50120130405017 2

International Journal of Computer Engineering and Technology (IJCET), ISSN 0976-6367(Print),
INTERNATIONAL JOURNAL OF COMPUTER ENGINEERING &
ISSN 0976 - 6375(Online), Volume 4, Issue 5, September - October (2013), © IAEME

TECHNOLOGY (IJCET)

ISSN 0976 – 6367(Print)
ISSN 0976 – 6375(Online)
Volume 4, Issue 5, September – October (2013), pp. 147-154
© IAEME: www.iaeme.com/ijcet.asp
Journal Impact Factor (2013): 6.1302 (Calculated by GISI)
www.jifactor.com

IJCET
©IAEME

ROBUST IMAGE ALIGNMENT USING SVM CLUSTERING FOR TAMPERING
DETECTION

Ms. Sheetal Kusal, Prof.Jyoti Rao
Computer Dept, D.Y.P.I.E.T, Pimpri, Pune, India,

ABSTRACT
The widespread use of technologies such as emails, social networks are available on Internet. In
the internet communication, systems should able to protect visual content against malicious
manipulations that could be performed during their transmission. One of the main issues addressed in
this is the authentication of the image received in a Communication. Tampering detection has great
significance in authentication of image. This task is usually performed by localizing regions of the
image which have been tampered. To localize the tampering aligned image should be first registered at
the sender by making use of the information provided by a specific component of the image hash
associated to the image. In this a robust alignment method is proposed which makes use of an image
hash component based on the Bag of Features (BOF) paradigm. These Bags of Features are clustered for
effective image alignment. And SVM clustering is preferred for more accuracy. The proposed signature
is attached to the image before transmission and then analyzed at destination to recover the geometric
transformations which have been applied to the received image. The proposed image hash encodes the
spatial distribution of the image features to deal with highly textured and contrasted tampering patterns.
A block-wise tampering detection which uses histograms of oriented gradients (HOG) representation is
proposed. A quantization of the histogram of oriented gradient space is used to build the signature of
each image block for tampering purposes. The proposed approach obtains good margin in terms of
performances with respect to state-of-the art methods.

Key terms: BOF, Forensic Hash, Geometric transformation, image alignment, tampering.

147


1. INTRODUCTION
The widespread uses of classic and newest technologies are available on Internet. In the context
of internet communication, systems should able to protect the visual content against malicious
manipulations that could be performed during their transmission. One of the main addressed in this is
the authentication of the image received in a communication. Altering /manipulating images are not new
it has been around since the early days of photography. Due to the advent of technology, such as digital
cameras, powerful PCs and the availability of digital image editing software, image manipulation has
become common and easy. Thus in the context of Internet communications methods useful to establish
the validity and authenticity of a received image are needed. It is the process of Manipulating, altering,
modifying the image for malicious purposes to change semantic meaning of image is called as
tampering. Tampering detection falls into two categories as follows: 1) watermarks 2) image hashes.
The problem of tampering detection can be addressed using watermarking based approach. A watermark
can be embedded into the image and during tampering detection; it is extracted to verify if there was a
malicious manipulation on the received image. Damage into the watermark indicates a tampering of the
image under consideration. A clear disadvantage in using watermarking is the need for distorting the
content unfortunately; watermarking authentication is not backward compatible with previously encoded
contents; unmarked contents cannot be authenticated later. Embedded watermarks increases the bit-rate
required when compressing a media file. To overcome this problem signature based approaches have
been introduced. In this approach image hash is not embedded into the image (e.g., in watermarking); it
is associated with the image as header information and must be small and robust against different
operations. Different hash based approaches have been recently proposed in literature. Most of them
share the same basic scheme: i) a hash code based on the visual content is attached to the image to be
sent; ii) the hash is analyzed at destination to verify the reliability of the received image. An image hash
is a distinctive signature which represents the visual content of the image in a compact way (just few
bytes). The image hash should be robust against allowable operations and at the same time it should
differ from the one computed on a different/tampered image. An image hash is a short signature of the
image that preserves its semantic information under allowable changes made to. That is, it should be
robust to allowable modifications (like small rotations, compression, scaling, addition of noise etc) and
sensitive to distinct images or illegal manipulations to the original like tampering. A common approach
of image hashing is extracting features which have perceptual importance and should survive
compression. The authentication data are generated by compressing these features or generating their
hash values. The user checks the authenticity of the received content by comparing the features or their
hash values to the authentication data. In order to perform tampering detection, the receiver should be
able to filter out all the geometric transformations added to the tampered image by aligning the received
image to the one at the sender.
2. RELATED WORK
The need of methods useful to establish the validity and authenticity of an image received
through an Internet communication is important. To deal with this problem different solutions have been
recently proposed in literature [1, 2, 3, 4, 5, and 6]. In [3] propose a robust alignment method which
makes use of an image signature based on the Bag of Features paradigm. In [3] Battiato et all proposed a
method, an image hash based on the Bag of Visual Words paradigm [4] is attached as signature to the
image before transmission. Then forensic hash is analyzed at destination to detect the geometric
transformations which have been applied to the received image. In [5] Lin et all proposed a novel
148


