Fusion by Biometrics
主講人：李佳明、陳明暘
指導教授：林維暘
Outline









Introduction
Biometric system
Feature extraction
The advantage of verification in biometrics
The flow of verification
Fusion methods
Experiment Results
Conclusion
Reference
Introduction


Multimodal biometrics systems consolidate the
evidence presented by multiple biometric
sources and typically provide better recognition
performance compared to systems based on a
single biometric modality.
Multi-biometrics systems provide anti-spoofing
measures by making it difficult for an intruder
to spoof multiple biometric traits
simultaneously.
Biometrics
Comparison of various
biometric technologies
(H=High, M=Medium, L=Low)
Biometrics
Universality
Uniqueness
Permanence
Collectability
Performance
Acceptability
Circumvention
Face
H
L
M
H
L
H
L
Fingerprint
M
H
H
M
H
M
H
Hand geometry
M
M
M
H
M
M
M
Keystrokes
L
L
L
M
L
M
M
Hand veins
M
M
M
M
M
M
H
Iris
H
H
H
M
H
L
H
Retinal scan
H
H
M
L
H
L
H
Signature
L
L
L
H
L
H
L
Voice
M
L
L
M
L
H
L
facial
thermogram
H
H
L
H
M
H
H
Odor
H
H
H
L
L
M
L
DNA
H
H
H
L
H
L
L
Gait
M
L
L
H
L
H
M
Ear recognition
M
M
H
M
M
H
M
The advantage of Multimodal Biometric

Multiple biometric sources enhance matching
performance.

Reducing failure to enroll rate.

Difficult to spoof multiple traits simultaneously.
A biometric system
A biometric system

A biometric-based authentication system
operates in two modes

1. Enrollment mode

2. Authentication mode
A biometric system

1. Enrollment：


A user’s biometric data is acquired using a biometric
reader and stored in a database.
The stored template is labeled with a user identity
(e.g., name, identification number, etc.) to facilitate
authentication.
A biometric system

2. Authentication：

A user’s biometric data is once again acquired and
the system uses this to either identify who the user
is, or verify the claimed identity of the user.


Identification：Comparing the acquired biometric
information against templates corresponding to all users in
the database.
Verification：Comparison with only those templates
corresponding to the claimed identity.
A biometric system
Feature extraction




Fingerprint
Face
Hand Geometry
Iris
Feature extraction

Fingerprint friction ridge details are generally
described in a hierarchical order at three
different levels：




Level 1 - pattern
Level 2 - minutia points
Level 3 - pores and ridge contours
Automated Fingerprint Identification Systems
(AFIS) currently rely only on Level 1 and Level
2 features.
Feature extraction

Level 1 features, or patterns, are the macro
details of the fingerprint such as ridge flow
and pattern type.
Feature extraction

Level 2 features, or points, refer to the Galton
characteristics or minutiae, such as ridge
bifurcations and endings.
Feature extraction

Level 3 features, or shape, include all
dimensional attributes of the ridge such as
ridge path deviation, width, shape, pores,
edge contour, incipient ridges, breaks, creases,
scars, and other permanent details.
Feature extraction

Fingerprint image resolution. The same
fingerprint captured at three different image
resolutions



(a) 380 ppi (Identix 200DFR)
(b) 500 ppi (CrossMatch ID1000)
(c) 1,000 ppi (CrossMatch ID1000).
Feature extraction


Different levels of fingerprint features detected.
Level 3 features are matched using the ICP
algorithm.
Feature extraction
Feature extraction

Reference point (X), the region of interest, and
80 sectors (B = 5, k = 16) superimposed on a
fingerprint
Feature extraction


Face recognition is an important biometric
identification technology. Facial scan is an
effective biometric attribute/indicator.
The performance of face recognition systems
dependent on consistent conditions such as
lighting, pose and facial expression.
Feature extraction

Preprocessing


幾何位置的調整 - 取人臉影像中兩個control point，
分別為左眼的中心點和右眼的中心點，利用這兩個控
制點。
明亮度的調整 - histogram equalization，此步驟是為
了縮小各張影像之間亮度的改變所造成的差異
Feature extraction


擷取三個人臉區域，在每個區域裡面，全部的
影像灰階值都會被儲存在一個向量，該向量就
是該區域的特徵向量。
利用了 Principal Component Analysis (PCA)將
特徵向量降維。
Feature extraction

Automatic feature extraction for 3D face
matching.
Feature extraction

Segmentation of facial scan.
Feature extraction

For a frontal facial scan, nose tip usually has
the largest z value.
Feature extraction

Pose angle quantization.

Example of directional maximum.
Feature extraction

extracted nose profiles.
Feature extraction

Feature extraction results which lead to
correct 3D face matches.
Feature extraction



Some biometrics may not be acceptable for
the sake of protecting an individual's privacy.
As hand geometry information is not very
distinctive, it is distinctive enough for
verification but not for identification.
It is simple method of sensing which does not
impose undue requirements on the imaging
optics.
Feature extraction


Hand geometry sensing device.
5 images of the same hand are taken.
Feature extraction


Hand shape alignment
We represent the shape of a hand by a set of
ordered points in the Euclidean plane.
Feature extraction

The fourteen axes along which feature values
are computed.
Feature extraction



The iris is a protected internal organ whose
random texture is complex, unique, and stable
throughout life .
It can serve as a kind of living passport or
password that one need not remember but can
always present.
"Biometric Personal Identification System
Based on Iris Analysis." U.S. Patent No.
5,291,560 issued March 1, 1994 (J. Daugman).
Feature extraction

Finding an Iris in an Image



minimum of 70 pixels in iris radius. Iris radius of 80
to 130 pixels has been more typical.
Monochrome CCD cameras (480 x 640) have been
used.
using a coarse-to-fine strategy terminating in singlepixel precision estimates of the center coordinates
and radius of both the iris and the pupil.
Feature extraction

The outline overlay shows results of the iris
and pupil localization.
Feature extraction

Iris Feature Encoding by 2D Wavelet
Demodulation.


