Image Forgery Detection: Survey and Future Directions

Abstract

In this age of digitization, digital images are used as a prominent carrier of visual information. Images are becoming increasingly ubiquitous in everyday life. Unprecedented involvement of digital images can be seen in various paramount fields like medical science, journalism, sports, criminal investigation, image forensic, etc., where authenticity of image is of vital importance. Various tools are available free of cost or with a negligible amount of cost for manipulating images. Some tools can manipulate images to such an extent that it becomes impossible to discriminate by human visual system that image is forged or genuine. Hence, image forgery detection is a challenging area of research. It is evident that good quality work has been carried out in the past decade in the field of image forgery detection. However, there is still a need to pay much attention in this field, as image manipulation tools are becoming more and more sophisticated. The main purpose of this paper is to review the various existing methods developed for detecting the image forgery. A categorization of various forgery detection techniques has been presented in the paper.

Full text

Image Forgery Detection: Survey
and Future Directions
Kunj Bihari Meena and Vipin Tyagi
1 Introduction
One famous proverb says “A picture is worth a thousand words”. Now everybody
understands the essence of this idiom. But due to the availability of sophisticated
tools for image manipulation, it is very easy to tamper the image by anyone with
a modicum of computer skills. Hence, authenticity of image is challenged openly,
therefore somewhere the above idiom loses its essence.
According to Merriam-Webster dictionary, digital image forgery is defined as
“falsely and fraudulently altering a digital image”. Image forgery is not a new concept;
it started way back in 1840. French photographer Hippolyte Bayard created the first
tampered image (Fig. 1) entitled with, “Self Portrait as a Drowned Man”, in which,
Bayard has professed to commit suicide [1].
More than a century ago, during American Civil War, a photo of American com-
manding general, General Ulysses S. Grant came into existence, which claimed that
General Grant was sitting on horseback in front of his troops, at City Point, Virginia
[2]. Later on, it has been found that image was not authentic; rather it was a composite
of three images formed using negatives of the photographs.
Almost a decade ago, Iran has been accused of doctoring an image from its missile
tests; the image [3] was released on the official website, Iran’s Revolutionary Guard,
which claimed that four missiles were heading skyward simultaneously. Recently, in
July 2017, a fake image of Russian president Vladimir Putin was circulated over the
social media related to the meeting with American president Donald Trump during
the G20 summit 2017. This faked image garnered several thousand likes and retweets
[4].
K. B. Meena ·V. Tyagi (B)
Jaypee University of Engineering and Technology, Raghogarh, Guna, MP, India
e-mail: dr.vipin.tyagi@gmail.com
© Springer Nature Singapore Pte Ltd. 2019
R. K. Shukla et al. (eds.), Data, Engineering and Applications,
https://doi.org/10.1007/978-981- 13-6351- 1_14
163
dr.vipin.tyagi@gmail.com

164 K. B. Meena and V. Tyagi
Fig. 1 First fake image [1]
Image has remarkable role in various areas such as forensic investigation, crim-
inal investigation, surveillance systems, intelligence system, sports, legal services,
medical imaging, insurance claim, and journalism.
Substantial amount of research has been carried out in the last one decade in the
field of forgery detection. Figure 2shows the bar chart of a number of publications
versus four types of image forgery detection techniques (copy-move, image splicing,
resampling, retouching) for last two decades, over the years 1998–2017, collected
from Google Scholar. Few observations from this bar chart are: startling growth has
been seen in copy-move forgery detection in last one decade, and a significant focus
is also given on image splicing detection in the last one decade over the first decade.
However, less focus was given on retouching detection, one reason behind this may
be that retouching is the least pernicious type forgery because generally, retouched
images are not used for illegal purposes.
Forgery detection techniques are broadly categorized into two categories; active
(non-blind, Fig. 3) and passive (blind) [5]. Active forgery detection techniques need
some prior information about the image which may have been embedded in the image
at the time of capturing the image or during image acquisition or later stages. Digital
watermarking [6–8] and digital signature [9,10] are the examples of active forgery
detection techniques, and these approaches can be used to test the authenticity of the
Fig. 2 Number of
publications in the field of
image forensics over the last
two decades
0
100
200
300
400
500
copy-move splicing retouching resampling
Number of Publications
Type of Forgery
1998-2007 2008-2017
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 165
Active Image Forgery
Digital Watermarking
Fragile Semi-fragile Robust
Digital Signature
Fig. 3 Active image forgery detection techniques
Passive
Forgery
Tampering
operation based
Dependent
Image splicing
Copy-move
Independent
Resampling
Retouching
Sharpening and
Bluring
Brightness and
Contrast
Compression
Image filtering
Source camera
identification based
Lens abberation
CFA
interpolation
Sensor
irregularities
Image feature
based
Fig. 4 Classification of passive forgery detection techniques
image based on embedded information. On the basis of application, digital water-
marking further can be categorized as fragile, semi-fragile, and robust watermarking
[11]. In practicality, it is very rare that images produced for forensic investigation
like fingerprint images, crime scene images, photographs of criminals, etc., would
contain the watermark or signature, hence it can be concluded that active forgery
detection techniques are not useful for forensic investigation of digital images.
On the contrary, passive forgery detection techniques do not need any prior knowl-
edge about the image; rather these techniques identify manipulations by extracting
intrinsic features of the image on the basis of the type of doctoring or photo-capturing
dr.vipin.tyagi@gmail.com

166 K. B. Meena and V. Tyagi
device identification. Passive forgery further can be classified (Fig. 4) as dependent
forgery and independent forgery. In dependent forgery, either tampering can be done
in the same image by copying and pasting (cloning) [12] some area within the image
or more than one image can be combined (image splicing) [13–15] to get convincing
composite. On the other hand, independent forgery is the forgery in which some
properties of the same image are manipulated. An example of independent forgery
includes resampling, retouching, image rotation, scaling, resizing, addition of noise,
blurring, image compression, etc. No involvement of prior knowledge about image
makes passive forgery more practical in real life.
2 Existing Surveys
In the last decade, many surveys have been carried out on image forgery detec-
tion. Lanh et al. [16] discussed various techniques to detect image forgery based
on camera. They have given conclusive remark that camera-based techniques are
better than other forgery detection techniques, in terms of reliability. Farid [17], cat-
egorized image forgery tools into five groups, pixel-based techniques, format-based
techniques, camera-based techniques, physically based techniques, and geometric-
based. He has elaborated each method in detail. Recently, Warif et al. [18], reviewed
copy-move forgery detection techniques. They have categorized copy-move forgery
detection mainly into two classes: block-based and keypoint-based approach. Table 1
shows existing survey papers available on Google Scholar during 2007–2017.
Detailed classification of forgery detection methods is shown in Fig. 4, wherein
blind forgery detection techniques have been categorized as tampering detection
based and source camera identification based techniques. Tampering detection tech-
niques have been discussed in this paper. For complete survey on source camera
identification based techniques, readers may refer to [16,23,24].
3 Tampering Detection Techniques
In context to digital image, tampering means any manipulation or alteration in image
to change its semantic meaning for illegal or unauthorized purposes. A tremen-
dous amount of images are produced before digital image forensic for investigating
whether the image is authentic (no alteration in semantic meaning of image) or tam-
pered. Since photo-editing tools are becoming increasingly ubiquitous, anybody can
tamper with the image and may use it with malicious intention. Figure 4shows
various tampering detection techniques.
In this section, four major types of tampering detection techniques (image splicing,
copy-move forgery, image resampling, image retouching detection) are presented.
Out of these four tampering detection techniques, first two are exploited for detecting
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 167
Table 1 Existing survey papers available on Google scholar
S. n. Author(s) Contribution Observations
1 Lanh et al. [16] Reviewed various techniques in digital
camera image forensics
• Intrinsic features based methods of camera hardware are more reliable and better
in terms of accuracy as compared to methods based on other camera software
parts
• Camera identification methods outperform as compared to other forgery
detection methods
• Hardware dependent characteristics such as aberration and CRF are potentially
more reliable than methods based on scene content like lighting and image
statistics
2 Farid [17] Categorized the image forgery detection
techniques into five groups (pixel-based,
format-based, camera-based, physically
based and geometric-based)
• Some forensic tools may not detect advanced forgeries but other forgery
detection techniques are much reliable to challenge image fakery
• Due to the advancement of image manipulation tools, an arms race between the
forger and forensic analyst is inescapable
3 Mahdian and
Saic [19]
Reviewed various method based on blind
image forgery
• Existing methods produce considerably higher false positive rates than which are
reported in the existing papers
• Existing methods are not fully automated, need human interpretation
• Need to develop more reliable and robust methods
4 Christlein et al.
[12]
Reviewed state of the art approaches
pertaining to copy-move forgery
• Keypoint-based methods better than block-based methods in term of
performance (execution time), however, block-based methods give better
detection accuracy
5 Birajdar and
Mankar [20]
Reviewed forgery detection techniques
with more emphasis on passive tampering
detection
• Existing methods are not automated, outputs need a human interpretation
• Existing methods are not effective when small regions are copy-moved
• Copy-move forgery detection, have shown high time complexity and false
positives
• Need to extend forgery detection on audio and video
(continued)
dr.vipin.tyagi@gmail.com

