An Effect of Compressive Sensing on Image Steganalysis

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 03 | Mar 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 1111
An Effect of Compressive Sensing on Image Steganalysis
I.Swathi1, K.Sravani2, M.Seshagirirao3, T.S.R.Krishnaprasad4
1, 2, 3 U.G Student, Department of Electronics and Communication Engineering, SR Gudlavalleru Engineering College,
Andhra Pradesh, India.
4 Associate Professor, Department of Electronics and Communication Engineering, SR Gudlavalleru Engineering
College, Andhra Pradesh, India.
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - As from several obtained measurements which
are suggested from the Nyquist sampling theory, the
compressive sensing theory states that the signal can be
regained with a high nature ofprospectwhen itshowssparsity
in some domain. In this project, a new frame work is proposed
to that compressive sensing is used to recover the signal by
adapting the known sparsifyingbasisviaL0 minimization. The
natural image of the intrinsic sparsity fis mainly effected by
the overlapped sparsely represented image, which is to be
patched with the help of adapted sparsifying basis in the form
of L0 norm to reduce the blocking artifacts and conveying the
CS solution space. To exhibit the executed scheme tractable
and robust a split Bregman iteration based technique is
generated to solve the non convex L0 minimzation problem
efficiently. The experimental results should be taken in to
account on CS recovery execution which leads to give the
proposed algorithm with a compelled performance for the
growth of current state of the art schemes which leads to
produce best convergence property.
Key Words: Compressive sensing, image recovery,
sparsifying basis optimization.
1. INTRODUCTION
Steganalysis was extensively studied over the last decade to
detect the secret signal presence (including payloads)
embedded in host images obtained from known sources. In
steganography, the medium used to embed the message
stated as cover and can be text or an image. Covers with
embedded secret messages are also definedasStegoobjects.
In many cases, the objective of steg analysis can be
formulated as a binary classification problem that
distinguishes between cover and stego objects. In this case,
we are interested in recovering secret messages. Existing
approaches to web analytics generally involve two major
steps. H. Feature extractionandclassification-baseddecision
making. as a compression method. There is a main problem
with CS is its weak robustness. Conventional CS
reconstruction relies on matricestogainthefeatureofsignal
along with the processing of signals. The goal istocapture as
much information as possible in as few samples as possible
without introducing aliasing. Reconstructionalsopresentsa
special challenge because it solves systems of
underdetermined linear equations.
1.1 Framework
With the increasing exchange of digital data over networks,
data security has become a major concern aroundtheworld.
The proliferation of digital data raises concerns about data
being attacked or tampered with by unauthorized persons.
Since the Internet provides a means of communication for
disseminating information, the technique of hiding secret
messages in various multimedia content is animprovedand
more effective technique. Steganography is a data hiding
technique aimed at sending messages over channels where
other information is already being sent. The main objective
of web analytics is to detect messages hidden in cover
objects, such as digital media,anddeducewhetherthe media
is embedded with sensitive data.
1.2 Method of Approach
To recover the secret signal MATLAB Tool (R2019A) is used
for the execution. The nominal image steganalysis can only
used to know the presence of detection of the image but it
can’t focus on secret signal recovery of original messages
which are embedded in a host image.. To execute this issue
the image steganalysis can be taken in to account with the
help of compressive sensing cs domain, where block CS
measurements matrix produce the thetransformcoefficient
of stego- image which reflects the statistical differences
between the cover and stego – images. The reconstruction
image by using compressive sensing is used to find the
suitable values of model parameters in order to gain the
most executed one.
1.3 Execution of the work
The execution of CS- based criterion on stego images to
recover the secret signals from the extracted features, the
stegnographic algorithms are combinedwiththeapplication
of most effective features sensed by the BCS measurement
matrix and to recover thehiddensignalsbymulti-hypothesis
predictions with a Tikhonov regularizationintheCSdomain.
The output shows the reconstructed fingerprintimageswith
40 features were detected by BCS as compared to other
embedding algorithms, and JPHS is the weakest among
embedding algorithms.

