International Journal of Scientific and Research Publications, Volume X, Issue X, Month 2018 1

ISSN 2250-3153

Image Steganography Detection for Varied Images

Tejasvi Singh, Gopal Dixit, HarshDeep Chandra

*
Electronics and Communication Department, JSS Academy Of Technical Education, Noida

Abstract- A picture's secret code is similar to image converting it into a secret code. It's like a very devious
steganography. It's a technique for concealing text, pictures, or approach to protect confidentiality. Thus, there are several
movies within other images. The great thing is that our eyes clever techniques that have been around for a while to
cannot see this buried information. Deep learning is a potent conceal information. However, as a result of the
technology that has been used recently for image steganography. abundance of content we share on social media and the
This work aims to investigate and discuss several deep learning internet, they are becoming increasingly significant. These
techniques applied to image steganography. These techniques days, steganography and cryptography are the two main
fall into three categories: conventional techniques, techniques strategies. Using cryptography is similar to making your
based on convolutional neural networks, and techniques based data extremely difficult to decipher. It's similar to
on general adversarial networks. We'll go over each kind. In this encoding your communication in a secret code that is
study, we also provide an extensive overview of the datasets visible to us.
utilized, the methodology employed for the experiments, and the
metrics applied. We also created a useful table to make it simple Steganography, on the other hand, functions more like a
to make it simple to locate the crucial information, we even ninja technique. It uses text, images, music, videos, and
created a helpful table. Our intention is to assist other files to conceal information from us directly. Another
researchers by compiling the most recent developments, technique is digital watermarking, which involves
obstacles, and prospects for this topic. It functions as a kind of concealing confidential information to reveal the owner.
manual for anyone curious about deep learning-based image Thus, although cryptography is the super hero of
steganography. information concealment, steganography is gaining
traction as a covert ninja that conceals things from view!
Data can be secretly played hide-and-seek with
Index Terms- CNN, EfficientNet b0, Image Steganalysis, steganography. It's the skill of enclosing a tiny, secret data
Multiclassifier, piece—like a message or image—inside a much bigger,
public data piece—like a file, video, image, or text. For
instance, we can secretly insert one image into another
I. INTRODUCTION using image steganography. Imagine for a moment that
you have a picture and that you wish to conceal something
When it comes to sending information through images, from view. Steganography fills that need. It cleverly
videos, and sounds—particularly in the context of the modifies the original image so that the concealed
Internet of Things (IoT)—technology has advanced information is hidden, in contrast to cryptography where
remarkably quickly in recent years. This data is typically the secret code is visible. The twist is that Steganalysis is a
transmitted over networks that may not be very secure. different player. It is similar to this game's detective.
Sharing images and videos on the internet is common Steganalysis attempts to determine whether a picture has a
among individuals and groups, but maintaining the hidden message and, if so, it attempts to reveal the
privacy and security of the data is a challenge. because location and content of the secret message. That makes it
there are many tools available for spying and other covert similar to the game of hiding and seeking secrets in
actions that can tamper with the data. Data encryption is images! Deep learning is a cutting-edge technique that is
one way that this issue can be resolved. It's similar to currently all the rage because of how useful it is for a
encoding your data with a unique key into a secret code. variety of tasks, including image recognition, speech
When it's received, the original information can be recognition, picture sorting, and even handling medical
restored using a different key to decode the secret code. images. Nowadays, there's this relatively new field called
There's a catch, though: the secret code appears to be a steganography where we may encrypt images and other
jumbled mess to the human eye, which makes one media. And what do you know? Convolutional neural
suspicious. Steganography is a very new and exciting field networks (CNNs) and generative adversarial networks
of study. It resembles a method used by secret agents to (GANs) are two techniques that deep learning is
conceal information. Steganography conceals the data so employing to enter the steganography market.
well that it is invisible to the human eye, as opposed to

http://dx.doi.org/10.29322/IJSRP.X.X.2018.pXXXX www.ijsrp.org
International Journal of Scientific and Research Publications, Volume X, Issue X, Month 2018 2
ISSN 2250-3153

