Global Journal of Computer Science and Technology Vol. 10 Issue 7 Ver. 1.
0 September 2010 P a g e | 61
Detecting Copy-Move Forgery in Digital Images:
A Survey and Analysis of Current Methods
GJCST Computing Classification
B.L.Shivakumar1 Lt. Dr. S.Santhosh Baboo2 I 4.1
Abstract-As one of the most successful applications of image our email. Digital image forgery detection techniques are
analysis and understanding, digital image forgery detection has classified into active and passive approach [3]. In active
recently received significant attention, especially during the approach, the digital image requires some pre-processing
past few years. At least two trend account for this: the first such as watermark embedding or signature generation at the
accepting digital image as official document has become a
time of creating the image, which would limit their
common practice, and the second the availability of low cost
technology in which the image could be easily manipulated. application in practice. Moreover, there are millions of
Even though there are many systems to detect the digital image digital images in internet without digital signature or
forgery, their success is limited by the conditions imposed by watermark. In such scenario active approach could not be
many applications. For example, detecting duplicated region used to find the authentication of the image. Unlike the
that have been rotated in different angles remains largely watermark-based and signature-based methods; the passive
unsolved problem. In an attempt to assist these efforts, this technology does not need any digital signature generated or
paper surveys the recent development in the field of Copy- watermark embedded in advance [4]. There are three
Move digital image forgery detection. techniques widely used to manipulate digital images [3]. 1)
Keyword-Image forgeries, Digital forensics, Copy-Move Tampering – tampering is manipulation of an image to
forgery detection, block matching achieve a specific result. 2) Splicing (Compositing) - A
I. INTRODUCTION common form of photographic manipulation in which the
digital splicing of two or more images into a single
F rom the early days an image has generally been accepted
as a proof of occurrence of the depicted event.
Computer becoming more prevalent in business and other
composite 3) Cloning (Copy-Move)
II. COPY-MOVE FORGERY
field, accepting digital image as official document has Copy-Move is a specific type of image manipulation, where
become a common practice. The availability of low-cost a part of the image itself is copied and pasted into another
hardware and software tools, makes it easy to create, alter, part of the same image (Fig 1).
and manipulated digital images with no obvious traces of
having been subjected to any of these operations. As result
we are rapidly reaching a situation where one can no longer
take the integrity and authenticity of digital images for
granted. This trend undermines the credibility of digital
images presented as evidence in a court of law, as news
items, as part of a medical records or as financial documents
since it may no longer be possible to distinguish whether a
given digital images is original or a modified version or
even a depiction of a real-life occurrences and objects.
Digital image forgery is a growing problem in criminal (a) (b)
cases and in public course. Currently there are no Fig 1. An example of copy-move forgery [5]: (a) the original
established methodologies to verify the authenticity and image with three missiles (b) The forged image with four
integrity of digital images in an automatic manner. missiles
Detecting forgery in digital images is an emerging research Copy-Move forgery is performed with the intention to make
field with important implications for ensuring the credibility an object “disappear” from the image by covering it with a
of digital images [1]. In the recent past large amount of small block copied from another part of the same image.
digital image manipulation could be seen in tabloid Since the copied segments come from the same image, the
magazine, fashion Industry, Scientific Journals, Court color palette, noise components, dynamic range and the
rooms, main media outlet and photo hoaxes we receive in other properties will be compatible with the rest of the
______________________________
image, thus it is very difficult for a human eye to detect.
About-1Associate Professor ,Department of Computer Applications. S.N.R. Sometimes, even it makes harder for technology to detect
Sons College Coimbatore -641006, India the forgery, if the image is retouched with the tools that are
About-2Associate Professor ,PG and Research Dept. of Computer available.
