Artificial Neural Network Based Digital Image Forgery Detection

Artificial Neural Network based Digital Image
Forgery Detection
Rajeev Rajkumar Thounaojam Angaleima Chanu Ningthoujam Nejit Singh Khwairakpam Robinson Singh
Department of Computer Science & Engineering
Manipur Institute of Technology Takyelpat, Imphal.
{ rajeev.bono, leimaathouna, nejit123, robsimkhwairakpam } @gmail.com
Abstract— It is very easy now a days to tamper digital images to

hide or create misleading images with the development of Pre Processing
technology and availability image processing tools. Digitally
forged photographs are so real that they do not leave any
evidence of having been tampered with and can be
indistinguishable from authentic photographs. Digitally
Processed Image forgery makes the digital image data highly Keypoint Based Blocked Based
correlated. In this paper, forged image is detected by training the
artificial neural network using Auto Regressive (AR) coefficients
as the feature vector for identifying the location of digital forgery
in a sample image. 300 feature vectors from different images are Feature
used to train an Artificial Neural Network (ANN) and the ANN is Extraction
tested with another 300 feature vectors.
Keywords—Auto Regressive, Neural network,

Matching
I. INTRODUCTION
Filtering
A copy-move forgery is a specific type of image manipulation
which denotes an image where part of its content has been
copied and pasted within the same image as shown in figure
2. Typical motivations are either to hide or to conceal
unwanted portion of an image, or to emphasize particular Post-processing
objects. A copy-move forgery is straight forward to create.
Additionally, both the source and the target regions stem from
the same image, thus properties like the color temperature,
illumination conditions and noise are expected to be well-
matched between the tampered region and the image. Most Fig 1: Common processing pipeline for the detection of
techniques follow a common pipeline, as shown in Fig. 1. copy-move forgeries. The feature extraction differs for
Copy-move forgery detection methods are either key point- keypoint-based features (top) and block-based features
based methods [3], [11],or block-based methods [4]–[7], [9], (bottom).
[10], [12]–[18].Preprocessing of the images is possible in
both the two cases. However ,for instance most methods
operate on gray- scale images, and as such require that the A forgery shall be reported if regions of such matches cluster
color channels be first merged. into larger areas. Both the keypoint and block-based methods
include further filtering for removing spurious matches. In
For every such region, a feature vector is computed and addition to this an optional post processing step of the
similar feature vectors are subsequently matched. By contrast, detected regions may also be performed, in order to group
keypoint-based methods compute their features only on matches that jointly follow a transformation pattern.
image regions with high entropy, without any image
subdivision. Similar features within an image are afterwards
matched.
The same procedure is repeated for every sub-block collected
to get 300 feature vectors and they are treated as feature
vectors for forged images. Sample feature vectors displayed in
Figure 1 and Figure 2, collected from manipulated images and
original images show significant differences. They are
oscillatory for the forged images.
(i) (ii)
Fig 2: (i) Original image (ii) a photograph of Lincoln's head was III. TRAINING ARITIFICIAL NEURAL NETWORK
superimposed onto a portrait of the southern leader John For training the neural network, two groups of feature vectors
Calhoun are collected, one group from the original images database and
. the other group from the forged images database. 300 feature
vectors (150 from original and 150 from forged images
II. FEATURE EXTRACTION database) are used to train the Back propagation Neural
The Digital Image data matrix is first divided into sub-blocks Network. The Aritificial Neural Network(ANN) used consists
each of size 50 X 50. Then every sub-block is reshaped and of 20 input neurons and 1 output neuron with 8 hidden
normalized to make the image data value vary between 0 to 1, neurons. Remaining 300 are used to test the ANN.
which can be attain by dividing each pixel value by 255
(maximum gray value). The resultant matrix is arranged as the IV. EXPERIMENTAL RESULT
row vector [1X2500]. The row vector is treated as 1D signal
and is modelled using ‘Auto Regressive’ modelling with 20 The images with digital forgery of the type 2 [Without copy
coefficients. This is called Auto Regressive coefficients. This move] are collected as the Image data. The Image data is
is randomly collected from different sub-blocks of different divided into sub-blocks of size 50X50. From every sub-block,
images. About 300 such feature vectors collected are treated feature vector is extracted as described above and is tested for
as feature vectors for the original images. The images used for originality using trained ANN. Original sub block is left as
the same are of high quality TIFF (Tagged Image File Format) such. The sub-blocks identified as forged are made white for
files or BMP (Bitmap) files. highlighting the area of forgery in the image. The sample
experiment results are displayed in figure 3.
The sub-blocks collected are again subjected to rotation, up-
sampling , down-sampling with 3 different levels and using
different techniques of interpolation like ‘Bilinear’, ’Bi-cubic’
and ‘Nearest’. Modified sub-blocks are reshaped, normalized
and modelled using ‘Auto Regressive’ modelling with 20
coefficients.
Figure 1. Sample feature vectors for the forged images

Figure 2. Sample feature vectors for the Original images
TABLE I. RESULTS OF EXPERIMENT V. CONCLUSION