168 K. B. Meena and V. Tyagi
Table 1 (continued)
S. n. Author(s) Contribution Observations
6 Qazi et al. [21] Surveyed various blind forgery detection
techniques and classified into mainly
three groups (copy-move, splicing, and
retouching)
• DCT and PCA based techniques, exhibit high computational complexity and low
accuracy rate
• DCT-based techniques are not effective when considering highly textured and
small forged regions
7 Ansari et al. [22] Various approaches of pixel-based image
forgery detection have been reviewed
• Some algorithms are unable to detect forgery effectively. Some are having high
time complexity
• Need to develop an efficient and accurate image forgery detection algorithm
8 Warif et al. [18] CMFD techniques are organized into two
approaches, namely: block-based and
keypoint-based
• Existing techniques are not fit to solve real-world big data problems
• To increase processing speed, dimension reduction techniques like PCA, DWT,
and SVD has been suggested
• Keypoint-based methods like SIFT and SURF are more reliable when
geometrical transformation operations are taken into consideration
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 169
dependent forgery and the last two tampering detection techniques are used for
examining independent forgery.
3.1 Image Splicing Detection or Photo Composite Detection
In image splicing forgery, some part of image is copied and pasted on the other image
to get forged image (Fig. 5). Image splicing is a basic step to create a photomontage
from a set of images. To make composite image more realistic, postprocessing (scal-
ing, cropping, retouching, rotating, etc.) operation may be applied on each of the
components, furthermore, after performing splicing operation, again postprocessing
operation can be applied to hide any discernible effects.
Although experts can identify image splicing forgery by just looking a forged
image, in some cases. However, experienced forger can make composite image so
elegant that it is almost impossible to say anything about the genuineness of an
image, merely by looking at the image. When image splicing operation is carried
out, some image statistics get disarranged. However, these statistical changes may
not be perceptible to the human visual system. These statistic disarrangements of
an image cannot be mitigated, even when expert burglar performs blending [25] and
matting [26] operations on the forged image as a postprocessing operation.
Bicoherence features were proposed by Farid [27] to highlight the traces of tam-
pered signal. In this paper, Farid has taken the assumption that in the frequency
domain, a natural signal has weak higher order statistical correlations. Then after
applying polyspectral analysis (bispectrum/bicoherence), he showed that “unnatu-
ral” correlations are introduced if the signal is passed through a nonlinearity. Farid
has applied this technique to detect the splicing in human speech. Later on, in [28],
Post
processing
(Optional)
Cropping/ resizing/
rotating (Optional)
Composite image
Splicing
Operation
Fig. 5 Process of image splicing
dr.vipin.tyagi@gmail.com

170 K. B. Meena and V. Tyagi
Ng et al. exploited bicoherence feature for detecting image splicing in images. They
have proposed two methods, the first method exploits the dependence of the bico-
herence features on the image content and the second offsets the splicing-invariant
component from bicoherence. Detection accuracy of their method was from 62 to
70%. In their method, features were evaluated with Support Vector Machine (SVM).
Later on, the same authors proposed [29] a new method to detect image spicing based
on bipolar signal perturbation. They concluded that image splicing process increases
the value of the bicoherence magnitude and phase features.
Considering lighting inconsistencies as a fundamental key feature for detecting
image splicing, Farid et al. proposed a method for estimating the direction of an illu-
minating light [30]. Hilbert–Huang Transform (HHT) based technique was proposed
in [31]. They have been exploiting SVM classifier and claimed 80.15% detection
accuracy. Further work was carried out by Li et al. [32] and they used SVM classi-
fier on moment features and HHT-based features together. They achieved detection
accuracy of 85.87%, which is higher than that of the prior work (70% as reported in
[28] and 80.15% in [31]) on the same evaluation dataset [33].
A natural image model was proposed by Shi et al. [34], in which statistical features
extracted from the test image and 2D arrays were generated by applying multi-size
block discrete cosine transform (MBDCT). The statistical features include moments
of characteristic functions of wavelet subbands and Markov transition probabilities
of difference 2D arrays. Dong et al. [35] devised a method for image splicing detec-
tion based on the statistical features extracted from image run length representation
and image edge statistics. The support vector machine (SVM) is used as a classifier
and achieved detection accuracy was 84.36%. Wang et al. [36], have given a new
technique, using image chroma component. Hsu et al., in their research paper [13],
presented a fully automatic method to detect splicing of digital images by incorpo-
rating three features: geometry invariant CRF estimation, consistency checking, and
image segmentation. Unfortunately, the method was not robust enough and showed
recall and precision about 70% only. Kakar et al. [37], developed a new approach to
detect image forgery based on discrepancies in motion blur and spectral analysis of
image gradients. They showed that their technique outperforms other contemporary
techniques, which are applicable to motion blur. Carvalho et al. [14], designed a
new method based on inconsistencies in the color of the illumination of image, by
exploiting SVM meta-fusion classifier.
Rao et al. developed a new approach [38] to unveil splicing in image, by exploiting
motion blur cues. Authors claimed that their approach can expose splicing even under
space-variant blurring situations. A new method with detection accuracy of 98.2% to
detect image splicing using Markov features in spatial and discrete cosine transform,
invented by El-Alfy and Qureshi [39]. They further improved the performance of
the proposed approach by applying the PCA (Principle Component Analysis) to
select the most relevant features before building the detection model. Meanwhile,
two other sophisticated methods were developed for unveiling splicing, in [40,41].
Noise discrepancies in multiple scales are utilized as indicators for image splicing
forgery detection in the paper [42] by Pun et al., they gave conclusive remark that their
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 171
proposed method retains good detection accuracy in diverse situations like spliced
area with different noise variance.
In [43], Park et al., introduced a method for image splicing detection, using
the characteristic function moments for the inter-scale co-occurrence matrix in the
wavelet domain with accuracy of 95.6%. They have tested their method on three pop-
ular datasets, Columbia, CASIA1, and CASIA2. Concurrently, Zhang and Lu [44],
obtained an approach for unveiling image splicing by incorporating Markov model,
in the block discrete cosine transform (DCT) domain and the Contourlet transform
domain. In their illustrated method, authors have exploited SVM classifier to classify
the authentic and spliced images for the gray image dataset. Verdoliva et al. devised
an approach [45] by using autoencoder-based anomaly as a key feature.
Recently, Shen et al. [45] developed an algorithm for detecting image splicing
by exploiting textural features based on the Gray-Level Co-occurrence Matrices.
A support vector machine (SVM) is employed for classification purpose. The illus-
trated algorithm achieves the detection rates of 98% on CASIA v1.0, 97% on CASIA
v2.0 and 91.88% on Columbia Image Splicing Detection Evaluation Dataset, with
96-D feature vector. Meanwhile, Farid [46] described three geometric techniques
for detecting traces of digital manipulation in images. His proposed techniques were
based on vanishing point, reflection, and shadow’s location. Table 2shows compar-
ison of various algorithms for image.
3.2 Copy-Move Forgery Detection
In copy-move forgery one segment of image is copied and pasted in the other part
of same image. Main intention of copy-move forgery is to hide some visual clues
or replicating the things in image to mislead peoples. The prominent reason behind
the surge in copy-move forgery is simplicity of this forgery. Good collection of
tampered images throughout the history of image processing is available in [3].
Common workflow of copy-move forgery detection techniques has been shown in
Fig. 6.
A survey on copy-move forgery detection techniques has been carried out in [18].
They have reviewed various research paper published in Web of Science (WOS)
during years 2007–2014.
Silva et al. [53] developed a method for detecting copy-move forgery based on
multiscale analysis and voting processes of a digital image. This method detects
key points by exploiting Speeded-Up Robust Features (SURF) technique; Nearest
Neighbor Distance Ratio (NNDR) is used for feature matching. Illustrated method
can work under rotation, resizing or any combination of both. Unfortunately, it might
not find a sufficient amount of key points in a small or homogeneous region.
Gabor filter based approach for copy-move forgery detection has been introduced
by Lee et al. [54], which incorporates lexicographical sorting as a feature matching
technique. Time complexity of this method was (O(PNlogN) +O(2JPN)). Mean-
while, in [55], Ardizzone et al. developed a copy-move forgery detection approach
dr.vipin.tyagi@gmail.com