Each isolated iris pattern is then demodulated to
extract its phase information using quadrature 2D
Gabor wavelets.
This process is repeated all across the iris with
many wavelet sizes, frequencies, and orientations,
to extract 2,048 bits.
Feature extraction

Advantages of the Iris for Identification







Highly protected, internal organ of the eye.
Externally visible;
high degree of randomness .
Pre-natal morphogenesis (7th month of gestation)
Limited genetic penetrance of iris patterns
Patterns apparently stable throughout life
Encoding and decision-making are tractable
Feature extraction

Disadvantages of the Iris for
Identification






Small target to acquire from a distance
Located behind a curved, wet, reflecting surface
Obscured by eyelashes, lenses, reflections
Partially occluded by eyelids, often drooping
Deforms non-elastically as pupil changes size
Illumination should not be visible or bright
A biometric system has four important components
1. Sensor module :

Acquire the biometric data of an individual.
2. Feature extraction module :

Acquire data is processed to extract feature values.
A biometric system has four important components
3. Matching module :

Feature values are compared against those in the
template by generating a matching score.
4. Decision-making module :

The user’s identity is established or a claimed identity
is either accepted or rejected based on the matching
score generated in the matching module.
Fusion in biometrics
(1) Fusion at the feature extraction level :



1. The data obtained from each sensor is used to
compute a feature vector.
2. Concatenate the two vectors into a single new
vector.
3. Feature reduction techniques may be employed.
Multimodal biometric system
A prototype multimodal biometric system.
Fusion in biometrics
(2) Fusion at the matching score level :


Each system provides a matching score indicating
the proximity of the feature vector with the
template vector.
These scores can be combined to assert the veracity
of the claimed identity.
Fusion in biometrics
(3) Fusion at the decision level:


Each sensor can capture multiple biometric data and
the resulting feature vectors individually classified
into the two classes –– accept or reject.
A majority vote scheme can be used to make the
final decision.
Fusion in biometrics

Fusion in the context of biometrics can take
the following forms :

(1) Single biometric multiple representation.

(2) Single biometric multiple matchers.

(3) Multiple biometric fusion.
Fusion in biometrics

(1) Single biometric multiple representation.


This type of fusion involves using multiple
representations on a single biometric indicator.
Typically, each representation has its own classifier.
Fusion in biometrics

(2) Single biometric multiple matchers.

It is also possible to incorporate multiple matching
strategies in the matching module of a biometric
system and combine the scores generated by these
strategies.
Fusion in biometrics

(3) Multiple biometric fusion.


By integrating matching scores obtained from
multiple biometric sources.
These include majority voting, sum and product
rules, k-NN classifiers, SVMs, decision trees,
Bayesian methods, etc.
Fusion in biometrics

(4) Others


1. Store multiple templates in database.
Example : A fingerprint biometric system may store
multiple templates of a users fingerprint (same
finger) in its database. When a fingerprint
impression is presented to the system for
verification, it is compared against each of the
templates, and the matching score generated by
these multiple matchings are integrated.
Fusion in biometrics

(4) Others


2. A system may store a single template of a users
finger, but acquire multiple impressions of the
finger during verification.
3. Another possibility would be to acquire and use
impressions of multiple fingers for every user.
Experiment Results



50 users
Biometrics : fingerprint , face , hand geometry
Sum rule
Experiment Results
Experiment Results
Experiment Results
Experiment Results
Conclusion


Multimodal biometric systems provide better
recognition performance.
Different users tend to adopt differently to
individual biometric indicators. These weights
can be learnt over time by examining the
stored template of the user.
Reference
1. http://en.wikipedia.org/wiki/Biometric
2. Score Normalization in Multimodal Biometric
Systems (by Anil Jain , Karthik Nandakumar ,
Arun Ross)
3. Information fusion in biometrics (by Arun Ross ,
Anil Jain)
4. http://biometrics.cse.msu.edu/
Reference
5. http://www.cl.cam.ac.uk/~jgd1000/
6. A. K. Jain and N. Duta, "Deformable matching of hand
shapes for verification", Proceedings of IEEE
International Conference on Image Prcoessing,
October 25-28, Kobe, Japan, 1999.
7. A. K. Jain and N. Duta, "Deformable matching of hand
shapes for verification", Proceedings of IEEE
International Conference on Image Prcoessing,
October 25-28, Kobe, Japan, 1999.
8. A.K. Jain, Y. Chen and M. Demirkus, "Pores and
Ridges: High Resolution Fingerprint Matching Using
Level 3 Features", IEEE Transactions on Pattern
Analysis and Machine Intelligence, Vol. 29, No. 1, pp.
15-27, January 2007.