25 3. Proposed Algorithm ----------------------------------------------------------------- 27 4. Performance Graphs ---------------------------------------------------------------- 32 5. Summary ----------------------------------------------------------------------------- 39 CHAPTER V – FREQUENCY DOMAIN TECHNIQUES IN AUDIO STEGANOGRAPHY ------------------------------------------------------------------ 41 1. Introduction -------------------------------------------------------------------------- 41 2. Lifting Wavelet Based Techniques ---------------------------------------------- 41
xii
2.1 Weighted Pattern Matching Lifting Wavelets Transform -------------- 42 2.2 Compressed Sensing for Security and Payload Enhancement in
Audio Steganography -------------------------------------------------------- 55 3. Summary ----------------------------------------------------------------------------- 76 CHAPTER VI – DISCUSSION AND CONCLUSIONS ------------------------ 77 1. Summary of Results -------------------------------------------------------------- 77 2. Future Directions ------------------------------------------------------------------ 80 REFERENCES ----------------------------------------------------------------------- 82
xiii
LIST OF TABLES
Chapter Description Page No.
IV – 1 PESQ values for different values of k --------------------------- 35
IV – 2 PESQ values for different methods ------------------------------ 35
V – 1 SNR/dB in case of text messages in WPM-LWT ------------ 50
V – 2 Possible φ's for Security Enhancement ------------------------ 67
V – 3 Compressibility ratio ------------------------------------------------ 72
xiv
LIST OF FIGURES
Chapter Description Page No.
I– 1 Magic triangle - three contradictory requirements of
Steganography -----------------------------------------------------
02
III – 1 Transparency through compressed sensing ------------------- 22
IV – 1 Vertical division of quantization levels for ---------- 27
IV – 2 Algorithm for embedding the information ------------------------ 29
IV – 3 Flow Chart for information embedding --------------------------- 31
IV – 4 Algorithm for information extraction --------------------------- 32
IV – 5 Flow chart for information extraction ------------------------------ 33
IV – 6 Original vs. Stego Signal for q=16, k=2 -------------------------- 37
IV – 7 Zoom area shown in rectangle in Figure IV-6 ------------------ 37
IV – 8 Zoom area shown in rectangle in Figure IV-7 ------------- 38
IV – 9 Cross Correlation between Original audio and Stego one 38
V – 1 Embedding Phase in WPM-LWT ------------------------------------ 45
V – 2 Identification of Embedding Locations ----------------------------- 46
V – 3 Extraction phase in WPM-LWT -------------------------------------- 47
V – 4 Robustness of proposed scheme against AWGN with
different carrier signals ---------------------------------------------
49
V – 5 Robustness of other schemes ------------------------------------- 49
xv
V – 6 Spectrogram analysis of three male voices ----------------------- 50
V – 7 Spectrogram analysis of three female voices -------------------- 52
V – 8 Time domain sound wave analysis of original and stego
signal --------------------------------------------------------------------
52
V – 9 Secret message transmission using CS with WPM-LWT ----- 56
V – 10 Preparing Audio Segments for Embedding ----------------------- 57
V – 11 Encryption and compression of secret image -------------------- 58
V – 12 Bits Embedding location selection and Bits Embedding ------- 62
V – 13 Audio Reconstruction --------------------------------------------------- 63
V – 14 Reconstruction of secret message at Receiver ------------------ 64
V – 15 Secret message recovery for (a) Original (b)
n=512,m=250 (c) n=512,m=200 (d) n=512,m=150 (e)
n=512,m=120 (f) n=512,m=90 ----------------------------------
70
V – 16 Secret Message Recovery for (a) Original (b) m=150 (c)
m=120 (d) m=100 --------------------------------------------------------
70
V – 17 Secret Message Recovery for (a) Original (b) m=150 (c)
m=120 (d) m=100 --------------------------------------------------------
71
V – 18 AWGN Vs Normalized Correlation ---------------------------------- 72
V – 19 Spectrogram analysis of three male voices ----------------------- 74
V – 20 Spectrogram analysis of three female voices -------------------- 75
V – 21 Time Domain analysis of original and stego signal ------------- 76
1
CHAPTER I
INTRODUCTION
Steganography is the art of writing messages in a hidden manner in
such a way that no one, except the authentic recipient knows of the
existence of the hidden message (R. Chandramouli et al., 2001). The word
steganography is derived from the Greek words “stegos” meaning “cover”
and “grafia” meaning “writing” defining it as “covered writing” (K. Bennett et
al., 2004). The first recorded use of the term was first used by Johannes
Trithemius in 1499 in his book Steganographia (Johannes Trithemius et al.,
1499). Secret data transmission is an evergreen research area.
Steganography is one of the techniques used for secret data transmission.
For example, during the Second World War (David Kahn et al., 1967), a
phrase was used by a German soldier to send a secret message. The
sentence was;
“Apparently neutral’s protest is thoroughly discounted and ignored. Is man
hard hit. Blockade issue affects for pretext embargo on by-products, ejecting
suets and vegetable oils.”
This contains the hidden message which can be built by using the
second letter of every word of the sentence. The message was
“Pershing sails for NY June 1.”
The main idea behind transmission through steganography is to
establish the communication in such a way that it remains imperceptible to
2
the intruders. For this purpose, a cover signal is used which carries the
secret information bits.
Cover signal can be of any type. It can be text (Majumder et al.,
2013), it could be an image (Nagaraj et al, 2013), audio (Avval et al., 2014)
or video (Dasgupta et al., 2013). In the same way, the secret message
embedded into cover signal can be of any type like text, image, audio and
video. So, all types of steganographic combinations exist by keeping one
thing in mind, that the cover signal should have sufficient number of bits that
it can hide the message bits. This theme is explained in fig-1. This figure is
named as “magic triangle”. This shows the simplest requirements of
information hiding in any digital media (Zhang et al., 2007). This explains
the representation of the trade-offs between the capacity of the
steganography data and robustness, while keeping the perceptual quality at
an acceptable level. Mainly the steganography is of two types (Kaur et al.,
2012).
Fragile
Figure I-1: Magic triangle - three contradictory requirements of Steganography
Imperceptibility/ inaudibility
Data Rate Robustness
3
Robust
Fragile means to hide information in a cover file that is destroyed if
the cover file has been modified. This method is not suitable for the copyright
holder of the cover file because it can be easily removed, but it is useful in
situations where one have to prove that this cover has not been tampered.
Fragile steganography tend to be easier to implement than a robust one.
Robust Steganography aims to hide information into a cover which cannot
easily be destroyed. A system considered as robust if the changes required
to removes the mark makes the file useless.
It is not possible to attain high data rate of the embedded message
as well as, high robustness at the same time. So, a high robustness means a
low data rate of the secret message. High data rates mean, most of the
times the signal is very fragile to signal modifications.
1. PROBLEM STATEMENT
Secret data transmission is one of the oldest and hottest areas of
research in all the times. Cryptography, steganography and watermarking
are the most popular techniques in this regard. Unlike cryptography,
steganography and watermarking benefits from the perception limitations of
human auditory and visual systems, which fail to recognize difference
between host and stego-signals/watermarked signals, respectively. (Huang
et al., 2010) Usually, in steganography the media files such as, image, audio
or video are used as host signals to hide the message data. In general,
using an image or video as steganography cover signal is more popular than
4
the audio signal. This is because human visual system (HVS) is far less
sensitive to noise/change in the signal than human auditory system (HAS).
The steganography algorithms are in need of some features which
depend on the transmission media and applications. The most important
requirements are imperceptibility (transparency), robustness (security
against certain attacks) and high embedding capacity. A number of digital
audio steganography techniques have been proposed in the literature.
Among them, least significant bit (LSB) based audio steganography
technique is the first and foremost technique in which the data is embedded
in LSB of the cover audio in time domain (Bender et al., 1996). The LSB
technique is considered to be the most imperceptible but least robust at the
same time since in many attacks LSBs of the signal are
destroyed/eliminated. Further to increase the robustness and embedding
capacity, higher bits like 3rd and 4th LSBs have also been used but it was
noted that the perceptual quality of the output signal is compromised (Cvejic
et al., 2004).
The prime objective of this dissertation is to find out and investigate
some new techniques for digital audio steganography so that
imperceptibility, security and robustness can be jointly optimized, while the
payload (capacity) may be maximized at the same time and the unintended
user/intruder cannot sense it throughout the secure transmission span. In
short, new algorithms are proposed and investigated for the solution of the
above constrained optimization problem that can optimize security,
imperceptibility and robustness, while maximizing the payload. The
effectiveness of the algorithms is depicted through computer simulations.
5
2. CONTRIBUTION OF THE THESIS
This dissertation contains four major contributions in the field of
audio steganography for imperceptibility, robustness, security and data rate
enhancement.
In first contribution, a technique is proposed to for message hiding
where an audio signal is used to hide the secret information. The theme of
the scheme used is first developing different groups of equal length of all the
possible amplitudes of cover audio signal. All amplitude levels of all groups
have some specific meaning and number of possible amplitudes in each
group is related with the possible secret bits attached with each sample. On
the other hand, secret data bits are also kept in groups of equal length,
called chunks, and one chunk is embedded in one sample of cover audio.
Main contribution towards this research work is the algorithm that is used to
embed chunks of secret message in cover audio by using changing slope
method (CSM). Changing slope means the change of slope of the line
joining any two consecutive samples of cover audio. The benefit achieved
here is the value of perceptual evaluation of speech quality (PESQ) which is
4.492. This value shows that graphs of original audio and stego signal are
very close that means a high level of imperceptibility is achieved.
In second contribution, a novel technique is proposed and
investigated to resolve the issue of robustness. The technique is named as
weighted pattern matching (WPM) that is used with three levels of integer-to-
integer lifting wavelet transform (LWT). Integer-to-integer LWT is a lossless
transformation technique with a lesser complexity as compared to other
6
conventional discrete wavelet transform (DWT). Embedding positions of
three levels detail-coefficients between LSB’s and MSB’s provide robustness
and WPM provides maximum normalized correlation between the secret
message and detail-coefficients which eventually ends up in improved
imperceptibility.
In third contribution, we have employed compressed sensing (CS)
for secret data transmission in conjunction with lifting wavelet transform. The
secret message is compressed by CS prior to embedding. From CS two type
of benefits are achieved that are payload compression (enhancement) and
the security. After compression, the message (image) is embedded in the
host audio using multilevel lifting wavelets as well as WPM. From
simulations, it is depicted that the proposed scheme significantly increases
the payload and security.
In fourth contribution, the idea of using compressed sensing (CS)
with lifting wavelets is investigated for a pure audio environment, that is, the
message and the cover signals are both audio. The secret audio message is
compressed by CS prior to embedding in the cover. After compression, the
message (image) is embedded in the host audio using multilevel integer
lifting wavelets as well as WPM. Simulations depict the enhancement in
security and payload of the proposed scheme significantly.
3. ORGANIZATION OF THE THESIS
Chapter II provides an overview of steganography schemes
investigated in the literature over the past two decades. It reviews audio,
7
image and video steganography and their algorithms in different
steganographic systems.
Chapter III gives a brief literature review of compressed sensing and
their applications in different domains.
Chapter IV is dedicated to changing slope method, a novel
technique of audio steganography in time domain and simulation results.
Chapter V contains the proposed frequency domain techniques for
audio steganography and their applications.
Chapter VI summarizes and concludes the work and gives
suggestions for the future work.
8
CHAPTER II
LITERATURE REVIEW ABOUT STEGANOGTRAPHY
1. INTRODUCTION
This chapter presents a comprehensive review on steganography, its
history, applications and various techniques of it. The concept of
steganography is not new; it is more than thousand years old. The old
Greeks used steganography for secret message transmission. Now, in
modern steganography, secret message can be embedded into some kind of
cover signal in an imperceptible way. This secret message can be retrieved
from the stego signal by the person who has the knowledge of the receiving
technique and some stego key, if applicable.