Applications, DG Vaishnav College,Chennai - 600 106, India
P a g e | 62 Vol. 10 Issue 7 Ver. 1.0 September 2010 Global Journal of Computer Science and Technology
III. COPY-MOVE FORGERY DETECTION TECHNIQUES As shown in Figure 3(c), the block B1, B2, and block B3
which are copies of blocks A1, A2, and block A3,
The simplest way to detect a Copy-Move forgery is to use
respectively. Therefore, VA1 =VB1, VA2 =VB2, and VA3
an exhaustive search. In this approach, the image and its
=VB3, where VX denotes the vector corresponding to block
circularly shifted version are overlaid looking for closely
X. As shown in sorted list, Figure 3(d), identical vectors are
matching image block. This approach is simple and effective
adjacent each other.
for small-sized images. However, this method is
computational expensive and even impractical for image of
medium-sized. In this method for an image size it
2
would take steps, since the comparison and image
processing require the order of operations for one shift..
Another technique for detecting forgery is based on
autocorrelation. All Copy-Move forgery introduces a
correlation between the original segment and the pasted one.
However, this method does not have large computational
complexity and often fail to detect forgery.
However, in most other approaches the detected image is
divided into overlapping blocks. The idea here is to detect
connected blocks that are copied and moved. The copied
region would consist many overlapping blocks. The distance
between each duplicated block pair would be same since
each block are moved with same amount of shift. The next
challenge would be extracting features form these blocks,
which would yield to very similar or same values for
duplicated block. Several authors presented to use different
features to represent the image block. These blocks are
vectorized and inserted into a matrix and the vectors are (a)
lexicographically sorted for later detection. The
computational time depends upon factor such as number of
blocks, sorting techniques and the number of feature.
Suppose an image size is , it is divided into
2
overlapping blocks of size b × b. The blocks are
represented as vectors of 2 dimensions, and sorted in a
lexicographical order (Fig 2). Vectors corresponding to
blocks of similar content would be close to each other in the
list, so that identical regions could be easily detected.
Input image
Dividing into
overlapping blocks
Approaches
Feature Extraction DCT, DWT, PCA
SVD, FMT etc
Computational Time
Lexicographically Number of blocks
Sorting Number of feature
Sorting Method
Locate the copy-move
region
The detection result
Fig. 3 (a). An original image, (b). Forged image (c) Three
Fig. 2. Configuration of a block Copy-Move Digital Image
pairs of identical blocks are marked by squares, (d). Feature
Forgery Detection System
vectors corresponding to the divided blocks are sorting in a
The image given in Figure 3(a) is the original image and list [13]
Figure 3(b) is the tampered image by Copy-Move Forgery.
Global Journal of Computer Science and Technology Vol. 10 Issue 7 Ver. 1.0 September 2010 P a g e | 63
Over the past 10 years, research has focused on how to Recently, Jing Zhang et al. [12] proposed a new approach
make Copy-Move forgery detection system fully automatic. based on the idea of pixel-matching to locate copy-move
Meanwhile, some significant advances have been made in regions. In this approach, DWT (Discrete Wavelet
this field. Nevertheless, many of the finding have important Transform) applied to the input image to yield a reduced
consequences for engineers who design algorithms and dimension representation. Then the phase correlation is
system for Copy-Move forgery detection. In the following computed to estimate the spatial offset between the copied
part of the paper we survey and highlight the summary of region and the pasted region. The task is to locate the Copy-
research on Copy-Move forgery detection. Move region by the idea of pixel-matching, which is shifting
the input image according to the spatial offset and
A. Region duplication detection: without Scaling and
calculating the difference between the image and its shifted
Rotation.
version. At the end, the MMO (Mathematical
Fridrich et al. [6] suggested the first method for detecting Morphological Operations) are used to remove isolated
the copy-move forgery detection. In their method, first the points so as to improve the location. The proposed technique
image is segmented into overlapping small blocks followed has lower computational complexity and it is reasonably
by feature extraction. They employed discrete cosine robust to various types of Copy-Move post processing.
transform (DCT) coefficients for this purpose. The DCT However, the performance of this method relies on the
coefficients of the small blocks were lexicographically location of Copy-Move regions.
