A Quick Detection Of Misuse Of Celebrities Images Over The

Social Media
*Mahendra Bilagi
mahendra.bilagi@gmail.com

ABSTRACT
The images of celebrities are publicly
available in the internet. There are some cases
where celebritys photo is edited in and misused
for advertisement or some other purpose. It is the
challenging problem for detecting misused
celebrity image over billons of internet images. A
SURF & SIFT methods are used to detect images
over social networks. But it consumes more time
and an in efficient technique. Moreover SIFT do
not perform well for real time applications. This
paper proposes a model that uses a feature
descriptor method which takes an image and
output feature vectors. Feature descriptor
encoded interesting information into number
series that can be used to differentiate a feature
from another. Then it will enhance a performance
of SURF algorithm by paralleling using Parallel
MATLAB.
Keywords
Image descriptor; Feature vector; Cloud SURF
(Speeded Up Robust Feature); SIFT;Parallel
programming; Numerical finger print; Parallel
MATLAB; Shared resources;encoding; Number
series;celebrates; image descriptor;
Objectives:
To study several image descriptor
techniques.
To learn parallel computing using Parallel
MATLAB toolkit.

**Nikhil Jain
nikhiljain10008@gmail.com

To propose a method that recognizes part

of the image in the cloud which is
compared through client
To improve performance of SURF image
descriptor technique by paralysing it.
To confirm confidentiality of images in
the social media

1. INTRODUCTION
Today, the utilization of Digital pictures is
expanding quickly in just about each zone of
human life like in instruction, programming
organizations, TV, organizations, news-casting,
restorative
imaging,
furthermore,
online
networking. It is anything but difficult to learn
and comprehend anything outwardly as opposed
to just perusing or tuning in. Another perspective
is that by and large visual data is accepted to be
valid. Be that as it may as the innovation
advances and loads of modern picture preparing
instruments are accessible. it turns out to be
anything but difficult to alter visual data. Digital
picture fabrication is a developing issue in
criminal cases and out in the open course. As of
now there are no settled strategies to confirm the
credibility and respectability of Digital pictures
of celebrities over social media in a programmed
way. Distinguishing falsification in computerized
pictures is a rising examination field with
imperative ramifications for guaranteeing the
validity of Digital picture.

The goal of this work can be separated into three

main stages. First, interest points are designated
at distinct Locations in the image, such as
corners, middle, blobs, and T-junctions. The most
valued property of an interest point detector is its
repeatability, i.e. whether it Reliably finds the
same interest points under different viewing
conditions. Next, the neighborhood of every
interest point is represented by a feature vector.
This Descriptor has to be distinct and, at the same
time, strong to noise, detection errors, and
symmetrical and photometric deformations.
Finally, the descriptor vectors are matched
between different images. The matching is often
based on a distance between the trajectories, e.g.
the Euclidean distance. The Measurement of the
descriptor has a direct impact on the time this
takes, and a lower number of measurements is
therefore essential.
Detection of digital image forgery having huge
number of applications related Forensic science
article interrogative section although which is
very helpful for media, publication, law,
Military, Medical image, Astronomical image,
research and World Wide Web publications.
Photo image forgery is classified in to two
categories. The first class of image forgeries
includes images tampered by copying one area in
an image and pasting it onto another area. It is
called as Copy-Move Forgery or Cloning. The
second class of forgeries is copying and pasting
areas from one or more images and pasting on to
an image being forged. The image processing
community formally refers to this type of image
as an image composition, which is defined as
the digitally manipulated combination of at least
two source images to produce an integrated
result. It is also called as Copy-Create Image
Forgery.
Advanced picture falsification identification
strategies are characterized into active and
passive methodologies. In active approach, the
digital picture requires some re-handling, for