2. TYPES OF STEGANOGRAPHY
After a careful literature review, it is observed that the digital
steganography techniques can be classified into a number of categories
given below based on the type of media (Pawar et al., 2014).
1. Audio steganography
2. Image steganography
3. Video steganography
4. Text steganography
9
2.1 Audio Steganography
In literature various techniques have been proposed for digital audio
steganography (Mat M.L. et al., 2011). These techniques can be mainly
categorized into two domains. Some of the techniques are implemented in
time domain and some of them are in the frequency domain. Time domain
methods include low bit encoding, echo hiding, while frequency domain
includes phase coding and spread spectrum hiding (Djebbar F. et al., 2012).
Low bit encoding techniques have been widely used in digital steganography
(Muhammad A. et al., 2011, Singh et al., 2010, Deepak D. et al., 2012) used
the low bit encoding for hiding the secret message. He used different
channels of a wave (.wav) file to store different characters of same message.
Then he created more randomness by scrambling bits of each information
character before placing each character bit in LSB’s (Least significant bit) of
their respective channel bytes. This scrambling fashion is already known at
receiver side to build up original character from the received signal. In
(Bhagyashri A. P. et al., 2013) another form of low bit encoding is visited. In
this paper, a public key encryption (PKE) algorithm is used to embed secret
information and achieved security for hidden information in this manner.
Ashwini Mane also gives an LSB technique for data hiding (Mane et
al., 2012). They proposed an algorithm for data embedding and used this
algorithm for offline data security. The algorithm embeds secret data into
LSBs of audio signal source by using a secret key and store stego file on
drive. Then retrieval of the secret information is only possible by knowing the
key.
10
In method of echo hiding, the information is embedded in the echo
part of the cover audio signal (Mane et al., 2011). The echo is actually a
resonated copy of the data which is added to the cover audio signal and
hence, the issue of additive noise is mitigated here. In echo hiding the
parameters to be considered are namely the initial amplitude, the offset
(delay) and the decay rate. These are to make the echo least audible or
imperceptible. Low detection ratio and lenient detection are the main
disadvantaging factors of this scheme. Due to these factors this scheme
could never got much attention of the researchers in steganography.
Phase coding is another method used by researchers to embed
secret data into the phase of cover audio. Using frequency domain Phase
coding, (Salah et al., 2011) hid the watermark information in a WAV file. By
reading the WAV file, first the header information is separated and then the
FFT of the rest of the data is taken by part by using Butterfly method. After
that, every result has two parts the real and imaginary. Then by using these
real and imaginary parts, phases of those results are calculated. Then these
phases were modified and the information was embedded. Then the real
and imaginary parts were recalculated according to new phases the IFFT
was taken using direct method. Finally the header was attached which was
split in the start and a new wave (.wav) file was written which contains all the
information.
According to (Bandyopadhyay et al., 2012), after split the header of
audio file, rest of the data portion is divided into segments of equal lengths.
The length of segments is equal to the size of message to be encoded. Then
11
the DFT of each segment was taken and the matrix of phases was obtained
as a result. Static phase coding was used in which fix phase of (π/2) is
added if secret data bit was 1 and the fix phase (π/2) is subtracted if secret
data bit was zero.
Researchers also used the class of Genetic algorithms for secure
data transmission with audio as cover signal. In this regard, Bhowal used
Genetic algorithm for image transmission by using audio signal as cover
(Bhowal et al., 2011). They used the technique to replace some bits of audio
sample with some of image bits and then applied the Genetic algorithm on
the remaining bits and the closest guess of original audio sample was found.
In this way, they took the benefit of HAS and achieved imperceptibility for
cover audio. Genetic algorithm in audio cover has also been used in (Zamani
et al., 2009).
Some researchers use different type of time domain algorithms
which actually choose some amplitude after some calculation on cover
signal and the new amplitude also contain their secret data. Like in
(Bhattacharyya et al., 2011), a mod 4 method is used to change the
amplitude of the cover signal and during this process, secret data is also
embedded. Like in example, they take the cover signal sample’s amplitude
as 23 and after applying the algorithm, output modified amplitude becomes
22 which also carry the two information bits 10.
2.2 Video Steganography
Digital video steganography is a more capacitive way to embed the
secret data as compared to digital audio steganography technique (Cheddad
12
et al., 2009). This is because the video consists of a number of frames that
are moved in a sequence per second, usually called frame rate, which is
measured in frames per second (fps). Each frame is kind of an image of a
certain resolution like 480p, 720p, 1080p etc.
In an article by Rahangdale, authors first collected information about
the cover video and selected a video frame and separated it from the original
one. After that, a hash based approach is used to find the 4 LSB positions of
that frame to embed the secret information. After embedding the secret
information, this stego frame again combined with the remaining video.
(Rahangdale et al., 2014)
Kelash used the technique in which they made segments of the
cover video sequence into frames and calculated histograms of each frame,
then based on these histograms comparisons, consecutive histogram
variations to be computed and by seeing these variations, appropriate pixels
to be selected for data embedding. (Kelash et al., 2014)
Saurabh used the concept that in a video of 30 frames per second, it
is very difficult for intruder that which frame carries actual hidden data
especially when only one LSB of selected frames are used for data hiding,
because every frame is difficult to analyze when data rate is very high.
(Saurabh et al., 2010)
In (Delforouzi et al., 2006), authors have employed five levels of
packet integer lifting wavelet transform (ILWT) to decompose the cover
audio into the sub-bands. After that, according to each sub-band the hearing
13
threshold was calculated for its corresponding ILWT domain sample. Data
bits were embedded in the LSBs of the ILWT coefficients based on the
calculated thresholds. Consequently, back in the time domain inverse ILWT
is applied on the modified coefficients to construct the stego audio signal. In
this research, more than 200kbps embedding capacity was achieved
provided with lossless data recovery.
Shahreza and Shalmani have proposed a novel digital audio
steganography technique, based on the ILWT, respectively. The data was
embedded in the LSBs of the detail coefficients after decomposing the cover
audio signal by means of ILWT. In this scheme, 20% of the input speech
signal embedding capacity was achieved with an acceptable ratio of
imperceptibility and a successful recovery has been accomplished.
In another study, authors wavelet packet transform (WPT) was
utilized by the researchers to audio cover decomposition into equal levels.
Once the due scaling factor was applied and sophisticated conversion from
audio to bits was done, the LSBs of the embedding candidate detailed
coefficients, were selected. After that, the bits block matching between the
LSBs of the host details coefficients and the message bits was performed in
order to select the optimal positions for embedding the message bits. After
that the altered coefficients were descaled and inverse WPT was performed
in order to recover the stego audio signal. Further, in this research the
authors claimed a very high embedding capacity that was about 300 kbps,
with at least 50dB signal to noise ratio (SNR) for the resultant
imperceptibility. However, the proposed scheme could not withstand the
14
attacks with the results, robustness was being compromised due to the
multiplicative scaling. (Shahadi et al., 2011).
To overcome the above cited problem same authors proposed
another very high capacity digital audio steganography scheme with a slight
modification in the existing scheme (Shahadi et al., 2014). In this technique
weighted block matching (WBM) was performed to find the suitable positions
for data embedding. Eventually, the authors showed by the simulations that
an embedding capacity of 300Kbps achieved with a transparency of 35dB
for the cover to noise ratio. Although the robustness was better than
previous approach, still the scheme was not robust against certain attacks.
Moreover, the transparency factor is also compromised that is reduced from
50dBs to 35dBs.
2.3 Image Steganography
Digital image steganography has become a prominent field in data
hiding for information protection. It involves the process of hiding information
into an image and output is a stego-image (Cheddad et al., 2009).
In (Qin C. et al., 2012), the authors proposed and investigated a new
reversible data hiding algorithm based on adaptive embedding technique
with integer transform. It embeds as high data rate as 2.17bits per pixel with
a PSNR of 20.71dB.
(Ghasemi et al., 2011) used transform domain to improve the
robustness of steganography system and the perceptual quality was achieve
with the help of genetic algorithm (GA) and optimal pixel adjustment process.
15
Simulation results show that this novel scheme gives a PSNR of 39.94 dB
and 50% capacity.
Many combinations of soft computing techniques are also used in
digital image steganography. Like in (Ghasemi et al., 2010), a combination of
support vector machine (SVM) and fuzzy c-means (FCM) is used to hide
data in images. FCM have the strength of clustering and SVM used in
classification. Combination of both soft-computing techniques used as the
main engine of this purpose that enhanced the imperceptibility and payload
capacity in the images.
2.4 Text Steganography
Text steganography is an art of hiding secret information in written
natural language text. (Chapman et al., 2001, Shirali-Shahreza et al., 2010)
proposed a new technique for information hiding by using Persian (Farsi)
and arabic texts. Arabic letters “Ya” and “Kaf” are two different characters
having the same shape in Unicode standard. At the middle and beginning of
the words, Unicode standard use different codes for these letters. Thus, by
using these two characters, information can be hidden in these two letters.
Java programming language has been used to implement this since it has a
built-in Unicode support. (Shirali-Shahreza et al., 2010)
Another method is proposed by using Arabic Unicode texts for data
hiding in (Azawi et al., 2011). In this article a text steganography technique
for Arabic Unicode texts was introduced. Two special characters are used in
Arabic Unicode text for joining and prevent joining of two Arabic letters. Zero
16
width joiner character (JWZ) that are used to join the two Arabic letters and
zero width non-joiner characters (JWNZ) that are used to prevent joining of
two letters. Two regular expressions (REs) were used to generate a
sequence of these two special characters that consists of JWZ and JWNZ
characters for information hiding.
Rabah used a method for text steganography and altered the
features of text for data hiding. He used to elongate or shorten the letters a
little that cannot be felt by the reader and embed secret message in it
(Rabah K. et al., 2004).
Another approach has been used for text steganography in Persian
and Arabic text presented in (Shirali-Shahreza et al., 2006). Many Arabic
and Persian letters have dots. These dots can hide information bits by their
little bit vertical movement, but this method is not applicable for English text
because only two letters “i ” and “ j ” have dots. That is why capacity for
hidden text in English by using this method is very low.
Abbreviation is also used to hide data in SMS. In (Shirali-Shahreza
et al., 2007), abbreviated words of SMS represent “0” and original form
represents “1. SMS steganography have many advantages. It is very cheap
way to transmit secret message and does not attract people because most
of the people use abbreviation in sending mobile SMS. On the other hand, it
needs a little bit of processing and therefore can easily be implemented on
low cost mobiles. The main disadvantage of this method is that if somebody
has the same algorithm routine, then message can easily be decoded.
17
Arabic language based text steganography method is presented in
(Shirali-Shahreza et al., 2008). Author used to embed information in the
word “La” and used its two different forms to embed “0” and “1” respectively.
This is also the text feature based steganography.
In (Souvik et al., 2013), the author used a different approach for text
based steganography. In this method, properties of sentences are ignored
and characteristics of the English language are used. This results in
increased computational complexity. On the other hand, flexibility and
freedom from sentence construction point of view is also increased.
3. SUMMARY
This chapter reveals that digital audio steganography is a prominent
area of research in the field of information security. For sake of feature
optimization and mitigating certain perceptual measures a number of
techniques have been investigated. It, however, can be observed by the
extensive literature review that some parameters are fixed and other are
optimized. For example, at a constant embedding, robustness and
perceptual measure can be optimized. Similarly, capacity and
imperceptibility can be enhanced, while robustness ignored and also
capacity and robustness with constrained imperceptibility. Till so far, no such
scheme has been proposed that can jointly optimize all of the features of
magic triangle expressed in Figure-1 of chapter 1.