2. LITERATURE SURVEY
The spatial domain based image steganography used to
transform domain one in terms of embedding capacity, but
the stego-image contains high amount of redundant data.
The main usage of compressive sensing is to construct the
acquired signal efficiently . In the field of image restoration
development, the compressive sensingtheoryplaysa crucial
role, which has drawn quite an amount of attention which is
an alternative to the present method of sampling with the
help of compression. To explode the redundancy ina existed
signal, the CS used for sampling and compression at the
same time. CS theory exhibits the signal can be decoded
from the suggested fewer measurements by the Nyquist
sampling theory , when the signal is sparse in some domain,
which has substantially changes the way of engineers
thinking of data acquisition.
2.1 Proposed System
The proposed algorithm is compared with four types of
cs recovery methods i.e., wavelet method (DWT) , total
variation (TV) method , multi-hypothesis (MH) method ,
collaborative sparsity (CoS) method , which deal with image
signals in the wavelet domain, the gradient domain, the
random projection residual domain, and the hybrid space-
transform domain, respectively. Itisworth emphasizingthat
MH and CoS are known as the current state-of-the-art
algorithms for image CS recovery. The proposed algorithm
can be executed in dual nature that to eliminate the ringing
effects and to preserve much shaper edges and finer details,
which shows much clearer and better visual results thanthe
other competing methods Our work also offers a fresh and
successful instance to corroborate the CS theory applied for
natural images.
2.2 Existing system
As steganography may also reason image content material
modified imperceptibly, it's miles anticipated such
modifications may be detected via wayofmeansoftheusage
of remodel area coefficients in a first-class scale, in which CS
can play an crucial function to signify the modificationsfrom
the neighborhood remodel coefficients added via way of
means of the embedded mystery messages. Therefore, we
purpose to BCS to come across DCT or Discrete Wavelet
Transform (DWT)steganographic embeddinginformationin
images, specially to reconstruct the authentic mystery sign.
In the sector of photograph restoration,possiblythemost up
to date subject matter is the latest improvement of
Compressive Sensing (CS) principle, whichhasdrawnpretty
an quantity of interest as an opportunity to the
contemporary method of sampling observed via way of
means of compression.Byexploitingtheredundancyexisted
in a sign, CS conducts sampling and compression on the
equal time. CS principle indicates that a signmaybedecoded
from many fewer measurements than cautioned via way of
means of the Nyquist sampling principle, whilst the sign is
sparse in a few area, that results in suppose for information
acquisition.
3. CS APPLICATION SECURITY
3.1 Image Encryption
The CS together with chaos theory is used for hybrid image
compression-encryptionalgorithms. Whenanimageisin the
process of CS it has executed with a block of Arnold
transformation which is followed by XOR operation to
permute the measurement positions then dissipate the
Gaussian distribution.
3.2. Image Watermarking
A robust image watermarking scheme for image tamping
identification and localization based on CS and distributed
source coding principles was described earlier. The basic
idea is to form a hash, which is robustly embedded as a
watermark in the image The amount of extracted data isnot
large enough, then the CS is used to retrieve the coefficients
by developing the sparseness in the DCT domain
3.3 Image Hiding
A data hiding is based on subsampling and CSwasscheduled
to the relying on the CS properties including sparsity with
random projection then the secret data can be embedded in
the observation domain of the image. The method was
further extended to an image steganography algorithm in
terms of some details on design procedures and
experiments, but they are roughly consistent in the train of
thought.
3.4 Image Hashing
Kang el al. proposed a secure androbustimageschemeusing
CS and visual information fidelity. The CS makes the hash
size keep small while the visual 2510 VOLUME 4,2016 Y.
Zhang et al.: Review of CS in Information Security Field
information fidelity helps to be robust against most image
manipulations. The foundation is that if the tampering can
solving a convex optimization problem with constraints
forced by the transmitted hash, provided that it is sparse
enough CS for a robust image hash sparse which is to be
represented for robust image hashing with tampering
recovery capability and strong robustness that against
content preserving
3.5 Image Authentication
An optical encryption technique based on CS was employed
to create a cancellable biometric authentication scheme,
which encrypts a finger vein image using a compressive
imaging system when capturing image. A new point of view,
encrypted sensing, based on DRPE and CS, was proposedfor
biometric authentication, which further improves the
security to protect the biometric template