example, watermarking, sign, and so on. Passive

methodology is distinct to active approach; this
methodology does not require any watermark
inserted ahead of time.
The recovery of pictures from these vast
accumulations has made a test of proficient
recovery of right pictures as indicated by the
utilization. The preparing of pictures for the is not
imperative only for purposes of data retrieval
additionally for use of substance information
mining procedures by organizations facilitating
these images on the Web and trying to adapt the
substance by not just giving picture curve comes
about additionally by mining data in pictures for
promoting, client profiling and different
purposes.
SURF calculations can each be isolated into two
unmistakable parts: interest point recognition,
which endeavors to discover stable interest
focuses in the picture, and intrigue point
depiction, which produces descriptors for each
interest point found in the identification step. For
the extraction of SURF components, in this
paper, we utilize the OpenSURF to distinguish
picture treating in non-level area for its
productivity in registering.
2. RELATEDWORK
During the previous five years, a few scientists
have created distinctive procedures to distinguish
duplicate move falsification. The duplicate move
fabrication presents a relationship between the
first picture district and the stuck area.[1]. S.
Bayram, H. T. Sencar, and N. Memon define how
copy-move forgeries creates in the images. It is
easy to learn and understand anything visually
rather than only reading or listening. Another
aspect is that generally visual information is
believed to be true. But as the technology
advances and lots of sophisticated image
processing tools are available, it becomes very
easy to edit visual information. In [2], the
advanced picture imitation identification routines
are grouped into Active Digital Image Legal
sciences and Passive Digital Image Forensics or

Blind Computerized Image Forensics. SURF with

KD tree method are used for the detection of a
particular forged region. In this proposed work,
HAC method is applied with SURF for a more
accurate result in terms of all attacks In [3],
G.Li.et. al. presented a framework of sorted
neighborhood Methodology in view of DWT
(Discrete Wavelet Transformation) and SVD
(singular
value
decomposition).
These
calculations in light of block matching what's
more, are computationally intricate. Some
calculations are frail to find the duplicate move
district after copying controls, for example, lossy
pressure, obscuring or blend of these operations.
[4], The KD-tree is usually utilized structure for
scanning for closest neighbors. The KD tree preforms information into a information structure
permitting us to make proficient extent inquiries.
It is a two fold tree that stores purposes of a kdimensional space in the clears out. In every
interim point, the tree separates the kdimensional space into two sections with (k-1)
dimensional hyper plane. Assume a KD-tree
comprises of N highlight vectors, it requires O(N
log2 N) operations to be developed and O(log2
N) to be sought.
Popescu et al [5] proposed a duplicate move
picture forgery location calculation algorithm
utilizing square coordinating methodology and
Principal Component Analysis (PCA). To
recognize pictures through turn, scaling and
different operations rapidly and productively,
picture alter discovery in view of Radon and
Fourier-Mellin
change
is
introduced.
Forthcoming Y. Shih et. Al.,[6], talk about the
methods of duplicate images fabrication and look
at four identification techniques for duplicate
spread fabrication recognition, which depend on
PCA, DCT, spatial space, and measurable area.
We examine their adequacy and affectability
under the impacts of Gaussian obscuring and
lossy JPEG compressions.
Amerini et al.[7]. identify the duplication
imitation furthermore gauge the change by
RANSAC. They created g2NN closest neighbor

looking for different duplicate glue recognition.

Their technique is strong to all change assaults.
Their method likewise works adequately to splice
assault recognition. In[8], where the creators
usedMPEG7 image mark apparatuses for
removing the features. Least Median of Squares
(LMedS) calculation is utilized rather than
RANSAC for evaluating the geometrical
changes.
In[9], Yanjun Cao, TiegangGao, present a
proficient and robust way to deal with identify
such particular artifact. Firstly, the unique picture
is separated into altered size squares, and discrete
cosine change (DCT) is connected to every piece,
in this way, the DCT coefficients speak to each
block. Secondly, each cosine changed piece is
spoken to by a circle square also, four elements
are extricated to lessen the measurement of every
piece. At long last, the element vectors are
lexicographically sorted, and copied picture
pieces will be coordinated by a preset limit
esteem. So as to make the calculation more
hearty, a few parameters are proposed to uproot
the off-base comparative squares. In [10]., the
author connected Gaussian pyramid on picture to
break down it into distinctive scales. After that
every scale space is partitioned into separate
roundabout squares. From every roundabout
piece highlights are removed by Hu moment of
length 4. Components are then sorted, and
coordinating is performed on it. The execution of
this technique is not changed in the vicinity of a
rotational change.
In [11], different keypoints are chosen from a
picture. Highlight descriptors are computed from
each of these keypoints. For identifying the
duplication, fabrication coordinating is performed
on keypoint highlight descriptors. Lowe [20]
developed SIFT, which identifies keypoints and
highlights froman picture. Filter application
exists in different fields. It performs better in
examination to past descriptors
SIFT is used in the region duplication discovery
technique. the author recommended that the

