Content uploaded by Sajjad Dadkhah
Author content
All content in this area was uploaded by Sajjad Dadkhah on Jul 08, 2015
Content may be subject to copyright.
Efficient Copy-Move Forgery Detection for
Digital Images
Somayeh Sadeghi, Hamid A. Jalab, and Sajjad Dadkhah
Abstract—Due to availability of powerful image processing soft-
ware and improvement of human computer knowledge, it becomes
easy to tamper images. Manipulation of digital images in different
fields like court of law and medical imaging create a serious problem
nowadays. Copy-move forgery is one of the most common types
of forgery which copies some part of the image and pastes it to
another part of the same image to cover an important scene. In
this paper, a copy-move forgery detection method proposed based
on Fourier transform to detect forgeries. Firstly, image is divided to
same size blocks and Fourier transform is performed on each block.
Similarity in the Fourier transform between different blocks provides
an indication of the copy-move operation. The experimental results
prove that the proposed method works on reasonable time and works
well for gray scale and colour images. Computational complexity
reduced by using Fourier transform in this method.
Keywords—Copy-Move forgery, Digital Forensics, Image Forgery.
I. INTRODUCTION
IMAGE as a communication media became very popular
immediately after invention of photography and plays crit-
ical role in real life, but from time to time image does not
tell the truth. With the entrance of digital data in current years
and improvement of human computer knowledge, expansion
of digital images increased and validity of digital data faces a
big problem [1]. Availability of the digital image processing
tools such as Photoshop or GIMP which are available free
makes it easy to change features of images which are flexible
to manipulation; these powerfultools caused suspicions on the
integrity of digital images that we face every day in our life
[2]. Consequently, digital image forgery invented to find out
the integrity of the image, and it became an important issue as
people tried to change the content of the image and present the
forged image as original one to achieve their illegal purposes.
Digital image forgery is important because of the usage of
digital images in many social areas like courts when they are
used as evidence, or in medical field to help physician makes
decisions base on digital images. Digital Image Forensics can
be subdivided into three branches as image source identifica-
tion; Computer generated image recognition and Image forgery
detection, and base on latest technology, digital image forgery
categorized in three groups; Copy-Move, Image splicing and
S. Sadeghi is with the Faculty of Computer Science and Information
Technology, University of Malaya, 50603 Kuala Lumpur, Malaysia. (email:
ssomayeh@siswa.um.edu.my)
H.A Jalab is with the Faculty of Computer Science and Information
Technology, University of Malaya, 50603 Kuala Lumpur, Malaysia. (email:
hamidjalabb@um.edu.my)
S.Dadkhah is with the Faculty of Computer Science and Information
System, University Technology Malaysia 54100 Kuala Lumpur, Malaysia
(email: dsajjad2@live.utm.my)
Image retouching. Copy-Move forgery or Region-Duplication
forgery is the most important type of forgery, in Copy-Move
some part of the image copies and pastes into another part of
the same image to create a new thing or to hide an important
scene [2]. Image splicing is the procedure of creating a
fake image by cutting one part of an image and paste it to
another image. It works on combining few images to create
one tampered image. One of the problems is that, when the
backgrounds in the images are different the objects in result
may appear unclear [3]. Image Retouching doesnt obviously
change the image, so it can be considered as the less corrupting
type of digital image forgery, it just enhance some features of
image. It is famous among magazine photo editors and most
of magazine covers use this technique to change some features
of an image but it is ethically wrong [3].
With the creation of digital image forgery,many researchers
developed different techniques to detect forgery, detection of
Copy-Move forgery is difficult compare to other forgeries
because the source and destination of forgery is same image,
also the original image segment and the pasted one have same
important properties such as dynamic range, noise component
and colour palette. An example for this type of forgery can
be seen in Fig.1, where (a) shows the original image and (b)
shows the tampered image [4].
Fig. 1: Sample of Image Forgery
In this paper,we proposea novelmethod for identifyingthe
location of copy-move tampering and authenticating an image
by applying Fourier transform. The image is first converted
to gray scale image and reduced in size based on the resize
criterion value. Fourier transform applied on image to perform
correlation, and correlation can be used to locate features
within an image, and finally it helps to find similar correlations
World Academy of Science, Engineering and Technology 71 2012
755
values in image and convert the result value to a matrix of real
numbers and shows location of matched blocks.