approach based on distributed source coding for the problem of backward compatible image
authentication. The main idea is to provide a Slepian-Wolf encoded quantized image projection as
authentication data. In [6] Lu et all proposed a new framework to perform multimedia forensics by using
compact side information to reconstruct the processing history of a multimedia document. They
presented a framework as FASHION, stands for Forensic hash for information assurance. There are
different robust alignments techniques have been proposed by computer vision researchers [9]-[11]. In
[10] Irani and Anandan presented direct Methods which recovers the unknown parameters directly from
measurable image quantities at each pixel in the image and direct methods are used for motion or shape
estimation. In [9] Torr and Zisserman presented feature based methods where it first extract a sparse set
of distinct features from image to build an initial estimation of geometry and then this geometry is used
to get of image correspondence in regions of image where there is less information. Richard Szeliski in
his tutorial [11], reviews image alignment and image stitching algorithms. But these techniques are
unsuitable in the context of forensic hashing, since a basic requirement is that the image signature
should be as compact as possible to reduce the overhead of the network communications. To fit the
underlying requirements, Lu et all [6] have proposed to exploit information extracted through Radon
transform and scale space theory in order to estimate the parameters of the geometric transformations
(i.e., rotation and scale). Lu et all proposed a new framework for multimedia forensics called FASHION.
FASHION stands for Forensic hash for information assurance, which bridges two research areas,
namely, blind multimedia forensics and robust image hashing, to achieve more efficient and accurate
forensic analysis. To make more robust the alignment phase with respect to manipulations such as
cropping and tampering, an image hash[7] based on robust invariant features has been proposed by
Wenjun Lu and Min Wu in [7]. They proposed a new construction of forensic hash based on visual
words representation. They encode SIFT features into a compact visual words representation for robust
estimation of geometric transformations and propose a hybrid construction using both SIFT and blockbased features to detect and localize image tampering. The technique proposed by Sujoy Roy and Qibin
Sun in [8] extends the idea by employing the Bag of Features (BOF) model to represent the features to
be used as image hash. The exploitation of the BOF representation is useful to reduce the space needed
for the image signature, by maintaining the performances of the alignment component. In [1] the
proposed method, an image hash based on the Bag of Visual Words paradigm is attached as signature to
the image before transmission. Then forensic hash is analyzed at destination to detect the geometric
transformations which have been applied to the received image. This is a more robust approach based on
a cascade of estimators has been introduced; it is able to better handle the replicated matchings in order
to make a more robust estimation of the orientation parameter. Besides, the cascade of estimators allows
a higher precision in estimating the scale factor. An effective way to deal with the problem of wrong
matching’s has been proposed in [2], where a filtering strategy based on the scale invariant feature
transform (SIFT) dominant directions combined in cascade with a robust estimator based on a voting
strategy on the parameter space is presented.
3. PROPOSED METHOD
In this section, we briefly review the idea of image alignment component and tampering
localization component. And present the how SVM clustering is better for tampering localization.

