An Overview of Biometrics

parisfawnΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 6 μήνες)

53 εμφανίσεις


Luciano Rila/RHUL

1

An Overview of Biometrics

Luciano Rila


Luciano Rila/RHUL

2

Contents


biometric systems

1.
Introduction

2.
Biometric identifiers

3.
Classification of biometrics methods

4.
Biometric system architecture

5.
Performance evaluation


Luciano Rila/RHUL

3

Contents

biometric technologies

6.
Signature recognition

7.
Voice recognition

8.
Retinal scan

9.
Iris scan

10.
Face
-
scan and facial thermogram

11.
Hand geometry


Luciano Rila/RHUL

4

Personal identification

Association of an individual with an identity:




Verification (or authentication): confirms
or denies a claimed identity.




Identification (or recognition): establishes
the identity of a subject (usually from a set
of enrolled persons).


Luciano Rila/RHUL

5

Personal identification objects



Token
-
based:
“something that you have”



Knowledge
-
based:
“something that you know”



Biometrics
-
based:
“something that you are”


Luciano Rila/RHUL

6

Biometrics

Bio + metrics:

The statistical measurement of biological data.

--

Biometric Consortium definition:

Automatically recognising a person using
distinguishing traits.


Luciano Rila/RHUL

7

Some applications



Financial security (e
-
fund transfers, ATM, e
-
commerce, e
-
purse, credit cards),



Physical access control,



Benefits distribution,



Customs and immigration,



National ID systems,



Voter and driver registration,



Telecommunications (mobile, TV)


Luciano Rila/RHUL

8

Biometric identifiers


Universality



Uniqueness



Stability



Collectability


Performance


Acceptability



Forge resistance


Luciano Rila/RHUL

9

Biometric technologies


Covered in ISO/IEC 27N2949:



recognition of signatures,



fingerprint analysis,



speaker recognition,



retinal scan,



iris scan,



face recognition,



hand geometry.



Luciano Rila/RHUL

10

Other biometric methods



Found in the literature:



vein recognition (hand),



keystroke dynamics,



palmprint,



gait recognition,



body odour measurements,



ear shape.



Luciano Rila/RHUL

11

Classification of biometrics
methods



Static:



fingerprint



retinal scan



iris scan



hand geometry




Dynamic:



signature recognition



speaker recognition



Luciano Rila/RHUL

12

Biometric system
architecture


Basic modules of a biometric system:



Data acquisition



Feature extraction



Matching



Decision



Storage



Luciano Rila/RHUL

13

Biometric system model

Raw data

Extracted
features

template

Authentication decision


Data
collection

Signal
processing

matching

storage

score

decision

Application


Luciano Rila/RHUL

14

Data acquisition module


Reads the biometric info from the user.



Examples: video camera, fingerprint
scanner/sensor, microphone, etc.



All sensors in a given system must be similar to
ensure recognition at any location.



Environmental conditions may affect their
performance.



Luciano Rila/RHUL

15

Feature extraction module



Discriminating features extracted from the
raw biometric data.


Raw data transformed into small set of
bytes


storage and matching.



Various ways of extracting the features.



Pre
-
processing of raw data usually
necessary.



Luciano Rila/RHUL

16

Matching module



The core of the biometric system.



Measures the similarity of the claimant’s
sample with a reference template.



Typical methods: distance metrics,
probabilistic measures, neural networks, etc.



The result: a number known as match
score.



Luciano Rila/RHUL

17

Decision module



Interprets the match score from the
matching module.



Typically a binary decision: yes or no.



May require more than one submitted
samples to reach a decision: 1 out of 3.



May reject a legitimate claimant or accept
an impostor.


Luciano Rila/RHUL

18

Storage module



Maintains the templates for enrolled users.



One or more templates for each user.



The templates may be stored in:



a special component in the biometric device,



conventional computer database,



portable memories such as smartcards.




Luciano Rila/RHUL

19

Enrolment



Capturing, processing and storing of the
biometric template.



Crucial for the system performance.



Requirements for enrolment:



secure enrolment procedure,



check of template quality and “matchability”,



binding of the biometric template to the
enrollee.



Luciano Rila/RHUL

20

Possible decision outcomes



A genuine individual is accepted.