The rest of paper is organized as follows. In next section,
the related work is reviewed. In section 3, the details of the
proposed method are presented with the general flowchart of
the program. Section 4 presents the experimental results and
discussion. Final conclusions are drawn in last section.
II. RELATED WORK
Copy-move forgery is a very general and important type
of forgeries because the source and destination of the forged
image is same. In this type of forgery, some part of the image
copies and pasts on other part of the same image to cover
and important scene for illegal purposes [5]. Consequently
detection of copy-move forgery invented to search the copied
regions and their pasted ones, but detection may vary base
on whether there has been any post-processing on copied
part before paste it to another part. Usually attackers will do
some operations such as rotation, filtering, JPEG compression,
resizing and noise addition to the original part before pasting,
and these operations make it difficult to detect copy-move
forgery, therefore forgery detector should be robust to all
manipulations [2].
To accomplish this task severalcopy-move forgery detection
techniques have been proposed. Fridrich, Soukal, and Jan Luk
in [5] described Copy-Move forgery base on similarity. They
believed there is a relationship between the original image
parts and pasted one, and this relationship can be used for a
successful detection of copy-move. Since the tampered image
will possibly be saved in the JPEG format, the image parts
might match approximately not accurately. As a result, they
found there are some requirements for detection algorithm,
which are: 1. Algorithm must permit for an approximate match
of small image parts. 2. When there is possibility of false
positive, it should work in a reasonable time. 3. Another thing
is that the forged parts should be a connected component rather
than a collection of individual pixels.
Accordingly, they have developed a technique for detection
based on exact match, and it works to find segments in
the image which match exactly. This technique is useful
for forensic analysis, but its applicability is limited [5]. In
2003, Fridrich analyzed the exhaustive search and proposed
a block matching detecting method based on discrete cosine
transform (DCT) and lexicographic sort used for detecting the
forged areas [5]. Mahdian and Stanislav in [6] proposed a
method based on blur moment invariant which is robust to all
manipulations but the main disadvantage of this method was
detection time which takes 45 minutes for a 256x256 pixel
image to detect which area has been duplicated.
An efficient non-intrusive method is proposed in [7], in this
method image is divided into sub blocks and a separate noise
image is created by using noise pattern of each sub blocks,
these noise images are used to approximate the overall noise
of the image which is useful later to guess the noise pattern of
different blocks. Finally blocks with similar noise histogram
are suspected to duplicated area. This method can segment
an image into complete objects more accurately compare to
previous methods but it cant work on different images as
detection can happen only if the background of the image is
simple. Detection based on DCT was proposed by Jie, Huax-
iong, Gao and Hai (2011), in this method Fridrichs method
based on DCT has improved by reducing false matching rate.
This method works by comparing image block features and
find out if number of matched blocks in certain region is
more than threshold. In order to improve the accuracy of
matching a lexicographicalsorting algorithm based on distance
proposed. It is robust to post image processing like adding
noise and blurring, but it is not robust to rotation [8]. Reference
[7] shows a new solution proposed based on dyadic wavelet
transform usage. It is robustto post processingbut this method
also has its drawbacks, it works only on images with simple
background.
According to XiaoBing KANG and ShengMin WEI [9] ,
in proposes detection based on Singular value decomposition
(SVD) can be done easily even when tamperer does some
manipulations such as additional noise, scaling or rotation to
image part before pasting to another part, and it works well for
lossy format such as JPEG, detection based on SVD happens
by dividing image into overlapping blocks and apply SVD on
each block base on SVD formula in (1) where A is image
matrix and U is a (m x m) orthogonal matrix, V is an (n x n)
orthogonal matrix and S is an (m x n) diagonal matrix with
singular values on the diagonal.
A=USVT(1)
From SVD, singular values will extract and arranged in a
matrix, then it needs to change features in each block into k-d
tree and search for similar blocks for each query using (2)
where u and v are values of orthogonal matrixes.