keypoint coordinating technique experiences a

few issues. For evacuating these issues, SIFT
cluster matching is proposed, where articles are
coordinated instead of the point. Focuses are
assembled here utilizing agglomerative various
leveled grouping. In [12], Frank Y. Shih et. Al.,
examine the systems of duplicate spread picture
phony and think about four location strategies for
duplicate spread fraud discovery, which depend
on PCA, DCT, spatial space, and measurable
area. We examine their adequacy and sensitive
under the impacts of Gaussian obscuring and
lossy JPEG compressions.
PreetiYadav, YogeshRathore[12],proposed an
enhanced calculation in view of Discrete Wavelet
Transform (DWT) is utilized to recognize such
cloning phony. In this procedure DWT (Discrete
Wavelet Transform) is connected to the
information picture to yield a diminished
dimensional representation. After that compacted
picture is isolated into covering squares. These
pieces are then sorted and copied squares are
distinguished. Because of DWT use, location is
initially did on least level picture representation
so this Copy-Move identification procedure
expands exactness of discovery procedure.
Qiao, Sung, Liu and Ribeiro [13] introduced
another approach for duplicate move falsification
location. Their system depends on multidetermination and multi-introduction curvelet
change. Curvelet change is generally performed
in recurrence area to have a superior
effectiveness. They changed over the picture into
grayscale. In curve let change, the grayscale
picture is disintegrated into an arrangement of
sub-groups. At that point they apportioned every
sub-band into a few piece what's more, performed
ridgelet investigation on them. Ridgelet change
consolidates Radon change and the 1-D wavelet
change. Be that as it may, it is computationally
perplexing. To

Diminish the intricacy they utilized quick discrete

curvelet change. This gives a pyramid structure
with numerous introductions at different scales,
which expands the recognition execution and
precision. Multi-directional decay gives exact
connection between neighboring introductions.
They utilized these pyramid organized multi
oriented
highlight
vectors
to
perform
coordinating. To decrease the computational
multifaceted nature, they sorted the element
vectors
lexicographically.
Their
strategy
effectively distinguished copied districts even
after JPEG pressure, scaling and pivots. Be that
as it may, it can't be connected on packed
pictures. They must be decompressed some time
recently this system can be utilized. Additionally,
the pictures must be in grayscale to perform this
investigation.
3. METHODOLOGY
3.1 Overview:
This proposed method is based on a SURF
algorithm for the detection of keypoints and for
extracting their corresponding feature descriptors.
Matching is performed in-between selected
keypoints by applying best bin first search
procedure. For detecting the duplicated regions,
HAC technique is applied. An input image is
inserted to the detection system, and the output is
imaged with duplicated regions.

3.1. Region Duplication Detection Algorithm.

If the image suffers from duplication forgery,
then it contains at least two same regions, one is
copied and the other is pasted region. The overall
technique for detecting the duplicated region is

as

follows.

Input: image.
Output: detected duplicate regions with image.
(1) If RGB image, then converted into gray scale.
(2) Applying SURF method.
(a) Keypoints are detected from an image(1, 2,
3, . . . ,).
(b) From the above detected keypoints, features
areextracted (1, 2, 3, . . . ,).
(c) This matrix is stored in a variable = .
(3) For each = 1 to ,for = 1 to .
(a) If = then go to step (b); else: return.
(b) Dot products are calculated between each
feature descriptor.
End of For .
(c) Inverse cosine angle of dot products will be
computed.
(d) Sorting is applied on the result, and values are
stored
[Value, index] = sort (cos1(dot prods)).
(e) If (Value(1)/Value(2)) <0.6, then match exists,
and index will be stored.
Else: index = 0.
End of For .
(4) For each keypoint
(a) If match exists, then go to step (b);
else: return.
(b) If the matched points are far from 10 10
square region, then go to step (c);
else: return.
(c) Store the coordinates of matched points in
bydata matrix and set: flag = 1.
End of For.
(5) If flag >0, then
(a) euclidean distance computed between each
pair of objects of ;