172 K. B. Meena and V. Tyagi
Table 2 Comparative study of existing techniques of image splicing detection
S. n. Algorithm Features
extracted
Classifier used Accuracy Dataset
1 Ng et al. [28] Bicoherence
features
SVM 62–70% CISDE
2 Fu et al. [31] Hilbert–Huang
Transform
(HHT), and
wavelet
decomposition
SVM 80.15% CISDE
3 Shi et al. [34] Moments of
characteristic
functions of
wavelet
subbands and
Markov
transition
probabilities of
difference 2-D
arrays
SVM with
RBF kernel
91.87% CISDE
4 Chen et al. [47] 2D phase
congruency
and statistical
moments of
characteristic
function
SVM 82.32% CISDE
5 Dong et al.
[35]
Statistic
moments of
characteristic
function of
image run
length
histograms
SVM 80.46–84.36% CISDE
Wang et al.
[36]
Image chroma
component
SVM 84.2% CISDE
6 Li et al. [32] HHT and
moments
feature
SVM 85.87% CISDE
7 Hsu and
Chang [13]
Geometry
invariant CRF
estimation,
consistency
checking, and
image
segmentation
SVM 70% precision,
70% recall
CUISDE
(continued)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 173
Table 2 (continued)
S. n. Algorithm Features
extracted
Classifier used Accuracy Dataset
8 He et al. [48] Approximate
run length
along with
edge gradient
direction
SVM with
RBF kernel
80.58% CISDE
9 He et al. [49] Markov
features in
DCT and
DWT domain
SVM-RFE 93.55% CISDE and
CASIA v1
10 Carvalho et al.
[14]
Inconsistencies
in the color of
the
illumination of
images
SVM
meta-fusion
85–86% Dataset of 200
images taken
from the
Internet
11 Xu et al. [50] The DCT
Markov
features in
chroma
channel
SVM – CUISDE
12 Qureshi et al.
[39]
Markov
features in
spatial and
discrete cosine
transform,
Principal
Component
Analysis
(PCA)
SVM with
RBF kernel
98.2% CISDE
13 Bahrami et al.
[40]
Partial blur
type
inconsistency
Block-based
partitioning
94.6% Dataset of
1200 tampered
images
14 Zhao et al. [41] 2D noncausal
markov model
SVM 93.36% CISDE
15 Park et al. [43] Characteristic
function
moments for
the inter-scale
co-occurrence
matrix in the
wavelet
domain
SVM with
RBF kernel
95.3–95.6% CASIA1 and
CASIA2
(continued)
dr.vipin.tyagi@gmail.com

174 K. B. Meena and V. Tyagi
Table 2 (continued)
S. n. Algorithm Features
extracted
Classifier used Accuracy Dataset
16 Han et al. [51] Markov
feature
SVM 94.87–98.50% CASIA1 and
CASIA2
17 Zhang et al.
[44]
Markov
feature in
block Discrete
Cosine
Transform
(DCT) domain
and the
Contourlet
transform
domain
SVM-RFE 96.69% Dataset of
1150 forged
color images
18 Rao and Ni
[52]
Deep learning
technique, and
Convolutional
Neural
Network
(CNN)
SVM 96.38% CASIA v1.0,
CASIA v2.0
and
CISDE
19 Shen et al. [45] Textural
features based
on the
gray-level
co-occurrence
matrices
SVM 97–98% CASIA v1.0
and CASIA
v2.0
based on matching triangles, by applying mean vertex descriptors. This approach
shows better performance in case of complex scenes; however, a lot of false matches
occur with regular background.
Cozzolino et al. devised an algorithm [56] by considering dense-field techniques
and Zernike moments as keypoint. Their algorithm utilizes nearest neighbor search
algorithm and PatchMatch as a feature matching technique. Experiments were per-
formed on copy-move forgery detection techniques in [57] by Li et al. and devised
robust method for copy-move forgery detection by employing vlFeat software as
feature extraction tool. Furthermore, Authors improved the accuracy of the obtained
results by employing RANSAC via the gold standard algorithm. Their experiment
showed the average precision as 0.86, however, method has high computation com-
plexity, hence lead to low detection speed.
Pun et al. [58] proposed an algorithm to investigate copy-move forgery by com-
bining keypoint feature and block-based feature. Experimental results show that their
proposed scheme can achieve much better detection results for copy-move forgery
images under various challenging conditions, such as geometric transforms, JPEG
compression, and downsampling, compared with the existing contemporary copy-
move forgery detection schemes. Also, they have measured precision value as 96.6%
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 175
Input: Image to be checkedOutput: Forgery detected in image
Pre-processing
(Optional)
Feature Extraction
Phase
Feature Matching
Phase
Visualization
(Optional)
Fig. 6 Common workflow of copy-move forgery detection techniques
and recall value as 100%. Block-based technique to detect copy-move forgery was
given by Lee et al. [54], by using a histogram of orientated gradients, which can
deal with images distorted by small rotations, blurring, adjustment of brightness,
and color reduction. However, their approach fails if high rotation and scaling are
introduced by forger.
A rotation-invariant method to detect copy-move forgery based on circular pro-
jection, was presented by Gürbüz et al. [59]. Meanwhile, Zhao et al. [60] proposed a
technique to detect copy-move forgery by incorporating split-half recursion matching
strategy to match SIFT keypoints. Method first calculates the affine transformation
between two matched regions. And then, the ZNCC coefficients are measured to
detect duplicate region.
Wenchang et al. introduced a new method [61] to detect copy-move forgery by
employing new concept particle swam optimization (PSO) along with SIFT keypoint.
In their experiment, authors have employed the best bin first (BBF) algorithm for fea-
ture matching. The method showed the precision of 99%, but unable to detect forgery
when duplicated region is very small. Zandi et al. proposed a technique [62] based
on interest point detector. In this paper, authors first detect all the interest points and
then describe features using Polar Cosine Transform. After that, an improved version
of the adaptive matching is employed. Furthermore, falsely matched pairs are dis-
carded by an effective filtering algorithm. Moreover, to enhance the result, they have
iterated process regarding the prior information. Authors claimed that their method
can be exploited in other image processing areas, such as scene recognition or image
retrieval, etc. However, method is vulnerable to resizing attack. Behavior knowl-
dr.vipin.tyagi@gmail.com

176 K. B. Meena and V. Tyagi
edge space-based fusion was employed for copy-move forgery detection by Ferreira
et al. [63]. In this work, they have overcome the limitations of fusing approaches
by introducing new behavior knowledge space representation. Furthermore, authors
also proposed multiscale behavior knowledge representation to deal with resizing
and noise addition issues. Drawback with this method, however, is that it does not
work well when image has several homogeneous regions.
Copy-move forgery detection technique based on scaled ORB has been proposed
by Zhu et al. [64]. Their technique first, establishes a Gaussian scale space then,
extracts FAST keypoints and the ORB features, in each scale space. Furthermore,
technique employs RANSAC algorithm to remove the falsely matched keypoints.
Experimental result shows that technique is effective for geometric transformation.
However, approach is time-consuming when operated on high-resolution images. Bi
et al. [65] designed a copy-move forgery detection technique by incorporating Multi-
Level Dense Descriptor (MLDD) as a feature extraction method. They have utilized
hierarchical feature matching method. Further, some morphological operations are
applied to generate the final detected forgery regions. Approach work effectively
with geometric transforms, JPEG compression, noise addition, and downsampling.
Ustubioglu et al. devised an algorithm for copy-move forgery detection by uti-
lizing DCT-phase terms to restrict the range of the feature vector elements and also
employed Benford’s generalized law to determine the compression history of the
image. The method uses element-by-element equality between the features. The
method was also robust against postprocessing operations.
A new keypoint-based copy-move forgery detection for small smooth regions was
developed by Wang et al. [66] by introducing the superpixel content based adaptive
feature point detector. They also employed robust EMs-based keypoint features and
fast Rg2NN based keypoint matching. However, serious limitation of method was
the high computational complexity. Copy-move forgery detection technique has been
devised by combining cellular automata(CA) and local binary patterns(LBP) by
Tralic et al. [67]. The combination of CA and LBP allows a simple and reduced
description of texture in the form of CA rules that represents local changes in pixel
luminance values.
Recently in [68], a copy-move forgery detection method based on CMFD-SIFT,
has been proposed by Yang et al. In their method, keypoints are detected by using a
modified SIFT-based detector. This method improves the invariance to mirror trans-
formation. Table 3shows pros and cons of various algorithms for copy-move forgery
detection based on several components such as feature extraction techniques, feature
matching techniques, performance of algorithm and dataset used.
3.3 Image Resampling Detection
Resampling is mathematical technique to change the resolution (number of samples)
of image, mainly for the purpose of increasing the size of image (upsampling) for
printing banners and hoardings, etc., or for minimizing the size of image (downsam-
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 177
Table 3 Comparative study of existing techniques of copy-move forgery detection techniques
published between 2015 and 2017
S. n Algorithm Feature
extraction
technique
Feature
matching
method
Performance Pros/cons Dataset
1 Silva
et al.
[53]
Multiscale
analysis
SURF
NNDR CPU time
1.881 s/image
Pros:
works
under
rotation,
resizing,
and these
operations
combined.
Cons: not
suitable
for small
or homo-
geneous
region
CMH
dataset
CMEN
datasets
2 Lee
[69]
Gabor
filter
Lexicographical
sorting
O(PNlogN)
+O(2JPN))
Pros: work
even when
image is
distorted
by slight
rotation
and
scaling,
JPEG
compres-
sion,
blurring,
and
brightness
adjust-
ment
CoMoFoD
dataset
CMFDA
3 Ardizzone
et al.
[55]
Matching
Triangles
Mean Vertex
Descriptors
10 s/image Pros:
better per-
formance
in case of
complex
scenes
Cons: a lot
of false
matches
image
with
regular
back-
ground
CMFDA
(continued)
dr.vipin.tyagi@gmail.com

