iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
A Review Paper on Digital Image Forgery
Detection Techniques using Matlab Tool
Authors: Priyanka Kundal1; Er. Sorab Kumar2
Department of computer science & Engineering, Sri Sai College of Engineering &
Technology Badhani, Punjab India1,2
Abstract 1.1 Introduction
Images have the capability to render a large amount With the advent of widespread use and applications
of information. With the widespread popularity of of social media, images have become the most
social networking services such as WhatsApp, common data format which is shared among various
Facebook, Instagram, Snapchat, Twitter etc., there users instantly. The images have become extremely
has been a huge increase in the volume of image data useful and effective due to the large amount of
that is shared instantly. This has also increased the information that that convey and contain. Images are
cases of fake or forged images being shared which also the fundamental components of videos which
can have serious after-effects. Image forgery can are also called frames, while with frames per second
affect the image of persons, organizations and generating videos. This is the primary reason for the
communities and in some cases cause social unrest widespread sharing of images on social media
and violence. Due to the size and complexity of the platforms and elsewhere. Images are a function of
data being shared, it is almost infeasible for manual two spatial variables x and y mathematically given
detection of image forgery. Therefore, it has become by:
mandatory to design automated systems which can
𝑰 = 𝒇(𝒙, 𝒚) (1.1)
detect image forgery in very less time and with high
accuracy. Since the data size to be analyzed by time Here,
critical applications is enormous indeed, therefore
the conventional techniques prove to be infeasible to I represents the digital image
detect image forgery with high level of accuracy
f represents a function of
which makes it mandatory for using automated tools.
In this proposed work specific approach will be used x and y are the spatial coordinates
prior to classification of the image as forged or
Images are made up of picture elements or pixels
unforged. For this task Matlab Tool will be used due
which render three critical pieces of information
to the availability of in-built mathematical tools for
which are:
engineering problems.
Keywords: Image Forgery, RGB-Gray scale a) Intensity or Grayscale value
conversion, binarization, Feature Selection, b) Spatial coordinates
Baye’s Classifier, Classification. c) RGB value
Priyanka Kundal; Er. Sorab Kumar, Volume 9 Issue 5, pp 137-142, May 2021
iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
Since the images are shared extensively among users humans but are no more than streams of numerical
and are one of the most common data formats used, values for a computing machine. The different
hence they are the ones which come under the parameters of the image may vary in accordance with
purview of attackers who try to morph or manipulate the variations in the image. Now it may be
the image so as to render a different information challenging to access whether the variations occur
compared to the one that is actually intended. This due to natural causes or due to sabotaged
gives rise to image forgery or image morphing which manipulations. The natural causes can be:
can play havoc on individuals, or groups or
organizations. This may as well lead to social unrest
in several situations where the image data can go 1) Degradations while capturing
viral and without the authenticity being checked can 2) Degradations while storage
be again re-distributed or shared among a lot of other 3) Degradations while transmission
people thereby starting a chain reaction which is 4) Degradations while reception
often difficult to control. Thus it is necessary to 5) Degradations while re-creation from bits
understand the basics of images, their salient features
and the changes which images undergo while they Another significant attribute is the fact that it may be
are manipulated. challenging to predict whether the image is actually
morphed or not by a computing machine since the
variations arising out of the natural processes may
also be competent enough to match the one arising
out of deliberate effects.
1.2 Types of Digital Image Forgery
The determination of forged or unforged images
needs the fundamental attributes of image forgery
techniques. This can be basically categorized as
image forgery with active approaches or image
forgery with passive approaches. The distinction
among the above is given by:
Active Approach: - This approach generally has the
following attributes:
Fig.1.1 Pixel Level Representation of images
1) Pre-Processing
The figure above represents the pixel wise 2) Visible changes in the texture of the image
representation of an image I where the pixel P 3) A-priori information of image parameters
renders information about the intensity, coordinates 4) Generally performed during image creation,
and the RGB value. The pixels are often seen as such as watermarking applications.
matrices comprising of rows and columns of values.
These value may have a visual perception for
© 2021, iJournals All Rights Reserved www.ijournals.in Page 138
© 2020, iJournals All Rights Reserved www.ijournals.in
iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
Passive Approach: - This technique has the 1.3 Challenges In Detection Of Image Forgery
following attributes:
The major challenge in the detection of image
1) No visible evidences of morphing forgery is the largeness of the datasets and the
2) No apriori information needed complexity of the images which are under
3) Generally a blind approach consideration. A typical example of a forged and an
unforged image is shown in the figure below.
The most common forms of passive approaches are:
Copy-Move Forgery: This technique is often
referred to as cloning. In this type of manipulation,
the image under consideration is analysed, then some
part of the image is selected and then that part is used
to shroud some other distinct part. By nature of the
attack, its is challenging to make the forgery
conspicuous since the properties of the same image is
used to shroud some other part of the image.