18
CHAPTER III
INTRODUCTION TO COMPRESSIVE SENSING
1. INTRODUCTION
This chapter consists of brief explanation of the tool, named
compressed sensing (CS), used in this dissertation for the sake of efficient
digital audio steganography. Originally the concept of compressive sensing
was developed by Candes (Candès et al., 2006) and Donoho (Donoho et al,
2004), respectively. In this technique random projections of the signal are
taken and from a small number of measurements using optimization
techniques, they are recovered. Conventional schemes of sampling, like
pulse coded modulation (PCM), a signal of any type follows the most
common theorem, called Shannon’s sampling theorem/Nyquist rate which
says to sample a signal minimum at twice the maximum frequency present in
the signal. This principle is used in almost every field of signaling. However,
for the signals which are not bandlimited pass through a low-pass filter to
make them bandlimited and then apply the Nyquist theorem. For example, in
a standard analog to digital conversion, Nyquist theorem used and the signal
is uniformly sampled at this rate. However, CS simply opposes the Nyquist
theory and gives another dimension in sampling research (Candès et al.,
2006).
19
2. COMPRESSED SENSING
According to compressed sensing theory any signal (no matter
audio, image etc.) can be recovered from far fewer samples then the Nyquist
rate used traditionally in many sampling techniques of decomposing an
analog signal into discrete signal. To make this realistic, CS relies on two
principles: sparsity and incoherence, which pertains to the signals of interest
and sensing modality respectively (Candès et al., 2008).
In the light of sparsity, it is said that the sampling rate of a signal
may be much smaller than suggested by Nyquist. This can also be stated
like that a number of degrees of freedom may exist on which a discrete-time
signal depends on which is comparably much smaller than its original length.
More precisely, CS exploits the fact that many analog signals when
expressed in the proper basis are sparse or compressible in the sense that
they have very concise representations.
Incoherence extends the interchangeability (duality) between time
and frequency domains and reveals the novel idea that a signal in one
domain having a sparse representation can be spread out in its original
domain, just as a delta function or a spike in the time domain is spread out in
the frequency domain that becomes an infinite rectangular function. Put
differently, incoherence says that unlike the signal of interest, the
sampling/sensing waveforms have an extremely dense representation in the
basis (Duarte et al., 2006)
20
This phenomenon can be critically observed and can be stated that it
is possible to design efficient technique using compressed sensing or
sampling that capture the useful information contents embedded in a sparse
signal and compress it into a small amount of data. Above mentioned
requirement of these techniques is based on correlation between the original
signals with a small number of fixed waveforms that are incoherent with the
sparsifying basis (Candès et al., 2006).
2.1 The Sensing Problem
The idea of sensing mechanisms, in which information about a
signal is obtained by linear functional recording the values, can be
expressed as;
⟨ ⟩ (Eq. III.1)
If , then there will be an opportunity in the under-sampled
situations in which the number of available measurements is much
smaller than the dimension of the signal . For many reasons, such
problems are extremely common. For instance, in many cases the number of
sensors is limited or measurements are extremely complex or expensive and
so on. Now the question is that, is accurate reconstruction of the problem
possible for measurements? Moreover, is it possible to design such a
sensing waveform that can capture almost all the information? Also, the
process of recovering from produces infinitely many
solutions in general. But one could imagine a way out of the problem which
can be seen next.
21
2.2 Signal Representation and Sparsity
In compressed sensing, signal representation and sparsity play a
significant role. Let represent a real signal with an assumption that the
signal is sparse in the orthogonal basis { } where is the
length of the signal, then can be represented by a linear combination of
basis functions as
∑ (Eq. III.2)
where be the coefficients to represent in terms of bases. Now we can
see the need of sparsity. When a signal is sparse in some domain, then
discarding a small portion of coefficients will not consequence in a severe
loss. That means it is possible to overcome a significant number of the
coefficients without any significant loss. According to Figure II-1(a & c), there
is an unnoticeable difference in the perceptual quality of both images of a
one megapixel grid while throwing away 97.5% of the coefficients and then
obtaining back the result. (Vetterli et al., 2002).
Figure III-1 (a) shows the original grayscale image with eight bits per
pixel and Figure.III-1 (b) its wavelet transform coefficients (arranged in
random order for enhanced visibility). Out of these huge number of
coefficients, relatively few wavelet coefficients capture most of the signal
energy; that shows that a number of this type of images can be highly
compressed. In Figure-III-1 (c) the signal is reconstructed by far few number
of coefficients in the range [0-255]. From the naked eye test, the difference
22
between the original and recovered image is hardly noticeable. For a perfect
recovery, a small increase in the coefficient can make it easily recoverable.
Figure III-1: Transparency through Compressed Sensing
2.3 Incoherent Sampling
Suppose a given pair of orthobases of . The first basis is
used for sensing the object as ⟨ ⟩ and the second is used to
represent . The restriction to pairs of orthobases is not essential and will
merely simplify our treatment. The coherence between the sensing basis
and the representation basis is
√ |⟨ ⟩| (Eq. III.3)
In other words, the coherence measures the largest correlation
between any two elements of and . The value of the coherence lies in
the range √ . Compressed sensing concerned into low coherence pairs.
23
2.4 Under-sampling and Sparse Signal Recovery
To measure all the coefficients of is ideal, but we only get to
observe a subset of these and collect the data
⟨ ⟩ (Eq. III.4)
Where { } is the subset of the cardinality . Now
norm minimization can be used to recover the signal. The proposed
reconstruction of is given by , where is the solution of the
following convex optimization problem
‖ ‖ subject to ⟨ ⟩ , (Eq. III.5)
Now by means of the following theorem presented below, we can
say that can be recovered by the above convex optimization problem
(Candes et al., 2007).
2.4.1 Theorem: Fix and suppose that the coefficient
sequence of in the basis is S-sparse. Select measurements in the
domain uniformly at random. Then if
(Eq. III.6)
for some positive constant , the solution to ‖ ‖ subject to
⟨ ⟩ , is exact with overwhelming probability.
24
3. SUMMARY
In this chapter, an overview of the compressed sensing technique is
presented. It is a very useful technique and its importance can be seen by its
huge number of applications in networks, communications, imaging and so
on.
25
CHAPTER IV
CHANGING SLOPE METHOD: A TIME DOMAIN APPROCH FOR AUDIO STEGANOGRAPHY
1. INTRODUCTION
Time domain steganography is one of the techniques to embed data
in time domain in cover audio. In this chapter, we focused on the design of
proposed time domain technique of audio steganography termed as
“Changing slope method” which helps us to embed secret message in a
cover audio in an imperceptible way. The more the imperceptibility in an
audio will result in less chance of intruder.
2. SYSTEM MODEL
Let { } be the given cover audio signal, where be
the total number of audio samples. Each audio sample contains bits.
Let { } be the total information that we want to embed
into given cover audio signal, where be the total number of information
bits. So we divide these information bits into chunks of bits, where is the
total number of information bits attached to each sample of cover audio. So,
there are total of ⁄ information chunks, where is multiple of and
. So we can represent those chunks as { }, where
contains bits of information from to , where This
26
division of bits we termed as horizontal division of information bits. Now each
chunk represents a binary number of bits.
Another type of division used in the proposed algorithm is vertical
division which is performed on actual sample space of cover audio signal. In
this division we divide quantization levels into groups of equal height. There
are a total of quantization levels and we divide these levels into groups
containing quantization levels each. So there will be a total of
vertical divisions. These vertical groups can be represented as
{ }, where any contain consecutive quantization levels
from to and . Each is distinct from each
other such that
(Eq. IV.1)
And
(Eq. IV.2)
Where
{ } (Eq. IV.3)
Figure1 shows the quantization level division for , where each
level named as in terms of ’s. ( )th level of our cover
audio sample space, where and . e.g. for ,
represents
(Eq. IV.4)
As shown in Figure IV-1 below.
27
Figure IV-1: Vertical division of quantization levels for
Each will represent a specific information which is binary value of
. e.g. are all the levels which represents binary information
and this is binary code of .
3. PROPOSED ALGORITHM
Our proposed algorithm maps each sample to amplitude which
represents the hidden information and HAS (human auditory system) cannot
sense any type of change. The algorithm has the following number of steps
shown in Figure IV-2 box below. This algorithm is applied to an offline cover
audio file. So, in step1, step2 and step3 just read the cover audio file and
collect all necessary information to process further on the algorithm.
𝑡 𝑡 𝑡 0
𝑁 0
𝑁
𝑁
𝑁 3
𝑁 0
𝑁
𝑁
𝑁 3
𝑁3 0
𝑁𝑝 3
28
As each sample is gray coded in PCM (Pulse coded modulation)
audio, so to find the actual level number, we convert it into decimal in Step4
and Step5. This conversion is applied to whole samples of cover audio so
that each sample can be mapped on its appropriate . Now we have the
cover audio in the form of ’s as
{ } (Eq. IV.5)
Where
(Eq. IV.6)
Step7 to Step15 are recursive. In these steps, we hide our
information chunks between samples of cover audio signal. Step7 computes
the of every two consecutive samples of cover audio and we use this
mean for computing the new mean that will also contain our hidden
information chunks. This new mean is represented by as shown in
Step11 and Step12.
In Step8, is calculated by using operator, e.g
which is actually just remainder. This remainder helps to adjust the secret
information in specified chunk at specified place which is also nearest of the
original amplitude. In Step11 and Step12, is the decimal equivalent of
which is th chunk of information.
29
In Step13, we use two points and to find the updated
value of . We use two point form of line for this purpose as shown
below
(Eq. IV.7)
Step 1: Read the cover audio file
Step2: 𝒌 = number of bits containing any 𝐜𝒍
Step3: 𝒏 = no of audio samples available
Step4: Convert each gray coded sample into its binary equivalent.
Step5: Convert each binary coded sample into its decimal equivalent.
Step6: 𝒙
Step7: 𝑚𝑒𝑎𝑛 (𝐍𝑖𝒙 𝑗𝒙 𝐍𝑖𝒙 𝑗𝒙 ) ⁄
Step8: 𝒓 𝑚𝑒𝑎𝑛 𝑚𝑜𝑑 𝒌 𝟏
Step9: 𝑛𝑚𝑒𝑎𝑛 𝑚𝑒𝑎𝑛 𝒓
Step10: if 𝑚𝑒𝑎𝑛 , then go to Step11, otherwise go to Step12
Step11: 𝑛𝑚𝑒𝑎𝑛 𝑛𝑚𝑒𝑎𝑛 5𝐜 𝑥,Go to Step13.
Step12: 𝑛𝑚𝑒𝑎𝑛 𝑛𝑚𝑒𝑎𝑛 5𝐜 𝑥 ,Go to step 13.
Step13: 𝐍𝑖𝒙 𝑗𝒙 (𝑛𝑚𝑒𝑎𝑛 𝐍𝑖𝒙 𝑗𝒙) 𝐍𝑖𝒙 𝑗𝒙
Step14: 𝒙 𝒙
Step15: if 𝒙 𝒏 go to step 7. Otherwise go to step 16.
Step16:- convert each decimal audio sample into gray code and save file.
Step17: send wav file to receiver.
Figure IV-2: Algorithm for embedding the information
30
(Eq. IV.8)
( ) (Eq. IV.9)
After updating , algorithm will run for
and
and this process continues up till last sample.
We can see algorithm working by taking some example. Let us
suppose that, , , and . At Step7,
5, then 5 and raw form of at Step9. Now as
. So, we will move to Step12. At this step, 5. So,
updated . Now we will repeat for next sample updating with this
new value. At last, after completing samples updating, before sending file to
receiver, again convert each sample into its gray code PCM .WAV file as our
cover audio was also gray coded. Then send it to receiver via some media.
