Fake Multi Biometric Detection using Image Quality Assessment

IJSRD - International Journal for Scientific Research & Development| Vol. 2, Issue 07, 2014 | ISSN (online): 2321-0613
All rights reserved by www.ijsrd.com 514
Fake Multi Biometric Detection Using Image Quality Assessment
R.Brindha1
V.Mathiazhagan2
1
M.E. Student 2
Assistant Professor
1,2
Department of Electronics and Communication Engineering
1,2
Sri Eshwar College of Engineering, Coimbatore
Abstract— In the recent era where technology plays a
prominent role, persons can be identified (for security
reasons) based on their behavioral and physiological
characteristics (for example fingerprint, face, iris, key-
stroke, signature, voice, etc.) through a computer system
called the biometric system. In these kinds of systems the
security is still a question mark because of various intruders
and attacks. This problem can be solved by improving the
security using some efficient algorithms available. Hence
the fake person can be identified if he/she uses any synthetic
sample of an authenticated person and a fake person who is
trying to forge can be identified and authenticated.
Key words: Biometric Systems, Image quality assessment,
Security, Attacks, Principal Component Analysis (PCA),
Linear Discriminant Analysis (LDA), Quadrature
Discriminant Analysis, Feature extraction, classification.
I. INTRODUCTION
Biometrics refers to metrics related to human characteristics
and traits. Biometrics authentication (or realistic
authentication) is used in Computer Science as a form of
identification and access control. It is also used to identify
individuals in groups that are under surveillance. Biometric
identifiers are the distinctive, measurable characteristics and
used to label and describe individuals. Biometric identifiers
are often classified as physiological versus behavioral
characteristics. Physiological characteristics are related to
the shape of the body. Examples such as fingerprint, palm
veins, face recognition, DNA, palm print, hand geometry,
iris recognition, retina and odor/scent. Behavioral
characteristics are related to the pattern of behavior of the
person, including but not limited to typing rhythm, gait, and
voice.
No single biometric will meet all the requirements
of every possible application. The synthetic or reconstructed
samples can easily break the security of the system. Hence
multi bio metric systems are used to provide better security
against various attacks. To achieve this Image Quality
Assessment for liveness detection technique is used. This
method distinguishes between fake and real samples.
II. IMAGE QUALITY ASSESSMENT FOR LIVENESS
DETECTION
Image quality is a characteristic of an image that measures
the perceived image degradation (typically, compared to an
ideal or perfect image). Imaging systems may introduce
some amounts of distortion or noises in the signal, so the
quality assessment is an important problem.
There are several techniques and metrics that can
be measured objectively and automatically evaluated by a
computer program. Therefore, they can be classified as full-
reference (FR) methods and no-reference (NR) methods. In
FR image quality assessment methods, the quality of a test
image is evaluated by comparing it with a reference image
that is assumed to have perfect quality. NR metrics try to
assess the quality of an image without any reference to the
original one.
For example, comparing an original image to the
output of JPEG compression of that image is full-reference –
it uses the original as reference.
Fig. 1: Image Quality Assessment for fake biometric
detection
Liveness detection methods are usually classified
into one of the two groups (see Fig.1)
 Hardware based techniques
 Software based techniques
A. Hardware Based Techniques
It adds some specific device to the sensor in order to detect
particular properties of a living trait (e.g., fingerprint sweat,
blood pressure, or specific reflection properties of the eye).
B. Software-Based Techniques
In this case the fake trait is detected once the sample has
been acquired with a standard sensor (i.e., features used to
distinguish between real and fake traits are extracted from
the biometric sample, and not from the trait itself).
The two types of methods present certain
advantages and drawbacks over the other and, in general, a
combination of both would be the most desirable protection
approach to increase the security of biometric systems. As
a coarse comparison, hardware-based schemes usually
present a higher fake detection rate, while software-based
techniques are in general less expensive (as no extra device
is needed), and less intrusive since their implementation is
transparent to the user. Furthermore, as they operate
directly on the acquired sample (and not on the biometric
trait itself), software-based techniques may be embedded in
the feature extractor module which makes them potentially
capable of detecting other types of illegal break-in attempts
not necessarily classified as spoofing attacks. For instance,
software-based methods can protect the system against the
injection of reconstructed or synthetic samples into the
communication channel between the sensor and the feature
extractor.
III. SECURITY PROTECTION METHOD
The problem of fake biometric detection can be seen as a
two-class classification problem where an input biometric
sample has to be assigned to one of two classes: real or fake.

