base paper
TRANSCRIPT
AN ENS EMBLE APPROACH FOR IMAGE AUTHENTICATION US ING LOCAL AND GLOBAL
FEATURES
ABSTRACT
An ensemble approach is developed to
detect the insertion of forged objects in an image
using Salient Region Comparison, and to determine
the background of an object, in case if an image is
cropped or damaged. The test image is compared
with that of a reference image and divided into many
salient regions; each salient region is then compared
to detect the insertion of any objects in the test image.
In addition, if an object of an original image is
cropped, using the technique Individual Pixel Value
Comparison the background of an image can be
detected. In order to achieve accuracy of the cropped
image the homogeneous and inhomogeneous regions
are found which helps in estimat ing the structure of
the removed/cropped objects.
INTRODUCTION
1.1 OVERVIEW
With the widespread use of image edit ing
software, ensuring credibility of the image contents
has become an important issue. Saliency comparison
is a technique that extracts a short sequence from the
image to represent its contents, and therefore can be
used for image authentication. Image forgery by
insertion of objects is a very serious issue faced
today. Some objects are been inserted by
unauthenticated users so as to forge the image. This
in turn creates chances of misinterpreting a wrong
decision in the area of forensic science. Any digital
image, or even non-digital images in some cases,
present details to a viewer that is not necessarily
readily d istinguishable from reality. Basically, an
individual often finds it difficu lt to determine
whether an image has been digitally altered or not.
Very few ways exist of distinguishing a picture that
has been altered from a picture that has undergone no
such alteration. Because of this fact, viewers of
images are often rightfully suspicious of the details
conveyed by images. By being too suspicious
however, one might neglect to realize the beauty or
significance of an image that contains no digital
alteration, and this beauty or significance might have
been comprehended if one simply was not so
suspicious of image details. In order to avoid such
wrong impacts an ensemble method is proposed to
detect the insertion of forged images. Initially the
image is divided into many salient regions. The
salient regions are computed for both the original and
test images. Each salient reg ion of the test and
reference images is compared to detect the insertion
in saliency map. In addition re-filling is done to
detect the background. Re-filling is very important
criteria to detect and pretend what the background
would be if the object is cropped or hided. In forensic
science some cases becomes forced to detect the
background. For such cases the proposed feature is
done by hiding or removing the homogeneous and
the non-homogeneous mixture of the images.
1.2 PROPOS ED METHOD
An ensemble approach is developed to
detect the insertion of forged objects in an image
using Salient Region Comparison, and to determine
the background of an object, in case if an image is
cropped or damaged. The test image is compared
with that of a reference image and divided into many
salient regions; each salient region is then compared
to detect the insertion of any objects in the test image.
In addition, if an object of an original image is
cropped, using the technique Individual Pixel Value
Comparison the background of an image can be
detected. In order to achieve accuracy of the cropped
image the homogeneous and inhomogeneous regions
are found which helps in estimat ing the structure of
the removed/cropped objects.
1.3 ADVANTAGES OF PROPOS ED METHOD
It consumes less time.
The insertions of the forged images are
identified along with the histograms.
Determining the accurate values.
Refilling is beneficial, to suspect the
back ground.
This is highly applicable in the areas of
forensic science.
ARCHITECTURE DIAGRAM
PROPOS ED SCHEME
A .IMAGE INS ERTION IDENTIFICATION
The proposed image salient reg ion detection
involves the following techniques
A.1 Salient region
Salient region ext racts features of objects in
images that are distinct and representative. It can also
be set as the objects in the image that attracts the
visual attention of the user. Here all the objects of the
image are taken into account as individual salient
region so as to ensure the minute insertion of objects.
The image is given a certain thresh hold value on
which the regions are categorized. The higher the
thresh hold value larger number of salient reg ions can
be generated. The steps undergone in salient region
detection are shown in the Fig 1. Detection of salient
image regions is useful for applications like image
segmentation, adaptive compression, and region-
based image retrieval. Most saliency .
Detection methods take a similar center-
versus-surround approach. One of the key decisions
to make is the size of the neighborhood used for
computing saliency.
Fig : 1 salient reg ion detection
A.2 Conversion of RGB to gray scale image
A grayscale digital image is an image in
which the value of each pixel is a single sample, that
is, it carries only intensity information. Images of this
sort, also known as black-and-white, are composed
exclusively of shades of gray, varying from black at
the weakest intensity to white at the strongest.
Grayscale images have many shades of gray in
between. Grayscale images are also
called monochromatic, denoting the presence of only
one (mono) co lor (chrome). The intensity of a pixel is
expressed within a given range between a min imum
and a maximum, inclusive. This range is represented
in an abstract way as a range from 0 (total absence,
black) and 1 (total presence, white), with any
fractional values in between.