Now let us see how we can recover our information chunks back on
receiver side. In Figure4 below, Step1 to Step6 is the basic information
collection steps which are necessary to run the algorithm. Step7 computes
the because our information resides in this . is remainder. Step9
computes which is decimal value of th information chunk. When these
chunks are computed for all received audio samples, then we will convert all
chunks into binary. Then we will arrange these chunks and our required info
will be recovered.
31
Start
𝒌 Number of attached bits of information
𝒏 Number of audio sample of cover audio | 𝒙
Read wav file
Convert each Gray coded
Sample into binary
Convert each binary Sample into its decimal equivalent
if
𝒙 =1 to 𝒏
Yes
𝑚𝑒𝑎𝑛 ( 𝑖𝑥′𝑗𝑥 𝑖𝑥 ′𝑗𝑥 ) ⁄
𝒓 𝑚𝑒𝑎𝑛 𝑚𝑜𝑑 𝒌
𝑚𝑒𝑎𝑛
if Yes
𝑛𝑚𝑒𝑎𝑛 𝑚𝑒𝑎𝑛 𝒓
𝑛𝑚𝑒𝑎𝑛 𝑛𝑚𝑒𝑎𝑛 5𝐜 𝒙
No
𝑛𝑚𝑒𝑎𝑛 𝑛𝑚𝑒𝑎𝑛 5𝐜 𝒙
𝑖𝑥 ′𝑗𝑥 (𝑛𝑚𝑒𝑎𝑛 𝑖𝑥′𝑗𝑥) 𝑖𝑥′𝑗𝑥
No Convert each decimal audio Sample into its gray code
Save wav file
Stop
Figure IV-3: Flow Chart for information embedding
32
4. PERFORMANCE GRAPHS
In this section, the results of our proposed algorithm are presented.
For this purpose, a PCM coded wave file is used shown in Figure6. It
contains samples and bits. So, there are a total of
55 levels in the whole signal sample space. We used
of information bits attached secretly by using our proposed
algorithm. So, and .
So, there will be a total of vertical divisions and
55 5. So, we represent our chunks as
Step1: Read received wav file
Step2: 𝒌 = number of bits containing any 𝐜𝒍
Step3: 𝒏 = no of audio samples available
Step4: Convert each gray coded sample of wav files into binary
Step5: Convert each binary sample into its decimal equivalent
Step6: 𝒙
Step7: 𝑚𝑒𝑎𝑛 (𝐍𝑖𝒙 𝑗𝒙 𝐍𝑖𝒙 𝑗𝒙 ) ⁄
Step8: 𝒓 𝑚𝑒𝑎𝑛 𝑚𝑜𝑑 𝒌 𝟏
Step9: 𝐜 𝒙 𝒓
Step10: 𝒙 𝒙
Step11: if 𝒙 𝒏 , then goto step 7
Step12: Convert each value info vector into its binary equivalent and
arrange in on sequence.
Step13: Show information
Figure IV-4: Algorithm for information extraction
33
{ }
Each new signal level which is modified after processing through our
algorithm can differ a maximum of from the original level of the audio
signal. This creates a very close graph to the original one if as shown
Start
𝒌 Number of attached bits of information
𝒏 Number of audio sample of cover audio|𝒙
Read wav file
Convert each Gray coded
Sample into binary
Convert each binary Sample into its decimal
equivalent
if
𝒙 =1 to 𝒏
Yes
𝑚𝑒𝑎𝑛 ( 𝑖𝑥′𝑗𝑥 𝑖𝑥 ′𝑗𝑥 ) ⁄
𝒓 𝑚𝑒𝑎𝑛 𝑚𝑜𝑑 𝒌
𝐜 𝒙 𝒓
No
Convert decimal information into
its binary
Show information
Stop
Figure IV-5: Flow chart for information extraction
34
in Figure IV-6, original and stego signal are overlapped with each other and
seems to be as one audio signal. To see the difference between both of the
levels we zoom our graph up till the area shown in the form of rectangle in
Figure IV-6. After zooming the area shown in Figure IV-6, the result
becomes Figure IV-7. But in Figure IV-7, not seen any major difference and
graph is seems to be one again. Now we zoom again up to the rectangle
shown in Figure IV-7 and result after zooming becomes Figure IV-8.
In Figure IV-8, we can see two graphs but they are still very close.
So, the zooming factor shown in Figure IV-8 used again and Figure IV-9 is to
be formed. In this figure, we can see the difference between both graphs.
This difference is so small that to see that difference, we use various levels
of zooming. That’s why, such a small difference cannot be detected by
Human auditory system (HAS).
Figure IV-9 shows the cross correlation between the original audio
and the stego one. This cross correlation also seems to be as auto
correlation of same audio signal and this thing also shows the
imperceptibility of our stego algorithm.
And the last one and the reliable one technique up till so for to judge
the human audio imperceptibility is Perceptual Evaluation of Speech Quality
(PESQ) algorithm (ITU-T Recommendation, 2005, Perceptual Evaluation of
Speech Quality (PESQ), 2001) which is an International Telecommunication
Union (ITU) standard. PESQ values normally vary between 1.0 and 4.5.
Value of PESQ will be less than 1.0 in worst cases and more the grater
values of PESQ shows the more best quality of audio signals. Our PESQ
35
values for different cases are also shown in Table IV-1 and result is very
good for lower values of . Comparisons of other techniques are available in
Table IV-2
Table IV-1: PESQ values for different values of k
PESQ
1 4.496798
2 4.494659
3 4.489622
4 4.470681
5 4.432074
6 4.254979
Table IV-2: PESQ values for different methods
Method Author PESQ
VOIP Yijing Jiang et. al. (2016) 3.2
36
FFT Siwar Rekik et. al. (2012) 4.14
DWT-FFT Siwar Rekik et. al. (2012) 3.68
G.711 A-law Wojciech Mazurczyk (2012) 4.015
GSM-FR Wojciech Mazurczyk (2012) 3.269
SpeexI Wojciech Mazurczyk (2012) 3.992
SpeexII Wojciech Mazurczyk (2012) 3.527
37
Figure IV-6 : Original vs. Stego Signal for q=16, k=2
Figure IV-7 : Zoom area shown in rectangle in Figure IV-6
38
Figure IV-8: Zoom area shown in rectangle in Figure IV-7
Figure IV-9: Cross Correlation between Original audio and Stego one
39
5. SUMMARY
In this chapter, a changing slope method (CSM) is proposed for
audio steganographic system. The benefits achieved in our proposed
scheme are as follows.
Firstly, the information is not hidden into the amplitudes but the
means of consecutive amplitudes. If one amplitude changes, then only the
leading and trailing information chunks will be destroyed and this error will
not propagate further.
Secondly, the average PESQ values for the existing networks are
3.8 (The PESQ Algorithm as the solution for speech quality evaluation on
2.5G and 3G Networks, 2009). Whereas, the PESQ estimates for our
proposed scheme is 4.496 for , 4.494 for and 4.489 for
which is very much close to 4.5 the maximum value. The measures of PESQ
also show that human auditory system cannot detect the change in stego
signal.
Thirdly, the cross correlation also seems just like the autocorrelation
and this thing also reflect that original and stego voices are very close to
each other. This thing also results in minimum change in audio quality and
introduces imperceptibility.
Fourthly, the size of the carrier and stego files are the same. So, the
memory also reflects that no data is embedded into the file. Change in size
of wave file can result in gaining intruder attention and our scheme may fail.
40
Lastly the graph of stego signal is very much close to the original
audio. For that reason, the stego file is very imperceptible not only for HAS
but also from the graph shown.
41
CHAPTER V
FREQUENCY DOMAIN TECHNIQUES IN AUDIO STEGANOGRAPHY
1. INTRODUCTION
In previous chapter, we have designed a changing slope method to
improve the imperceptibility (i.e. to increase the valve of PESQ) but the
method was not robustness. A small change in amplitude can result in loss
of data. That’s why, the method was only suitable for offline storage. In this
chapter, two strategies are proposed. First is used to increase robustness
and imperceptibility. Second is used to increase robustness, imperceptibility
as well as security.
2. LIFTING WAVELETS BASED TECHNIQUES
In order to make the hidden data imperceptible as well as robust, a
number of techniques have been investigated with lifting wavelets transform
(LWT), compressed sensing (CS) with other wavelets. Wavelets produce two
types of coefficients, approximate and detail coefficients. Details contain the
small energy parts of voices. This is the reason that by modifying details of a
voice, overall effect on the voice is small. Approximates contain the major
part of voice energy. Modification of approximates give a serious loss in
voice energy and imperceptibility effects a lot. Following the detail about
each technique is given.
42
2.1 Weighted Pattern Matching Lifting Wavelets
Transform (WPM-LWT)
Weighted pattern matching is used for maximum similarity and
minimum change between the replaced secret information and cover audio
signal. In this way, the maximum similarity will be shown and minimum
change will not alert the intruder. The system model will be as follows
2.1.1. Embedding: The embedding procedure is shown in fig-1 while its
description is given as:
Let be a cover audio of samples and each
sample having bits. So carrier signal has bits. be
the secret text message having characters of bits each. The following
steps are performed to embed the secret message in cover audio .
Cover audio segmented into segments such that , where
is the number of samples in each segment and in powers of . i.e
(Eq. V.1)
Where
(Eq. V.2)
Text message segmented into segments such that
where be the length of each message segment. i.e.
(Eq. V.3)
Where
43
(Eq. V.5)
Also .
,LWT of each cover audio segment taken three times. First time LWT will
produce
⁄ coefficients as detail and
⁄ as approximate represented by
and , respectively. Second pass of LWT applying on produce
and of
⁄ coefficients each. Third pass applying on produce
⁄ coefficients each for and respectively. So we have a total of
⁄
⁄
⁄
⁄ coefficients as detail and
⁄ as approximate. i.e
5 detail coefficients and 5 approximate coefficients. And these
5 detail coefficients are used as our region of interest (ROI).
and are concatenated to form a new vector called .
Map sign vector denoted by and 3 formed to keep the
signs of and , respectively.
Change all –ve values of and with their respective +ve
values and convert it to binary.
Now find the maximum correlation of message segment and
by using the following steps.
o Find the first most significant bit of each coefficient of
which is equal to binary 1 represented by
,where and
⁄
⁄
⁄ , where
each represents the first MSB equal to 1 of coefficient of
segment.
44
o Now, two more factors are used to control the robustness
against LSB attack and tolerance of change can be represented
as and respectively. Where and
size of , if these condition does not meet in any
coefficient of a segment , then this coefficient will not
consider for correlation competition. Where points the
location count from LSB of coefficient of segment and
points the location difference from in the direction of LSB of
coefficient of segment. i.e if , and ,
then the bits of coefficient of segment to find correlation
can be seen in shaded area of figure below, so area from 5
location to location is the ROI (region of interest). A
necessary condition is that message segment. points the
5 location.
o Find correlation of with each ROI present in the coefficients of
each segment of cover audio.
o Find the index with which the correlation of is maximum.
After finding the maximum correlation index, replace the message
segment with that particular coefficients ROI present in and
replace this index in for recovery purpose.
Now convert and into their equivalent decimal values.