(IJSRD/Vol. 2/Issue 07/2014/113)
The probability of an image can be obtained through feature
extraction. Some of the general image quality measures is
considered. In order to keep its generality and simplicity,
the system needs only one input: the biometric sample to be
classified as real or fake (i.e., the same image acquired for
biometric recognition purposes). Furthermore, as the
method operates on the whole image without searching for
any trait-specific properties, it does not require any pre-
processing steps (e.g., fingerprint segmentation, iris
detection or face extraction) prior to the computation of the
IQ features. This characteristic minimizes its computational
load. Once the feature vector has been generated the sample
is classified as real (generated by a genuine trait) or fake
(synthetically produced), using some simple classifiers
Fig. 2: Block diagram
IV. FACE RECOGNITION
From a digital image or a video frame from a video source a
person can be identified through the face recognition
system. This can be done by comparing selected facial
features from the image and a facial database. For this
purpose fusion of Principal Component Analysis (PCA) and
Linear Discriminant Analysis (LDA) is used.
These algorithms identify facial features by
extracting landmarks, or features, from an image of the
subject's face. These features are then used to search for
other images with matching features. Three different types
of attacks were considered:
 Print, illegal access attempts are carried out with
hard copies of high-resolution digital photographs
of the genuine users.
 Mobile, the attacks are performed using photos and
videos taken with the iPhone using the iPhone
screen.
 Highdef, similar to the mobile subset but in this
case the photos and videos are displayed using an
iPad screen with resolution 1024 × 768.
 Such a variety of real and fake acquisition
scenarios and conditions makes the REPLAY-
ATTACK DB (unwanted changes in the
background; different sizes of the heads; a black
frame due to an imperfect fitting of the attack
media on the capturing device screen) a unique
benchmark for testing anti-spoofing techniques for
face-based systems.
It is a set of two tasks:
 Face Identification: Given a face image that
belongs to a person in a database, tell whose image
it is.
 Face Verification: Given a face image that might
not belong to the database, verify whether it is from
the person it is claimed to be in the database
Fig. 3: Flow diagram for face module
A. Steps
 The input image is loaded and converted from
RGB to gray scale
 The image is preprocessed by using Gaussian filter
to remove noise and some specific distortions
 Then the gradient (horizontal & vertical texture)
and magnitude of the image is calculated
 The edges of the input image is detected by using
canny method for edge detection
 After the edges are detected the preprocessed
image is smoothened by using low pass filter
(average filter)
 The smoothened image is then normalized in both
horizontal and vertical direction
 The normalized image is then reconstructed using
anisotropic diffusion
 Finally the reconstructed image is obtained
B. Canny Method for Edge Detection
 The Canny method for edge detection uses an edge
operator to detect a wide range of edges in images.
Canny edge detection is a four step process.
 A Gaussian blur is applied to clear the image from
noise
 A gradient operator is applied for obtaining the
gradients' intensity and direction
 Non-maximum suppression determines the pixel
that is capable of detecting their neighbours
 Hysteresis thresholding finds the start and end of
the edges
C. Principal Component Analysis (PCA)
The Principal Component Analysis (PCA) is one of the most
successful techniques that have been used in image
recognition and compression.
 The purpose of PCA is to reduce the large
dimensionality of the data space to the smaller
intrinsic dimensionality of feature space, which is
needed to describe the data economically
 PCA can do prediction, redundancy removal,
feature extraction, data compression
In PCA, every image in the training set can be
represented as a linear combination of weighted
eigenvectors called as Eigen faces. A training image set