178 K. B. Meena and V. Tyagi
Table 3 (continued)
S. n Algorithm Feature
extraction
technique
Feature
matching
method
Performance Pros/cons Dataset
4 Cozzolino
et al.
[56]
Dense-
field
techniques
and
Zernike
moments
nearest neighbor
search algorithm
and PatchMatch
11 s/image Pros:
achieves
higher
robustness
on
rotations
and scale
changes
Cons:
slow per-
formance
Database
of 80
images
along with
[12]
5 Li et al.
[57]
VlFeat
software,
RANSAC
K nearest
neighbors
Precision is
86%
Pros: good
detection
accuracy
Cons:
slow per-
formance
CMFDA
along with
MICC-
F600 and
MICC-
F2000
6 Pun
et al.
[58]
SIFT Morphological
operation
Precision
96.6% and
recall 100%
Pros:
better
accuracy
CMFDA
7 Lee
et al.
[54]
Histogram
of
orientated
gradients
Lexicographical
sorting
Fc factor >
90%
Pros:
robust
against
small
rotations,
blurring,
adjust-
ment of
bright-
ness, and
color
reduction
Cons: not
suitable
with high
rotation
and
scaling
CoMoFoD,
second
dataset of
30 high-
resolution
images
(continued)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 179
Table 3 (continued)
S. n Algorithm Feature
extraction
technique
Feature
matching
method
Performance Pros/cons Dataset
8 Gürbüz
et al.
[59]
Circular
projection
technique
Lexicographical
sorted
Accuracy
99%
Pros:
robust
against
scaling
and
mirroring
opera-
tions.
Cons: less
effective
while
rotation
angle is
big
20 test
images
with sizes
of 326 ×
245
9 Zhao
et al.
[60]
SIFT g2NN – Pros:
effective
with
rotation,
scaling
and
multiple
copy
operations
Cons:
accuracy
decreases
with com-
pression
MICC
F2000
10 Wenchang
et al.
[61]
Particle
Swam
Optimiza-
tion with
SIFT
Best bin first Precision
99%
Cons: fails
to detect
too small
region
CMFDA
11 Zandi
et al.
[62]
Interest
point
detector
Adaptive
matching
436 ms/image Pros:
effective
under
various
challeng-
ing
conditions
SBU-
CM161
(continued)
dr.vipin.tyagi@gmail.com

180 K. B. Meena and V. Tyagi
Table 3 (continued)
S. n Algorithm Feature
extraction
technique
Feature
matching
method
Performance Pros/cons Dataset
12 Ferreira
et al.
[63]
BKS
fusion
Multiscale
Behavior
Knowledge
O(N2) Cons: less
efficient,
also fails
when
image has
several
homoge-
neous
regions,
CPH
dataset
13 Zhu
et al.
[64]
FAST
keypoints
and the
ORB
features
hamming
distance
270 ms/image Pros:
effective
for
geometric
transfor-
mation
Cons:
time con-
suming
for forgery
detection
of high-
resolution
images.
Dataset of
107 real
images
and 107
tampered
images
14 Bi et al.
[65]
Multi-
level
Dense
Descriptor
Hierarchical
Feature
Matching
F score >
91%
Pros:
robust
against
various
attacks
CMFDA
dataset
15 Bi et al.
[70]
Multiscale
feature
Adaptive patch F =95.05% Pros: good
perfor-
mance on
downsam-
pling and
multiple
copies
CMFDA
dataset
16 Ustubioglu
et al.
[71]
DCT-
phase
term
Element-by-
element
equality
Accuracy
96%
Pros:
robust
against
various
postpro-
cessing
operations
Comofod
and Kodak
databases
(continued)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 181
Table 3 (continued)
S. n Algorithm Feature
extraction
technique
Feature
matching
method
Performance Pros/cons Dataset
17 Wang
et al.
[66]
Superpixels
classifica-
tion and
adaptive
keypoints
Reversed g2NN 221 s/image Pros:
effective
with
geometric
transforms
Cons:
higher
computa-
tional
complex-
ity
CMFDA
dataset
18 Zheng
et al.
[72]
Zernike
moments
and SIFT
along
g2NN F =84.91% Pros: can
detect
smooth
regions
CMFDA
dataset
CoMoFoD
19 Tralic
et al.
[67]
Cellular
Automata
(CA) and
LBP
Euclidean
distance
F > 0.92 Pros: low
computa-
tional
complex-
ity
CoMoFoD
20 Yang
et al.
[68]
Adaptive
SIFT
AHC algorithm F > 90% Pros:
improves
the
invariance
to mirror
transfor-
mation
CoMoFoD
21 Huang
et al. [73]
FFT,
SVD, and
PCA
Exhaustive
search
Accuracy
98%
Pros: high
detection
accuracy
CASIA
v1.0
pling) for email and website use. In general, almost all sort of digital image forgery
(more specifically image splicing) involve scaling, rotation or skewing operations
to manipulate the image. In these operations use of resampling and interpolation
processes is inevitable. Hence, it is possible to detect the image forgery by tracing
the symptoms of resampling in image. Several papers have been published in the
past decade to detect the forgery in image on the basis of resampling.
Popescu et al. proposed a method [74] to expose digital forgeries by detecting
traces of resampling. In blind forgery detection, no prior information is available
about image, like which particular postprocessing attack has been applied, which
interpolation is used to resample the image or part of image. However, to identify
traces of resampling, interpolation details might be a basic telltale cue of resam-
pling detection. Hence, authors in exploited expectation/maximization algorithm
dr.vipin.tyagi@gmail.com