sorted to check whether the adjusted blocks are similar or Ye et. al.[20] described a passive approach to detect digital
not. In their paper, the method shown was robust to the forgeries by checking the inconsistencies based on JPEG
retouching operations. However, the authors did not employ blocking artifacts. There approach consists of three main
any other robustness tests. steps: i) Collection of DCT statistics ii) Analyses of
On the other hand, A.C.Popescu et. al. [7] applied a statistics for quantization tables estimation and iii)
principle component analysis (PCA) on small fixed-size Assessment of DCT blocks errors with respect to the
image to yield a reduced dimension DCT block estimated quantization tables. The experimental result in
representation. Each block was represented as 16x16 and the their paper shows that the blocking artefact measure of
coefficients in each block were vectorized and inserted in a JPEG compression version is 97.1. In this paper, the authors
matrix and the corresponding covariance matrix was failed to mention how to remove the suspicious tampered
constructed. The matrix constructed stores floating numbers. regions for estimating quantization table. However, Battiato
By finding the eigenvectors of the covariance matrix, a new et. al [21], suggests that such techniques are strictly related
linear basis was obtained. Duplicated regions are then with the amount of forged blocks in comparison with the
detected by lexicographically sorting all of the image total number of blocks.
blocks. Their method was robust to compression up to JPEG All the above copy-move methods are most effective for
quality level 50 and the time complexity of sorting was detection when the region is pasted without any change
time. (scaling or rotation) to another location in the image.
Subsequently, G.Li et. al. [8] proposed a method which However, in practice, the duplicated region is often scaled or
reduced the time complexity for sorting was reduced to rotated to better fit it into the surroundings at the target
. The given image was decomposed into four location. Since, scaling or rotation change the pixel values, a
sub-bands by applying discrete wavelet transform (DWT). direct matching of pixel is unlikely to be more effective for
The singular value decomposition (SVD) was then applied the detection.
on these blocks of low-frequency component in wavelet
B. Region duplication detection: with Scaling and
sub-band to yield a reduced dimension representation. The
Rotation.
SV vector was lexicographically sorted to detect duplicated
region. Their method was robust to compression up to JPEG Recently, Bayram et. al [19] suggested a method by
quality level 70. Later on W. Luo et al. [9] suggested a new applying Fourier Mellin Transform (FMT) on the image
method based on the pixel block characteristics. The image block. They first obtained the Fourier transform
was first divided into small overlapped blocks and measured representation of each block, re-sampled the resulting
block characteristics vector form each block. Then the magnitude values into log-polar coordinates. Then they
possible duplicate region was detected by comparing the obtained a vector representation by projecting log-polar
similarity of the block. In this approach the time complexity values onto 1-D and used these representations as our
for sorting was further reduced to Their method features. In their paper, the authors showed that their
was robust to compression up to JPEG quality level 30 and technique was robust to compression up to JPEG quality
against Gaussian blurring and additive noise with SNR 24 level 20 and rotation with 10 degree and scaling by 10%.
dB. Hwei-Jen Lin et. al. [13] proposed a method in which each
Myna et al. [10] proposed an approach based on the block B of size (=16x16) by a 9-dimensional feature
application of wavelet transform that detects and performed vector. Unlike other techniques, where the feature vector
exhaustive search to identify the similar blocks in the image extracted stored floating numbers, this method stored them
by mapping them to log-polar coordinates and using phase as integer value. The feature
correlation as the similarity criterion.
P a g e | 64 Vol. 10 Issue 7 Ver. 1.0 September 2010 Global Journal of Computer Science and Technology
vectors extracted are then sorted using the radix sort, which potential transform between the original and duplicated
makes the detection more efficient without degradation of regions are estimated and the duplicated regions are
detection quality. The difference (shift vector) of the identified using correlation map. Even though using SIFT
positions of every pair of adjacent feature vectors in the keypoints guarantee geometric invariance and their method
sorting list was computed. The accumulated number of each enables to detect rotated duplication, these methods still
of the shift vectors was then evaluated and the large have a limitation on detection performance since it is only
accumulated number was considered as possible presence of possible to extract the keypoints from peculiar points of the
a duplicated region. The feature vectors corresponding to the image.