149


3.1 ALIGNMENT COMPONENT

Fig.1 registration component
As previously stated, one of the important steps of tampering detection systems is the alignment
of the received image. Image registration is important since all the other tasks (e.g., tampering
localization) usually assume that the received image is aligned with the original, and hence could fail if
the registration is not properly done. The registration component is already proposed by [1] which
shown in Fig. 1. Adopted a BOF-based representation be used as hash component for the alignment. We
employ a transformation model and a voting strategy to retrieve the geometric manipulation used in [1].
In the system, a codebook is generated by clustering the set of SIFT [4] extracted on training images.
The clustering procedure points out a centroid for each cluster. The K-means clustering technique is
used for finding the centroid. Here we are suggesting Support vector machine (SVM) clustering
algorithm, which have more accuracy. SVMs are able to handle simple, linear, classification tasks and
more complex problems, i.e. nonlinear, classification. Both separable and non separable problems are
handled by SVMs in the linear and nonlinear case. The inspiration behind SVMs is to map the original
data points from the input space to a high dimensional or infinite dimensional, feature space such that
the classification problem becomes simpler in the feature space. The mapping is done by a suitable
choice of a kernel function. The computed codebook is shared between sender and receiver. It should be
noted that the codebook is built only once [1], and then used for all the communications between sender
and receiver (i.e., no extra overhead for each communication). The sender extracts SIFT features [2] and
sorts them in descending order with respect to their contrast values. Then top n SIFT are selected and
associated to the id label corresponding to the closest centroid belonging to the shared codebook. Hence,
the final hash for the alignment component is created by considering the id label, the dominant direction
Q, and the key point coordinates (x, y) for each selected SIFT. The source image and the corresponding
hash component for the alignment (hs) are sent to the destination. The method assumes that the image is
sent over a network consisting of possibly untrusted nodes, where the signature is sent upon request by a
trusted authentication server which encrypts the hash in order to guarantee its integrity. In the untrusted
communication the image could be modified for malicious purposes. Then image reaches to destination,
the receiver generates the related hash signature for registration (hr) by using the same procedure
employed by the sender. Then entries of the hashes hs and hr are matched by considering the id values.
Note that an entry of hs may have more than one association with entries of hs (and vice versa) due to
possible replicated elements in the hash signatures. After matching's are obtained, the alignment is
achieved by employing a similarity transformation of keypoint pairs corresponding to matched hashes
entries as shown in equation 1 and 2:
Xr = Xsbcosa - Ysbsina + Tx (1)
Yr= XSbsina + Ysbcosa + Ty (2)
These transformations are used to model the geometrical manipulations [1] which have been done on the
source image during the untrusted communication.
150


3.2 TAMPERING LOCALIZATION
Once the alignment has been performed, the image is analyzed to detect tampered regions.
Tampering localization [1] is the process of localizing the regions of the image that have been
manipulated for malicious purposes to change the semantic meaning of the visual message. The
tampering manipulation typically changes the properties (e.g., edges distributions, colors, textures, etc.)
of some image regions. This technique makes use of some useful routines already available for HOG(Histogram of gradient) like feature computing. More specifically, image gradients are computed by
using the simple 1D filters [1 0 1] and [1 0 1] T. These gradients are then used to obtain magnitude and
orientation for each image pixel. The image is then divided into non-overlapped blocks 32x32. The
orientation bins can be in the range [0,180] (unsigned gradient) or [0,360] (signed gradient). Each pixel
of the considered block votes for a bin of the histogram through a function of the gradient magnitude of
the pixel. Finally obtained histograms are normalized and quantized. Obtained histograms are
normalized and then quantized in order to obtain a more compact hash for each block (e.g., 15 bit for
each block [8]). The comparison of histograms of corresponding blocks is usually performed through a
similarity measure (e.g., Euclidean distance) and a thresholding procedure.
3.3 METHODOLOGY
System consists of two parts. First part of system consists of alignment component and overall
registration framework and second part consists of tampering localization component. Aim of the image
alignment component is to model the geometrical manipulations which have been done on the source
image during the untrusted communication. Image manipulations are estimated by scaling, rotation and
translation parameters by using key point co ordinates and hash signatures. System uses a cascade
approach; first an initial estimation of the rotation and translation parameters is accomplished through a
voting procedure. Afterward, the scaling parameter is estimated by means of the rotation and translation
parameters which have been previously estimated on the reliable information. A codebook is generated
by SVM clustering set of SIFT features on training images.SVM clustering points out support vectors
from each cluster then these support vector for each clusters represent the codebook to be used during
hash generation. This codebook is shared between sender and receiver by trusted network for further
use.
3.3.1 VOTING PROCEDURE
As we have seen earlier the aim of the alignment phase is the estimation of the quadruple (a, b,
Tx, Ty) by exploiting the correspondences ((Xs, Ys); (Xr, Yr)) related to matchings between hs and hr.
We employ cascade approach; first an initial estimation of the parameters (a, Tx, Ty) is accomplished
through a voting procedure in the quantized parameter space a*Tx*Ty which already employed by
Battiato et all [1].
3.3.2 SUPPORT VECTOR MACHINE
The key concept of SVMs, which were originally first developed for binary classification
problems, is the use of hyper planes to define decision boundaries separating between data points of
different classes. The idea behind SVMs is to map the original data points from the input space to a high
dimensional, or infinite-dimensional, feature space such that the classification problem becomes simpler
in the feature space. The mapping is done by appropriate choice of a kernel function.
SVM Algorithm:

151


Input:
D= xi, yi with i= 1 to n or with xi, yi.
Xi = the input vectors = xi ∈Rd
Yi = the class labels = yi ∈ {−1, +1}.
Output:
f(x) = sign (wTϕ(x) + b).
Algorithm:
Consider a training data set D with n points in d dimensional space.
D= xi, yi with i= 1 to n or with xi, yi.
Xi = the input vectors = xi ∈Rd
SVM maps the d-dimensional input vector x from the input space to the dh dimensional feature space
using a nonlinear function ϕ (·),
ϕ (·): Rd →Rdh.
The separating hyper plane in the feature space is then defined as
wTϕ(x) + b = 0
with b ∈R and w an unknown vector with the same dimension as ϕ(x). A data point x is assigned to the
first class,
if f (x) = sign (wTϕ(x) + b) equals +1
or
to the second class if f (x) equals −1.
Nonlinear function Rd = dh maps the input space to a so-called higher dimensional feature space. It is
important to note that the dimension Rdh of this space is only defined in an implicit way (it can be
infinite dimensional). The term b is a bias term. In primal weight space the classifier takes the form or it
can be The classifier is written as
f (x) = sign (wTϕ(x) + b).
but, on the other hand, is never evaluated in this form. The SVM classifier takes the form
#‫ܞܛ‬

܎ሺ‫ܠ‬ሻ ൌ ‫ܖ܏ܑܛ‬ሺ෍ હܑ‫۹ܑܡ‬ሺ‫ܑܠ ,ܠ‬ሻ ൅ ‫܊‬ሻ
ܑୀ૚

where #SV represents the number of support vectors and the kernel function K (.,.) is positive definite.
Various types of kernel functions can be chosen
linear SVM: K(x, z) = xT z,
polynomial SVM of degree d:
K(x, z) = (τ + xT z) d, _ ≥ 0,
Radial basis function (RBF):
K(x, z) =

ୣ୶୮ሺିԡ୶ି୸ԡሻమ
మ
ρଶ

MLP: K(x, z) = tanh (k1xT z + k2)
where K(·, ·) is positive definite for all ρ values in the RBF kernel case and τ ≥ 0 values in the
polynomial case, but not for all possible choices of k1, k2 in the MLP case. A main advantage of SVM
152


classification is that SVM performs well on datasets that have many attributes, even when there are only
a few cases that are available for the training process. However, several disadvantages of SVM
classification include limitations in speed and size during both training and testing phase of the
algorithm and the selection of the kernel function parameters.
4. EXPERIMENTAL RESULTS
This section reports a number of experiments on which the proposed approach has been tested
and compared with respect to [1].In our experiments a collection of 25 dissimilar images was used.
Some of these images have been tampered by performing splicing, content removal, and content
rearranging, to generate perceptually indistinguishable image queries. We have also performed image
transformations as following: cropping, rotation, scaling, translation, local tampering, and global
tampering. As we are employing overall procedure used by Battiato et al. In [1] Battiato et all performed
K-means clustering for forming the codebook. The codebook has been learned through k-means
clustering on the overall SIFT descriptors extracted on training images. In our proposed method we are
employing SVM clustering. The registration results obtained employing the proposed alignment
approach with hash component of different size (i.e., different number of SIFT) are shown in graphs.

Fig.2 tampering detection comparison between SVM and k-means. Results have been obtained by
increasing no of SIFT points and accuracy.
As shown in Fig.2 by increasing the number of SIFT points the number of unmatched images
decreases (i.e., image pairs that the algorithm is not able to process because there are no matchings
between hs and hr). In all cases the percentage of images on which our method is able to work is higher
than the one obtained by the approach proposed in [1]. The tests reported in the following reveal that our
method strongly outperforms compared to [1] in terms of accuracy and robustness with respect to the
different transformations. As time required by SVM is more compared to K-means. Using SVM
clustering, no of SIFT points required are less compared to [1].