D(UN)=(
n
i=1
(ui−vi)2)1
2(2)
At this time, blocks similarity matching will be done to find
similar blocks. The main idea of this step is that a duplicated
region consists of many neighbouring duplicated blocks. If
two similar blocks can be fined in the analyzed space and
if their neighbourhoods are also similar to each other, there
is a high probability that they are duplicated and they are
tagged as duplication area, then the output of the method is a
duplicated regions map which showing the image regions that
are expected duplicated. Based on their experimental results,
their proposed method gives robustness against post processing
like blur filtering, Gaussian noise addition, etc. Detection time
for this method for a 256 x 256 colour image is 120 second
which is better compare to existed method in [6].
In this paper, we propose a detection algorithm based on
Fourier transform to extract transformed image matrix, and
find location of similar blocks in the image using inverse
Fourier transform.
III. PROPOSED METHOD
When a forgery occurs in a digital image, it shows that
statistical characteristics of image have changed; therefore it
World Academy of Science, Engineering and Technology 71 2012
756
is obvious that statistical characteristics of forged area are
different from original area. For detecting forged area, the
statistical characteristics of each small sections of the image
calculated and compared with each other. Fig. 2 shows the
general procedures of detecting copy-move forgery in digital
images [10].
Fig. 2: Copy-move forgery detection procedure
Most of detection techniques focused on block matching,
and the procedure is common by dividingimage into same size
blocks and extracting image features by different techniques,
afterwards searching for similar block to find which region
is duplicated base on block matching, this is happen if many
similar blocks in a specific distance can be fined, and these
suspected blocks are connected to each other to identify
tampered area. The proposed method works by resizing image
to specific resize scale and convert the input image to gray
scale image if it is colour to reduce time of detection. We
have defined block size as a square with K x K pixels and
assumed to be smaller than the size of the duplicated regions
which have to be detected. Here we have defined block size
20 by default to divide image based on block size to same size
overlapping blocks, and number of blocks calculated from (M
x K + 1)(N x K + 1) where (M, N) are image pixels and
K is the size of the block we have defined before. Fourier
transform applied on image to extract features of each blocks,
when Fourier transform of the image calculated, a function is
created with the intensity signal across the image, and function
is decomposed into a sum of orthogonal basis functions by
using Fourier transform. f (m, n) is a function of two discrete
spatial variables m and n, and the two-dimensional Fourier
transform of f (m, n) is defined by the relationship in (3)
F(ω1,ω
2)=
∞
m=−∞
∞
n=−∞
f(m, n)e−jω1me−jω2n(3)
The variables ω1andω2 are frequency variables, and F
(ω1, ω2) is frequency-domain representation of f (m, n). F
(ω1,ω2) is a complex-valued function that is periodic both
in ω1andω2, with period 2πand period range −π≤ω1,
ω2≤π. Fourier transform applied on image blocks to
perform correlation, subsequently the correlation of the blocks
computed to locate features within image, then correlation
are sorted in a lexicographically order because it can make
matching more effective, and sorted correlation stored in a
matrix named Q with the size of (M - K + 1) x (N - K + 1) x
K2. After all blocks sorted properly, the algorithm continues
into the matching step by testing each pair of blocks whether
they are matching. For each row in matrix Q, correlations
values are computed for the block matching to current row
with the blocks matching to rows around the current row, if the
computed maximum correlation value exceeds threshold which
is block-matching threshold, then two blocks are duplicated.
When similar blocks detected then the inverse of a transform
is performed on a transformed image to produce the original
image, and the inverse of two-dimensional Fourier transform
of the image is done by (4)
F(m, n)= 1
4π2π
ω1=−ππ
ω2=−π
F(ω1,ω
2)ejω1mejω2ndω1dω2
(4)
Where ω1andω2 are frequency variables, and and F (ω1,
ω2) is frequency-domain representation of f (m, n).
IV. EXPERIMENTAL RESULT
The proposed method has been implemented using Matlab
7.9. Experimental environment is on a personal computer of
2.00GHz processors with 1GB memory. The block size is 20 x
20 pixels. Tests have been performed on various images with
different size of duplication region and different formats. In
the first experiment, several copy-move tampered images have
been examined with proposed algorithm.
Fig.3 demonstrates an ordinary forgery, in Fig.3 (b) tam-
pered image shown in which some part of the leaves copied
and pasted on the trunk to cover it, and result of detection
showninFig.3(c).