shift vectors with large accumulated numbers were detected, Recently, Seung_Jin Ryu et. al[15] suggested a method to
whose corresponding blocks are then marked to form a detect duplicated region using Zernike moments. The
tentative detected result. The final result was obtained by authors proposed to use Zernike moments over other
performing connected component analysis and medium technique since they found it to be superior to the others in
filtering on the tentative detected result. Even though, the terms of their insensitivity to image noise, information
proposed technique reduced the time complexity to content, and ability to provide faithful image representation.
with help of radix sort, the method failed to detect all copied A detailed review of relevant studies in Zernike moments is
region of smaller size. According to their experimental beyond the scope of this paper. For details the readers are
results, the scheme performed well when the degree of referred to the papers [16-18]. In their experiment, 12
rotation was 90, 180 and 270 degree. The figure 3 [13] different images were used to detect Copy-Move forgery
shows duplicated region with and without rotation. with various manipulations such as rotation etc. In the
proposed method the image was divided into
overlapped sub-blocks of and calculated the magnitude
of Zernike moments to extract vectors of each sub-block.
The vectors were then sorted in lexicographically order.
Finally, the suspected region is measured by Precision,
Recall, and F1 –measure which are often-used measures in
the field of information retrieval. The experimental result in
their paper show that their system could detect duplicated
region rotated some angle before it is pasted, the system is
weak against scaling or the other tempering based on affine
Fig. 3 (a) (b) transform.
a) Duplicated regions form several identical shift IV. CONCLUSION
vector u. As Copy-Move forgeries have become popular, the
b) Duplicated region from several (different) shift importance of forgery detection is much increased.
vector(u1-u4) , rotated through 90 degree. Although many Copy-Move Forgery detection techniques
have been proposed and have shown significant promise,
H. Huang et al. [11] presented a method to detect region robust forgery detection is still difficult. There are at least
duplication based on local image statistical features known three major challenges: tampered images with compression,
as scale invariant features transform (SIFT). SIFT tampered images with noise, and tampered images with
descriptors of an image are invariant to changes in rotation. In this paper we reviewed several papers to know
illumination, rotation, scaling etc. First the SIFT descriptors the recent development in the field of Copy-Move digital
of the image is extracted, and descriptors are then matched image forgery detection. Sophisticated tools and advanced
between each other to seek for any possible forgery in manipulation techniques have made forgery detection a
images. Even though this method enables to detect challenging one. Digital image forensic is still a growing
duplication, this scheme still have a limitation on detection area and lot of research needed to be done.
performance since it is only possible to extract the keypoints
from peculiar points of the image V. REFERENCES
More challenging situation for detection of copy-move 1) H.T. Sencar, and N.Memon, “Overview of State-
forgery is to detect the duplicated region which is rotated of-the Art in Digital image Forensics”, World
some angle before it is pasted. The method presented by Scientific Press, 2008
[13] to detect duplicated regions in limited rotation angles. 2) H . Farid, “A Survey of image forgery detection”,
More recently Xunyu Pan et. al[14] suggested a method to IEEE Signal Processing Magazine, Vol. pp. 16-25,
detect duplicated regions with continuous rotation regions. 2009
As described in [14] the new method was based on the 3) B.L.Shivakumar and S.Santhosh Baboo, “Digital
image SIFT features Image Forgery Detection”, SAJOSPS, Vol. 10(2),
First the SIFT features are collected from the image, and the pp. 116-119, 2010
image is segmented into non-overlapping examination 4) Lou Weigi, Qu Zhenhua, Pan Feng, and Herang
blocks. The matches of SIFT keypoints in each non- Jiwu, “ Survey of Passive Technology for Digital
overlapping pixel blocks are computed. After which the
Global Journal of Computer Science and Technology Vol. 10 Issue 7 Ver. 1.0 September 2010 P a g e | 65
Image Forensics”, Frontiers of Computer Science 17) Khotanzad, A., Hong, Y.H., “Invariant image
in China, Vol. 1(2), pp. 166-179, May 2007 recognition by Zernike moments”,.IEEE Trans.
