25-08-2017, 09:32 PM
Human Recognition Using Multiple Fingerprints
1Human Recognition.pdf (Size: 348.7 KB / Downloads: 57)
Abstract
In this project, relative eciencies of two biometric systems are being compared. One
biometric system is based on ngerprint of one xed nger and second biometric system is based
on ngerprints of two randomly selected ngers. Harris Corner Points and Zernike moment of
that Harris corner points are used as key features for matching two ngerprints. Results are
based on comparative reading of Receiver Operating Curve of two biometric systems at optimal
thresholds for both threshold for distances between Zernike moments of Harris Corner points of
both database image and query image, and threshold for Matching Score.
Introduction
Existing resources are divided into two types: Public and Private, based on the right to access.
Public resources are open to all but Private resources needs authentication to preserve privacy of
the owner.
These are three traditional modes of authentication: i)Possessions: This is the Lowest level of
security. Keys, Photo Identity cards, passport, smartcard are the examples of possession level
authentication. ii)Knowledge: This is the Second level of security. Password, ATM PIN(personal
identication number) comes under knowledge level security, something you can remember. But
both Possession and Knowledge is transferable, so they can not be used for a place which needs high
security. iii)Biometrics: Biometrics is the highest level of security system.
Biometrics
In this mode of authentication, we use a person's characteristics, something he/she does, something
he/she is.
There are two types of biometric systems, based on Physiological or Behavioural Characteristics.
Physiological characteristics are Face, Iris, Ear, Fingerprint, Palm print etc. Behavioural character-
istics are Voice, Keystroke and Signature.
In any biometric system[1], there are four important modules: (i) Data Acquisition Module: col-
lecting data from users (ii)Feature Extraction Module: In this module input data get processed
to extract key feature, which could be used for matching. (iii)Matching Module: In this module
key features of two inputs get compared to nd out if they are same or not. Based on matching
algorithms match score get calculated. (iv)Decision Module: At the end, based on some thresholds
results get dened, if images are matched or not.
Multi-modal Biometrics
Multi-modal biometric systems were introduced to reduce problems faced with uni-modal biometric
systems. Multi-modal biometric system uses two or more entries of same or dierent biometrics to
provide decision.
As it is discussed earlier that a biometric system possesses four modules, so Multi-modal biometric
system can be developed by merging biometric data at any one of the four levels. At the level
of, i)Data Acquisition Module: two input les can be superimposed to get resultant input data
for system; ii)Feature Extraction Module: Features extracted from two input data can be merged;
iii)Matching Module: Match Scores obtained for inputs, can be merged; iv)Decision level module:
Decisions coming for two inputs can be merged using logic gates.
Fingerprints
A ngerprint[9] image is composed of many lines, these lines are known as ridges. Our nger surface
has two type of structures in it, upper skin layer segment is called as ridges and lower skin layer
segment is known as valley. Due to dierent orientations, ridges form many shapes which is called
as minutia points. Two mostly used, for matching process, type of minutia points are ridge ending
and ridge bifurcation. Other kind of minutia points arei)dots,(ii)islands, (iii)lakes, (iv)crossover,
(v)spurs, (vi)bridges, etc. Fingerprint Matching systems can be divided into two types:i)
Related Work
L. Sha et al.[3] has proposed a Two Stage Fusion Scheme for multiple ngerprint impressions. They
used multiple impressions of the same nger to consolidate the outcome of ngerprint matching. In
their algorithm, during matching process of query ngerprint with a template ngerprint, they are
utilizing the complementary information gained from multiple impression of the template ngerprint.
They have implemented fusion at two levels one at matching score level and second at decision level.
Most of the existing ngerprint biometric systems uses minutia points for matching. In those al-
gorithm two reference minutia points get selected one from query image and other from template
image, then another set of minutia points get selected based on those two set of minutia points
system aligns two ngerprints and perform matching process. But output for such system is only
satisfactory for small regions. E. Zhu et al.[4] proposed a new system with global alignment of
minutiae points. For that they used multiple reference points instead of single reference point as
general ngerprint biometric systems.
Other related works will be discussed in next section during detailed discussion of the proposed
system.
Results
Performance of the proposed system is still not good but I am still modifying scoring algorithm to
get better results. ROC(Receiver Operating Curve) curve for both biometric systems; rst system
with single ngerprint and second system with two random ngers.
Final result is based on the combined left thumb and left index nger. Result is still not very good
but better than the single ngers. ROC curve for two ngers.