A genuine individual is rejected (error).



An impostor is rejected.



An impostor is accepted (error).


Luciano Rila/RHUL

21

Errors


Balance needed between 2 types of error:


Type I
: system fails to recognise valid user
(‘false non
-
match’ or ‘false rejection’).


Type II
: system accepts impostor (‘false match’
or ‘false acceptance’).


Application dependent trade
-
off between
two error types.


Luciano Rila/RHUL

22

Pass rates


Luciano Rila/RHUL

23

Tolerance threshold


Error tolerance threshold is crucial and
application dependent.


Tolerance too large gives Type II error
(admit impostors).


Tolerance too small gives Type I errors
(reject legitimate users).



Equal error rate for comparison: false non
-
match equal to false match.


Luciano Rila/RHUL

24

Biometric technologies


Signature recognition


Voice recognition


Retinal scan


Iris scan


Face biometrics


Hand geometry


Luciano Rila/RHUL

25

Signature recognition


Signatures in wide use for many years.


Signature generating process a trained
reflex
-

imitation difficult especially ‘in real
time’.


Automatic signature recognition measures
the dynamics of the signing process.


Luciano Rila/RHUL

26

Dynamic signature recognition


Variety of characteristics can be used:


angle of the pen,


pressure of the pen,


total signing time,


velocity and acceleration,



geometry.


Luciano Rila/RHUL

27

Signature recognition:
advantages


disadvantages


Advantages:


Resistance to forgery


Widely accepted


Non
-
intrusive


No record of the
signature



Disadvantages:


Signature
inconsistencies


Difficult to use


Large templates
(1K to 3K)



Luciano Rila/RHUL

28

Fingerprint recognition


Ridge patterns on fingers uniquely identify
people.


Classification scheme devised in 1890s.


Major features: arch, loop, whorl.


Each fingerprint has at least one of the
major features and many ‘small’ features.


Luciano Rila/RHUL

29

Features of fingerprints


Luciano Rila/RHUL

30

Fingerprint recognition (cont.)


In a machine system, reader must minimise
image rotation.


Look for minutiae and compare.


Minor injuries a problem.


Automatic systems can not be defrauded by
detached real fingers.


Luciano Rila/RHUL

31

Fingerprint authentication



Basic steps for fingerprint authentication:



Image acquisition,



Noise reduction,



Image enhancement,



Feature extraction,



Matching.




Luciano Rila/RHUL

32

Fingerprint processing

a)
Original

b)

Orientation

c)

Binarised

d)

Thinned

e)

Minutiae

f)

Minutia graph


Luciano Rila/RHUL

33

Fingerprint recognition:
advantages


disadvantages


Advantages:


Mature technology


Easy to use/non
-
intrusive


High accuracy


Long
-
term stability


Ability to enrol
multiple fingers



Disadvantages:


Inability to enrol
some users


Affected by skin
condition


Association with
forensic applications



Luciano Rila/RHUL

34

Speaker recognition


Linguistic and speaker dependent acoustic
patterns.


Speaker’s patterns reflect:



anatomy (size and shape of mouth and throat),



behavioral (voice pitch, speaking style).



Heavy signal processing involved (spectral
analysis, periodicity, etc)



Luciano Rila/RHUL

35

Speaker recognition systems



Text
-
dependent: predetermined set of
phrases for enrolment and identification.



Text
-
prompted: fixed set of words, but user
prompted to avoid recorded attacks.



Text
-
independent: free speech, more
difficult to accomplish.



Luciano Rila/RHUL

36

Speaker recognition:
advantages


disadvantages


Advantages:


Use of existing
telephony infrastruct


Easy to use/non
-
intrusive/hands free


No negative
association



Disadvantages:


Pre
-
recorded attack


Variability of the
voice


Affected by noise


Large template
(5K to 10K)



Luciano Rila/RHUL

37

Eye biometric



Retina:



back inside of the eye
ball.



pattern of blood vessels
used for identification.





Iris:



coloured portion of the
eye surrounding the pupil.



complex iris pattern used
for identification.



Luciano Rila/RHUL

38

Retinal pattern



Accurate biometric measure.



Genetically independent: identical twins
have different retinal pattern.



Highly protected, internal organ of the eye.