Dist(, ) = (2 1)2 + (2 1)2; (1)

where, = (1, 1 ), = (2, 2)
(b) linkage function is applied for linking the
objects into hierarchal tree;
(c) the smallest height is taken for cutting the
hierarchal tree into clusters;
(d) a line is drawn between the matched objects
from different clusters;
(e) objects of different clusters are shown from
the different colors.
3.2 Keypoint Detection and Feature
Extraction.
Bay
etal.proposed
SURF
method
whosecomputation is faster thanSIFT. How
keypoints are detected and feature descriptor is
generated from SURF are discussed below.
3.2.1. Integral Image.
Integral image increases the computation speed
as well as the performance, its value is calculated
from an upright rectangular area. vertices are ,
, , and . Suppose that an input image and a
point (; ) are given. The integral image is
calculated by the sum of the values between the
point and the origin. The following formula is
used to calculate the integral image:
i y j y

I ( x , y )= I ( x , y ) ;
i=0 j=0

IMAGE FORGERY

ACTIVE METHOD

PASSIVE METHOD
TEMPERING

WATER MARKING
IMAGE
RETOUCHING

COPY MOVE

IMAGE
SPLICING

INPUT IMAGE

KEY POINT
DETECTION
FEATURE EXTRACTION

DOT PRODUCTS BETWEEN

FEARTURE DESCRIPTORS

LINE DRAWN BETWEEN THE

MATCHED OBJECT OF
DIFFERENT CLUSTERS

SORT THE INVERSE COSINE ANGLE

OF DOT PRODUCTS

DUPLICATE REGION DISPLAYED

SORT THE INVERSE COSINE ANGLE

OF DOT PRODUCTS

RATIO < 0.6

3.2.2. Keypoint Detection.

MATCH EXIST, STORE THE

COORDINATES

ALL KEY POINTS

PROCESSED

EUCLIDEAN DISTANCE
COMPUTED BETWEEN MATCHED
KEY POINTS

This step requires scale space generation for the

extraction of keypoints. In SURF, Laplacian of
Gaussian is approximated with a box filter.
Convolution is applied to an image with varying
size box filter for creating the scale space. After
constructing the scale space, determinant of the
Hessian matrix is calculated for detecting the
extremum point. If determinant of the Hessian
matrix is positive, that means both the Eigen
values are of the same sign either both are
negative or both are positive. In case of the
positive response, points will be taken as
extrema; otherwise, it will be discarded.
Hessian matrix is represented by
H(x,) =

TWO DIFFERENT KEY POINTS ARE HIERARCHICAL TREE IS

CREATED ON THE BASIS OF
TAKEN
DISTANCE

CLUSTERS ARE CONSTRUCTED

L xx ( x , ) Lxy ( x , )
L xy ( x , ) L yy ( x , )

where(, ) is the convolution of the Gaussian

second order derivative with the image in point
, and similarly (, ) and
(, ). These derivatives are called
Laplacian of Gaussian. The approximate
determinant of the Hessian matrix is calculated
by

Det ( H )=

Dxx D yy

( 0.9 Dxy )2

3.2.3. Orientation Assignment.

At first, a circular area is constructed around the
keypoints. Then, Haar wavelets are used for the
orientation assignment. It also increases the
robustness and decreases the computational cost.
Haar wavelets are filters that detect the gradients
in and directions. In order to make rotation
invariant, a reproducible orientation for the
interest point is identified. A circle segment of
60 is rotated around the interest point. The
maximum
value
is
chosen
as
a
dominant orientation for that particular point.
3.2.4. Feature Descriptor Generation.
For generating the descriptors, first construct a
square region around an interest point, where
interest point is taken as the center point. This
square area is again divided into 4 4 smaller
subareas. For each of these cells, Haar wavelet
responses are calculated. Here, termed as
horizontal response and as vertical response.
For each of these sub regions, 4 responses are
collected as