182 K. B. Meena and V. Tyagi
(EM) [75] to determine if a signal has been resampled. Two models were developed,
one for those samples that are correlated to their neighbors, and the second model
corresponds to those samples that are not correlated. Their method is effective to
unveil the sign of linear or cubic interpolation. However, it fails to detect other more
sophisticated nonlinear interpolation techniques.
Kirchner [76] introduced a method based on fixed linear predictor. Method extracts
periodic artifact and detect resampling. Meanwhile, Mahdian and Saic [77] proposed
an algorithm to detect interpolation and resampling with 100% detection accuracy.
The method was based on derivative operator and radon transformation. Their method
was effective to detect the traces of scaling, rotation, skewing transformations.
Li et al. developed an algorithm in [78] to detecting resampling based on periodic-
ity introduced by resampling and JPEG compression. They employed EM algorithm
to obtain the probability map of an image. Further, Fourier-transformed and matched
with affine-transform templates employed to detect resampling. They have experi-
mentally concluded that image is not undergone resampling if the periodicity of the
probability map obtained. Moreover, they have examined their method on the dataset
of 100 grayscale images and claimed the detection accuracy better than [74].
Lien et al. [79] illustrated a new approach to detect forgery by observing the
detectable periodic distribution properties generated from the resampling and inter-
polation processes. Their approach divided resampling as horizontal and vertical and
then applied detection technique. Experimentally, authors have claimed 95% detec-
tion accuracy of their method which in turn can verify one image of resolution 512 ×
512 only in 50 s on their mentioned system. In [80] Qian et al. developed a method
to detect blind image forgery using resampling history detection algorithm. Instead
of calculating the exact resampling energy spectrum of second-order derivative rate,
authors have proposed a special distance measurement for measuring how far apart
two sub-images are away from each other in terms of resampling difference. Method
can detect the resampling even when rotation has been performed after resampling.
In [81], Birajdar et al. invented a new technique that blindly detects global rescal-
ing operation and estimates the rescaling factor based on the autocovariance sequence
of zero-crossings of second difference of the tampered image. The method is robust
to detect rescaling operation for images that have been subjected to various forms of
attacks like JPEG compression and arbitrary cropping with accuracy of 99.5%.
Recently, David and Fernando [82] devised a new approach for the detection of
resampling by incorporating new tools and concepts from RMT (Random Matrix
Theory). RMT provides useful tools for modeling the behavior of the eigenvalues
and singular values of random matrices. Striking positive aspect of the method was
very low computational complexity. Meanwhile, Qian et al. also proposed a method
for detecting resampling forgery in digital image by using linear parametric model.
In their method, first resampling is detected in 1D signal then further they have
extended it for 2D image.
Table 4summarizes the pros and cons of various algorithms developed for resam-
pling detection, based on several components such as feature extraction techniques,
detection accuracy of algorithm, and dataset used.
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 183
Table 4 Comparative study of existing techniques of resampling detection
S. n. Algorithm Feature
description
Detection accu-
racy/performance
Pros/cons Dataset
1 Popescu
and Farid
[74]
EM algorithm Accuracy 80% Pros: work with
GIF format also
Cons: fail to
detect other
more
sophisticated
nonlinear
interpolation
Database of
200 grayscale
images in
TIFF format
with 512 ×
512 size
cropped from
a smaller set
of 25, 1200 ×
1600 images
2 Kirchner
[76]
Fixed linear
predictor
Accuracy 100%
for upsampling
Pros: fast and
reliable
Database of
200
uncompressed
8 bit grayscale
with
resolution
3112 ×2382
pixels
3 Mahdian
and Saic
[77]
Derivative
operator and
radon
transformation
100% Pros: capable of
detecting traces
of scaling,
rotation,
skewing
transformations
Dataset of 40
images
corrupted by
various trans-
formations
4 Wang and
Ping [83]
Singular value
decomposition
79.838% Pros: robust
against scaling
manipulation
Cons: less
accurate with
rotating
transformation
and
compression
UCID
5 Li et al.
[78]
EM algorithm,
Fourier
transform and
affine
transform
Better detection
accuracy than
[74]
Pros: better on
resampling
detection in
JPEG
compression
Cons: very time
consuming
Database of
100 gray-level
images of
various
resolutions
and formats
(TIFF, BMP,
PNG etc.)
(continued)
dr.vipin.tyagi@gmail.com

184 K. B. Meena and V. Tyagi
Table 4 (continued)
S. n. Algorithm Feature
description
Detection accu-
racy/performance
Pros/cons Dataset
6 Lien et al.
[79]
Pre-calculated
resampling
weighting
table
Accuracy 95%
and CPU time
50 s/image
Pros: better
detection
accuracy
Dataset with
160 gray
images with
resolution 512
×512
7 Qian
et al. [80]
DFT 0.5203 s/image Pros: effective
with rotation
after resampling
Dataset of 500
images
cropped with
different
resampling
rates
8 Feng
et al. [84]
SVM 100% Pros: shows
better
performance for
downsampling
BOSS
database
9 Hou et al.
[85]
Local linear
transform
96.15–98.75% Pros: good
resampling
detection
performance
Dataset of
1000 colored
bmp images
cropped into
512 ×512
pixels
10 Birajdar
and
Mankar
[81]
Autocovariance
sequence,
DFT
99.5% Pros: work well
with various
forms of attacks
like JPEG
compression
and arbitrary
cropping
UCID dataset
USC-SIPI
dataset
11 David
and
Fernando
[82]
Asymptotic
eigenvalue
distribution
and Random
Matrix Theory
(RMT)
0.0066 s/image Pros: Low
computational
complexity
Dresden
Image
Database of a
total of 1317
raw images
[86]
12 Qiao
et al. [87]
Probability of
residual noise
and LRT
detector
0.0996 s/image Pros: effective
with uncom-
pressed/compressed
resampled
images
500
uncompressed
non-
resampled
images and
500
compressed
resampled
JPEG images
with Quality
Factor (QF)
from 50 to 90
(continued)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 185
Table 4 (continued)
S. n. Algorithm Feature
description
Detection accu-
racy/performance
Pros/cons Dataset
13 Su et al.
[88]
Inverse
filtering
process with
blind
deconvolution
90% Pros: does not
affect with
JPEG block
artifacts
Cons: Not
effective to
detect blurred
images
UCID
14 Peng
et al. [89]
AR
coefficients
and
normalized
histograms
98.3% Cons:
performance
degrades with
increasing JPEG
compression
ratio
BOSS dataset
15 Bayar
and
Stamm
[90]
Convolutional
Neural
Network
(CNN)
91.22% Pros: can detect
resampling in
recompressed
images
Dataset of
6500 images
of size at least
2688 ×1520
3.4 Image Retouching Detection
Retouching can be defined as “polishing of an image”. In general, retouching refers
to subsequently improving the surface of an image. Contrast enhancement is a widely
used technique to remove obvious visual clues from the forged image as a postpro-
cessing operation. However, more involvement of retouching can be seen in enter-
tainment media, magazine covers, etc., where retouching is not used maliciously.
Contrast enhancement operations are tantamount to pixel value mappings, which
introduce some statistical traces [91]. Therefore, retouching can be exploited as a
tool for image forgery detection.
Stamm et al. proposed a method [92] for detecting contrast enhancement in an
image on the basis of gray value histogram. They have developed a model for the
histogram of an unaltered image and then exploited this model to detect manipulated
artifacts. Detection accuracy of the algorithm was claimed to be about 99%. Cao
et al. [93] developed a technique to detect sharpening alteration in digital images.
Authors have measured gradient aberration of the gray histogram generated from
unsaturated luminance regions of an image and exploited to unveil traces of sharp-
ening manipulation. Cao et al. [94] proposed a new method to detect unsharp mask-
ing sharpening based on the feature overshoot artifacts occurred around side-planar
edges. By experimental study, authors claimed, their method to be accurate to detect
sharpening on small size images even when post-JPEG compression and noising
attacks employed. Same authors further explained a method [95] for detecting the
contrast enhancement in digital images. This time, they have utilized the histogram
dr.vipin.tyagi@gmail.com