153


5. CONCLUSION
The main contribution of this work is related to the alignment of images in the context of
distributed forensic systems. This work is concerned to establish the minimal number of SIFT needed to
guarantee an accurate estimation of the geometric transformations. With a proper learning process we
could observe SVM clustering algorithm out performs K-means clustering algorithm. SVM is very
effective when no of SIFT points are less and efficiently localize tampering regions. So SVM increases
overall performance and which is more efficient than K-means.
REFERENCES
1)
2)

3)

4)
5)
6)
7)
8)
9)
10)
11)
12)

13)

14)

S. Battiato, G. M. Farinella, E. Messina,G. Puglisi, “Robust Image Alignment for Tampering
Detection, IEEE Transactions On Information Forensics And Security, Vol. 7, No. 4, August 2012
S. Battiato, G.M. Farinella, E.Messina, And G. Puglisi, “Robust Image Registration And Tampering
Localization Exploiting Bag Of Features Based Forensic Signature,” In Proc. ACM Multimedia
(Mm’11), 2011.
S. Battiato, G. M. Farinella, E. Messina, And G. Puglisi, “Understanding Geometric Manipulations
Of Images Through Bovw-Based Hashing,” In Proc. Int. Workshop Content Protection Forensics
(Cpaf2011), 2011.
H. Farid, “Digital doctoring: how to tell the real from the fake,” Significance, vol. 3, no. 4, 2006.
Y.-C. Lin, D. Varodayan, and B. Girod, “Image authentication based on distributed source coding,”
in IEEE ComputerSociety International Conference on Image Processing, 2007.
W. Lu, A. L. Varna, and M. Wu, “Forensic hash for multimedia information,” in SPIE Electronic
Imaging Symposium -Media Forensics and Security, 2010.
W. Lu and M. Wu, “Multimedia forensic hash based on visual words.” in IEEE Computer Society
International Conference on Image Processing, 2010.
S. Roy and Q. Sun, “Robust hash for detecting and localizing image tampering,” in IEEE Computer
Society International
Conference on Image Processing, 2007.
P. H. S. Torr and A. Zisserman, “Feature based methods for structure and motion estimation,” in
Proc. Int.Workshop Vision Algorithms, held during ICCV, Corfu, Greece, 1999, pp. 278–294.
M. Irani and P. Anandan, “About direct methods,” in Proc. Int. Workshop Vision Algorithms, held
during ICCV, Corfu, Greece, 1999, pp .267–277.
R. Szeliski, “Image alignment and stitching: a tutorial,” Foundations and Trends in Computer
Graphics and Computer Vision, vol. 2, no. 1, 2006.
Soraya. Tighidet and Abdelkrim Khireddine, “Evaluation of The Results of Supervised
Segmentation Using A Robust Tool: The Case of Zernike Moments Method Applied To Medical
Imaging” International journal of Computer Engineering & Technology (IJCET), Volume 3, Issue
1, 2012, pp. 106 - 114, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375, Published by IAEME
Nagham Hamid, Abid Yahya , R. Badlishah Ahmad, Osamah M. Al-Qershi, “An Improved Robust
And Secured Image Steganographic Scheme” International journal of Electronics and
Communication Engineering &Technology (IJECET), Volume 3, Issue 2, 2012, pp. 484 - 496,
ISSN Print: 0976- 6464, ISSN Online: 0976 –6472, Published by IAEME.
Sandip S. Patil, Asha P, “Classification of Emotions From Text Using SVM Based Opinion
Mining” International journal of Computer Engineering & Technology (IJCET), Volume 3, Issue 1,
2012, pp. 330 - 338, ISSN Print: 0976 – 6367, ISSN Online: 0976 – 6375, Published by IAEME
154

50120130405017 2

Recommended

Recommended

More Related Content

What's hot

What's hot (19)

Viewers also liked

Viewers also liked (9)

Similar to 50120130405017 2

Similar to 50120130405017 2 (20)

More from IAEME Publication

More from IAEME Publication (20)

Recently uploaded

Recently uploaded (20)

50120130405017 2