called basis function obtains these eigenvectors from
covariance matrix. The weights are found out after the
selection of a set of most relevant Eigen faces.
Recognition is performed by considering the test image
onto the subspace spanned by the Eigen faces and then
classification is done by measuring the Euclidean distance.
Fig. 4: Face recognition
 Mean and Eigen vectors finds the projected image
 Find the mean centered data
 Find covariance matrix’s Eigen vector using
surogate method to reduce computational cost
 Surogate matrix of mean subtracted
 Eigen vectors of surrogate matrix
 Sorting Eigen vectors to select the most dominant
Eigen vectors
 Normalization
 Dimensionality reduction
 Calculating the mean of training data
 Perform PCA to get Eigen faces in reduced
dimensions
 Projecting training and testing images in Eigen
space
 Recognizing images using Euclidean distance
 Calculating Euclidean distance between the
projected test image and the projection
 Distance of all centered training image is calculated
 Test image is said to have minimum distance with
its corresponding image in the training database
 Finding the recognized number whose distance is
minimum
 Generating the desired result
D. Linear Discriminant Analysis (LDA)
Linear discriminant analysis (LDA) is a kind of Fisher
analysis which is used to separate or distinguish two or more
classes of objects or events. The resulting combination may
be used as a linear classifier or, more commonly, for
dimensionality reduction before later classification.
 There has been a tendency in the computer vision
community to prefer LDA over PCA
 This is mainly because LDA deals directly with
discrimination between classes while PCA does not
pay attention to the underlying class structure
 This paper shows that when the training set is
small, PCA can outperform LDA
 When the number of samples is large and
representative for each class, LDA outperforms
PCA
 Only linearly separable classes will remain
separable after applying LDA.
 It does not seem to be superior to PCA when the
training data set is small.
E. Methodology
 Generate the set of MEFs for each image in the
training set
 Given a query image, compute its MEFs using the
same procedure
 Find the k closest neighbours for retrieval (e.g.,
using Euclidean distance)
F. Most Expressive Features (MEF):
The features (projections) obtained using PCA.
G. Most Discriminating Features (MDF):
The features (projections) obtained using LDA.
V. IRIS RECOGNITION
Iris recognition is an automated method that identifies the
irises of an individual person’s eyes from any videos or
images. It uses mathematical pattern-recognition techniques
to detect the original sample of the authenticated person
Fig. 5: Iris Recognition
For the iris modality the protection method is tested
under two different attack scenarios, namely:
 Spoofing attack and
 Attack with synthetic samples
For each of the attacks a specific pair of real-fake databases
is used. Databases are divided into totally independent (In
terms of users): train set, used to train the classifier; and test
set, used to evaluate the performance of the protection
method. The classifier used for the two scenarios is based
on Quadratic Discriminant Analysis (QDA). It has better
performance than the Linear Discriminant Analysis (LDA).
It maintains the simplicity of the system.
Fig. 6: Flow diagram for iris module
A. Steps
RGB to gray scale

 The edges of the input image is detected by using
canny method for edge detection
 The features are extracted using Gabor filter
method
 Then the iris template is generated
B. Quadratic Discriminant Analysis (Qda)
Quadratic discriminant analysis (QDA) is closely related to
linear discriminant analysis (LDA), where it is assumed that
the measurements from each class are normally distributed.
Unlike LDA however, in QDA there is no assumption that
the covariance of each of the classes is identical. When the
normality assumption is true, the best possible test for the
hypothesis that a given measurement is from a given class is
the likelihood ratio test matrices will substitute the
population quantities in this formula.
C. Methodology
 Histogram Equalization
 Iris template is created - generates a biometric
template from an iris in an eye image.
 Normalization is performed
 Feature encoding performed
 Range of pupil & iris radii is defined
 Scaling factor to speed up Hough transform is
defined
 Segment iris - automatic segmentation of the iris
region from an eye image is performed. Also
isolates noise areas such as occluding eyelids and
eyelashes.
 The iris boundary is found
 Array for recording noise regions is formed
 Top eyelid is found
 Bottom eyelid is found
 For CASIA, eliminate eyelashes by thresholding
 The circular Hough transform is performed
 Find the maximum in the Hough space, and hence
the parameters of the circle
 Canny edge detection is performed
 Convolving each row of an image with 1d log-
Gabor filters
 Encode - generates a biometric template from the
normalized iris region,
 Also generates corresponding noise mask
 Convolve normalized region with Gabor filters
VI. FINGERPRINT RECOGNITION
Fingerprint recognition or fingerprint authentication refers
to the automated method of verifying a match between two
human fingerprints. Fingerprints are one of many forms of
biometrics used to identify individuals and verify their
identity.
The analysis of fingerprints for matching purposes
generally requires the comparison of several features of the
print pattern. These include patterns, which are aggregate
characteristics of ridges, and minutia points, which are
unique features found within the patterns. It is also
necessary to know the structure and properties of human
skin in order to successfully employ some of the imaging
technologies.
Fig. 7: Fingerprint recognition
A. Patterns
The three basic patterns of fingerprint ridges are the arch,
loop, and whorl:
 Arch: The ridges enter from one side of the finger,
rise in the center forming an arc, and then exit the
other side of the finger.
 Loop: The ridges enter from one side of a finger,
form a curve, and then exit on that same side.
 Whorl: Ridges form circularly around a central
point on the finger.
Fig. 8: Flow diagram for fingerprint module
B. Steps
RGB to gray scale
 Histogram Equalization is performed
 Ridge is segmented by determining ridge frequency
values across the image
 Finding the median frequency value used across the
whole
 The minutiae extraction is done.
As in the iris experiments, the database are divided
into a: train set, used to train the classifier; and test set, used
to evaluate the performance of the protection method. In
order to generate totally unbiased results, there is no overlap
between both sets (i.e., samples corresponding to each user
are just included in the train or the test set). The same QDA