Image splicing: This is one of the most common
types of image forgery where one part of an image
Fig.1.3 Example of Forgery
under consideration is shrouded by a part of some
other image. It is more detectable compared to the The figure above illustrates the fact that the images
copy and move approach but with the advent of new which are under consideration are generally edited
image editing tools, detection is challenging under using sophisticated tools which makes it extremely
different conditions of lighting etc. difficult to recognize the forgeries with naked eyes.
Moreover, for real time applications, automated
Retouching: Retouching can be a subtle change in
systems need to be designed which can work in the
the image with enhancement, variation in colour or
bound time limits to meet the requirements of real
illumination of the image.
time critical applications. This being said, the major
The basic categorization of image forgery techniques hindrances in the detection of image forgery are the
is given below: addition features listed below:
Scene complexity: Often the scene in the
background is chosen such that it becomes identical
to the actual image under consideration. This may
include images of scenes of natural or artificial
backgrounds.
Uneven lighting: The lighting is another factor
which is responsible for the accurate classification of
the images in categories. This being said, the ruling
Fig 1.2 Image Forgery Classifications
© 2021, iJournals All Rights Reserved www.ijournals.in Page 139
© 2020, iJournals All Rights Reserved www.ijournals.in
iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
factor is the number of pixels captured while Araz Rajab Abrahimet al. in [1] presented a
capturing the image. If this is less than the prescribed mechanism for forgery detection in images. The
or minimum number of pixels needed, a type of noise features or parameters which were utilized were the
called Poisson Noise occurs which leads to blurring histogram oriented gradient (HOG) and the local
effects in the image. binary pattern (LBP) features. The basic edge based
features and the texture based colour features were
Blurring and degradation: Blurring effects used for recognizing the image patterns. The
generally are caused due to: classification was done using a feed forward neural
network. The main limitation of the proposed work is
1) Defocussing effects
the absence of comprehensive statistical features
2) Dearth of pixels or their orientation while
along with the texture based features. This happens
image creation
to be so since the statistical features often help to
detect image forgeries which are generally not
Aspect ratios: The aspect ratios also play a critical
distinct by analysing just the colour and texture
role in deciding the classification of a data sample
based features. Moreover, the feed forward neural
into forged or unforged category. Typically, the
network may render low time complexity but its
aspect ratio is the dimensions of pixels in the x and y
limitation is often in terms of the efficacy of training
spatial dimensions. Differences in aspect ratio may
due to no feedback mechanism for errors.
lead to misleading classifications of the data sets.
Thales Pomari et al. in [2] proposed an approach for
Similar Background: Similar background mat lead
image forgery using Illumination Inconsistencies and
to the confusion for the classifier to classify the
Deep Learning. This approach primarily trains a deep
image into the correct category. This is due to the
neural network based on the features which signify
similarity in the pixel values that are overlapping in
inconsistency in illumination or indirectly the
the case of the images which are unforged and the
grayscale value of the image. The sudden changes in
ones which are forged.
the illumination of the image can often render insight
into the image that is then fed to the designed deep
Non-correlation of Pixels with Classification:
neural network which is generally a convolution
Pixel based classification is practically infeasible
neural network. The evident limitation of the
since pixels of forged and unforged images show
proposed approach is the heavy reliance in the
almost no correlation. Hence simple pixel based
inconsistency in the illumination and grayscale value
classification is not feasible.
of images. This predominantly trains the neural
2.1 Literature Review network based in the intensity value of the pixels of
the two categories of the images which are forged
This section presents the previous work done in the and unforged. The lack of inputs from different
domain. It highlights the salient features of the perspectives such as texture or statistical and
approach as well as mentions the limitation found in probabilistic parameters often turn out to be a one
the approach. dimensional approach for the classification problem
limiting the accuracy of the image.
© 2021, iJournals All Rights Reserved www.ijournals.in Page 140
© 2020, iJournals All Rights Reserved www.ijournals.in
iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
Jason Bunk et al. in [3] proposed a technique that CNN are not present, the image classification suffers
used Resampling Features and Random Walker heavily which is clearly undesirable.
segmentation to train a deep neural network. The
Yuan Rao et al. in [5] proposed a technique utilizing
approach tries to segment out the parts of the image
which seem to be morphed based on the Random the convolutional neural network for the detection of
splicing image forgery and copy-move image
Walker segmentation approach. This part is followed
forgery. The approach again uses no separate image
by the computation of resampling features. The
pre-processing technique and uses the one-shot CNN
segmentation and resampling feature computation
learning approach. The limitations of this approach
from the segmented parts are fed to the deep neural
again stems from the fact that low level details
network to classify the image. The major limitation
regarding to the image are sent for analysis to high
of the approach is the necessity of segmentation of
level neurons which again pass it forward to the next
the image parts with apparent or visible
level of neural layers with the process of
modifications. While this can lead to satisfactory
thresholding the features in the previous layer before
results for active forgery approaches, the passive
passing it on the subsequent layer. This thresholding
approaches may clearly bypass such a segmentation
process often reduces the accuracy of the features
based approach or render false positive or false
retained for the classification process. Moreover, the
negative values. Additionally, segmentation may
lack of encapsulated neurons that can analyse the low
result in fringing of the edges of segmenting parts
level orientation features reduces the classification
which can result in plummeting values of accuracy of
accuracy of the system. The certain neural layers that
classification.
are not receptive of the low level features or even
Clemens Seibold et al. in [4] proposed a technique non-receptive to the thresholded feature values affect
for detecting facial morphing using the deep learning the classification accuracy adversely.
approach of convolutional neural networks (CNN).