3.6 Video and Audio Security
Cossalter al. stated that for the videosurveillanceofassured
privacy tracking system for the CS data Credit systema fixed
camera has installed to analyze the video frames to detect a
possible moving object in the scene wherethereconstructed
un necessary original content of the video sequence
3.7. Cloud Security Scenario
In cloud securityscenario,privacy-assuredmultimedia cloud
computing based on CS and sparse representation was
investigated, whichdiscussedsomecompressivemultimedia
applications, including multimedia compression,adaptation,
editing/manipulation, enhancement, retrieval, and
recognition. The guaranteed original signal canbeinprivacy
since each cloud only has a small amount of information in
terms of both the measurements and asymmetric support-
set.
4. BLOCK DIAGRAM
Since steganography can imperceptibly change image
content, it is hoped that such changes can be detected at
finer scales usingtransform-domaincoefficients.HereCS can
play an important role in characterizing the changes from
the local transform coefficients introduced by the
embedding. secret message. Therefore, the main aim of BCS
is to detect the DCT or DST steganographic embedded data
in images respectively to reconstruct the original secret
Fig-1: Flowchart of the proposed framework.
4.1 Comparison of the Extracted Features
To conveniently compare the effect of the number of
functions in the stegaanalyzer, we extract 2000 stegoimages
with an embedding rate of 0.15 bpp using the embedding
algorithms F5, PQ, Outguess, and JPHS. On the other hand,
we train all functions using support vector machines(SVMs)
to derive the best fitness function that performs well in
different applications. We then applied an improvednature-
inspired CS algorithm and compared it with the optimized
DPSO algorithm. Each algorithm has 35 iterations and a
population size of 30
4.2 Quality of Recovered Secret Signal
In our approach, we tried to find a CS-based covert signal
recovery criterion using features extracted from stego
images. The final experiment combined four steganography
algorithms, applied the most effective features captured by
the BCS measurement matrix, and recovered the hidden
signal by multi-hypothesis prediction using his Tikhonov
regularization in the CS domain to a reconstructed
fingerprint image with 40 featureswithdifferentalgorithms.
Compared with other embedding algorithms, the
reconstructed fingerprint image of nsF5 can achieve the
closest approximation to the original signal with 40features
detected by BCS, and JPHS is the weakest among the
embedding algorithms. .
Fig-2: Comparison between SBI and IST for solving our
proposed l0 minimization From left to right: progression
of the PSNR (dB)
4.3 Performance Assessment
In our experiments, the CS measurements are produced by
applying a Gaussian random projectionmatrixto theoriginal
image signal at block level, The value of λ is related to the
overlapped step size, which will be given inthefollowing. All
the experiments are performed in Matlab 7.12.0 on a Dell
OPTIPLEX computer with Intel(R) Core(TM) 2 Duo CPU
E8400 processor (3.00GHz), 3.25G memory, and Windows
XP operating system.
4.4 Algorithm Convergence
Here, empirical evidence isillustratedthegoodconvergence
with a proposed algorithm , the plots of evolution of PSNR
versus iteration numbers for four test images with various
subrates (subrate=30% and subrate=40%). As from the
observed growth of iteration number , all the PSNR curves
increases monolithically, stable and flat which exhibits the
good convergence property
Fig -3: Convergence of the proposed algorithm. From left to
right: Progression of the Peak Signal to NoiseRatio(PSNR)in

decibels (dB) results achieved by proposed algorithm for
four test images w.r.t the iteration number in the cases of
subrate=30% and subrate=40%.
5. RESULT AND ANALYSIS
5.1 Objective or Numerical Measures
These measures are used to compare the cover images and
their corresponding stego-images based on some numerical
criteria that do not require extensive subjective studies.
Hence, in recent times, these measures are more commonly
used for image quality assessment .These include Peak
Signal-to-Noise ratio, PSNR value to the imperceptibility of
stego-images. That is, PSNR-10 log10 R 2 MSE dB
5.2 Input Images
Fig -4: Input Images
5.3 Output Images
Fig -5: Output Image of House
Fig -6: Output Image of Barbara
Fig -7: Output Image of Leaves
Fig -8: Output Image of Parrot
Fig -8: Output Image of Monarch
Fig -9: Output Image of Vessel

6. CONCLUSIONS
In this project, we propose to characterize the inherent
sparsity of natural images by a patch-based redundant
sparsity representation using adaptively learned
sparsification bases. This particular type of surrogate
representation is formulated by non-convex minimization
for compressed image acquisition recovery. This can be
efficiently solved by a developed technique based on split
Bregman iteration. Experimental results on a wide range of
natural images for CS restoration show that the proposed
algorithm achieves significant performancegainsovermany
current state-of-the-art schemes and exhibits superior
convergence properties. This project presents a secure
cryptographic scheme that combines the strengths of
compressed sensing schemes. This method also provides an
effective way to compress data. This combined chemistry
meets requirements such as capacity, security, and
robustness for secure data transmissionoveropenchannels.
The proposed method can also be applied toaudioandvideo
data as a future improvement.
REFERENCES
[1] L. Rudin, S. Osher, and E. Fatemi, “Nonlinear total
variation based noise removal algorithms,” Phys. D., 60
(1992) 259–268.
[2] D. Geman and G. Reynolds, “Constrained restorationand
the recovery of discontinuities,” IEEE Trans. on Pattern
Analysis and Machine Intelligence, 14 (3) (1992) 367–
383
[3] D. Mumford and J. Shah, “Optimal approximation by
piecewise smooth functions and associated variational
problems,”Comm. on Pure and Appl. Math., 42 (1989)
577–685
[4] A. A. Efros and T. K. Leung, “Texture synthesis by non-
parametric sampling,” Proc. of Int. Conf. ComputerVision,
Kerkyra, Greece, (1999) 1022–1038.
[5] A. Buades, B. Coll, and J. M. Morel, “A non-local algorithm
for image denoising,” proc. of Int. Conf. on Computer
Vision and Pattern Recognition, San Diego, CA, USA,
(2005) 60–65.
[6] H. Takeda, S. Farsiu, and P. Milanfar, “Kernel regression
for image processingandreconstruction,”IEEETrans. on
Image Process., 16 (2) (2007) 349–366.

An Effect of Compressive Sensing on Image Steganalysis

Recommended

Recommended

More Related Content

Similar to An Effect of Compressive Sensing on Image Steganalysis

Similar to An Effect of Compressive Sensing on Image Steganalysis (20)

More from IRJET Journal

More from IRJET Journal (20)

Recently uploaded

Recently uploaded (20)

An Effect of Compressive Sensing on Image Steganalysis