classifier already considered in the iris related experiments
is used here.
C. Methodology
 The image is renormalized so that the *ridge
regions* have zero mean, unit standard deviation
 Normalizes image values to 0-1, or to desired mean
and variance plot of ridge orientation data
 Placement of orientation vectors is determined
 Frequencies calculated for non-ridge regions are
masked out
 Median frequency over all the valid regions of the
image is found
 The fingerprint ridge frequency is estimated within
a small block of a fingerprint image an orientation
within the block. This is done by averaging the
sines and cosines of the doubled angles before
reconstructing the angle again. This avoids
wraparound problems at the origin.
 The image block is rotated so that the ridges are
vertical now crop the image so that the rotated
image does not contain any invalid regions. This
prevents the projection down the columns from
being mucked up
 The columns are summed down to get a projection
of the grey values down the ridges
 Peaks in projected grey values is found by
performing a gray scale
 Dilation and then finding where the dilation equals
the original values
 The spatial frequency of the ridges is determined
by dividing the distance between the 1st and last
peaks by the (No of peaks-1). If no peaks are
detected, or the wavelength is outside the allowed
bounds, the frequency image is set to 0
 Fixed angle increment between filter orientations in
degrees this should divide evenly into 180
 The filtering is done finally
VII. IMAGE QUALITY MEASURES
A. Error Sensitivity Measures
Traditional perceptual image quality assessment approaches
are based on measuring the errors (i.e., signal differences)
between the distorted and the reference images.
B. Correlation-Based Measures
Normalized Cross-Correlation (NXC), Mean Angle
Similarity (MAS) and Mean Angle- Magnitude Similarity
(MAMS).
C. Edge-Based Measures
Total Edge Difference (TED) and Total Corner Difference
(TCD).
D. Spectral Distance Measures
The Fourier transform is another traditional image
processing tool which has been applied to the field of image
quality assessment.
E. Gradient-Based Measures
Gradient Magnitude Error (GME) and Gradient Phase Error
(GPE).
F. Information Theoretic Measures
Visual Information Fidelity (VIF) and the Reduced
Reference Entropic Difference index (RRED).
VIII. APPROACHES
A. Distortion-Specific Approaches
These techniques depend on the obtained information about
the sort of visual quality losses due to certain distortion
B. Training-Based Approaches
This approach can be applied to images by training a model
using clean and distorted images
C. Natural Scene Statistic Approaches
Here to train the initial model no distorted images are used
instead a natural scene distortion free image is used
IX. EXPERIMENT AND RESULTS
The evaluation experimental protocol has been designed
with a two-fold objective:
 First, the multi-biometric dimension of the
protection method is evaluated. For this purpose
three biometric modalities have been considered in
the experiments:
 Second, the multi-attack dimension of the
protection method is evaluated. It does not detect
the spoofing attacks alone but also the fraudulent
access of the synthetic samples or reconstructed
samples.
A. Database
Databases used in this case are:
 Real database: CASIA-IrisV1. This dataset is
publicly available through the Biometric Ideal Test
(BIT) platform of the Chinese Academy of
Sciences Institute of Automation (CASIA).
 Synthetic database: WVU-Synthetic Iris DB. Being
a database that contains only fully synthetic data, it
is not subjected to any legal constraints and is
publicly available through the CITER research
center.
B. Output
Real samples of Tom - face, iris and fingerprint
Fig. 11: Database image