3 Conclusion
This approach uses the concept of one shot learning
for the convolutional neural networks. The major It can be concluded from the different research based
limitations of the proposed work are the absence of on image forgery that images are widely spread
distinctive pre-processing techniques which can through social media applications and hence there is
remove the chances of noise and disturbance effects. a chance for them being viral rapidly. The size of the
Moreover, once shot learning doesn’t rely on the data being shared makes it infeasible for
separate computation of features and simply relies on conventional or manual classification of images as
the CNN based features. This happens since this forged or unforged. Thus automated tools are
neural network approach doesn’t use the orientation mandatory which can filter out forged image thereby
of the image based features and the neurons do not mitigate the possibilities of consequences arising out
have the capability to handle this type of feature of them. The proposed approach will use the image
analysis. This approach also tries to carry out a pre-processing and feature selection prior to
search for a specific or particular type of feature classification of the image as forged or unforged. As
extraction and based on it, the classification is done. per the concern with classifier in future research
This means that if the features compatible with the Baye’sclassifier will be used which works on the
© 2021, iJournals All Rights Reserved www.ijournals.in Page 141
© 2020, iJournals All Rights Reserved www.ijournals.in
iJournals: International Journal of Software & Hardware Research in Engineering (IJSHRE)
ISSN-2347-4890
Volume 9 Issue 5 May 2021
principle of Baye’s theorem of conditional [10]DavideCozzolino ; Diego Gragnaniello ; Luisa
probability. Such a probabilistic approach is effective Verdoliva, “Image forgery detection through
since image pixels do not exhibit correlation with residual-based local descriptors and block-
being forged or unforged. . For this task Matlab Tool matching”, IEEE 2014
will be used due to the availability of in-built [11] GK Birajdar, VH Mankar, “Digital image
mathematical tools for engineering problems. forgery detection using passive techniques: A
survey”, Elsevier 2013
[12] G Lynch, FY Shih, HYM Liao, “An efficient
4. References
expanding block algorithm for image copy-move
forgery detection”, Elsevier 2013
[1] Araz Rajab Abrahim, MohdShafryMohd Rahim,
[13] M Hussain, G Muhammad, SQ Saleh, AM
Ghazali Bin Sulong “Splicing image forgery
Mirza, “Image forgery detection using multi-
identification based on artificial neural
resolution Weber local descriptors”, IEEE 2013
networkapproach and texture features”, IEEE 2018
[14] MF Hashmi, AR Hambarde, “Copy move
[2] Thales Pomari et al. et al., “Image Splicing
forgery detection using DWT and SIFT features”,
Detection Through Illumination Inconsistencies and
IEEE 2013
Deep Learning”, IEEE 2018
[15] G Muhammad, M Hussain, G Bebis, “Passive
[3] Jason Bunk et al., “Detection and Localization of
copy move image forgery detection using
Image Forgeries Using Resampling Features and
undecimated dyadic wavelet transform”, Elsevier
Deep Learning”, IEEE 2017
2012
[4] Clemens Seibold et al., “Detection of Face
[16] Gonzalez and Woods, “Digital Image
Morphing Attacks by Deep Learning”, Springer 2017
processing”, 2nd Edition, Person Publications.
[5] Yuan Rao ;JiangqunNiet, “A deep learning
approach to detection of splicing and copy-move
forgeries in images”, IEEE 2016
[6] Belhassen Bayar, Matthew C. Stamm et al.,“A
Deep Learning Approach to Universal Image
Manipulation Detection Using a New Convolutional
Layer”, IEEE 2016.
[7] Jiansheng Chen ; Xiangui Kang ; Ye Liu ; Z. Jane
Wang, “Median Filtering Forensics Based on
Convolutional Neural Networks”, IEEE 2015.
[8] Chi-Man Pun , Xiao-Chen Yuan , Xiu-Li Bi,
“Image Forgery Detection Using Adaptive
Oversegmentation and Feature Point
Matching”,IEEE 2015.
[9]Jian Li et al., “Segmentation-Based Image Copy-
Move Forgery Detection Scheme”, IEEE 2014
© 2021, iJournals All Rights Reserved www.ijournals.in Page 142
© 2020, iJournals All Rights Reserved www.ijournals.in