5) Nizza, M., Lyons, P.J.: In an iranian image, a Pattern Analysis and Machine Intelligence, Vol.
missile too many. In: The Lede, The New York 12(5), pp. 489-497, 1990.
Times News Blog(2008) 18) Teh, C.H., Chin, R.T, “On image analysis by the
http://thelede.blogs.nytimes.com/ 2008/07/10/in- methods of moments”, IEEE Trans. Pattern
an-iranian-image-a-missile-too-many/. Analysis and Machine Intelligence Vol.10(4) , pp.
6) Fridrich, D. Soukal, and J. Lukás, “Detection of 496–513,1988.
copy move forgery in digital images,” in Proc. 19) Sevinc Bayram, Taha Sencar, and Nasir Memon,
Digital Forensic Research Workshop, Aug. 2003. “An efficient and robust method for detecting
7) A. C. Popescu and H. Farid, “Exposing Digital copy-move forgery,” in Proceedings of ICASSP
Forgeries by Detecting Duplicated Image 2009, 2009.
Regions,” Technical Report, TR2004-515, 20) S.M. Ye, Q.B. Sun, and E.C. Chang, “Detecting
Department of Computer Science, Dartmouth digital image forgeries by measuring
inconsistencies of blocking artifact”, in Proc. IEEE
8) G. Li, Q. Wu, D. Tu, and S. Sun, “A Sorted International Conference on Multimedia and Expo
Neighborhood Approach for Detecting Duplicated 2007, Beijing, China, pp.12-15, July 2007.
Regions in Image Forgeries based on DWT and 21) Sebastiano Battiato AND Giuseppe Messina,
SVD,” in Proceedings of IEEE International “Digital Forgery Estimation into DCT Domain – A
Conference on Multimedia and Expo, Beijing Critical Analysis”, in Proc of the First ACM
China, July 2-5, 2007, pp. 1750-1753. workshop on Multimedia in forensics, China, pp.
9) W .Luo, J. Huang, and G. Qiu, “Robust Detection 37-42 , 2009.
of Region Duplication Forgery in Digital Image,”
in Proceedings of the 18th International Conference
on Pattern Recognition, Vol. 4, 2006, pp. 746-749.
10) A. N. Myna, M. G. Venkateshmurthy, and C. G.
Patil, “Detection of Region Duplication Forgery in
Digital Images Using Wavelets and Log-Polar
Mapping,” in Proceedings of the International
Conference on Computational Intelligence and
Multimedia Applications (ICCIMA 2007), Vol. 3,
pp. 371-377, 2007.
11) H . Huang, W. Guo, and Y. Zhang, “Detection of
Copy-Move Forgery in Digital Images Using SIFT
Algorithm,” in Proceedings of IEEE Pacific-Asia
Workshop on Computational Intelligence and
Industrial Application, Vol. 2, pp. 272-276, 2008.
12) Jing Zhang, Zhanlei Feng and Yuting Su, “A New
Approach for Detecting Copy-Move Forgery in
Digital Images”, in: IEEE Singapore International
Conference on Communication Systems,
Guangzhou, China, pp. 362-366, 2008
13) Hwei-Jen Lin, Chun-Wei Wang, Yang-Ta Kao,
“Fast Copy-Move Forgery Detection”, in WSEAS
Transaction on Signal Processing, Vol 5(5), pp.
188-197, May 2009.
14) Xunyu Pan and Siwei Lyu, “Detecting Image
Region Duplication Using SIFT Features”, in:
International Conference on Acoustics, Speech, and
Signal Processing, Dallas, TX, 2010
15) Seung-Jin Ryu, Min-Jeong Lee and Heung-Kyu
Lee, “Detection of Copy-Rotate-Move Forgery
using Zernike Moments”, in: 12th International
Workshop on Information Hiding, Calgary,
Alberta, Candada, 2010
16) Kim, H.S., Lee, H.K., “Invariant image watermark
using Zernike moments”, IEEE Trans. Circuits and
Systems for Video Technology, Vol. 13(8), pp.
766-775, 2003.