Fake samples of tom- Fingerprint, face and iris
Fig. 15: Database image
C. Local Binary Pattern
Local binary patterns (LBP) are a type of feature used for
classification in computer vision. Three neighborhood
examples used to define a texture and calculate a local
binary pattern (LBP)
The LBP feature vector, in its simplest form, is
created in the following manner:
 Divide the examined window into cells (e.g. 16x16
pixels for each cell)
 For each pixel in a cell, compare the pixel to each
of its 8 neighbour’s (on its left-top, left-middle,
left-bottom, right-top, etc.). Follow the pixels along
a circle, i.e. clockwise or counter-clockwise
 Where the center pixel's value is greater than the
neighbour’s value, write "1". Otherwise, write "0".
This gives an 8-digit binary number (which is
usually converted to decimal for convenience)
 Compute the histogram, over the cell, of the
frequency of each "number" occurring (i.e., each
combination of which pixels are smaller and which
are greater than the center).
 Optionally normalize the histogram
 Concatenate (normalized) histograms of all cells.
This gives the feature vector for the window
 The feature vector can now be processed using the
Support vector machine or some other machine-
learning algorithm to classify images. Such
classifiers can be used for face recognition or
texture analysis
X. CONCLUSION
Better recognition compared to existing method. Better
Security, the multi-biometric system increases the security
level. Uni-biometric system is easy to attack but the multi-
biometric system is not so easy because attacker cannot
obtain multiple traits of the same individual. More secure
than other system. Multiple fingerprint scanner support.
Multiple IRIS Scanner support. The existing recognition
systems can only detect the presence but it is not able to
distinguish between the real and fake samples. So the
system proposed can overcome this limitation.
REFERENCES
[1] Javier Galbally, Sébastien Marcel, Member, IEEE, and
Julian Fierrez,” Image Quality Assessment for Fake
Biometric Detection: Application to Iris, Fingerprint,
and Face Recognition, vol. 23, no. 2, February 2014
[2] M. G. Martini, C. T. Hewage, and B. Villarini, “Image
quality assessment based on edge preservation,” Signal
Process. Image Commun. vol. 27, no. 8, pp. 875–882,
2012
[3] J. Galbally, C. McCool, J. Fierrez, S. Marcel, and J.
Ortega-Garcia, “On the vulnerability of face
verification systems to hill-climbing attacks,” Pattern
Recognit., vol. 43, no. 3, pp. 1027–1038, 2010
[4] J. Galbally, F. Alonso-Fernandez, J. Fierrez, and J.
Ortega-Garcia, “A high performance fingerprint
liveness detection method based on quality related
features, “Future Generat. Comput. Syst., vol. 28, no. 1,
pp. 311–321, 2012
[5] J. Galbally, J. Ortiz-Lopez, J. Fierrez, and J. Ortega-
Garcia, “Iris liveness detection based on quality related
features,” in Proc. 5th IAPR ICB, Mar./Apr. 2012, pp.
271–276.
[6] Maltoni, D. Maio, A. Jain, and S. Prabhakar, Handbook
of Fingerprint Recognition. New York, NY, USA:
Springer-Verlag, 2009.
[7] R. Cappelli, D. Maio, A. Lumini, and D. Maltoni,
“Fingerprint image reconstruction from standard
templates,” IEEE Trans. Pattern Anal.
[8] Mach. Intell., vol. 29, no. 9, pp. 1489–1503, Sep. 2007.

Fake Multi Biometric Detection using Image Quality Assessment

More Related Content

What's hot (19)

Similar to Fake Multi Biometric Detection using Image Quality Assessment (20)

More from ijsrd.com (20)

Recently uploaded (20)

Fake Multi Biometric Detection using Image Quality Assessment