186 K. B. Meena and V. Tyagi
peak/gap artifacts feature to detect global contrast enhancement applied to the pre-
viously JPEG-compressed images. Their proposed method is effective for detecting
forgery when contrast enhancement is employed as the last step of manipulation.
However, method fails to detect forgery when image is highly compressed.
In [91], Lin et al. explained that the contrast enhancement can disturb the inter-
channel similarities of high-frequency components, and then proposed a new method
to detect the cut past forgery by detecting symptoms of contrast enhancement. Unfor-
tunately, this method also fails when image is compressed after forgery. Ding et al.
[96], proposed a new method to detect the special characteristic of the texture mod-
ification caused by the USM sharpening by employing edge perpendicular binary
coding.
Recently, Zhu et al. presented a new approach [97] to detect image sharpening
operation based on the overshoot artifact metric. First, they have detected edges using
canny operator, then, non-subsampled contourlet transform (NSCT) is employed to
classify image edge points. In the final stage, they have measured the overshoot
artifact for each edge points and then, on the basis of overshoot artifacts judgment
were made on sharpening operation. Table 5shows various algorithms for retouching
detection, based on several components such as; feature extracted, classifier applied,
detection accuracy, and dataset used for testing the algorithm.
4 Datasets Available
Table 6shows several publicly available datasets, which are frequently used by
researchers.
5 Conclusion and Future Directions
In this paper, various existing methods on blind image forgery detection are reviewed.
A broad classification of image forgery detection techniques is given. More specifi-
cally, a comprehensive overview of four main types of forgery detection techniques
such as image splicing, copy-move, resampling, and retouching detection is given.
Various existing methods have been reviewed in each category and observed that
existing techniques suffer from one or more following limitations. (1) Detection
accuracy (2) High computation complexity (3) Vulnerable against various attacks
such as rotation, scaling, JPEG compression, blurring, and brightness adjustment,
etc. (4) A lot of false matches with regular background.
Apart from abovementioned limitations, one major issue of these detection tech-
niques is the limited scope of utilization, for example, method developed for copy-
move forgery cannot work with image splicing or resampling and vice versa. In
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 187
Table 5 Comparative study of existing techniques of retouching detection
S. n. Algorithm Extracted
feature
Classifier Detection
accuracy
Dataset used
1 Stamm and
Ray [92]
Gray value
histogram
Thresholding
classifier
Global
contrast 99%,
Local contrast
98.5%,
Histogram
equalization
99%
341 images
captured using
different
digital cameras
2 Cao et al. [93] Ringing
artifacts
Fisher linear
classifier
Precision 0.85 Dataset of 403
JPEG images
3 Cao et al. [94] Overshoot
strength
Thresholding
classifier
88% Dataset of 400
JPEG images
with the size
from 1200 ×
900 to 2832 ×
2128 pixels
4 Lin et al. [91] Interchannel
correlation
Thresholding
classifier
90% Dataset of 100
uncompressed
color images
of size 1600 ×
1200
5 Cao et al. [95] Histogram
peak/gap
artifacts
Thresholding
classifier
100% BOSS public
dataset and
UCID
6 Ding et al. [96] Rotation-
invariant
LBP
SVM 90% UCID
7 Zhu et al. [97] Multiresolution
overshoot
artifact
NSCT 92% UCID
spite of burgeoning research in the field of image forgery detection, no detection
method can be used as a solution for detecting all kind of forgeries. Hence, there
is a great need to develop a robust, sophisticated forgery detection technique which
could eliminate aforementioned limitations. Furthermore, researchers may extend
these techniques to detect forgeries in videos.
dr.vipin.tyagi@gmail.com

188 K. B. Meena and V. Tyagi
Table 6 Description of various available datasets related to forgery detection
S. n Dataset Forgery type Total
images
Resolution Description
1 CISDE [33] Splicing 1845 128 ×128 Contains 933
forged images
and 912
authentic
images, all are
gray images in
PNG format
2 CUISDE [98] Splicing 361 757 ×568,
1152 ×768
Contains 180
forged images
and 181
authentic
images, all are
colored images
in TIFF format
3 CASIA v1.0 [99] Splicing 1725 324 ×256 Contains 925
forged images
and 800
authentic
images, all are
colored images
in JPEG format
4 CASIA v2.0
[100]
Splicing 12614 240 ×
160–900 ×
600
Contains 5123
forged images
and 7491
authentic
images, all are
colored images
in JPEG format
Also contains
uncompressed
images and
JPEG images
with different Q
factors
5 CMFDA [12] Copy-move 48 420 ×
300–3888 ×
2592
Contains original
and forged image
applied with
JPEG
compression,
rotation and
scaling operation
(continued)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 189
Table 6 (continued)
S. n Dataset Forgery type Total
images
Resolution Description
6 CoMoFoD
dataset [101]
Copy-move 260 512 ×
512–3000 ×
2000
Contains original
and forged
images, applied
with translation,
rotation, scale,
distortion or a
combination of
them
7 MICC-F600
[102]
Copy-move 600 800 ×
533–3888 ×
2592
Contains original
and forged
images, that are
randomly taken
from
MICC-F2000
and SATS-130
datasets
8 MICC-F2000
[103]
Copy-move 2000 2018 ×
1536
Contains original
and forged
image, applied
with translation,
rotation, scale
9 SBU-CM161
[104]
Copy-move 240 800 ×580 Contains images
based on 16
original JPEG
images with
rotation, scaling,
compression
10 CPH [53] Copy-move 216 845 ×
634–296 ×
972
Contains images
with forgeries
created through
mixed operations
such as resizing,
rotation, scaling,
compression,
illumination
matching
11 SCUT-FBP [105] Retouching 500 384 ×512 Contains 500
different female
face images
along with the
attractiveness
rating scores
computed from
individual scores
from 70
observers
(continued)
dr.vipin.tyagi@gmail.com

190 K. B. Meena and V. Tyagi
Table 6 (continued)
S. n Dataset Forgery type Total
images
Resolution Description
12 BOSS public
dataset [106]
Retouching 800 2000 ×
3008–5212
×3468
Contains
unaltered
photograph
images in raw
format
13 UCID [107] Retouching 1338 384 ×512 Contains
uncompressed
images in TIFF
format on
various topics
such as natural
scenes,
man-made
objects, indoors
and outdoors
References
1. Revolvy.com: Hippolyte Bayard (French, 1801–1887). https://www.revolvy.com/topic/
Hippolyte Bayard&item_type =topic
2. Loc.gov: Civil War Glass Negatives and Related Prints. https://www.loc.gov/pictures/
collection/cwp/mystery.html (2008)
3. Photo Tampering Throught History. http://pth.izitru.com/
4. Tait, A.: How a badly faked photo of Vladimir Putin took over Twitter. http://www.
newstatesman.com/science-tech/social-media/2017/07/how-badly-faked-photo-vladimir-
putin-took-over-twitter (2017)
5. Tyagi,V.: Understanding Digital Image Processing. CRC Press (2018). ISBN 9781315123905
6. Wang, S., Zheng, D., Zhao, J., Tam, W.J., Speranza, F.: An image quality evaluation method
based on digital watermarking. IEEE Trans. Circuits Syst. Video Technol. 17, 98–105 (2007)
7. Singh, P., Chadha, R.S.: A survey of digital watermarking techniques, applications and attacks.
IEEE Int. Conf. Ind. Inform. 2, 165–175 (2013)
8. Arnold, M., Schmucker, M., Wolthusen, S.D.: Techniques and Applications of Digital Water-
marking and Content Protection. A Cataloging in Publication Record, Artech House Inc,
Norwood, MA, USA (2003)
9. Lu, C., Liao, H.M., Member, S.: Structural digital signature for image authentication: an
incidental distortion resistant scheme. IEEE Trans. Multimed. 5, 161–173 (2003)
10. Schneider, M., Chang, S.: A robust content based digital signature for image authentication.
In: IEEE International Conference on Image Processing. pp. 227–230 (1996)
11. Cox, I.J., Miller, M.L., Bloom, J.A., Kalker, T.: Digital Watermarking and Steganography
Second Edition
12. Christlein, V., Riess, C.C., Jordan, J., Riess, C.C., Angelopoulou, E.: An evaluation of popular
copy-move forgery detection approaches. IEEE Trans. Inf. Forensics Secur. 7, 1841–1854
(2012)
13. Hsu, Y., Chang, S.: Camera response functions for image forensics: an automatic algorithm
for splicing detection. IEEE Trans. Inf. Forensics Secur. 5, 816–825 (2010)
14. Carvalho, T.J.De, Member, S., Riess, C., Member, A., Angelopoulou, E., Pedrini, H., Rocha,
A.D.R.: Exposing digital image forgeries by illumination color classification. IEEE Trans.
Inf. Forensics Secur. 8, 1182–1194 (2013)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 191
15. Popescu, A.C., Farid, H.: Exposing digital forgeries by detecting traces of resampling. IEEE
Trans. Inf. Forensics Secur. 53, 758–767 (2005)
16. Lanh, T.V.L.T., Van Chong, K.-S., Chong, K.-S., Emmanuel, S., Kankanhalli, M.S.: A sur-
vey on digital camera image forensic methods. In: 2007 IEEE International Conference on
Multimedia and Expo, pp. 16–19 (2007)
17. Farid, H.: A survey of image forgery detection techniques. IEEE Signal Process. Mag. 26,
16–25 (2009)
18. Warif, N.B.A., Wahab, A.W.A., Idris, M.Y.I.: Copy-move forgery detection: survey, chal-
lenges and future directions. J. Netw. Comput. Appl. (2016)
19. Mahdian, B., Saic, S.: A bibliography on blind methods for identifying image forgery. Signal
Process. Image Commun. 25, 389–399 (2010)
20. Birajdar, G.K., Mankar, V.H.: Digital image forgery detection using passive techniques: a
survey. Digit. Investig. 10, 226–245 (2013)
21. Qazi, T., Hayat, K., Khan, S.U., Madani, S.A., Khan, I.A., Kołodziej, J., Li, H., Lin, W., Yow,
K.C., Xu, C.-Z.: Survey on blind image forgery detection. Image Process. IET. 7, 660–670
(2013)
22. Ansari, M.D., Ghrera, S.P., Tyagi, V.: Pixel-based image forgery detection: a review. IETE J.
Educ. 55, 40–46 (2014)
23. Ali, M., Deriche, M.: A bibliography of pixel-based blind image forgery detection techniques.
Signal Process. Image Commun. 39, 46–74 (2015)
24. Lukas, J., Fridrich, J., Goljan, M.: Detecting digital image forgeries using sensor pattern noise.
In: Proceedings of SPIE, vol. 6072, pp. 60720Y–60720Y–11 (2006)
25. Yatziv, L., Sapiro, G.: Fast image and video colorization using chrominance blending. IEEE
Trans. Image Process. 1120–1129 (2006)
26. Chuan, Y.Y., Curless, B., Salesin, D.H., Szeliski, R.: A bayesian approach to digital matting.
Comput. Vis. Pattern Recognit. (2001)
27. Farid, H.: Detecting digital forgeries using bispectral analysis. Mit Ai Memo Aim-1657 Mit
(1999)
28. Ng, T., Chang, S., Sun, Q.: Blind detection of photomontage using higher order statistics. In:
IEEE International Symposium on Circuits System, pp. 7–10 (2004)
29. Ng, T., Chan, S.F.: A model of Image Splicing. In: IEEE International Conference on Image
Process (2004)
30. Johnson, M.K., Farid, H.: Exposing digital forgeries by detecting inconsistencies in lighting.
In: Proceedings of 7th Workshop on Multimed Security—MM&Sec’05, pp. 1–10 (2005)
31. Fu, D., Shi, Y.Q., Su, W.: Detection of image splicing based on Hilbert-Huang transform and
moments of characteristic functions. Int. Work. Digit. Watermarking 177–187 (2006)
32. Li, X., Jing, T., Li, X.: Image splicing detection based on moment features and Hilbert-Huang
transform. IEEE Int. Conf. Inf. Theory Inf. Secur. (2010)
33. Columbia DVMM Research Lab,Image Splicing Detection Evaluation Dataset. www.ee.
columbia.edu/dvmm/researchProjects/AuthenticationWatermarking/Blind (2004)
34. Shi, Y.Q., Chen, C., Chen, W.: A natural image model approach to splicing detection. In:
Proceedings of 9th Workshop Multimedia Security, pp. 51–62 (2007)
35. Dong, J., Wang, W., Tan, T., Shi, Y.Q.: Run-length and edge statistics based approach for
image splicing detection. IWDW Int. Work. Digit. Watermarking. 5450 LNCS, 76–87 (2009)
36. Wang, W., Dong, J., Tan, T.: Effective image splicing detection based on image chroma. In:
IEEE International Conference on Image Process, pp. 1257–1260 (2009)
37. Kakar, P., Member, S., Sudha, N., Member, S., Ser, W., Member, S.: Exposing digital image
forgeries by detecting discrepancies in motion blur. IEEE Trans. Multimed. 13, 443–452
(2011)
38. Rao, M.P., Rajagopalan, A.N., Member, S.: Harnessing motion blur to unveil splicing. IEEE
Trans. Inf. Forensics Secur. 9, 583–595 (2014)
39. El-Alfy,E.S., Qureshi, M.A.: Combining spatial and DCT based Markov features for enhanced
blind detection of image splicing. Pattern Anal. Appl. 18, 713–723 (2015)
dr.vipin.tyagi@gmail.com

