Copy Move Forgery Detection Using GLCM Based Statistical Features
Copy Move Forgery Detection Using GLCM Based Statistical Features
Copy Move Forgery Detection Using GLCM Based Statistical Features
4, August 2016
ABSTRACT
The features Gray Level Co-occurrence Matrix (GLCM) are mostly explored in Face Recognition and
CBIR. GLCM technique is explored here for Copy-Move Forgery Detection. GLCMs are extracted from all
the images in the database and statistics such as contrast, correlation, homogeneity and energy are
derived. These statistics form the feature vector. Support Vector Machine (SVM) is trained on all these
features and the authenticity of the image is decided by SVM classifier. The proposed work is evaluated on
CoMoFoD database, on a whole 1200 forged and processed images are tested. The performance analysis
of the present work is evaluated with the recent methods.
KEYWORDS
GLCM, CMFD, SVM Classifier, Detection rate
1. INTRODUCTION
Digital images have a significant role in conveying the information. Digital Image manipulation
became very easy with the availability of advanced photo editing tools. But, due to the
manipulation the trustworthiness of digital images is lost. Hence, detection of image forgery is
important and is achieved in passive mode without embedding any signature in the original
image. Passive image forgery detection works on the discrepancies in the statistical features of
the forged image. Copy-Move tampering is a very common method of tampering digital image
where in some portion of an original image is copied and pasted at some other location in the
same original image. In general, this is done with intent to conceal a region in the image. The
copied portions are within the image, so the changes in texture, variations in intensity or any
statistical property may match with the remaining portion of the original image. Hence, it is
challenging for detecting the forged portion based on HVS [1]. An exhaustive search can be used
to identify the significant features of copied and pasted portions on the tampered image. This
mechanism needs more time for detection and is computationally complex [2]. Therefore,
similarity measure can be used on the identical image regions for detecting the forgery
successfully [2]. Figure 1(a) and 1(b) illustrates Copy move forgery.
DOI: 10.5121/ijci.2016.5419
165
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
a.Original Image
A comprehensive report on passive methods for forgery detection in images is available in [3].
Here, the works based on textural features are reviewed. Shikha Dubey et al. [4] used local
descriptors for textural features and block matching is performed using clustering technique. In
[5], the Gabor magnitude of the image is computed and a histogram is formed as a feature vector.
Gabor Wavelets and Local Phase Quantization [6] are used to extract texture features for image
forgery detection. In [7], features are extracted based on GLCM and Histogram of Oriented
Gradient (HOG) and KNN classifier is used for image forgery detection.
2. METHODS
2.1. GLCM
GLCM is the key process of this work. The Gray Level Co-occurrence Matrix (GLCM) provides
information on the occurrence of various combinations of pixel intensities in a gray image. It is a
statistical approach [8] of exploring the spatial relationship among pixels. GLCM computes in
what a way a pixel with intensity i occur horizontally, vertically or diagonally to a pixel with
intensity j.
GLCM exhibits certain properties regarding the spatial relationships of gray intensities in the
image.
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
The process involved in GLCM formation is shown in Figure 2. The statistical features that are
computed from GLCMs are as follows:
2
Energy = i , j P(i, j )
(1)
(2)
Homogeneity = i , j
1
P(i, j )
2
1 + (i j )
(3)
Inertia = i , j (i j ) P(i, j )
Correlation = i , j
(4)
(i )( j )P(i, j )
(5)
Shade = i , j (i + j 2 ) P(i, j )
(6)
Prominence = i , j (i + j 2 ) P (i, j )
(7)
Variance = i , j (i ) P (i, j )
(8)
and = i(i x )
P(i, j ) = ( j ) P(i, j )
2
Contrast = (i j ) P(i, j )
i
(9)
1
{P(i, j )}
1 + (i j ) 2
(11)
(12)
Dissimilarity = i j P(i, j )
i
(10)
(13)
(14)
2 Na
(15)
i =2
(16)
N a 1
(17)
i =0
HXY HXY 1
max{HX , HY }
1/ 2
= (1 exp[ 2.0( HXY 2 HXY )])
1
{P(i, j )}
Inverse Difference =
i
j 1+ i j
(18)
(19)
(20)
167
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
3. PROPOSED METHOD
A Copy-Move Forgery Detection (CMFD) method is proposed using GLCM and SVM. The
proposed method is detailed below and is shown in Fig.3.
i.
ii.
iii.
iv.
v.
vi.
The standard database CoMoFoD consists of original, forged and processed images is
considered in the performance analysis.
The images in the database are converted to gray scale.
The statistical features are computed on GLCMs developed from the gray scale images.
The Support Vector Machine is trained with those 20 statistical features for every image
in the database using RBF kernel.
Statistical features of the testing image are obtained in similar process using steps 2 and
3.
The SVM classifier classifies the image either to be authentic or forged.