45
Figure V-1: Embedding Phase in WPM-LWT
Cover Audio
Segmentation
[ 1 , 2 , 3 , , ]
if
Segmentation
[ 1 , 2 , 3 , , ]
Message
i = 1
Yes
Map Sign Vector of 𝐷3
Change Sign of -ve 𝐷3
LWT of
1 𝐷1
LWT of 1
2 𝐷2
LWT of 2
3 𝐷3 Map Sign Vector of 𝐷12
Change Sign of -ve 𝐷12
𝐷12
Change 𝐷12 into binary
Change 𝐷3 into binary Find max Correlation
Replace in Max
correlation index of 𝐷12
Maximum
Correlation index Replace position of Maximum
correlation index and its bit #
in 𝐷3
No
Stop
𝐷2 𝐷1
ILWT of 3 and 𝐷3
𝐷3
ILWT of 2 and 𝐷2
2
ILWT of 1 and 𝐷1
1
Stego segment , i=i+1
Place –ve sign using Map sign vectors of 𝐷3 and 𝐷12
𝐷12
46
By using and 3 , place –ve signs in and .
Break into and .
can be produced by applying ILWT on and .
In the same way, and stego segment can be produced by
applying the ILWT on , and , respectively. Proceed to
next audio cover segment and repeat the whole procedure.
After producing the all segments, concatenate the all segments
and produce the final stego signal.
Figure V-2: Identification of Embedding Locations
2.1.2 Extraction: Now at the receiver, to reproduce the stego message
, following steps are performed.
Make segments of received signal. i.e .
Now take 3 level LWT of each . First level of LWT produces
and . For second level LWT, used and produce and .
LWT of produces and .
Concatenating and in one vector .
Create map sign vectors and 3 for and
respectively that keeps the sign information alive for these vectors.
0 0 0 0 0 0 0 0 0 1 0 1 1 0 1 1
16 15 14 13 12 11 10 9 8 7 6 5 4 3 2 1 Locations of bits
𝑝𝑖𝑗 𝑀𝑖𝑗 𝐿𝑖𝑗
47
Figure V-3: Extraction phase in WPM-LWT
Change –ve elements present in and with their
respective +ve elements.
Change –ve elements present in and with their
respective +ve elements.
Stego Audio
Segmentation
[ 1 , 2 , 3 , , ]
if
i = 1, Temp=’’
Yes
Map Sign Vector of 𝑠𝐷3
Change Sign of -ve 𝑠𝐷3
LWT of
𝑠 1 𝑠𝐷1
LWT of 𝑠 1
𝑠 2 𝑠𝐷2
LWT of 𝑠 2
𝑠 3 𝑠𝐷3 Map Sign Vector of 𝑠𝐷12
Change Sign of -ve 𝑠𝐷12
𝑠𝐷12
Change 𝑠𝐷12 into binary
Change 𝑠𝐷3 into binary
Read Maximum correlation
index and its bit # in 𝑠𝐷3
No
Stop
Read bits in 𝑠𝐷12 from index #
and bit # coming from 𝑠𝐷3
Append bits coming from upper
level into Temp, i=i+1
Segmentation of
Temp[ 1 , 2 , 3 , , ]
Change each Temp segment
into char[ 1 , 2 , 3 , , ]
Predicted Message
48
Read maximum correlation index number and its respective
bit position from and get
Append bits to a string TEMP each time.
Change these TEMP segments into characters and it will be the
recovered message.
2.1.3 Results and Discussion in WPM-LWT : In this section the
proposed methodology is demonstrated through the MATLAB simulations.
The technique is analyzed in various aspects like transparency, load and
robustness etc.
Figure V-4 shows the robustness of proposed scheme against
additive white gaussian noise (AWGN). To show the level of robustness,
normalized correlation (NC) is used as the figure of merit (Shahadi et al.,
2014). It is actually the correlation between the original cover signal and the
stego audio signal. The maximum value of NC is 1, this is the case when
there is no difference between the original cover signal and the stego audio
signal. Its value varies between 0 and 1 otherwise. Formula for NC is given
below
∑
√∑
∑
(Eq. V.6)
Where L is total number of samples in audio signal and are
original cover signal and the received stego audio signal respectively.
49
Figure V- 4: Robustness of proposed scheme against AWGN with different
carrier signals
Figure V- 5: Robustness of other schemes
50
Figure V- 6 : Spectrogram analysis of three male voices
Table V-1: SNR/dB in case of text messages in WPM-LWT
Cover Audio Sampling Frequency/Hz SNR/dB of text message
Female1 44100 15.2757
Female2 44100 21.1788
Female3 44100 17.0529
Male1 44100 13.3452
Male2 44100 12.3933
51
Male3 44100 14.7456
Bird1 44100 16.0273
Bird2 44100 13.3434
Bird3 44100 12.7889
According to Figure V-4, the proposed scheme is highly robust
against AWGN. The stego signal is introduced different levels of AWGN but
the normalized correlation NC approaches to 1 at -25dBs for all kind of
carrier signals. Different types of carriers are investigated including human
voices (male, female) from different age groups and some natural sounds. It
is worth mentioning here that the robustness of scheme is even higher in the
case of human voices where the NC tappers off to 1 at -50dBs. That means
the proposed scheme is quite useful for day to day human communication.
In contrast to the scheme proposed in Figure V-5 (Shahadi et al. ,2014),
where users demonstrated that their scheme is robust at 35dB and higher,
our scheme saves 60dBs for natural sounds while 85dBs for human voices.
However, the maximum payload for the proposed scheme is 60kbps which
shows that the scheme can work well for the typical voice modems where
supported data rate is 56kbps.
52
Figure V- 7: Spectrogram analysis of three female voices
Figure V-8: Time domain sound wave analysis of original and stego signal
Another mathematical way to measure the signal quality, is to
calculate the signal to noise power ratio given by the formula in (Shahadi et
al., 2014). This is actually the measure of perceptual quality or
53
imperceptibility of the stego signal. In this formula, the stego signal is taken
as a noise. So higher the values of SNR mean stego signal is more
imperceptible and vice versa.
0∑
∑
(Eq. V.7)
Table V-1 shows the calculated values of said SNR for the above
mentioned voices. For the clarification in the results, all the voices are
sampled at 44100 Hz. The minimum achieved SNR in case of text message
is 12.3933dB (male voice) while the maximum achieved SNR in case of text
message is 21.1788 dB (female voice). This shows that in case of female
voice stego signal is more imperceptible compared to other voices.
Figure V-6 shows the difference between the spectrograms of
original cover audio signal and the stego signal. Analysis of spectrograms
shows that there are little changes between the original and stego signal in
the case of three different male voices. This shows that the proposed
scheme does not introduce much changes in the spectrogram. Unlike the
scheme proposed in (Shahadi et al., 2014), where these differences are
more significant that may attract the attacker.
Figure V-7 shows the difference between the spectrograms of
original cover audio signal and the stego signal. Analysis of spectrograms
shows that there are negligible changes between the original and stego
signal in the case of three different female voices. This shows that the
proposed scheme does not introduce much changes in the spectrogram. It is
apparent from the figures that female voices spectra have better
54
performance compared to male voices. This is perhaps because the female
voices exhibit higher pitch compared to male voices. So the frequency
components are crisper and confined in female cases compared to male
voices. Hence their spectra are less vulnerable to the changes offered
because of the proposed scheme. Since, in the proposed scheme the data is
embedded in detail components of LWT.
Figure V-8 shows the comparison of sound waves generated by
original and stego signal. This is carried out for both male and female voice
cases. It is apparent from naked eye view, there is no difference between
any two pair of original and stego signal in all the six examples. Hence the
proposed scheme offers high level of imperceptibility in time domain.
The technique exhibits a very high level of robustness against
common and well known attacks like AWGN, wave behavior and
spectrogram attacks etc. In case of AWGN the signal can survive at a very
poor level of SNR that is -50dB while most of the schemes in the literature
afford same performance at 35dB or higher. That is why, in our comparison
text messages are taken that are highly sensitive to the attacks and the
scheme was able to recover them even in very hostile conditions (below -
50dB). Moreover, upon investigation through MATLAB simulations it is clear
that the scheme contains a high level of imperceptibility as well. However,
the maximum achievable capacity in the proposed scheme is 60kbps, which
makes is still suitable for many voice communication channels like voice
modems that support upto 56kbps etc. The main focus of the scheme was to
55
make the digital audio steganography more robust and imperceptible, and
these goals are achieved well with a fair capacity payload.
2.2 Compressed Sensing for Security and Payload
Enhancement in Audio Steganography
This section has been used compressed sensing for payload and
security enhancement achievements. Now WPM provide the maximum
similarity index, LWT provide robustness and CS provide the payload
enhancement and security. The system model will be as follows.
2.2.1 Compressed Sensing Based WPM-LWT System Model:
This system model can be divided into different parts due to its model
complexity. So each transmitter part and receiver part is divided into sections.
2.2.2 Transmitter: Let be a cover audio of samples
and each sample having bits. So carrier signal has bits. be the secret
image of to be transmitted. The following steps performed for each of
the block shown in Figure V-10
2.2.3 Audio Segmentation: Audio contains samples can be
segmented into segments such that , where is the number of
samples in each segment and in powers of . i.e
(Eq. V.8)
56
Where be the segment of cover audio .
Figure V- 9: Secret message transmission using CS with WPM-LWT
2.2.4 Preparing Audio Segments: Audio segment preparing
involves third level of LWT so that the hidden message can be more
imperceptible as shown in Figure V-10. The outputs of preparing audio
segment section are approximation and detail coefficients where and
Audio Cover Signal
Preparing each Audio
Segment
Bits embedding location
selection
Secret Image
Encryption and
Compression
Bits Embedding
Reconstruct Audio
Transmit
Audio Segmentation
Segmentation
57
are third level approximation and detailed coefficients. is
concatenation of first and second level detail coefficients vectors of
segment .
Figure V-10: Preparing Audio Segments for Embedding
2.2.5 Encryption and Compression of Secret image: For the
sake of encryption and compression of our secret message, the concept of
compressed sensing (CS) to be used here. The Shannon’s sampling
theorem states that to recover a signal, the sampling rate must be at least
the Nyquist rate. Compressed sensing is based on the interesting fact that to
recover a signal that is sparse in some domain representation, one can
sample at the rate far below the Nyquist rate. This concept used here to
Segment
LWT of
58
decrease payload and enhance security as shown in Figure V-10
mathematically.
Figure V-11: Encryption and compression of secret image
Here in Figure V-10, is an image having dimensions
containing messages of each and are the bases vectors matrix of
the domain in which can be sparsely represented with same dimensions
as . Coefficients of the new domain are represented by . Now, is
of and which is sensing matrix having dimensions where
. So, which is our output of this module having dimensions . An
important fact is that matrix is only known at transmitting and receiving
side and hence introduce security and payload reduction.
𝑓 𝜓𝑥
𝑥 𝜓𝑇𝑓
𝑦 𝜑𝑥
59
2.2.6 Segmentation: This module includes the Segmentation of the
encrypted message having dimensions . As secret image is of bit
gray scale image. So there are 𝑠 bits in total to embed. Secret
message segmented into segments such that 𝑠 where be
the length of each message segment. So, now after segmentation, the
secret message will be of the form
3 (Eq. V.9)
Where each segment is of bits long. i.e.
(Eq. V.10)
2.2.7 Bits Embedding location Selection and Bits
Embedding: Bits embedding location selection receives inputs from two
modules which are “preparing audio segments” and “segmentation”. First
module outputs are , and while second module output is .
Bits embedding location depends upon the maximum correlation
between the and message segment . The following steps are
followed to perform the functionality given in Figure V-11
Map sign vector denoted by and 3 formed to keep the
signs of and , respectively.
Change all –ve values of and with their respective +ve
values and convert it to binary.
60
Now find the maximum correlation of message segment and
by using the following steps
o Find the first most significant bit of each coefficient of
which is equal to binary 1 represented by
,where and
⁄
⁄
⁄ , where
each represents the first MSB equal to 1 of coefficient
of segment.
o Now, two more factors are used to control the robustness
against LSB attack and tolerance of change can be
represented as and respectively. Where
and size of , if these condition does not
meet in any coefficient of a segment , then this
coefficient will not consider for correlation competition.