Now, the purpose of this essay is to examine current

developments in the field of covert identification and
concealment in images. They are interested in knowing Steganography in Multimedia Formats: Steganography
what approaches are being taken, what is popular, and techniques have been applied to various multimedia
what difficulties people are having. The data and formats, including images, audio, and video. Research
measurements that are frequently employed in these in this area focuses on developing methods tailored to
research are also being examined. To top it all off, they the specific characteristics and constraints of each
are discussing the positive and negative aspects of format, such as the human visual and auditory systems'
various approaches and comparing how well they work. sensitivity to changes.
It's similar to a comprehensive overview of all the
covert work being done on deep learning and image Steganography in Text and Documents: In addition to
hidden messaging! multimedia formats, steganography can be applied to
text and document formats. Techniques include
II. Literature Survey embedding data in whitespace, formatting tags, or other
non-content elements of the document, as well as
1. Steganography linguistic steganography, where the structure or content
of the text itself hides the secret message.
In the digital age, steganography has evolved to hide
messages within digital media, such as images, audio, Challenges and Future Directions: Challenges in
video, and text, often in conjunction with cryptographic steganography include balancing imperceptibility with
techniques. Here's an overview of some key research in robustness, developing techniques resistant to
steganography: steganalysis, and addressing ethical and legal
considerations surrounding the use of covert
Spatial Domain Techniques: Spatial domain communication methods. Future research directions
steganography methods directly manipulate the pixels of may include exploring new embedding algorithms,
an image or the samples of audio/video signals. studying the integration of steganography with
Examples include least significant bit (LSB) emerging technologies such as blockchain and artificial
embedding, where secret data is hidden in the least intelligence, and investigating applications in areas such
significant bit of each pixel/sample, and masking and as privacy-preserving communication and digital
filtering techniques, where changes are made to specific watermarking.
regions of the cover media to embed the secret
information. Overall, steganography continues to be a vibrant area of
research with applications in cybersecurity, privacy
Transform Domain Techniques: Transform domain protection, and covert communication, posing both
steganography involves applying mathematical challenges and opportunities for researchers and
transforms, such as the discrete cosine transform (DCT) practitioners alike.
or discrete wavelet transform (DWT), to the cover
media before embedding the secret data. This can Steganalysis
improve robustness and imperceptibility compared to
spatial domain techniques. Common methods include Steganalysis is the study of using steganography to
modifying the coefficients of the transformed domain identify hidden messages. This is similar to
and exploiting the frequency domain characteristics of cryptanalysis applied to cryptography. That's why we
the media. need to understand steganography. The purpose of
steganalysis is to identify suspicious objects, determine
Quantization Index Modulation (QIM): QIM is a whether they are encoded with the payload, and if
steganographic technique that embeds information by possible, send the payload back.
adjusting the quantization levels of the cover media. By
carefully selecting the quantization intervals, QIM can Unlike cryptanalysis, where relevant data (albeit
hide data imperceptibly while maintaining statistical encrypted) contains a message, steganalysis usually
properties of the cover media. starts with a bunch of suspicious data, but little
information about which documents (if any) contain
payment information. Steganalysis is usually the job of
a forensic statistician, who must first reduce this data
Adaptive Steganography: Adaptive steganography (usually large; in most cases, this will be all data on the
techniques dynamically adjust the embedding process computer) to the most probable subset. Steganalysis
based on features of the cover media and the secret algorithm may or may not be based on steganography
message. This can improve the robustness of the algorithms (SA). Based on this, steganalysis is divided
embedding process against steganalysis techniques into the following categories:
while minimizing perceptual distortion.