192 K. B. Meena and V. Tyagi
40. Bahrami, K., Member, S., Kot, A.C., Li, L., Li, H., Member, S.: Blurred image splicing
localization by exposing blur type inconsistency. IEEE Trans. Inf. Forensics Secur. 6013,
1–10 (2015)
41. Zhao, X., Wang, S., Li, S., Li, J.: Passive image-splicing detection by a 2-D noncausal Markov
model. IEEE Trans. Circuits Syst. Video Technol. 25, 185–199 (2015)
42. Pun, C.M., Liu, B., Yuan, X.C.: Multi-scale noise estimation for image splicing forgery
detection. J. Vis. Commun. Image Represent. 38, 195–206 (2016)
43. Park, T.H., Han, J.G., Moon, Y.H., Eom, I.K.: Image splicing detection based on inter-scale 2D
joint characteristic function moments in wavelet domain. EURASIP J. Image Video Process
30 (2016)
44. Zhang, Q., Lu, W.: Joint image splicing detection in DCT and contourlet transform domain.
J. Vis. Commun. Image Represent. (2016)
45. Shen, X., Shi, Z., Chen, H.: Splicing image forgery detection using textural features based on
the grey level co-occurrence matrices. IET Image Process. 11, 44–53 (2017)
46. Farid, H.: How to Detect Faked Photos (2017)
47. Chen, W., Shi, Y.Q., Su, W.: Image splicing detection using 2-D phase congruency and sta-
tistical moments of characteristic function. In: Security Steganography and Watermarking
Multimedia Contents IX, vol. 6505, pp. 1–8 (2007)
48. He, Z., Sun, W., Lu, W., Lu, H.: Digital image splicing detection based on approximate run
length. Pattern Recognit. Lett. 32, 1591–1597 (2011)
49. He, Z., Lu, W., Sun, W., Huang, J.: Digital image splicing detectionbased on Markov features
in DCT and DWT domain. Pattern Recognit. 45, 4292–4299 (2012)
50. Xu, B., Liu, G., Dai, Y.: Detecting image splicing using merged features in chroma space.
Sci. World J. (2014)
51. Han, J.G., Park, T.H., Moon, W.H., Eom, I.K.: Efficient Markov feature extraction method for
image splicing detection using maximization and threshold expansion. J. Electron. Imaging.
(2016)
52. Rao, Y., Ni, J.: A deep learning approach to detection of splicing and copy-move forgeries in
images. In: 8th IEEE International Workshop Information Forensics Security WIFS (2016)
53. Silva, E., Carvalho, T., Ferreira, A., Rocha, A.: Going deeper into copy-move forgery detec-
tion: Exploring image telltales via multi-scale analysis and voting processes. J. Vis. Commun.
Image Represent. 29, 16–32 (2015)
54. Lee, J.C., Chang, C.P., Chen, W.K.: Detection of copy-move image forgery using histogram
of orientated gradients. Inf. Sci. (Ny) 321, 250–262 (2015)
55. Ardizzone, E., Bruno, A., Mazzola, G.: Copy-move forgery detection by matching triangles
of keypoints. IEEE Trans. Inf. Forensics Secur. 10, 2084–2094 (2015)
56. Cozzolino, D., Poggi, G., Verdoliva, L.: Efficient dense-field copy-move forgery detection.
IEEE Trans. Inf. Forensics Secur. 10, 2284–2297 (2015)
57. Li, J., Li, X., Yang, B., Sun, X.: Segmentation-based image copy-move forgery detection
scheme. IEEE Trans. Inf. (2015)
58. Pun, C., Member, S., Yuan, X., Bi, X.: Oversegmentation and feature point matching. IEEE
Trans. Inf. Forensics Secur. 10, 1705–1716 (2015)
59. Gürbüz, E., Uluta¸s, G., Uluta¸s, M.: Rotation invariant copy move forgery detection method.
In: Proceedings of 9th International Conference on Electrical and Electronics Engineering,
pp. 202–206 (2015)
60. Zhao, F., Zhang, R., Guo, H., Zhang, Y.: Effective digital image copy-move location algorithm
robust to geometric transformations. In: IEEE International Conference on Signal Processing,
Communications and Computing (ICSPCC) (2015)
61. Wenchang, S.H.I., Fei, Z., Bo, Q.I.N., Bin, L.: Improving image copy-move forgery detection
with particle swarm optimization techniques. China Commun. 139–149 (2016)
62. Zandi, M., Mahmoudi-Aznaveh, A., Talebpour, A.: Iterative copy-move forgery detection
based on a new interest point detector. IEEE Trans. Inf. Forensics Secur. 11, 2499–2512
(2016)
dr.vipin.tyagi@gmail.com