Fig : 2 Grayscale conversions
Fig 2 depicts gray scale conversion. For
many applications of image processing, color
informat ion doesn't help us identify important edges
or other features. Color images include complexity in
code format ion. For learning image processing, it's
better to understand grayscale processing first and
understand how it applies to mult ichannel processing
rather than starting with full color imaging and
missing all the important insights.
A.4. Contrast limited adaptive equalization
The contrast of the gray scale
image is increased so that the dim objects can be
viewed clearly. Contrast is defined as the separation
between the darkest and brightest areas of the image.
Increasing the contrast increases the separation
between the dark and bright, making shadows darker
and highlights brighter. Decreasing contrast makes
the shadows highlights down to make them closer to
one another. Adding contrast usually adds “pop” and
makes an image look more v ibrant while decreasing
contrast can make an image look duller which is
shown in Fig 3. It is clear that the dimmer objects
gets brightened so that the forgery occurred can be
found to a greater extent.
Fig : 3 Contrast based adaptive equalization
A.5. Background subtraction
It also known as Foreground Detection is a
technique in the fields of image
processing and computer vision wherein an image's
foreground is extracted for further processing (object
recognition etc.). Generally an image's regions of
interest are objects (humans, cars, text etc.) in its
foreground. After the stage of image preprocessing
(which may include image denoising etc.) object
localization is required which may make use of this
technique. Background subtraction is a widely used
approach for detecting moving objects in videos from
static cameras. The rationale in the approach is that
of detecting the moving objects from the difference
between the current frame and a reference frame,
often called “background image”, or “background
model. The background detected objects are
differentiated from each other by watershed
algorithm shown in the Fig 4. This segregates each
object from its relative objects. It takes into account
of two spatial dimensions and one brightness
dimension. Watershed makes the user’s view clearer.
A.6. Image histogram
An image h istogram is a type of histogram that
acts as a graphical representation of the tonal
distribution in a digital image. It p lots the number of
pixels for each tonal value. By looking at the
histogram for a specific image a viewer will be able
to judge the entire tonal distribution at a glance.
Image histograms are present on many modern dig ital
cameras. Photographers can use them as an aid to
show the distribution of tones captured, and whether
image detail has been lost to blown-out highlights or
blacked-out shadows.
The horizontal axis of the graph represents the
tonal variations, while the vertical axis represents the
number of pixels in that particular tone. The left side
of the horizontal axis represents the black and dark
areas, the middle represents medium grey and the
right hand side represents light and pure white areas.
The vertical axis represents the size of the area that is
captured in each one of these zones. Thus, the
histogram for a very dark image will have the
majority of its data points on the left side and center
of the graph. Conversely, the histogram for a very
bright image with few dark areas and/or shadows will
have most of its data points on the right side and
center of the graph shown below in Fig 5.
Fig :5 Histogram
B. BACKGROUND REPLACEMENT
B.1 Homogeneous region and inhomogeneous
region
General idea of a h istogram to the homogeneity
domain is to detect the homogeneous region.
Uniform reg ions are identified via multilevel
thresholding on homogeneity histogram regions.
While processing the homogeneity histogram, both
local and global information is taken into
consideration. This is particularly help ful in taking
care of small objects and local variation of color
images. An efficient peak-finding algorithm is
employed to identify the most significant peaks of the
homogeneous region. Homogeneous region is define
as detecting the inner part of the forged object which
means complete inner region of the detected image or
object will be displayed using homogeneous
detection algorithm .All the similar images or objects
which is been detected as forged area will be
displayed with perfect content. Inhomogeneous
region is defined as the outline or border of the
homogeneous region which is detected earlier.
Inhomogeneous region is detected to identify the
exact border of the forged image or object.
Considering both the homogeneous region and the
inhomogeneous exact forged image or an object can
be identified shown in Fig 6.
Fig : 6 homogeneous and inhomogeneous region
B.2 Threshold value
Threshold value is the maximum value on
which a process can be carried out; here threshold
value indicates the maximum value upto which the
image can be divided into many salient regions. You
can also specify actions to be taken when a second
value (called the reset value) is reached. For some
metrics, it is appropriate to specify a reset value,
which resets the threshold and allows it to be
triggered again when the trigger value is reached. For
those thresholds, you can specify a command to be
run when the reset value is reached. For other metrics
(such as the File Status metric and the Text metric on
file monitors, and any message set on a message
monitor), you can specify to automatically reset the
threshold when the trigger command is run.
Thresholds are triggered and reset based on the value
at the time the metric collection is made. Specifying a
higher number of collection intervals in the Durat ion
field helps to avoid unnecessary threshold activity
due to frequent spiking of values.
