Forgery Authentication

8/9/2019 Forgery Authentication

1/7

Forgery Authentication Using Distortion Cue

and Fake Saliency Map

Abstract

The widespread availability of photo manipulation software has made it

unprecedentedly easy to manipulate images for malicious purposes. Image

splicing is one such form of tampering. In recent years, researchers have

proposed various methods for detecting such splicing. In this paper, we

present a novel method of detecting splicing in images, using discrepancies

in motion blur. We use motion blur estimation through image gradients in

order to detect inconsistencies between the spliced region and the rest of

the image. We also develop a new measure to assist in inconsistent region

segmentation in images that contain small amounts of motion blur.

Experimental results show that our technique provides good segmentation of

regions with inconsistent motion blur. We also provide quantitative

comparisons with other existing blur-based techniques over a database of

images. It is seen that our technique gives signicantly better detection

results.

Distortion is often considered as an unfavorable factor in most image analysis. Hoever! it is

undeniable that the distortion reflects the intrinsic property of the lens! especially! the e"treme

ide#angle lens! hich has a significant distortion. $n this paper! e discuss ho e"plicitlyemploying the distortion cues can detect the forgery ob%ect in distortion image and make the

folloing contributions& '( a radial distortion pro%ection model is adopted to simplify the


2/7

traditional captured ray#based models! here the straight orld line is pro%ected into a great

circle on the vieing sphere) *( to bottom#up cues based on distortion constraint are provided

to discriminate the authentication of the line in the image) +( a fake saliency map is used toma"imum fake detection density! and based on the fake saliency map! an energy function is

provided to achieve the pi"el#level forgery ob%ect via graph cut. ,"perimental results on

simulated data and real images demonstrate the performances of our method

Existing System

The power of the visual medium is compelling and so, malicious tampering

can have signicant impact on people!s perception of events. "isleading

images are used for introducing psychological bias, sensationali#ing news,

political propaganda, and propagating urban myths. The image in $ig. %,

ta&en from '%(, is an instance of the latter. This photograph was widely

circulated via e-mail, supposedly having been obtained from a camera found

in the debris of the World Trade )enter buildings after the attac&s of

*eptember %%, +%. The approaching aircraft in the bac&ground seems to

imply that this image was captured mere seconds before the impact.

owever, this image is clearly fa&e.

"any techniques have been developed to discover splicing and compositing

of images . *tatistical analyses and lighting inconsistencies. may be used in

order to detect image tampering. ther methods involve exploiting certain

features of images which are characteristic of the imaging systems, formats,

and the environment

Proposed System


3/7

This paper is an extension of our wor& reported . We change the motion blur

estimation technique from spectral matting to image gradients for faster

processing. We rene the segmentation process in order to provide better

results and deal with cases of more complex blurs.We develop new measures

in order to reduce the amount of human intervention needed in our

technique and improve its robustness. We also provide a detailed

quantitative analysis of the e/ciency of our technique and test our

technique on a database of images.

e present a novel -inary artitioning /ree analysis ork for Semi#supervised image

segmentation! hich shos the suitability for the application of forged ob%ect e"traction. From

an over#segmentation result and the generated -/! the proposed -/ analysis algorithm

automatically selects an appropriate subset of nodes to represent a more meaningful

segmentation result. 0ne is that a novel dissimilarity measure considering the impact of color

difference! area factor and ad%acency degree in a unified ay is proposed for region merging and

used in the -/ generation process. /he other is the proposed -/ analysis algorithm! in hich

the node evaluation is designed to reasonably identify forged regions! and the folloing to#

phase node selecting process guarantees a meaningful segmentation result possibly reserving

forged regions. As an unsupervised image segmentation approach! our approach improves the

segmentation performance from the vie of forged ob%ect e"traction.

Modules

1. Image Preprocessing

/he image division operator normally takes to images as input and produces a third hose pi"el

values are %ust the pi"el values of the first image divided by the corresponding pi"el values of the

second image. Many implementations can also be used ith %ust a single input image! in hich

case every pi"el value in that image is divided by a specified constant. /he pi"el values are


4/7

actually vectors rather than scalar values 1e.g.for color images( than the individual components

1e.g.red! blue and green components( are simply divided separately to produce the output value.

/he division operator may only implement integer division! or it may also be able to handle

floating point division. $f only integer division is performed! then results are typically rounded

don to the ne"t loest integer for output. /he ability to use images ith pi"el value types other

than simply 2#bit integers comes in very handy hen doing division.