Image Forgery Detection: Survey and Future Directions 193
63. Ferreira, A., Felipussi, S.C., Alfaro, C., Fonseca, P., Vargas-Munoz, J.E., Dos Santos, J.A.,
Rocha, A.: Behavior knowledge space-based fusion for copy-move forgery detection. IEEE
Trans. Image Process. 25, 4729–4742 (2016)
64. Zhu, Y., Shen, X., Chen, H.: Copy-move forgery detection based on scaled ORB. Multimed.
Tools Appl. 75, 3221–3233 (2016)
65. Bi, X., Pun, C.M., Yuan, X.C.: Multi-level dense descriptor and hierarchical feature matching
for copy-move forgery detection. Inf. Sci. (Ny) 345, 226–242 (2016)
66. Wang,X., Li, S., Liu, Y.: A new keypoint-based copy-move forgery detection for small smooth
regions. Multimed. Tools Appl. (2016)
67. Tralic, D., Grgic, S., Sun, X., Rosin, P.L.: Combining cellular automata and local binary
patterns for copy-move forgery detection. Multimed. Tools Appl. 16881–16903 (2016)
68. Yang, B., Sun, X., Guo, H., Xia, Z., Chen, X.: A copy-move forgery detection method based
on CMFD-SIFT. Multimed. Tools Appl. (2017)
69. Lee, J.C.: Copy-move image forgery detection based on Gabor magnitude. J. Vis. Commun.
Image Represent. 31, 320–334 (2015)
70. Bi, X.L., Pun, C.M., Yuan, X.C.: Multi-scale feature extraction and adaptive matching for
copy-move forgery detection. Multimed. Tools Appl. 1–23 (2016)
71. Ustubioglu, B., Ulutas, G., Ulutas, M., Nabiyev, V.V.: A new copy move forgery detec-
tion technique with automatic threshold determination. AEU Int. J. Electron. Commun. 70,
1076–1087 (2016)
72. Zheng, J., Liu, Y., Ren, J., Zhu, T., Yan, Y., Yang, H.: Fusion of block and keypoints based
approaches for effective copy-move image forgery detection. Multidimens. Syst. Signal Pro-
cess. 27, 989–1005 (2016)
73. Huang, D., Huang , C., Hu, W.: Robustness of copy-move forgery detection under high JPEG
compression artifacts. Multimed. Tools Appl. 76(1), 1509–1530 (2017)
74. Popescu, A.C., Farid, H.: Exposing Digital Forgeries by Detecting Traces of Resampling
Resampling Detecting Resampling Experiment Results (2005)
75. Dempster, A., Laird, N., Rubin, D.: Maximum lilelihood from in- complete data via the EM
algorithm. J. Roy. Stat. Soc. 99, 1–38 (1977)
76. Kirchner, M.: Fast and reliable resampling detection by spectral analysis of fixed linear pre-
dictor residue. In: Proceedings of 10th ACM Workshop Multimedia Security—MM&Sec’11
(2008)
77. Mahdian, B., Saic, S.: Blind authentication using periodic properties of interpolation. IEEE
Trans. Inf. Forensics Secur. 3, 529–538 (2008)
78. Li, S.P., Han, Z., Chen, Y.Z., Fu, B., Lu, C., Yao, X.: Resampling forgery detection in JPEG-
compressed images. In: Proceedings of 2010 3rd International Congress on Image Signal
Process CISP 2010, vol. 3, pp. 1166–1170 (2010)
79. Lien, C.-C., Shih, C.-L., Chou, C.-H.: Fast Forgery detection with the intrinsic resampling
properties. J. Inf. Secur. 1, 11–22 (2010)
80. Qian, R., Li, W., Yu, N., Hao, Z.: Image forensics with rotation-tolerant resampling detection.
In: IEEE International Conference on Multimedia Expo WorkshopsICMEW, pp. 61–66 (2012)
81. Birajdar, G.K., Mankar, V.H.: Blind method for rescaling detection and rescale factor esti-
mation in digital images using periodic properties of interpolation. AEU Int. J. Electron.
Commun. 68, 644–652 (2014)
82. David, V., Fernando, P.: A Random Matrix Approach to the Forensic Analysis of Upscaled
Images. IEEE Trans. Inf. Forensics, XX (2017)
83. Wang, R., Ping, X.J.: Detection of resampling based on singular value decomposition. In:
Proceedings of Fifth International Conference on Image Graph, pp. 879–884 (2009)
84. Feng, X., Cox, I.J., Doërr, G.: Normalized energy density-based forensic detection of resam-
pled images. IEEE Trans. Multimed. 14, 536–545 (2012)
85. Hou, X.D., Zhang, T., Xiong, G., Zhang, Y., Ping, X.: Image resampling detection based on
texture classification. Multimed. Tools Appl. 72, 1681–1708 (2013)
86. Gloe, T., Ohme, R.: The dresden image database for benchmarking digital image forensics.
In: ACM Symposium on Applied Computing, pp. 1584–1590
dr.vipin.tyagi@gmail.com

194 K. B. Meena and V. Tyagi
87. Qiao, T., Zhu, A., Retraint, F.: Exposing image resampling forgery by using linear parametric
model. Multimed. Tools Appl. (2017)
88. Su, Y., Jin, X., Zhang, C., Chen, Y..: Hierarchical image resampling detection based on blind
deconvolution q. J. Vis. Commun. Image Represent. 1–11 (2017)
89. Peng, A., Wu, Y., Kang, X.: Revealing traces of image resampling and resampling antiforen-
sics. Adv. Multimed. (2017)
90. Bayar, B., Stamm, M.C.: On the robustness of constrained convolutional neural networks to
JPEG post-compression for image resampling detection. In: IEEE International Conference
on Acoustics Speech Signal Process, pp. 2152–2156 (2017)
91. Lin, X., Li, C., Hu, Y.: Exposing image forgery through the detection of contrast enhancement.
In: International Conference on Image Process, pp. 4467–4471 (2013)
92. Stamm, M., Ray, K.J.: Blind forensics of contrast enhancement in digital images. In: Pro-
ceedings of International Conference on Image Process ICIP, pp. 3112–3115 (2008)
93. Cao, G., Zhao, Y., Ni, R.: Detection of image sharpening based on histogram aberration and
ringing Artifacts. In: IEEE International Conference on Multimedia and Expo, pp. 1026–1029
(2009)
94. Cao, G., Zhao, Y., Ni, R., Kot, A.C.: Unsharp masking sharpening detection via overshoot
artifacts analysis. IEEE Signal Process. Lett. 18, 603–606 (2011)
95. Cao, G., Zhao, Y., Ni, R., Li, X.: Contrast enhancement-based forensics in digital images.
IEEE Trans. Inf. Forensics Secur. 9, 515–525 (2014)
96. Ding, F., Zhu, G., Yang, J., Xie, J., Shi, Y.Q.: Edge perpendicular binary coding for USM
sharpening detection. IEEE Signal Process. Lett. 22, 327–331 (2015)
97. Zhu, N., Deng, C., Gao, X.: Image sharpening detection based on multiresolution overshoot
artifact analysis. Multimed. Tools Appl. (2016)
98. Hsu, Y.F., Chang, S.F.: Detecting image splicing using geometry invariants and camera char-
acteristics consistency. In: International Conference on Multimedia and Expo, pp. 549–552
(2006)
99. Dong, J., Wang, W.: CASIA tampered image detection evaluation database
100. Dong, J., Wang, W.: CASIA2 tampered image detection evaluation (TIDE) database
101. Tralic, D., Zupancic, I., Grgic, S., Grgic, M.: CoMoFoD—New database for copy-move
forgery detection
102. Amerini, I., Ballan, L., Caldelli, R., Bimbo, A. Del, Serra, G.: A SIFT-based forensic method
for copy—move attack detection and transformation recovery. IEEE Trans. Inf. Forensics
Secur. 1099–1110 (2011)
103. Amerini, I., Ballan, L., Caldelli, R., Del Bimbo, A., Del Tongo, L., Serra, G.: Copy-move
forgery detection and localization by means of robust clustering with J-Linkage. Signal Pro-
cess. Image Commun. 659–669 (2013)
104. Zandi, M., Mahmoudi-Aznaveh, A., Mansouri, A.: Adaptive matching for copy-move
forgery detection. In: IEEE International Workshop on Information Forensics and Security,
pp. 119–124 (2014)
105. Xie, D., Liang, L., Jin, L., Xu, J., Li, M.: A benchmark dataset for facial beauty perception.
http://www.hcii-lab.net/data/SCUT-FBP
106. Bas, P., Filler, T., Pevný, T.: Break our steganographic system: the ins and outs of organizing
BOSS. In: Proceedings of Information Hiding, Prague, Czech Repub, pp. 59–70 (2011)
107. Stich, M., Schaefer, G.: UCID—an uncompressed colour image database. In:Proceedings of
SPIE, Storage Retrieval Methods and Application Multimedia, pp. 472–480 (2004)
dr.vipin.tyagi@gmail.com