Filetype PDF | Posted on 15 Sep 2022 | 3 years ago
Authentication
digital resources
153x Filetype PDF File size 1.17 MB Source: romisatriawahono.net
COMPUTER SCIENCE REVIEW ( ) –
Available online at www.sciencedirect.com
ScienceDirect
journal homepage: www.elsevier.com/locate/cosrev
Current status and key issues in image
steganography: A survey
Mansi S. Subhedara,∗, Vijay H. Mankarb
aResearch Scholar, Department of Electronics & Telecommunication, Bapurao Deshmukh College of Engineering,
Sevagram, Wardha, 442102, Maharashtra, India
bDepartment of Electronics & Telecommunication, Government Polytechnic, Nagpur, 440001, Maharashtra, India
A R T I C L E I N F O A B S T R A C T
Article history: Steganography and steganalysis are the prominent research fields in information hiding
Received 5 October 2013 paradigm. Steganography is the science of invisible communication while steganalysis
Accepted11September2014 is the detection of steganography. Steganography means “covered writing” that hides
the existence of the message itself. Digital steganography provides potential for private
Keywords: and secure communication that has become the necessity of most of the applications
Information hiding in today’s world. Various multimedia carriers such as audio, text, video, image can act
Imagesteganography as cover media to carry secret information. In this paper, we have focused only on
Steganalysis image steganography. This article provides a review of fundamental concepts, evaluation
Imagequalitymeasures measures and security aspects of steganography system, various spatial and transform
domain embedding schemes. In addition, image quality metrics that can be used for
evaluationofstegoimagesandcoverselectionmeasuresthatprovideadditionalsecurityto
embeddingschemearealsohighlighted.Currentresearchtrendsanddirectionstoimprove
onexisting methodsaresuggested.
c
⃝2014ElsevierInc.Allrights reserved.
1. Introduction Steganography and cryptography are closely related con-
cepts. Though both the terms share a common goal, the
The word steganography is obtained from the Greek words way and the usage of both differ significantly. Steganogra-
“stegos” means “cover” and “grafia” means “writing”, defin- phyishiddenwritingwhereascryptographyissecretwriting
ing it as “covered writing”. Usually secure communication i.e. cryptography provides security with respect to content of
is achieved by the method of encryption. But nowadays, de- the message whereas steganography will hide the existence
mand for security is increasing day by day that leads to the of the message itself. Digital watermarking is another branch
use of steganography for information security. The idea of of information hiding. Both steganography and watermark-
data hiding or steganography was first introduced with the ing are the methods of data embedding, but there are sev-
example of prisoner’s secret message by Simmons in 1983 eral differences among them. A detailed comparison can be
[1–3]. Fig. 1 shows various disciplines of information hiding. found in [4–7]. A variety of multimedia carriers that includes
∗Corresponding author. Tel.: +91 9867967304.
E-mail addresses: mansi_subhedar@rediffmail.com, msubhedar@mes.ac.in (M.S. Subhedar), vhmankar@gmail.com (V.H. Mankar).
http://dx.doi.org/10.1016/j.cosrev.2014.09.001
c
1574-0137/⃝2014Elsevier Inc. All rights reserved.
2 COMPUTER SCIENCE REVIEW ( ) –
Security Systems
Information Hiding
Cryptography Steganography Watermarking
Linguistic Technical Robust Fragile
Steganography Steganography
Imperceptible Visible Fingerprint
Digital Images Video Audio Text
Fig. 1 – Disciplines of information hiding [4].
text, audio, video, imagecanbeusedforsteganography.Some Security means eavesdropper’s inability to detect hidden
of the ways to achieve text steganography involve modifica- information. Perceptual transparency ensures the retention of
tion of text layout, use of nth character from text or alteration the visual quality of the cover after data embedding. Tamper
of some of the rules such as spaces etc. Another approach resistance means to remain intact in the face of malicious at-
includes usage of a code consisting of combination of char- tacks. The embedding rate is measured as the number of em-
acter, line and page numbers. However, this technique lacks beddedbitspercarrier bit. The embedding efficiency is given by
in security. Hiding information in audio files can be done by the expected number of embedded message bits per modi-
using frequencies that are inaudible to human ear. Similarly, fied carrier bit. The change rate gives the average percentage
video files can also be thought of to embed secret informa- of modified carrier bits.