descriptor. As matching these keypoints with

their high dimensional feature vector 64 takes
time, therefore best bin first (BBF) method is
chosen for selecting two nearest neighbours . Dot
products are calculated between each keypoint
feature descriptor with the others. After that sort
the inverse cosine angles of dot products. Store
their values as well as their corresponding index
number. The ratio between two nearest
neighbours value is compared to a predefined
threshold. In this work, the threshold is set to 0.6,
because above this value the probability of false
matches arises. If the ratio is less thanthe given
threshold, they satisfy the similarity criterion and
match exists. In case of matching, their relative
index number will be stored. This procedure
continues for all keypoints.
3.4. Keypoint Clustering.

4 4 4 = 64.

HAC is also known as hierarchy of clusters, in

which each keypoint behaves as a single cluster
at the starting stage. Euclidean distance between
each keypoint with the remaining other keypoints
will be calculated. Merging is performed if two
clusters are dissimilar to each other. This step is
repeated until there is one cluster left or
dissimilarity criterion unsatisfied. Single,
average, and ward methods are types of linkage
used for merging and creating a hierarchal tree.
Single Linkage. It uses the smallest distance
between objects in two clusters,

3.3. Keypoint Matching.

d(A,B) = min (dist (xAi , xBj))

V sub region= [

dx , dy , |dx| , dy
]

So each sub region contributes 4 values.

Therefore, the descriptor is calculated as

A set of keypoints and their corresponding

feature descriptors are obtained from SURF. The
comparison is performed between each keypoint
with the remaining other keypoints feature

Average Linkage. It uses the average distance

between all pairs of objects in the two clusters,

d(A,B) =

nA

1
nAn

nB

dist ( x Ai x Bj )
B

images and correct it.

i=1 j=1

Ward Linkage. It is based on the increment or

decrementing the value of error sum of squares
(ESS). In other words, distance between the
clusters is the difference between the ESS for
unified cluster and ESS of the individual clusters,
d(A,B) = ESS(AB) [ESS(A) + ESS(B)],
where
nA

ESS(A) =

|x Ai x A|
i=1

Here, indicates the combined cluster,

indicates number
of objects in cluster , indicates number of
objects in
cluster B,

indicates th object in the cluster , and

indicates cancroids of cluster whose value is
calculated by

4.2 Example 2

nA

~XA =

1
x Ai
nA
i=1

4. Results
4.1 Example 1
4.3 References and Citations
There are many examples where people use these
illegal techniques to modify the rivals images for
revenge. So, this method provide to detect these

Footnotes should be Times New Roman 9-point, and justified to

the full width of the column.
Use the standard Communications of the ACM format for
references that is, a numbered list at the end of the article,
ordered alphabetically by first author, and referenced by numbers
in brackets [1]. See the examples of citations at the end of this
document. Within this template file, use the style named
references for the text of your citation.
The references are also in 9 pt., but that section (see Section 7) is
ragged right. References should be published materials accessible
to the public. Internal technical reports may be cited only if they

are easily accessible (i.e. you can give the address to obtain the
report within your citation) and may be obtained by any reader.
Proprietary information may not be cited. Private communications
should be acknowledged, not referenced (e.g., [Robertson,
personal communication]).

4.4 Page Numbering, Headers and

Footers
Do not include headers, footers or page numbers in your
submission. These will be added when the publications are
assembled.

5. FIGURES/CAPTIONS
Place Tables/Figures/Images in text as close to the reference as
possible (see Figure 1). It may extend across both columns to a
maximum width of 17.78 cm (7).
Captions should be Times New Roman 9-point bold. They should
be numbered (e.g., Table 1 or Figure 2), please note that the
word for Table and Figure are spelled out. Figures captions
should be centered beneath the image or picture, and Table
captions should be centered above the table body.

6. SECTIONS
The heading of a section should be in Times New Roman 12-point
bold in all-capitals flush left with an additional 6-points of white
space above the section head. Sections and subsequent subsections should be numbered and flush left. For a section head and
a subsection head together (such as Section 3 and subsection 3.1),
use no additional space above the subsection head.