Image Database
Gray scale Conversion
GLCM Features
GLCM Features
SVM Training
SVM
Classifier
Forged Image
Genuine image
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
Scaling
Rotation
Rotated angle
TPR
Scaled factor in %
TPR
95.31
40
75
75
70
84.37
40
68.75
95
89.5
90
62.50
105
96.87
The post-processed images with the below attacks are considered for evaluation.
i.
ii.
iii.
iv.
v.
vi.
The present method is appraised by considering 50 forged images in each post-processing attack
category, so at the outset 1200 forged and processed images are tested.
Table 2: TPR and FNR of our proposed method for various post-processing attacks
Attack Description
TPR in % FNR in %
No Attack
100
92
100
100
66
34
68
32
169
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
76
24
Color Reduction 32
98
Color Reduction 64
94
94
60
40
68
32
12
100
96
78
22
70
30
74
74
26
80
20
90
10
94
100
100
100
It is evident from Table 2 that the proposed method withstand attacks JPEG compression, Image
blurring, Color reduction, brightness change and Noise addition in a better manner when
compared to the attacks Contrast adjustment and Image blurring. It is evident from Table 3 that
our method outperforms the other two methods [4, 6] in terms of TPR under no attack.
Table 3: Comparative Analysis of the proposed method
Method
Method in [4]
RST invariant
95.48
Method in [6]
No
99.83
100
5. CONCLUSIONS
In recent times, GLCM features are exploited to identify forgery related to Human faces in digital
images. But, in our proposed method it is explored for all kinds of images such as buildings,
plants, vehicles, people and textures. The simulation results indicate that our proposed method
withstands all the post-processing attacks except Contrast Adjustment and Intensity Blurring. The
proposed method outperforms the two methods [4, 6]. Proposed method is also invariant to
170
International Journal on Cybernetics & Informatics (IJCI) Vol. 5, No. 4, August 2016
rotation and scaling attacks to some extent. In future, the work can be extended to localize the
tampered regions.
REFERENCES
[1]
Shivakumar B L, Santhosh Baboo S. Detecting Copy-Move Forgery in Digital Images: A Survey and
Analysis of Current Methods, Global Journal of Computer Science and Technology, 2010, 10 (7), pp.
61-65.
[2] Khan S, Kulakarni A, A reduced time complexity for detection of copy-move forgery detection using
Discrete Wavelet Transform, International of Computer Applications, 2010, 6 (7), pp.31-36.
[3] Mahdian B,Stanislav S. . A bibliography on blind methods for identifying image forgery, Signal
Processing: Image Communication, 2010, 25 (6), pp. 389-99.
[4] Shikha Dubey, A Sarawagi, Manish Srivastava, Image Forgery Detection based on Local
Descriptors and Block Matching using Clustering Technique International Journal of Computer
Applications, Vol.141, No.10, May 2016, pp.11-14.
[5] Jen Chun Lee, Copy-Move image forgery detection based on Gabor magnitude, Journal of Visual
Communication and Image representation, Vol.31, 2015, pp.320-334.
[6] Meera Mary Isaac, M Wilscy, Image forgery detection based on Gabor Wavelets and Local Phase
Quantization Proceedia Computer Science, Vol. 58, 2015, pp.76-83.
[7] Liya Baby, Ann Jose, Digital Image Forgery Detection Based on GLCM and HOG Features
International Journal of Advanced research in Electrical, Electronics and Instrumentation
Engineering, Vol.3, Issue.5, Dec.2014, pp.426-430.
[8] Mryka
Hall-Beyar,GLCM
Tutorial,February
2007
[Online]
Available:
http://www.fp.ucalgary.ca/mhallbey/tutorial.htm ( February 21, 2007).
[9] Vapnik, Vladimir. The nature of statistical learning theory. springer, 2000.
[10] http://www.vcl.fer.hr/comofod/comofod.html
AUTHORS
Gulivindala Suresh is a research scholar in the Department of ECE at JNTU-K
University College of Engineering, AP, India. He obtained his M.Tech from Biju Patnaik
University of Technology, Orissa, India. He obtained his B.Tech from JNTU, AP, India.
His research interests are Digital Image Processing and VLSI. He is a member of IETE.
Presently, he is working as Assistant Professor in the Department of ECE, GMR Institute
of technology, Rajam, AP, INDIA.
Srinivasa Rao Ch is currently working as Professor in the Department of ECE, JNTUK
University College of Engineering, Vizianagaram, AP, India. He obtained his PhD in
Digital Image Processing area from University College of Engineering, JNTUK,
Kakinada, AP, India. He received his M. Tech degree from the same institute. He
published 40 research papers in international journals and conferences. His research
interests are Digital Speech/Image and Video Processing, Communication Engineering
and Evolutionary Algorithms. He is a Member of CSI. Dr Rao is a Fellow of IETE.
171