Digital Forgery Detection %
Hit(Forgery+Oroginal) 75%
Thus an attempt to detect the digital forgery of merging
Miss (Forgery) 20%
category was done using ‘AR’ coefficients as feature vectors.
False (Forgery) 30% Percentage of success is comparable with the technique
suggested by [1] with less complexity and reliability. Also the
paper concludes that, instead of using forged images got by
rotating, scaling and resizing the original images, the database
of forged images helps in training the ANN better.
REFERENCES
[1] J. Redi, W. Taktak, and J.-L. Dugelay, “Digital image forensics: A

booklet for beginners,” Multimedia Tools Applicat., vol. 51, no. 1, pp.133–
162, Jan. 2011.
[2] H. Farid, “A survey of image forgery detection,” Signal Process. Mag.,
vol. 26, no. 2, pp. 16–25, Mar. 2009.
[3] I. Amerini, L. Ballan, R. Caldelli, A. D. Bimbo, and G. Serra, “A SIFT-
based forensic method for copy-move attack detection and transformation
recovery,” IEEE Trans. Inf. Forensics Security, vol. 6, no. 3, pp. 1099–
1110, Sep. 2011.
[4] M. Bashar, K. Noda, N. Ohnishi, and K. Mori, “Exploring duplicated
regions in natural images,” IEEE Trans. Image Process., accepted for
publication.
Figure 3.Sample results – experiment 1 [5] S. Bayram, I. Avcıbaş, B. Sankur, and N. Memon, “Image manipula-
tion detection with binary similarity measures,” in Proc. Eur. Signal
Processing Conf., Antalya, Turkey, Sep. 2005.
[6] S. Bayram, H. Sencar, and N. Memon, “An efficient and robust
method for detecting copy-move forgery,” in Proc. IEEE Int. Conf.
Acoustics, Speech, and Signal Processing, Apr. 2009, pp. 1053–1056.
[7] S. Bravo-Solorio and A. K. Nandi, “Exposing duplicated regions af-
fected by reflection, rotation and scaling,” in Proc. Int. Conf.
Acoustics, Speech and Signal Processing, May 2011, pp. 1880–1883.
[8] V. Christlein, C. Riess, and E. Angelopoulou, “On rotation invariance
in copy-move forgery detection,” in Proc. IEEE Workshop on
Informa- tion Forensics and Security, Seattle, WA, Dec. 2010, pp.
129–134.
[9] B. Dybala, B. Jennings, and D. Letscher, “Detecting filtered cloning
in digital images,” in Proc. Workshop on Multimedia & Security,
Sep.2007, pp. 43–50.
[10] J. Fridrich, D. Soukal, and J. Lukáš, “Detection of copy-move forgery
in digital images,” in Proc. Digital Forensic Research Workshop,
Cleveland, OH, Aug. 2003.
[11] H. Huang, W. Guo, and Y. Zhang, “Detection of copy-move forgery
in digital images using SIFT algorithm,” in Proc. Pacific-Asia
Workshop on Computational Intelligence and Industrial Application,
Dec. 2008, vol. 2, pp. 272–276.
[12] S. Ju, J. Zhou, and K. He, “An authentication method for copy areas
of images,” in Proc. Int. Conf. Image and Graphics, Aug. 2007,
pp.303–306.
[13] X. Kang and S. Wei, “Identifying tampered regions using singular
value decomposition in digital image forensics,” in Proc. Int. Conf.
Computer Science and Software Engineering, 2008, vol. 3,
pp.926–930.
[14] Y. Ke, R. Sukthankar, and L. Huston, “An efficient parts-based near-
duplicate and sub-image retrieval system,” in Proc. ACM Int. Conf.
Multimedia, Oct. 2004, pp. 869–876.
[15] G. Li, Q. Wu, D. Tu, and S. Sun, “A sorted neighborhood approach
for detecting duplicated regions in image forgeries based on DWT and
SVD,” in Proc. IEEE Int. Conf. Multimedia and Expo, Jul. 2007,
pp.1750–1753.
[16] A. Langille and M. Gong, “An efficient match-based duplication detec-
tion algorithm,” in Proc. Canadian Conf. Computer and Robot Vision,
Jun. 2006, pp. 64–71.
[17] C.-Y. Lin, M. Wu, J. Bloom, I. Cox, M. Miller, and Y. Lui, “Rotation,
scale, and translation resilient watermarking for images,” IEEE Trans.
Image Process., vol. 10, no. 5, pp. 767–782, May 2001.
[18] H. Lin, C. Wang, and Y. Kao, “Fast copy-move forgery
detection,”WSEAS Trans. Signal Process., vol. 5, no. 5, pp. 188–197, 2009.

Artificial Neural Network Based Digital Image Forgery Detection

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Artificial Neural Network Based Digital Image Forgery Detection

Uploaded by

Copyright:

Available Formats

Artificial Neural Network based Digital Image

Abstract— It is very easy now a days to tamper digital images to

Keywords—Auto Regressive, Neural network,

Figure 1. Sample feature vectors for the forged images

TABLE I. RESULTS OF EXPERIMENT V. CONCLUSION

[1] J. Redi, W. Taktak, and J.-L. Dugelay, “Digital image forensics: A

You might also like