http://dx.doi.org/10.29322/IJSRP.X.X.2018.pXXXX www.ijsrp.org
International Journal of Scientific and Research Publications, Volume X, Issue X, Month 2018 3
ISSN 2250-3153

extraction is nothing but dimension reduction. The

extracted features require a good understanding of the art.
A specific or focused steganalysis Image quality measurement, wavelet decomposition,
image statistical histogram moments, Markov empirical
SA Detector design (steganalysis algorithm) is known to transition matrix, spatial and frequency domain image
be based on SA. Steganalysis algorithms are based on statistical moments and co-occurrence matrices are some
SA. This type of steganalysis is based on analyzing the of the feature extraction methods.
results of images that have changed after drawing. The
advantage of using steganalysis is that the results are Classification: It is a method of classifying images
very accurate. The disadvantage of using this method is according to their importance. Monitoring learning is an
that it is limited to a specific embedding algorithm and important aspect of steganalysis. Supervised learning
image format. provides supervision-based learning. In this study, a series
of training inputs containing input features were given as
Blind steganography or universal steganalysis input to train the classifier. Predict course grades based on
the specifications given after training.
In general, not everyone knows about steganalysis and
SA. Therefore, anyone can create a test that does not Steganalysis uses the following categories:
rely on SA to detect the existence of passwords.
Compared to specific steganalysis, steganalysis is Multiple regression: There are regression coefficients. In
generally more common but less effective. General the training phase, minimum mean square error is used to
steganalysis is still more widely used than specific estimate regression coefficients.
steganalysis because it is independent of SA. Today,
scientific research focuses on universal steganalysis. It FLD: It is the combination of features that makes it stand
is intuitively shown as follows: out. During the classification process, different features
are arranged in the output space.
Experiment Design
SVM: This classification method learns from given
Statistical steganalysis examples. It is trained to recognize and provide the list
according to the given process.
Statistical analysis of pixels was performed to verify the
presence of hidden messages. It is also divided into two:
spatial domain steganalysis and transformation domain
steganalysis. Chapter Steganalysis Deep Learning Model

In the spatial domain, consider several pixels and In recent years, deep learning models such as
calculate the difference between them. This pair can be convolutional neural networks (CNN or ConvNet) have
2 adjacent pixels. They can be selected in one block or been introduced and used in the image steganalysis
two blocks. Finally, a histogram is drawn showing the environment. Deep learning combines two levels of
presence of hidden messages. machine learning (feature extraction and classification)
into a compact model.
In the transformation, the frequency number of the
coefficients is calculated and then histogram analysis is Initial Processing: Convert colored input images to
performed. With the help of these features, it is possible grayscale to reduce computational complexity while
to distinguish between cover images and steganographic retaining hidden information. Utilize a standard function
images. However, this method does not provide such as "rgb2gray()" to perform the conversion. Grayscale
information about the embedding algorithm. To conversion simplifies subsequent analysis while
overcome this problem, we can select functions preserving the integrity of concealed data within the
according to function. images.

Steganalysis Machine Learning Technology

Feature Extraction:Extracted relevant texture features
from both original (non-stego) and altered (stego) images
This method will capture the features of the image to
to characterize their unique properties. Utilized Statistical
select and store relevant information. These extracted
Feature Texture Analysis (SFTA), Local Binary Patterns
features are used to identify hidden messages in the
(LBP), and Gray-Level Co-occurrence Matrix (GLCM) to
image. They can also be used to train classes. It includes
capture texture information.
the following 2 levels:

Feature Extraction: It is the process of creating different

sets of images. These are called features. Feature

http://dx.doi.org/10.29322/IJSRP.X.X.2018.pXXXX www.ijsrp.org
International Journal of Scientific and Research Publications, Volume X, Issue X, Month 2018 4
ISSN 2250-3153

shortcomings in the game. In summary, deep learning

holds great promise in this world of hidden images that
is, if the remaining problems can be resolved. It like a
blueprint for devious picture concealing in the future.
Description:

SFTA: Derive statistical descriptors to quantify texture IV. REFERENCES

characteristics.
[1] F. Petticolas, Information hiding techniques for
LBP: Capture local texture patterns by comparing pixel steganography and digital watermarking,
intensities. StefenKatzenbeisser, Artech house books, 1999.