Where points the location count from LSB of
coefficient of segment and points the location
difference from in the direction of LSB of coefficient of
segment. i.e if , and , then the bits of
coefficient of segment to find correlation can be seen
in shaded area of figure below, so area from 5 location to
location is the ROI (region of interest). A necessary
condition is that message segment. points the
5 location
o Find correlation of with each ROI present in the
coefficients of each segment of cover audio.
61
o Find the index with which the correlation of is maximum.
o After finding the maximum correlation index, replace the
message segment with that particular coefficients ROI
present in and replace this index in for recovery
purpose. Now modified detail coefficients becomes
and respectively.
After finding the maximum correlation index, replace the message segment
with that particular coefficients ROI present in and replace this index
in for recovery purpose. Now modified detail coefficients becomes
and respectively.
62
Figure V-12: Bits Embedding location selection and Bits Embedding
2.2.8 Reconstruct Audio: After embedding secret message ,
reconstruction of voice is performed. Approximate, modified detail and map
sign vectors ( , , , and 3 ) are the inputs of this
module. To reconstruct the audio, first we have to convert the binary values
Preparing each Audio Segment
Segmentation
Map sign Vector of
Map sign Vector of
Change sign of negative
Change sign of negative
Change into binary
Change into binary
Change into binaryFind Max Correlation index
Find Max Correlation index
Replace position index in
Replace in Max correlation
index of
Reconstruct Audio
63
to equivalent decimals and place negative signs by the help of and
3 and use inverse transform as shown in Figure V-12.
Figure V-13: Audio Reconstruction
𝑀𝑆𝑉 𝑖 𝑀𝑆𝑉3𝑖 𝑐𝐷 �� 𝑐𝐷 𝑖
Change signs of 𝑐𝐷 �� using
𝑀𝑆𝑉3𝑖
Change signs of using
V
ILWT of and Break
into and
𝑐𝐴 ��
𝑐𝐷 �� ILWT of and
𝑐𝐷 �� ILWT of and
Stego Sound
64
Figure V-14: Reconstruction of secret message at Receiver
Received Audio Signal
Multilevel LWT of each
Segment
Min 𝑋 𝑙
Subject to
�� 𝜑𝑋 𝑙
𝜀
Recovered Secret Image
Received Audio
Segmentation
Multilevel IDWT of each X
𝑠𝐴 𝑡 𝑠𝐷 𝑡 𝑠𝐷 𝑡
Produce ��
𝑋
65
2.2.9 Receiver: At receiver, to recover the secret message from the
received audio , first audio segmentation is performed. Size of segment
should be the same as at transmitter. Received audio segments represented
by where . Now each . Secondly, multilevel LWT performed
uptill the level performed at transmitter which produces the 𝑠 (third level
LWT approximate), 𝑠𝐷 (third level LWT details) and 𝑠𝐷 (first and second
level LWT details). After that, 𝑠𝐷 provide the indexes present in 𝑠𝐷 in
which the located. Then after getting all the ’s, reassemble having
dimensions .
Now from these, we have to estimate the having dimensions
by the use of following problem solving using CVX
Minimize
(Eq. V.11)
Subject to
This passes through the ILWT process and finally get the having
dimensions .
2.2.10 Achievements of CS based WPM-LWT: There are three
main achievements can be seen through CS based WPM-LWT scheme
Multilevel high end security
High end payload compressibility
High normalized correlation
66
First level of security is the basic characteristics of the
steganography and second level of security comes from sensing matrix .
Possible number of ’s are very huge in number and only one of them is of
use to recover secret message. Possible number of ’s shown in Table V-2.
High payload compressibility shows the increase in payload in
percent showed in Table V-3. i.e, if compressibility is 5.6888, it means the
468.88% increase in payload
Normalized correlation measure the change in stego audio before
and after passing through AWGN. This shows that signal is robust against
high attack of AWGN.
2.2.11 Results of CS based WPM-LWT: In this section, the
authenticity of the proposed scheme is depicted through the MATLAB and
CVX simulations in terms of robustness, transparency and payload etc.
For the simulation, Is taken as multilevel two dimensional Discrete
wavelet transform and is the first rows of QR decomposition of a
Gaussian matrix of with { } as entries. So, is of . Nobody can
recover estimate of on receiver side without having . There are
possible ’s and it is near to impossible to check all of them. The possible
huge number of ’s show the level of security achieved here for different
simulations shown in Table V-2
67
Table V-2 : Possible φ's for Security Enhancement
Figure
Figure V-15 (b) 512 250
Figure V-15 (c) 512 200
Figure V-15 (d) 512 150
Figure V-15 (e) 512 120
Figure V-15 (f) 512 90
Figure V-16 (b) 512 150
Figure V-16 (c) 512 120
Figure V-16 (d) 512 100
Figure V-17 (b) 512 150
Figure V-17 (c) 512 120
68
Figure V-17 (d) 512 100
Table V-2 shows the level of security by showing the possible
number of ’s. the minimum number seen in table1 is 5 which
is a big huge number for all possible ’s. this shows that our proposed
scheme meets the security needs of steganography.
Figure V-14(a) shows an original Lina image of dimension 512x512
in grayscale. After applying compressed sensing and embedding it into to
cover audio, its recovered version is shown in Figure V-14(b) with
compressed sensing parameters n=512 and m=250. With m=250 means
104% increase in payload. The recovered image is very close to the original
image in terms naked eye test.
Further in Figure V-14(c) the compressed sensing parameter n is
kept same while m is taken as 200. With m=200 means 156% increase in
payload In this way we are taking lesser information (more compressed and
secure) for embedding in the cover audio. The result shows that the image is
degraded compared to Figure V-14(b) but still it is perceptually acceptable.
Similarly, in Figure V-14(d) the compression parameter m is taken as 150.
With m=150 means 241% increase in payload.
In Figure V-14(e) the compression parameter =120 is taken very
low. With m=120 payload increases 326.67%.
69
In Figure V-14(f) the is taken extremely low as 90. With =90
payload increases 468.88%. That means much secure and compressed but
degraded compared to the former cases.
In Figure V-15 another example is shown. In Figure V-15(a) original
sketch image is shown while in Figure V-15 (b) and (c) its recovered images
are depicted. In both cases the parameter is taken as 150 and 120 and
payload in both cases increases 241.33% and 326.66% respectively. The
recovered images still considerably recognizable. On the other hand, in
Figure V-15(d), for =100, the payload increases 412% but edges of the
images effect badly. This shows that as compression and payload increases,
the quality of the recovered images decreases.
Figure V-16 presented another example of text based images. Both
uppercase and lowercase letters are used in this example and all possible
letters of English used in the phrases. Figure V-16(a) shows the original
image of the text. In Figure V-16(b), (c) and (d), parameter is taken as 150
, 120 and 100 respectively. Figure V-16(b) and (c) shows some degradations
in quality of the images but are easily readable. But Figure V-16(d) shows
degraded badly but still have the recognizable text.
From these results, it is apparent that proposed scheme provides a
good payload with significant level of and security.
Another achievement provided by compressed sensing is the
compressibility ratio which also shows the payload enhancement per second
and can be formulized as;
70
Compressibility ratio =
(Eq. V.12)
Where be the original image and is the image passes through
CS process. Compressibility ratio for different images can be seen in Table
V-3.
Figure V-15: Secret message recovery for (a) Original (b) n=512,m=250 (c) n=512,m=200 (d) n=512,m=150 (e) n=512,m=120 (f) n=512,m=90
Figure V-16: Secret message recovery (a) Original (b) for m=150 (c) for m=120 (d) for m=100
(a)
(e)
(b) (c)
(d)
(f)
(a) (b)
(c) (d)
71
Figure V-17: Secret message recovery (a) Original (b) for n=512, m=150 (c) for n=512, m=120 (d) for n=512, m=100
For the sake of robustness test, normalized correlation (NC) is used
as a figure of merit. NC used to test the level of correlation between the
original and stego audio. Its value varies between 0 and 1. Formula for NC is
Normalized Correlation ∑
√∑
∑
(Eq. V.13)
Where, L is total number of samples in audio signal. and are
original cover signal and the received stego signal respectively. Figure V-17
shows the level of NC and shows the proposed scheme is highly robust
against AWGN. The stego signal is tested at different levels of AWGN but
the NC approaches to 1 at -25dBs for all kind of audio cover signals.
Different types of voices are investigated including human voices (male,
female) from different age groups and some natural sounds. It is worth
mentioning here that the robustness of scheme is even higher in the case of
(a)
(c)
(b)
(d)
72
human voices where the NC tappers off to 1 at -50dBs. That means the
proposed scheme is quite useful for day to day human communication. In
contrast to the scheme proposed in (Shahadi et al., 2014), where users
demonstrated that their scheme is robust at 35dB and higher, our scheme
saves 60dBs for natural sounds while 85dBs for human voices
Figure V-18: AWGN Vs Normalized Correlation
Table V-3: Compressibility ratio
Figure Compressibility ratio
Figure V-15 (b) 512 250 2.048
Figure V-15 (c) 512 200 2.56
73
Figure V-15 (d) 512 150 3.4133
Figure V-15 (e) 512 120 4.2666
Figure V-15 (f) 512 90 5.6888
Figure V-16 (b) 512 150 3.4133
Figure V-16 (c) 512 120 4.2666
Figure V-16 (d) 512 100 5.12
Figure V-17 (b) 512 150 3.4133
Figure V-17 (c) 512 120 4.2666
Figure V-17 (d) 512 100 5.12
The difference between the spectrograms of original cover audio
signal and the stego signal shown in Figure V-18 and Figure V-19. Analysis
shows that there are negligible small amount of changes between the
original and stego signal in the case of different female voices. This shows
that the proposed scheme does not introduce much changes in their
74
spectrograms. It is apparent from the figures that female voices spectra have
better performance compared to male voices. This is perhaps because the
female voices exhibit higher pitch compared to male voices. So the
frequency components are more crisp and confined in female cases
compared to male voices. Hence their spectra are less vulnerable to the
changes offered because of the proposed scheme. Since, in the proposed
scheme the data is embedded in detail components of LWT.
Figure V-19: Spectrogram analysis of three male voices
Figure V-20 shows the time domain changes in cover audio before
and after embedding the secret message. Six examples are taken to
analysis the time domain changes due to our proposed scheme and all the
examples shows that no significant change can be seen in the signals before
and after embedding in time domain.
75
In this technique, the compressed sensing is applied to the message
image prior to embedding in the cover audio by means of LWT. Here,
compressed sensing provides high level security because sensing matrix
is only known at transmitter and receiver and to estimate the real
among the huge number of combinations is nearly impossible.
The simulation results show that the proposed scheme promises a great
enhancement in payload, robustness, security and imperceptibility.
Figure V-20: Spectrogram analysis of three female voices
76
Figure V-21: Time domain analysis of original and stego signal
3. SUMMARY
Two techniques investigated in this chapter, first was WPM-LWT and
second was compressed sensing based WPM-LWT. First strategy provides
the imperceptibility and robustness while second provide the imperceptibility,
robustness, high payload as well as high level of security that cannot be
seen in some encryption systems.
77
CHAPTER VI
DISCUSSION AND CONCLUSIONS
Main contributions of the dissertation including achievements
regarding imperceptibility, robustness, security and payload enhancement
are summarized in this chapter. Some future directions have also been
pointed out.