May change during the life of a person.


Luciano Rila/RHUL

39

Retinal scan:

advantages


disadvantages


Advantages:


High accuracy


Long
-
term stability


Fast verification



Disadvantages:


Difficult to use


Intrusive


Limited applications



Luciano Rila/RHUL

40

Iris properties



Iris pattern possesses a high degree of
randomness: extremely accurate biometric.



Genetically independent: identical twins have
different iris pattern.



Stable throughout life.



Highly protected, internal organ of the eye.



Patterns can be acquired from a distance (1m).



Patterns can be encoded into 256 bytes.


Luciano Rila/RHUL

41

Iris recognition



Iris code developed by John Daugman at
Cambridge.



Extremely low error rates.



Fast processing.



Monitoring of pupils oscillation to prevent fraud.



Monitoring of reflections from the moist cornea
of the living eye.


Luciano Rila/RHUL

42

The iris code


Luciano Rila/RHUL

43

Iris recognition:

advantages


disadvantages


Advantages:


High accuracy


Long term stability


Nearly non
-
intrusive


Fast processing



Disadvantages:


Not exactly easy to
use


High false non
-
match rates


High cost



Luciano Rila/RHUL

44

Face
-
scan and facial
thermograms



Static controlled or dynamic uncontrolled
shots.



Visible spectrum or infrared (thermograms).



Non
-
invasive, hands
-
free, and widely
accepted.



Questionable discriminatory capability.


Luciano Rila/RHUL

45

Face recognition



Visible spectrum: inexpensive.



Most popular approaches:



eigenfaces,



Local feature analysis.



Affected by pose, expression, hairstyle,
make
-
up, lighting, eyeglasses.



Not a reliable biometric measure.


Luciano Rila/RHUL

46

Face recognition:

advantages


disadvantages


Advantages:


Non
-
intrusive


Low cost


Ability to operate
covertly



Disadvantages:


Affected by
appearance/environment


High false non
-
match
rates


Identical twins attack


Potential for privacy
abuse



Luciano Rila/RHUL

47

Facial thermogram



Captures the heat emission patterns derived
from the blood vessels under the skin.



Infrared camera: unaffected by external
changes (even plastic surgery!) or lighting.



Unique but accuracy questionable.



Affected by emotional and health state.


Luciano Rila/RHUL

48

Facial thermogram:

advantages


disadvantages


Advantages:


Non
-
intrusive


Stable


Not affected by
external changes


Identical twins
resistant


Ability to operate
covertly



Disadvantages:


High cost (infrared
camera)


New technology


Potential for privacy
abuse



Luciano Rila/RHUL

49

Hand geometry



Features: dimensions and shape of the
hand, fingers, and knuckles as well as their
relative locations.



Two images taken: one from the top and
one from the side.




Luciano Rila/RHUL

50

Hand geometry:

advantages


disadvantages


Advantages:


Not affected by
environment


Mature technology


Non
-
intrusive


Relatively stable



Disadvantages:


Low accuracy


High cost


Relatively large readers


Difficult to use for some
users (
arthritis, missing
fingers or large hands)



Luciano Rila/RHUL

51

Choosing the biometrics



Does the application need identification or
authentication?



Is the collection point attended or
unattended?



Are the users used to the biometrics?



Is the application covert or overt?



Luciano Rila/RHUL

52

Choosing the biometrics (cont.)



Are the subjects cooperative or non
-
cooperative?



What are the storage requirement
constraints?



How strict are the performance
requirements?



What types of biometrics are acceptable to
the users?



Luciano Rila/RHUL

53

References



ISO/DIS 21352: Biometric information management and
security, ISO/IEC JTC 1/SC 27 N2949.




Scheuermann, Schwiderski
-
Grosche, and Struif,
“Usability
of Biometrics in Relation to Electronic Signatures”
, GMD
Report 118, Nov. 2000.


Jain et al., “Biometrics: Personal Identification in Networked
Society,” Kluwer Academic Publishers.


Nanavati et al., “Biometrics: Identity Verification in a
Networked Society,” Wiley.



The Biometric Consortium:
http://www.biometrics.org/



Luciano Rila/RHUL

54

Any comments or questions?

luciano.rila@rhul.ac.uk