|
uttuswami
|
Posted: Sunday 24, June 2012 03:32:55 PM
|
|
|
This project is based on Score level fusion of two biometric tr
aits: Face and Finger
1) For face recognition we have used Principle Component Analys
is (i.e. Eigen vector generation) and for fingerprint recogniti
on also we have used the concept of eigen vectors.
2) For Face recognition we have used Open CV library (for some
basic image processing operations)
3) For finger print we have used Open Source AFIS library
4) Both Face and Finger modules generate scores which are fused
using Sum-Rule based weighted equation.
5) This project is completely finished (100% WORKING)
6) Language used to develop: C# (.net framework)
7) For face we have used C++
8) You can download this project from the following links:
https://rapidshare.com/files/3972644236/Multimodal_Biometri
c_Authentication_System.rar]https://rapidshare.com/files/397264
4236/Multimodal_Biometric_Authentication_System.rar
or
http://www.filefactory.com/file/1nva...ion_System_rar
or
http://www.mediafire.com/?1rkmdh2poa2tx20
or
http://www2.zippyshare.com/v/28167122/file.html
9) This project can be used for Research work and can be used f
or reference. This project work is my own work with some open s
ource content.
10) Price for the source code is negotiable.
Detailed Description:
Biometric systems make use of the physiological and/or behavior
al traits of individuals, for recognition purposes. These trait
s include fingerprints, hand-geometry, face, voice, iris, retina
, gait, signature, palm-print, ear, etc. Biometric systems that
use a single trait for recognition (i.e., unimodal biometric s
ystems) are often affected by several practical problems like n
oisy sensor data, non-universality and/or lack of distinctivene
ss of the biometric trait, unacceptable error rates, and spoof
attacks. Multimodal biometric systems overcome some of these pr
oblems by consolidating the evidence obtained from different so
urces. These sources may be multiple sensors for the same biome
tric (e.g., optical and solid-state fingerprint sensors), multip
le instances of the same biometric (e.g., fingerprints from diff
erent fingers of a person), multiple snapshots of the same biome
tric (e.g., four impressions of a user’s right index finger), mu
ltiple representations and matching algorithms for the same bio
metric (e.g., multiple face matchers like PCA and LDA), or mult
iple biometric traits (e.g., face and fingerprint).
A Unimodal Biometric System (UBS) is usually more cost-efficien
t than a multimodal biometric system. However, it may not alway
s be applicable in a given domain because of the limitations an
d problems like skin dryness, disease, data quality, pressure,
dirt, oil, etc. Implementing an authentication based on weighte
d multimodal system gives not only high efficiency and performa
nce but also allows the administrator to adjust ratio of weight
s as required. Generally feature matching or projecting input o
n template generates a score which may be non homogeneous. So i
n that case to fuse two or more traits, score level normalizati
on (numerical scaling) is performed to overcome the limitation
of incompatibility of scores. Whereas in our system; the input
is continuously projected on the template to record % (percenta
ge) of accuracy or confidence based on the least distance (Eucl
idean Distance) measurement in finding neighbors (specifically
in case of face verification). We record a hundred values; and,
in a divide and conquer fashion, mean accuracy scores are stor
ed. These scores are then multiplied with a floating point numb
er ‘n’ typically less than 1, which are then added with the mul
tiplication of another biometric score and ‘1-n’.
For face verification we have used High quality 1/4 CMOS sensor
- 480K pixels (Interpolated 8M pixels still image) and for read
ing finger prints we used optical sensor with 0.14 sec (continu
ous) / 0.20 sec (snap-shot) imaging speed. Face verification is
based on the fundamental concept of 2D model i.e. Principal Co
mponent Analysis. It is a mathematical procedure that performs
a dimensionality reduction by extracting the principal componen
ts of the multi-dimensional data. Fingerprint verification is b
ased on minutiae extraction and Eigen vectors formation.
A single modal biometric system has many limitations due to sen
sitivity to noise, pressure, dryness, data quality, oil, dirt,
etc and moreover it results in relatively high False Acceptance
Rate. A robust and efficient biometric authentication system s
hould have high genuine acceptance rate and a low false accepta
nce rate. Multimodal biometric systems are those which utilize,
or which have capability of utilizing, more than one physiolog
ical or behavioral characteristic for enrollment, verification,
or identification. The problem can be better understood with t
he help of following Venn diagram.
Venn diagram showing intersection of feature sets of a Genuine
person and Imposter
G is the set of feature vectors of a genuine person and I is th
e set of feature vectors of an Imposter. Intersection in this d
iagram depicts that some feature vectors of an Imposter may app
ear same as the feature vectors of genuine person due to intra-
class variability, sensitivity to noise, and vibrations in inpu
t (distortions), which may lead to false acceptance [1].
However, weighted fusion of two or more biometric traits will r
educe the effect of this overlapping (false acceptance), and wi
ll actually minimize the false acceptance and also increases ge
nuine acceptance rate. Adjusting threshold values changes the a
mount of overlapping i.e. G I. If the threshold for matching is
kept low than G I increases and if threshold is kept high then
G I decreases, but genuine acceptance also decreases.
|
|
Back to Top
^ |
|
|