On the New Monitor - Metrics page, the
threshold tabs provide a place for you to specify a
threshold value for each metric that you have selected
to monitor. For example, if you are creating a job
monitor, you can set your threshold values in the
following ways depending on the type of metric you
have selected. Every threshold has a current value,
which is displayed by the threshold indicator. The
value ranges from 0 - 100% and the position of the
bar in the indicator shows this value. The value of the
threshold is defined by two factors: the trigger, and
how long the threshold must have triggered for.
The number of salient region increases rapidly
by increasing the threshold value. For every increase
of 25 integers in threshold value the number of
salient regions increases by an average of 9 which is
depicted in the below in the table 1.
Threshold Value Detected Salient Regions
25 36
50 45
75 54
100 63
Table 1 threshold value for salient reg ion detection
Threshold value is the maximum value on
which a process can be carried out; here threshold
value indicates the maximum value upto which the
image can be divided into many salient regions. You
can also specify actions to be taken when a second
value (called the reset value) is reached. For some
metrics, it is appropriate to specify a reset value,
which resets the threshold and allows it to be
triggered again when the trigger value is reached. For
those thresholds, you can specify a command to be
run when the reset value is reached.
B.3 Filtering the image
Filtering the image is the process of removing
the noise from an image.
The algorithm proposed is an extension of Lee's local
statistics method modified to utilize local gradient
informat ion. It does not require image modeling, and
it will not smear edges and subtle details. For both
the additive and multiplicative noise cases, the local
mean and variance are computed from a reduced set
of pixels depending on the orientation of the edge.
Consequently, noise along the edge is removed, and
the sharpness of the edge is enhanced. UINTA
automatically discovers the statistical properties of
the signal and can thereby restore a wide spectrum of
images. The bilateral filter is a nonlinear filter that
does spatial averaging without smoothing edges; it
has shown to be an effective image denoising
technique. An important issue with the application of
the bilateral filter is the selection of the filter
parameters, which affect the results significantly. For
practical applications when the noise variance is
spatially varying and unknown, an adaptive filtering
algorithm is developed. Experiments show its good
potential for processing real-life images. Examples
on images containing 256×256 pixels are given to
substantiate the theoretical development. Filtering is
used to remove the unwanted noise, frequency etc.,
The filter thus removes the unwanted noise at a larger
scale by doing so the image will be free of
noising...clear view of the image is obtained.
C.CONCLUS ION AND FUTURE
ENHANCEMENT
The proposed method clearly detects the
insertion of objects and identification of tampered
location in the saliency map of the image. The
number of salient region is increased for ease
identification of insertion. Even the insertion of
minute objects in the image can be detected
accurately. It provides the way for the detection of
background of the image when the object in the
image is cropped. It is applicable also when the
objects in the image is damaged because of any
external noise. Gaussian vectors are found which
depicts accuracy in results. This paper is highly
beneficial in the areas of forensic science and crime
investigation. Also when secure images have to be
sent this technique can be used. In future the paper
can extended by suspecting the background of color
images and to identify the other forgeries such as
removal and color modification in the saliency of the
image.
D.REFERENCES
[1] V. Monga, A. Banerjee, and B. L. Evans, “A
clustering based approach to perceptual image
hashing,” IEEE Trans. Inf. Forensics Security, vol.1,
no. 1, pp. 68–79, Mar. 2006.
[2] S. Xiang, H. J. Kim, and J. Huang, “Histogram-
based image hashing scheme robust against
geometric deformations,” in Proc. ACM Mult imedia
and Security Workshop, New York, 2007, pp. 121–
128.
[3] Z. Tang, S.Wang,X. Zhang, W.Wei, and S. Su,
“Robust image hashing for tamper detection using
non-negative matrix factorization,” J. Ubiquitous
Convergence Technol., vol. 2, no. 1, pp. 18–26, May
2008.
[4]A. Swaminathan, Y. Mao, and M. Wu, “Robust
and secure image hashing,” IEEE Trans. Inf.
Forensics Security, vol. 1, no. 2, pp.215–230, Jun.
2006.
[5] Y. Lei, Y.Wang, and J.Huang, “Robust image
hash in Radon transform domain for authentication,”
Signal Process.: Image Communication ., vol. 26, no.
6, pp. 280–288, 2011.
[6]F. Khelifi and J. Jiang, “Perceptual image hashing
based on virtual watermark detection,” IEEE Trans.
Image Process ., vol. 19, no. 4, pp.981–994, Apr.
2010.
[7] V. Monga and M. K. Mihcak, “Robust and secure
image hashing via non-negative matrix
factorizations,” IEEE Trans. Inf. Forensics Security,
vol. 2, no. 3, pp. 376–390, Sep. 2007..
[8]Z. Tang, S. Wang, X. Zhang, W. Wei, and Y.
Zhao, “Lexicographical framework for image
hashing with implementation based on DCT and
NMF”.