tion. Since it is a moving stream of images and sounds, any
minordistortions may be unseen because of continuous flow 1.2. General model of steganography
of information. The advantage in this case will be high pay-
load capacity. Image is the most popular file format used for The concept of steganography is usually modeled by pris-
steganography as they possesses high degree of redundancy. oner’s problem. Fig. 2 exhibits the overall structure for the
With image steganography, better imperceptibility and pay- steganography system. Let ‘C’ denotes the cover medium
′
loadcapacitycanbeachieved.Steganalysisisanartofdetect- i.e. image A and C be the stego image obtained by data em-
ing covert communication [8]. In this paper, we focus only on bedding.Let‘K’representsanoptionalkeyand‘M’isthemes-
imagesteganographywithlittlemoreemphasisontransform sage we want to communicate. E suggests the embedding
m
domainsteganography. process and Ex is for the process of extraction. Compression
and encryption eliminate the redundancy in secret message
1.1. Fundamental concepts and result in enhanced security. Thus, data embedding pro-
cess can be represented as follows:
Cover image refers to the image used for carrying the embed- ′
E :C⊕K⊕M→C
ded bits, embedded data is known as payload and the image m
E E c,k,m ≈m, ∀c∈C,k∈K,m∈M. (1)
withembeddeddataiscalledasstegoimage.Steganalysisrefers x m
totheattackonsteganography.Thedistortioninducedonthe Imageisthemostoftenusedfileformatforsteganography
hostsignalbythedataembeddingprocessiscalledtheembed- and is only discussed here where the secret message
ding distortion. is embedded in cover image. Applications of steganography
Imperceptibility is innocuousness of the stego image. Stego include copyright control of materials, enhancing robustness
image should not have severe visual artifacts. Some of the of image search engines and smart id’s, feature tagging,
major requirements of steganography include capacity, ro- secret communication, video-audio synchronization, TV
bustness and security. Robustness indicates the amount of broadcasting, TCP/IP packets etc. [10,11]. Image quality
modification that the stego medium can withstand before measures are used for the evaluation of stego image
an adversary can destroy hidden information. Capacity refers quality obtained after embedding. Different methods exist
to the amount of information that can be hidden in cover for attacking the steganographic algorithm. The number of
medium without deteriorating the integrity of the cover im- steganography tools are available that includes Ezstego, F5,
age. It is represented in terms of bits per pixel (bpp). Embed- Hide and Seek, Hide4PGP, Mp3Stego, OutGuess, StegHide,
ding operation needs to preserve the statistical properties of Stegnos, S-tools etc. Various forms of steganalysis include
the cover image in addition to the perceptual quality. identifying the existence of secret message and finding its
COMPUTER SCIENCE REVIEW ( ) – 3
Fig. 2 – General model of steganography [9].
approximate length or even trying to retrieve it. Various medium is wrongly classified as cover, TN (True
stego attacks include image resizing attack, image tampering Negative): a cover medium is correctly classified as
attack, AWGN attack, JPEG compression attack, RS attack, cover, FP (False Positive): a cover medium is wrongly
filter attack, chi square attack, J. Fridrich’s RS steganalysis, classified as stego.
Jeremiah J. Harmsena’s Histogram attack etc. The algorithm Payload capacity: It is defined in terms of number of secret
used for data embedding should withstand against all these bitsthatcanbeembeddedperpixel.Ideallyitshould
types of attacks making eavesdropper unable to retain the beashighaspossiblewhilemaintainingtheaccept-
hiddenmessage. able quality of the stego image. It is also known as
Survey reports presented till now [7,10–12] explored var- hiding capacity or embedding capacity and is mea-
ious fundamental issues in steganography, spatial domain sured in terms of bits per pixel or bits per trans-
andtransformdomainapproachesfordatahiding,steganaly- form coefficient (for spatial and transform domain
sis techniques etc. However, there are many unattended ar- approachrespectively).
eas that considerably influence image steganography. This Imperceptibility: Steganography system should have high
paper focuses mainly on such areas e.g. security aspects of embedding capacity and capability to withstand
steganography system, cover selection measures and IQM against stego attacks. The stego image should not
analysis. haveseverevisualartifacts.Higherthefidelityofthe
The outline of the paper is as follows: Section 2 discusses stego image, the better.
performance evaluation measures and security aspects. Sec-
tion 3 presents overview of various image steganography 2.1. Steganographic security
techniques. Image quality measures and cover selection cri-
teria are illustrated in Sections 4 and 5 respectively. Section 6 Securityisalwaystheimportantcriterionwhiledesigningany
concludes the paper. application. There are numerous ways to define security of
steganography system e.g.
2. Performance evaluation measures Maximummeandiscrepancy(MMD)security: Withadvances
in steganography algorithms and methods to detect
In literature, many steganography schemes are presented them, issue of comparing them with a fair bench-
basedonvarietyofparameters.Someofthemworkinspatial mark is a critical task. The task of identifying the
domain and other in transform domain. Irrespective of the differences between cover and stego image is a two
approachusedfordataembedding,somecommonattributes sample problem that can be solved using MMD. It
need to be defined to achieve uniqueness in performance findsthediscrepancybetweenpdfofcoverandstego
rating. Some of them can be defined as follow: objects. It is given by,
MMD F,X,Y
Security against attack: The steganographic system may ( )
suffer from different types of stego attacks, allow- D D
,sup 1 f(x)− 1 f(y) (3)
ingeavesdroppertoretrievesecretmessagebitsem- D i D i
bedded in cover media. The system is said to be γ— f∈F i=1 i=1
secure if TP Rate − FP Rate ≤ γ, where 0 ≤ γ ≤ 1. And where X = x1,x2,...,xD and Y = y1,y2,...,yD are
is said to be perfectly secure if γ = 0. thesamplesfromprobabilitydistributionsPx andPy
respectively. f is a class of function which is built
TP FP
TPRate= S , FP Rate = S (2) fromsymmetric,positivedefinite function.