Fig. 3: (a) Original Image, (b) Tampered Image, (c) Detection Result
Lena gray scale image tampered and Fig.4(b) illustrates the
tampered image where some part of the hair has been copied
World Academy of Science, Engineering and Technology 71 2012
757
and pasted on the hat to cover some part of the hat, and
Fig.4(c) shows the detection map.
Fig. 4: (a) Original Image, (b) Tampered Image, (c) Detection Result
Fig.5 shows another forgery which is obvious, original
image is tampered by hiding the man in the image and created
tampered image in Fig.5 (b) along with detection result in
Fig.5(c).
Fig. 5: (a) Original Image, (b) Tampered Image, (c) Detection Result
In time performance, the complexity time of the proposed
algorithm is better compared with existing methods [6]. Kang
and Wei explained in their experimental results [6] they have
tested a gray scale image with dimension 256 x 256 and block
size 20, the detection time was 60 seconds, and compare
with our algorithm for the same image detection time is 8
seconds. Another image has been test, based on Kang and
Wei algorithm, the average runtime of the their algorithm for
one colour channel of a 256x256 image when block size B=20,
is approximately 120 seconds, and compare to our algorithm
for the same size of colour image it takes 11 seconds to find
duplicated areas.
V. CONCLUSION
With the rapid progress of image processing technology,
detection of digital image forgery is an interesting research
topic in forensics science. In this paper, a specific type of
forgery which is Copy-move forgery investigated and an effi-
cient detection method proposed based on Fourier transform.
The procedure of detection starts by dividing image into same
size overlapped blocks and apply Fourier transform on each
block, finally demonstrate the location of similar blocks by
using inverse Fourier transform. Proposed method is able to
locate duplicated areas in reasonable time compare to existing
methods, and computational complexity reduced. Our future
work is to enhance our method to detect duplicated area more
accurately and improve it to be able to detect another kind of
forgery which is image splicing.
ACKNOWLEDGMENT
The authors would like to thank University of Malaya for
their educational and financial support.
REFERENCES
[1] J. Fridrich, “Methods for tamper detection in digital images,” Proceeding
of ACM Conference on Multimedia and Security, p. 1923, 1999.
[2] C. L. Jing, “Image copy-move forgery detecting based on local invariant
feature,” Journal of Multimedia, vol. 7, 2012.
[3] Q. S. W. Chen and W. Su, “Image splicing detection using 2-d phase
congruency and statistical moments of characteristic function,” E. J.
Delp and P. W. Wong, editors, Proceedings of SPIE: Security and
Watermarking of Multimedia Content IX, vol. 6505, p. 65050, 2007.
[4] D. T. G. Li, Q. Wu and S.Sun, “A sorted neighborhood approach for
detecting duplicated regions in image forgeries based on dwt and svd,”
Proceedings of IEEE International Conference on Multimedia and Expo,
pp. 1750–1753, 2007.
[5] a. J. L. J. Fridrich, D.Soukal, “Detection of copy-move forgery in digital
images,” Proceeding of Digital Forensic Research Workshop, pp. 55–61,
2003.
[6] S. S. B. Mahdian, “Detection of copymove forgery using a method
based on blur moment invariants,” Proceedings of fifth International
conference on Forensic Science International, p. 180189, 2007.
[7] M. G. M. Najah, H. Muhammad and B. George, “Copy-move forgery
detection using dyadic wavelet transform,” Eighth International Con-
ference Computer Graphics, Imaging and Visualization, pp. 103–108,
2011.
[8] G. Q. H. Jie, Z. Huaxiong and H. Hai, “An improved lexicographical
sort algorithm of copy-move forgery detection,” Second International
Conference on Networking and Distributed Computing, pp. 23–27, 2011.
[9] S. W. X. Kang, “Identifying tampered regions using singular value
decomposition in digital image forensics,” International Conference on
Computer Science and Software Engineering, 2008.
[10] X. P. Z. H. Z. Zhang, Y. Ren and S. Zhang, “A survey on passive-blind
image forgery by doctor method detection,” Proceedings of the Seventh
International Conference on Machine Learning and Cybernetics, 2008.
World Academy of Science, Engineering and Technology 71 2012
758