6.1 Subsections
The heading of subsections should be in Times New Roman 12point bold with only the initial letters capitalized. (Note: For
subsections and subsubsections, a word like the ora is not
capitalized unless it is the first word of the header.)

6.1.1

Subsubsections

The heading for subsubsections should be in Times New Roman

11-point italic with initial letters capitalized and 6-points of white
space above the subsubsection head.

6.1.1.1 Subsubsections
The heading for subsubsections should be in Times New Roman
11-point italic with initial letters capitalized.

6.1.1.2 Subsubsections
The heading for subsubsections should be in Times New Roman
11-point italic with initial letters capitalized.

7. CONCLUSION
In this paper, a system was introduced for
recognizing the copy region based on SURF and
HAC. The essential picture utilized as a part of
SURF diminishes the time multifaceted nature.
SURF has less highlight descriptor dimensional

size. So that matching connected on SURF

descriptor is quicker and builds the calculation
speed also. The Haar wavelets are utilized for
highlight descriptors calculation from each key
point, so descriptors are vigorous to brightening
changes. The trial results demonstrate that SURF
highlight descriptors are invariant towards
diverse mix of scaling and turn. In the vicinity of
JPEG pressure, Gaussian clamor expansion and
gamma revision assault, this strategy gives great
result.
8. REFERENCES
[1]. S. Bayram, H. T. Sencar, and N. Memon, A
survey of copy move forgery detection
techniques, in Proceedings of the IEEEWestern
New York Image Processing Workshop, pp. 538
542, September 2008.
[2]. B.L.Shivakumar and S.SanthoshBaboo,
Digital Image Forgery Detection, SAJOSPS,
Vol. 10(2), pp. 116-119, 2010
[3]. 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 Proceedings of IEEE
International Conference on Multimedia and
Expo, Beijing China, July 2-5, 2007, pp. 17501753
[4]. V. Christlein, C. Riess, and E. Angelopoulou,
A Study on Features for the Detection of CopyMove forgeries, in GI SICHERHEIT, 2010.
[5]. A. C. Popescu and H. Farid, Exposing
Digital Forgeries by Detecting Duplicated Image
Regions, Technical Report, TR2004-515,

Department of Computer Science, Dartmouth

College, pp. 758-767, 2006.
[6]. S. Bayram, H.T. Sencar, N. Memon, An
efficient and robust method for detecting copymove forgery, in: IEEE International
Conference on Acoustics, Speech and Signal
Processing, IEEE Press, New York, 2009.
[7]. I. Amerini, L. Ballan, R. Caldelli, A. del
Bimbo, and G.Serra, A SIFT-based forensic
method for copy-move attackdetection and
transformation recovery, IEEE Transactions on
Information Forensics and Security, vol. 6, no. 3,
pp. 10991110.
[8]. P. Kakar and N. Sudha, Exposing
postprocessed
copy-pasteforgeries
through
transform-invariant features, IEEE Transactions
on Information Forensics and Security, vol. 7, no.
3, pp10181028, 2012.
[9]. Yanjun Cao a,*, TiegangGao, A robust
detection algorithm for copy-move forgery in
digital images,Forensic Science International
214 2012.

[10]. G. Liu, J. Wang, S. Lian, and Z. Wang, A

passive imageauthentication scheme for detecting
region-duplication forgerywith rotation, Journal
of Network and Computer Applications,vol. 34,
no. 5, pp. 15571565, 2011.
[11]. D. G. Lowe, Distinctive image features
from scale-invariantkeypoints, International
Journal of ComputerVision, vol. 60,no.2, pp. 91
110, 2004.
[12]. PreetiYadav, YogeshRathore, AartiYadav,
DWT Based Copy- Move Image Forgery
Detection, International Journal of Advanced
Research in Computer Science an Electronics
Engineering Volume 1, Issue 5, July 2012
[13]. Qiao, M., Sung, A., Liu, Q.,
&Ribeiro, B. (2011). Anovel approach
for detection of copy-move forgery.
Paperpresented at the ADVCOMP 2011,
The Fifth InternationalConference on
Advanced
Engineering
Computing
andApplications in Sciences.