1. SUMMARY OF RESULTS
In digital steganography, a secret message is concealed in the
digital content. In order to make this hidden information secure, robust and
imperceptible, the secret information is embedded in secret positions of the
digital cover media. Improving the security aspect of steganography system
is one of the challenges in the steganography domain. The steganography
schemes presented in this dissertation comprised of two parts. In the first
part, a time domain steganography scheme is presented in which the secret
data can be hide in slope of the cover audio while in second part, two
frequency domain audio steganography schemes are presented in which
secret message can be hide in wavelet coefficients.
In first part, a time domain scheme named “Changing Slope Method”
presented. This scheme contain the hidden bits in the slope of the line
joining the two consecutive amplitudes of a given digital 16 bit audio cover.
As a 16 bit audio can contain a maximum of 55 different levels and
78
a slight change in audio slope cannot be detected easily by human auditory
system (HAS). So, this method was highly imperceptible. Another advantage
of the scheme is that change in amplitude can result only in loss of leading
or trailing information chunk of hidden information and this error will not
propagate further. Secondly the PESQ score is 4.496 for k =1, 4.494 for k =2
and 4.489 for k= 3 which is very much close to 4.5 the maximum value.
While PESQ for existing networks is near about 3.8. Thirdly, the cross
correlation is just like the auto correlation of the original and stego signal
which also reflect that both voices are very close and cannot be segregated
by the human ear. Fourthly, the size of the stego file and the original file
cannot be changed even a single bit and this file memory minimize the
probability that no data is present in the file. This thing will not alert the
intruder which is provide another level of security. Fifthly, the graphs of
original and stego audios are also very much close and cannot be seen
without applying a high level zooming. This also saves the secure data from
intruder.
In the second part, two frequency domain schemes are presented to
solve the contradictory parameters of digital audio steganography i.e.
imperceptibility, robustness and payload.
In first scheme, a novel technique for digital audio steganography
using a weighted pattern matching approach in LWT is proposed. The
technique exhibits a very high level of robustness against common and well
known attacks like AWGN attack etc. In case of AWGN the signal can
survive at a very poor level of SNR that is -50dB while most of the schemes
in the literature afford same performance at 35dB or higher. That is why, in
79
our comparison text messages are taken that are highly sensitive to the
attacks and the scheme was able to recover them even in very hostile
conditions (below -50dB). Moreover, upon investigation through MATLAB
simulations it is clear that the scheme contains a high level of imperceptibility
as well. However, the maximum achievable capacity in the proposed
scheme is 60kbps, which makes is still suitable for many voice
communication channels like voice modems that support upto 56kbps etc.
The main focus of the scheme was to make the digital audio steganography
more robust and imperceptible, and these goals are achieved well with a fair
capacity payload.
In second scheme, the compressed sensing is applied to the
message image prior to embedding in the cover audio by means of LWT.
Here, compressed sensing provides high level security because sensing
matrix is only known at transmitter and receiver and to estimate
the real among the huge number of combinations is nearly
impossible. The simulation results show that the proposed scheme promises
a great enhancement in payload, robustness, security and imperceptibility.
Summary of the main contribution of the dissertation can be
highlighted and given below. The proposed scheme is;
Higher level of security due to compressed sensing technique which
cannot be seen even in encryption system while keeping the audio
imperceptibility.
Higher payloads which can be seen by higher compressibility ratio
due to compressed sensing.
Highly imperceptible and achieves a high level PESQ score.
80
Highly robust against AWGN attack so that text message can be
transmitted.
Higher normalized correlation between original and stego audios at
low SNR.
Having the same size of original and stego audios.
2. FUTURE DIRECTIONS
In future, a lot of directions can be explored for audio, images and
videos in the field of digital steganography. Few of these directions are given
as below
In the first part of dissertation, the scheme presented named
changing slope method investigated in time domain, this scheme can be
seen in frequency domain. For such scheme, many different types of
frequency domains can be investigated and their comparison can be
presented in case of different attacks. PESQ score can be also be visited for
little change.
In the Second part of dissertation, integer based lifting wavelets
used to transform used to transform the signal in frequency domain. Other
transformation techniques can also be investigated to improve performance
regarding NC or complexity.
On the other hand, we did not use any error correcting codes to
make our information more secure. Different error correcting codes can be
used to make the technique more robust.
81
It’s very hard to transmit a video as a secret message. In video
steganography, secure video in video cover can also be investigated by
applying some sort of higher level of compression techniques on payload.
82
REFERENCES
Al Azawi, A. F. and Fadhil, M. A., “An Arabic Text Steganography Technique Using Zwj And Zwnj Regular Expressions,” International Journal Of Academic Research Vol. 3. No. 3. May 2011.
Ali, A. H., Mokhtar, M. R. and George, L. E., "Recent Approaches for VoIP
Steganography." Indian Journal of Science and Technology” Vol 9, No. 38, 2016.
Anil, S., Mythili, T., and Hazra, S., "A Novel Rgb Based Steganography
Using Prime Component Alteration Technique." IIOAB JOURNAL Vol. 7, Issue 5, pp 58-73, 2016.
Asad, M., Gilani, J., Khalid, A., “An Enhanced Least Significant Bit
Modification Technique for Audio Steganography”.2011. Avval, A. H. A. and Mohanna, S., "A New Robust Audio Signal
Steganography Based On Multiple Chaotic Algorithms With Edge Detection Technique." Advances in Computer Science: an International Journal, Vol. 3, Issue 2, No. 8, 2014.
Banerjee, I., Bhattacharyya, S. and Sanyal, G., "Robust image
steganography with pixel factor mapping (PFM) technique." Computing for Sustainable Global Development (INDIACom), International Conference on 5-7 March. pp 692-698, 2014.
Bandyopadhyay S. K., Banik, B. G., “LSB Modification and Phase Encoding
Technique of Audio Steganography Revisited ”. International Journal of Advanced Research in Computer and Communication Engineering. Vol. 1, Issue 4, June 2012.
Bender W., Gruhl D., Morimoto N. and Lu A., “Techniques for data hiding.”,
IBM systems journal, Vol. 35, No. 3.4, pp 313-336, 1996. Bennett, K., "Linguistic Steganography: Survey, Analysis, and Robustness
Concerns for Hiding Information in Text", Purdue University, CERIAS Tech. Report, 2004.
Bhattacharyya, S., et al.“Audio Steganography Using Mod 4 Method(M4M)”.
Journal of Computing, New York ,USA, Vol. 3, Issue 8, August 2011. Bhattacharyya, S., Banerjee, I. and Sanyal, G. "A novel approach of secure
text based steganography model using word mapping method (WMM)." International Journal of Computer and Information Engineering, Vol. 4, No. 2, pp 96-103, 2010.
Bhowal, K., Sarkar, D., Biswas, S., and Sarkar P. P., “Secured Image
transmission with GA based Audio Steganography”. IEEE annual conference india(INDICON). pp 1-4, 2011.
83
Candes, E., and Romberg, J. Sparsity and incoherence in compressive
sampling. Inverse problems, Vol. 23, Issue 3, pp 969, 2007. Candès, E. J., and Wakin, M. B. An introduction to compressive sampling.
IEEE signal processing magazine, Vol. 25, Issue 2, pp 21-30, 2008. Candès, E. J. and Tao, T., “Near optimal signal recovery from random
projections: Universal encoding strategies,” IEEE Trans. Inform. Theory, Vol. 52, No. 12, pp. 5406–5425, Dec. 2006.
Candès, E., Romberg, J., and Tao, T., “Robust uncertainty principles: Exact
signal reconstruction from highly incomplete frequency information,” IEEE Trans. Inform. Theory, Vol. 52, no. 2, pp. 489–509, Feb. 2006.
Chapman, M., Davida, G., and Rannhard, M., “A practical and Effective
Approach to Large Scale Automated Linguistic Steganography”, Proceedings of the Information Security Conference, October 2001, pp. 156-165.
Chandramouli, R., and Memon, N.,"Analysis of LSB based Image
Steganography", IEEE ICIP, pp 1022-1022, Oct. 2001. Cheddad, A., Condell, J., Curran, K. and Kevitt, P. M., “Digital image
steganography: Survey and analysis of current methods”. Signal Processing., Vol. 90, Issue 3, pp 727-752, 2010.
Cheddad, A., Condell, J., Curran, K. and Kevitt, P. M., 2010. Digital Image
Steganography: Survey and Analysis of Current Methods. Journal of Signal Processing, Vol. 90, Issue 3, 2009.
Cvejic, N. and Seppnen, T., “Reduced distortion bit-modification for LSB
audio steganography.” Proceeding of 7th International Conference on Signal Processing Proceedings (ICSP), Beijing, China, Vol. 3, pp 2320-2323, 2004.
Dasgupta, K., Mondal, J. K., & Dutta, P. “Optimized Video Steganography
Using Genetic Algorithm (GA)”. Procedia Technology, Vol. 10, pp 131-137, 2013.
Deepak D. et al.,“Efficient Method to Increase Robustness in Audio
Steganography”. International Journal of Advanced Research in Electrical, Electronics and Instrumentation Engineering. Vol. 1, Issue 6, December 2012.
Delforouzi, A. and Pooyan, M. “Adaptive digital audio steganography based
on integer wavelet transform.” Circuits, Systems & Signal Processing, Vol. 27, Issue 2, pp 247-259, 2006.
Deshmukh, P. R. and Rahangdale, B.,” Hash Based Least Significant Bit
Technique For Video Steganography”, International Journal of
84
Engineering Research and Applications ,ISSN : 2248-9622, Vol. 4, Issue 1( Version 3), pp.44-49, January 2014.
Djebbar, F. et al.,“Comperitive study of digital audio steganography
technique”. EURASIP journal on audio, speech and music processing. 2012
Djebbar, F., Ayad, B., Hassmam, H., and Abed-Meraim, K. “A view on latest
audio steganography techniques”, International Conference on Innovations in Information Technology (IIT), IEEE, 2011.
Donoho, D. L., "Compressed Sensing," IEEE Transactions on Information
Theory, Vol. 52, pp.1289-1306, 2006. Duarte, M. F., Davenport, M. A., Wakin, M. B. and Baraniuk, R. G., 2006,
May. “Sparse signal detection from incoherent projections. In Acoustics, Speech and Signal Processing, 2006. ICASSP 2006 Proceedings”. 2006 IEEE International Conference on, Vol. 3, pp. 3 2006.
Ercelebi, E. and Batakc L., “Audio watermarking scheme based on
embedding strategy in low frequency components with a binary image.”, Dig. Signal Process., Vol. 19, Issue 2, pp 265-277, 2009.
Ghasemi, E., Shanbehzadeh, J., and Fassihi, N., "High capacity image
steganography using wavelet transform and genetic algorithm."Proceedings of the international multiconference of engineers and computer scientists. Vol. 1. 2011.
Hemalatha, S., Acharya, U. D. and Priya, R. K., "A secure color image
steganography in transform domain." International journal on cryptography and information security, Vol. 3, No. 1, pp 17-24, 2013.
Hemalatha, S., Acharya, D., U., Renuka, A. and Kamath, P., R., "An Integer
Wavelet Transform Based Steganography Technique for Color Images." International Journal of Information & Computation Technology. Vol. 3, Issue 1, pp 13-24, 2013.
Huang, X., Abe, Y., and Echizen I., “Capacity adaptive synchronized
acoustic steganography scheme.” J. Inform. Hiding Multimedia Signal Process., Vol. 1, Issue 2, pp 72-90, 2010.
Hussain, H. S., et al., "A novel hybrid fuzzy-SVM image steganographic
model." Information Technology (ITSim), 2010 International Symposium in. Vol. 1. pp 1-6, 2010.
ITU-T Recommendation P.862.3, "Application guide for objective quality
measurement based on Recommendation P.862, P.862.1 and P.862.2", November 2005.