2. Edge Classifcation

/he given image is represented in scale#space by repeatedly smoothing ith a 3aussian

filter of increasing si4e. '* levels in scale space are e"tracted. /hen! the features5

locations and scales are decided by using scale#adapted 6aplacian of 3aussian 16o3( !

hich can indicate gradient changes and corners in the image. For the 6o3 filter! the

output of the standard 6o3 filter is convolved ith a scale#adapted 3aussian filter!

hereas a similar approach is adopted for the 3uassian filter by using a multi#scale

3uassian operator. 6ikely candidates for feature points are selected based on having

certain values resulting from the 6o3 and Harris filters. For points having a Harris filter

response greater than a prespecified threshold! e sort the points in decreasing order of

6o3 filter response. /he 7 points ith the ma"imum responses 1sub%ect to meeting

e"clusion 4one criteria( are selected for signature e"traction. An e"clusion 4one of M

pi"els is used around each feature to maintain a minimum spatial distance beteen to

features. /his is important because it is 8uite likely that points shoing the highest

response to the filters tend to come from the same ob%ects or regions! and e ish to

avoid clustering of features in certain areas of the image.

Feature Extraction:

A circular region around each feature point 1ith a radius dependent on the scale( is

scaled to a radius of +* pi"els! alloing for robustness to scale changes. /his region is

then sub%ected to the trace transform ! hich is a generali4ation of the 9adon transform. $t

generali4es the 9adon transform by alloing for various functionals 1in addition to the

integral used in the 9adon transform( to be applied over straight lines in the circular

region. -y a proper choice of a functional! invariance to rotation and scaling can be


5/7

achieved. /he trace transform is computed by draing samples 1inde"ed by t( along lines

parameteri4ed by distance

RDP Model Generation

$n the 9Dmodel! the straight orld line is first pro%ected to a great circle on the vieing sphere

and then pro%ected to a conic on the image plane. /his pro%ection! here the purple lines denote

the pro%ection of a straight orld lines . /he conic is fitted from the pro%ected points of purple

orld line. /he points represent to endpoints and one midpoint of the candidate line and are

back#pro%ected into points ! respectively. /he points define a great circle ! hich is pro%ected to

the straight line in the space ith the center of the vieing sphere. $n contrast! the points lie on

the forgery line! hich could be presented as a conic section on the forgery ob%ect. /he three

points are back#pro%ected to points on the vieing sphere! hich defines another circle . 7ote

this circle is not a great circle on the vieing sphere! because the forgery line is not satisfying

this geometric constraint any more. $n other ords! the circle generated by the forgery line

determines a plane section! hich cuts the vieing sphere ithout passing the center of the vie

sphere! as shon as the red dotted circle . /hanks to this constraint of line! there are to effective

bottom#up cues! the volume cue and the distance cue! hich could be used to distinguished the

forgery line.

Volume Cue Extraction

:ith the geometrical constraint! if three points on the image plane are pro%ected from the straight

orld line! their back#pro%ected points on the vieing sphere follo the geometrical constraint

here ;olume is the volume of tetrahedron made up by the center of vieing sphere and points

of the great circle ! as shon as the purple shado. Statistically! the forgery line is highly

unlikely to satisfy this geometrical constraint! as the blue shado. So the volume cue is defined

as a bottom#up cue here is the length of the forgery line in the image. the curves of the volume

cue of each candidate line ith different virtual focal length of the 9D model! here the

volume cues of original lines are close to 4ero and the cues of forgery lines are more likely to be

aay from the 4ero. -ut in the e"periments! the volume cue of the forgery line! hich locates

aay from the image center! is not distinguished clearly ith the cues of original lines! such as


6/7

the right#most candidate line in the second sample. 7ote that ith the virtual focal length ! the

9D model degenerates and all the distortion values e8ual to 4ero! because the pro%ective

transformation of the Step * is invalidated and the incidence angle of every pi"el vanishes

identically .

Distance Cue Estimation

Same as the volume cue! the distance beteen the center and the plane! hich is defined by three

pro%ected points on the vieing sphere! is also considered as a cue. Suppose the matri" is

composed of a general point and three points on the vie sphere hich define the plane . Since

for points on e can read off the plane coefficients

FAKE SALIENCY MAP GENERATION

Segmentation of forgery ob%ects in an image remains a challenging problem! as the forged ob%ect

1or detected region( is usually noisy and incomplete. $n this section! e introduce a fake saliency

map to ma"imum fake detection density. Firstly! the untrustorthy likelihood is employed to

initiali4e seed. And then the fake saliency map is generated by the untrustorthy likelihood to

e"press the location of forgery ob%ect. Finally! the forgery ob%ect is segmented by minimi4ing theenergy function via graph cut! hich achieves a pi"el#level segmentation. 7ote that our fake

saliency map is not constrained to a specific choice of the forensic methods! and here e still

continue the distortion cues to describe this method.

SYSTEM SPECIFICTI!"#

$%&'%E#

Proceor : Intel Pentiu! "

RAM : # G$

Me!or% : #&' G$

Di(la% : Intel )*' E+ ,GA


7/7

SOFT-ARE:

OS : Microo.t /P (ro.eional

Tool : MATLA$ 01'

Forgery Authentication

Documents

Transcript of Forgery Authentication