TP +FN TN +FP
S S S S Advantages of MMD include numerical stability,
where, TP (True positive): a stego medium is cor- well scaled with data dimensionality and converge
1
rectly classified as stego, FN (False Negative): a stego independently on data dimension d with error √ ,
D
4 COMPUTER SCIENCE REVIEW ( ) –
Fig. 3 – (a) Confusion matrix (b) ROC curve.
whereDisthenumberofsamples[13].Also,MMD’s vergence which justifies the security. The bigger the
2 correlation parameter ρ, KL divergence is larger.
computational complexity is O D , that is faster
thansupportvectormachines(SVM).
ROCbasedsecurity: Anotherwaytoquantifysecurityiswith
reference to ROC. As shown in Fig. 3, it is the plot 3. Imagesteganography
of false positive rate versus true positive rate [6,14].
The true positive rate is plotted on Y axis and false Imagesteganographycanbebroadlyclassifiedintospatialdo-
positive rate on X axis. Larger the area under the main, transform domain, spread spectrum and model based
curve, better the performance of the steganalytic steganography as depicted in Fig. 4. In spatial domain, se-
system e.g. performance of curve C is better than B, cret message is embedded in pixel value directly whereas
andthatofBisbetterthanA.Eq.(2)describesoneof transformdomainmethodsachieveembeddingbyfirsttrans-
the conditions to be satisfied for the secure system. forming the image from spatial to frequency domain using
Correlation coefficient: The correlation coefficient between any one of the transforms such as discrete cosine transform
twoadjacentelementsC andC is ρ andthecorre- (DCT), discrete wavelet transform (DWT), Hadamard trans-
i i+1
lation coefficient between two arbitrary elements C form, Dual tree DWT, double density dual tree DWT (DD DT
i
andCj isρ|j−i|. The bigger the ρ, the stronger the cor- DWT), ridgelet transform, curvelet transform etc. and then
relation. Security will be improved if selected cover embeddingisdoneinsuitabletransformcoefficients.
image is with smaller ρ [15–17]. In statistics, Bhat- This section deals in detail with each of these methods.
tacharya distance measures the similarity between Various techniques can be employed to optimally choose the
two discrete or continuous probability distributions transform coefficients to hide data in. Soft computing tools
and is denoted by DB(Pc,Ps). Smaller the ρ, smaller can be considered for this purpose. As transform domain
the DB(P ,P ) when ρ ∈ [0,1]. BD between P and P
c s c s methods are more immune to image processing operations
is defined as, and are less susceptible to stego attacks, they are usually
1 det(R) preferred over spatial domain methods.
BD P ,P = ln (4) Spread spectrum steganography involves embedding in
( c s) 2 det(R ) det(R )
c s noise inherent to image acquisition process. Image restora-
where Pc is probability distribution of cover im- tion and error control techniques can be used while extract-
age, Rc is covariance matrix, σ2 is the variance of ing the data at the decoder side. It is a blind scheme as
c
marginal distribution and ρ is the parameter repre- original image is not required while extraction. This method
senting the degree of cover data dependency. outperforms in terms of payload capacity and invisibility.
Kullback–Leibler (K–L) divergence: KL divergence is also one Model based steganography is based on statistical model of
of popular security measures to analyze the the cover image. It is also known as statistics aware embed-
steganography system. It has been proposed by ding. Before selecting the locations for data hiding in cover
Cachin in 1998 [17]. Let X and Y represent cover and image, statistical global features of image are taken into ac-
stego image and p and q denote the probability count and then actual data embedding process is carried out
x y
distribution function of X and Y respectively. KL di- accordingly. Thus, it provides additional layer of security to
vergence between two probability distribution func- steganography. All these methods are discussed in detail in
tions is given by, forthcoming subsections.
p (g)
D p ∥q = p (g)log x (5) 3.1. Spatial domain steganography
x y x q (g)
g∈G y
where g ∈ G = {0,1,2,...,255} is the pixel value in In this method, the pixel value is directly modified for data
gray scale images. The embedding algorithm should hiding. The various approaches to achieve embedding in
be designed so as to get minimum value of KL di- spatial domain are shown in the Fig. 5.
no reviews yet
Please Login to review.