85
Jagdale, B., N., Bedi, R., K., and Desai, S., "Securing MMS with high performance elliptic curve cryptography." International Journal of Computer Applications. Vol. 7, pp 17-20, 2010.
Jain, M. and Kumar, A., 2016. “RGB channel based decision tree grey-alpha
medical image steganography with RSA cryptosystem”. International Journal of Machine Learning and Cybernetics, pp.1-11, 2016.
Jain, N., Meshram, S. and Dubey, S., 2012. “Image steganography using
LSB and edge–detection technique”. International Journal of Soft Computing and Engineering (IJSCE) ISSN, 223.
Johannes Trithemius, Steganographia, 1499. Juneja, M., Sandhu, P. S. and Walia, E., 2009. “Application of LSB based
steganographic technique for 8-bit color images”. World Academy of Science, Engineering and Technology, Vol. 50, pp 423-425, 2009
Kahn, D., “The Code Breakers, The Story of Secret Writing”, The Macmillan
Company. (ISBN 0-684-83130-9) New York, NY 1967. Kaur, R., Singh, G., & Singh, S., “An Overview Of Data Hiding Technique:-
Steganography”. International Journal Of Data & Network Security, Vol. 1, Issue 2 , pp 34-36, 2012.
Kaur, S., Bansal, S., and Bansal, R. K., "An Efficient Adaptive Data Hiding
Scheme for Image Steganography." Proceedings of the International Congress on Information and Communication Technology. Springer Singapore, 2016.
Kelash, H. M., Wahab, O. F. A., Elshakankiry, O. A. and El-Sayed, H. S.,
2014. Utilization of steganographic techniques in video sequences. International Journal of Computing and Network Technology, Vol. 2, Issue 1, pp 17-24, 2014.
Kiah, M. L. M., et al.“A review of audio based Steganography and digital
watermarking” International journal of the physical sciences, Vol. 6, Issue 16, pp 3837-3850, August, 2011.
Kumar, A., and Sharma, R., "A secure image steganography based on RSA
algorithm and hash-LSB Technique." International Journal of Advanced Research in Computer Science and Software Engineering. Vol. 3, Issue 7, 2013.
Majumder, A., and Changder, S. “A Novel Approach for Text Steganography:
Generating Text Summary Using Reflection Symmetry”. Procedia Technology, Vol. 10, pp 112-120, 2013.
Mane, A., Galsetwar, G. and Jeyakumar, A. “Data hiding technique: Audio
Steganography LSB technique”. International Journal of Engineering Research and Applications. Vol. 2, Issue 3, pp. 1123-1125, 2012.
86
Mondal, A., and Pujari, S. "A Novel Approach of Image Based
Steganography Using Pseudorandom Sequence Generator Function and DCT Coefficients." International Journal of Computer Network and Information Security, Vol. 7,Issue 3, 2015.
Nag, A., et al. "A novel technique for image steganography based on DWT
and Huffman encoding." International Journal of Computer Science and Security, (IJCSS), Vol. 4, Issue 6, pp 497-610, 2011.
Nagaraj, V., Vijayalakshmi, V., & Zayaraz, G. (2013). “Color Image
Steganography based on Pixel Value Modification Method Using Modulus Function”. IERI Procedia, Vol. 4, pp 17-24.
Nagaraj, V., Vijayalakshmi, V., and Zayaraz, G., "Modulo based Image
Steganography Technique against Statistical and Histogram Analysis." IJCA Special Issue on “Network Security and Cryptography” NSC, Vol. 4, pp 34-39, 2011.
Nithyanandam, P., Ravichandran, T. , Priyadharshini, E., and Santron, N. M.
"A Image Steganography Technique on Spatial Domain Using Matrix and LSB Embedding based on Huffman Encoding." i-Manager's Journal on Future Engineering and Technology, Vol. 6, No. 3, pp 25 2011.
Nithyanandam, P., Ravichandran, T., Priyadharshini, E., and Santron, N. M.
"An image steganography for colour images using lossless compression technique." International Journal of Computational Science and Engineering, Vol. 7, No. 3, pp 194-205 2012.
Patil, B. A. and Vrishali, A. C., “Review of an improved Audio Steganography
Technique over LSB through Random Based Approach”. IOSR journal of computer engineerin. Vol. 9, Issue 1, 2013.
Pawar, S. S., Kakde, V., "REVIEW ON STEGANOGRAPHY FOR HIDING
DATA." International Journal of Computer Science and Mobile Computing, Vol. 3, Issue. 4, pp.225 – 229, 2014.
Perceptual Evaluation of Speech Quality (PESQ), an Objective Method for
End-To-End Speech Quality Assessment of Narrowband Telephone Networks and Speech Codecs, February 2001.
Poonia, S., Nokhwal, M., and Shankar, A., "A secure image based
steganography and cryptography with watermarking." International Journal of Emerging Science and Engineering (IJESE), Vol.1, Issue 8, 2013.
Pradhan, A., et al., "Performance evaluation parameters of image
steganography techniques." Research Advances in Integrated Navigation Systems (RAINS), International Conference on. IEEE, 2016.
87
Pujari, S., and Mukhopadhyay, S., "An Image based Steganography
Scheme Implying Pseudo-Random Mapping of Text Segments to Logical Region of Cover Image using a New Block Mapping Function and Randomization Technique." International Journal of Computer Applications, Vol. 50, Issue 2, 2012.
Qin, C., Chang, C.C., Huang, Y.H. and Liao, L.T., 2013. An inpainting-
assisted reversible steganographic scheme using a histogram shifting mechanism. IEEE Transactions on Circuits and Systems for Video Technology, Vol.23, Issue 7, pp 1109-1118.
Rabah, K., "Steganography-The Art of Hiding Data", Information Technology
Journal, Vol. 3, Issue 3, 2004, pp. 245-269. Rajput, V., Tiwari, S. K., and Gupta, R. "An enhanced image security using
improved RSA cryptography and spatial orientation tree compression method." Signal Processing, Communication, Power and Embedded System (SCOPES), 2016 International Conference on. IEEE, 2016.
Rameshkumar, P., Monisha, M. and Santhi, B., 2014. “Enhancement of
information hiding in audio signals with efficient LSB based methods”. Indian Journal of Science and Technology, Vol.7 No. S5, pp.80-85.
Reddy, H. S., Manjunatha, N., Sathisha, A. K., and Raja, K. B., "Secure
steganography using hybrid domain technique." In Computing communication & networking technologies (ICCCNT), 2012 third international conference on, pp. 1-11. IEEE, 2012.
Reddy, H. S. and Raja, K. B.,"Wavelet based secure steganography with
scrambled payload." International Journal of Innovative Technology and Exploring Engineering (IJITEE), Vol. 1, Issue 2, pp 121-9, 2012.
Reddy, H. S. and Raja, K. B. "Hybrid Domain based Steganography using
BPS, LSB and IWT." International Journal of Computer Applications Vol. 54, Issue 3, 2012.
Roy, S., and Venkateswaran P., "A text based steganography technique with
indian root." Procedia Technology, Vol. 10, pp 167-171, 2013. Salah S. A., et al., “Watermarking in WAV files Based on Phase Coding”.
Engineering and Technology journal, Vol. 29, No. 4, pp 770-780, 2011.
Saraswathi, V., and Mrs Kingslin, S., "Different Approaches to Text
Steganography: A Comparison." Int. J. Res. Manag. Technol 9359, No.11, pp 124-127, 2014.
88
Seyyedi, S. A., Sadau, V., and Ivanov, N., "A Secure Steganography Method Based on Integer Lifting Wavelet Transform." IJ Network Security Vol. 18, Issue 1, pp 124-132, 2016.
Shahadi, H. I. and Jidin, R., “High capacity and in audibility Audio
steganography scheme.” Proceeding of 7th International Conference on Information Assurance and Security (IAS), pp: 104-109, 2011.
Shahadi, H., I., Jidin, R. and Way, W., H., 2014. A novel and high capacity
audio steganography algorithm based on adaptive data embedding positions. Research Journal of Applied Sciences, Engineering and Technology, Vol. 7, Issue 11, pp 2311-2323, 2014.
Shahreza, S., and Shalmani, M., “Adaptive wavelet domain audio
steganography with high capacity and low error rate.” Proceeding of IEEE International Conference on Information and Emerging Technologies (ICIET), pp: 25-29, 2007.
Shahreza, S. and Shalmani, M., “High capacity error free wavelet domain
speech steganography.” Proceeding of IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), pp: 1729-1732, 2008.
Shirali-Shahreza, M. H., Shirali-Shahreza, M., “Arabic/Persian Text
Steganography Utilizing Similar Letters With Different Codes,” The Arabian Journal for Science and Engineering, Vol. 35, Number 1B, April 2010
Shirali-Shahreza, M. H., Shirali-Shahreza, M.,"A New Approach to
Persian/Arabic Text Steganography", Proceedings of the 5th IEEE/ACIS International Conference on Computer and Information Science (ICIS 2006), Honolulu, HI, USA, 10-12 July, 2006, pp. 310-315.
Singh, P. K., Aggrawal, R. K., “ Enhancement of LSB based Steganography
for Hiding Image in Audio.” International journal on Computer Science and Engineering, Vol. 2, No. 5, pp 1652-P1658, (2010).
Singh, A. K., Dave, M., and Mohan, A., "Hybrid technique for robust and
imperceptible image watermarking in DWT–DCT–SVD domain." National Academy Science Letters Vol. 37, No. 4, pp 351-358, 2014.
Singh, S., “Hiding Image to Video” International Journal of engineering
science & technology Vol. 2, Issue 12 , pp 6999-7003, 2010. Sharma, S., and Kumari, U., "A High Capacity Data-Hiding Technique Using
Steganography." International Journal of Emerging Trends & Technology in Computer Science (IJETTCS), Vol. 2, Issue 3, 2013.
89
Shirali-Shahreza, M., & Shirali-Shahreza, M., H. (2007, November). Text steganography in SMS. In Convergence Information Technology, 2007. International Conference on (pp. 2260-2265). IEEE.
Shirali-Shahreza, M. (2008). A New Persian/Arabic Text Steganography
Using “La” Word. In Advances in Computer and Information Sciences and Engineering (pp. 339-342). Springer Netherlands.
The PESQ Algorithm as the Solution for Speech Quality Evaluation on 2.5G
and 3G Networks, 2009 The PESQ Algorithm as the Solution for Speech Quality Evaluation on 2.5G
and 3G Networks, 2009. Vetterli, M., Marziliano, P., and Blu, T., “Sampling Signals with Finite Rate of
Innovation,” IEEE Transaction on Signal Process, Vol. 50, no. 6, pp.1417–1428, 2002.
Yadav, P., Mishra, N., and Sharma, S., "A secure video steganography with
encryption based on LSB technique." Computational Intelligence and Computing Research (ICCIC), International Conference on. IEEE, pp 1-5, 2013.
Zaidan, A. A., Zaidan, B. B., Taqa, Y. A., Sami M. K., Alam, G. M. and Jalab
A. H., "Novel multi-cover steganography using remote sensing image and general recursion neural cryptosystem." International Journal of Physical Sciences, Vol. 5, No. 11: pp 1776-1786. 2010.
Zhang, H. J., Tang, H. J. “A Novel Image Steganography Algorithm Against
Statistical Analysis”, Proceedings of The Sixth International Conference on Machine Learning and Cybernetics, IEEE, Hong Kong, Vol. 7, pp. 3884-3888, 2007.
Zamani, M., Manaf, A. A., Ahmad, R. B., Zeki, A. M. and Abdullah, S., "A
genetic-algorithm-based approach for audio steganography." World Academy of Science, Engineering and Techqnology, Vol 54, 360-363, 2009.