SECRET DATA TRANSMISSION USING STEGANOGRAPHY

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.