GLCM: Quantify texture properties based on spatial relationships [2] F. Hartung and M. Kutter, Multimedia watermarking
between pixel intensities. techniques, Proceedings of the IEEE, vol. 87, no. 7, July
1999.
Result: Combine extracted features into a comprehensive feature
[3] Martin A., Sapiro G., Seroussi G., Is image
vector representing each image.
steganography natural?, IEEE Transactions on Image
processing, 2005,
Classification:Employ advanced classifiers to discern between
original and altered images based on their feature vectors. Train [4] Kadhim I. J., Premaratne P., Vial P. J., Halloran B.,
Gaussian Discriminant Analysis (GDA) and Naïve Bayes Comprehensive survey of image steganography:
classifiers using feature vectors from both image types. Techniques, Evaluations, and trends in future research,
Neurocomputing, 2019.
Optimization: Employ techniques like feature selection,
parameter tuning, and ensemble methods to enhance classifier [5] Hussain M., Wahab A. W. A., Idris Y. I. B., Ho A.
performance. Ensure robustness and generalization across T., Jung K. H., Image steganography in spatial domain:
diverse datasets and steganographic techniques. A survey, Signal Processing: Image Communication,
2018.

[6] Karampidis K., Kavallieratou E., Papadourakis G.,

III. CONCLUSION A review of image steganalysis techniques for digital
forensics, Journal of information security and
Image steganography is similar to a covert operation in which applications, 2018.
you conceal information by embedding it into an ordinary image,
right in front of the public. Deep learning, this expensive [7] Nissar A. Mir A. H., Classification of steganalysis
technology, is now being used to make it even more devious. techniques: A study, Digital Signal Processing, 2010.
Many approaches have been investigated, including the
traditional ones based on the LSB substitution as well as some [8] Y. Zhang and Z. Wang, “Hybrid malware detection
novel ones like PVD, DCT, and EMD. Conventional techniques approach with feedback-directed machine learning,”
have their limits because over-hiding might distort the image. On Information Sciences, vol. 63, no. 139103, pp. 1–
the other hand, to make it all high-tech, they have also 139103, 2020.
incorporated fascinating deep learning architecture like VGG, U-
Net, and Xu-Net. GANs, which are essentially extremely [9] Z. Yan, P. Yang, R. Ni, Y. Zhao and H. Qi, “CNN-
intelligent algorithms, have been all the rage lately. They are based forensic method on contrast enhancement with
quite good at creating surreptitious images that closely resemble JPEG post-processing,” Computers, Materials &
the originals. For their covert picture games, some users use Continua, vol. 69, no.3.
datasets like ImageNet, CelebA, or BOSSBase. The problem is
that each person has a different way of determining how effective [10] J. Fridrich and M. Golan and R. Du, Detecting LSB
their approach is, making it difficult to compare them. They steganography in color and gray-scale images, IEEE
discovered that GANs provide the best security and covert Multimedia Magazine, Special Issue on Security, pp.
maneuvers, much like superheroes. Conventional techniques are 22-28, October-November 2001.
less secure since it is simple to determine whether a message is
buried. [11] Subhedar M. S., Mankar V. H., Current status and
key issues in image steganography: A survey, Computer
Finally, this study examined current trends as well as all science review, 2014,
the amazing techniques being used for covert
photographs. They also discussed the difficulties and

http://dx.doi.org/10.29322/IJSRP.X.X.2018.pXXXX www.ijsrp.org
International Journal of Scientific and Research Publications, Volume X, Issue X, Month 2018 5
ISSN 2250-3153

[12] Sahu A. K., Swain G., A review on LSB [14] V. Holub and J. Fridrich, “Digital image
substitution and PVD based image steganography steganography using universal distortion,” in Proc. of
techniques, Indonesian Journal of Electrical Engineering the First ACM Workshop on Information Hiding and
and Computer Science, 2016. Multimedia Security, Montpellier, France, pp. 59–68,
2013.
[13] H. Wang and S. Wang, Cyber warfare
Steganography vsSteganalysis, ACM Commun. vol. 47,
pp. 76-82, October 2004.

http://dx.doi.org/10.29322/IJSRP.X.X.2018.pXXXX www.ijsrp.org