Biometrics - FarinHansford.com

brasscoffeeΤεχνίτη Νοημοσύνη και Ρομποτική

17 Νοε 2013 (πριν από 3 χρόνια και 8 μήνες)

74 εμφανίσεις

Biometrics

Dr. Dirk Colbry

Dirk.Colbry@asu.edu

http://www.dirk.colbry.com/



Center for Cognitive Ubiquitous Computing

CUbiC

Outline


What is Biometrics?


Why do we need Biometrics?


Common Biometric Examples


Research Difficulties


My Previous Research


Biometrics and CUbiC



What is Biometrics


Bio (Biology) Metric (Measurement)


Access Control


Something you have


Key


Something you know


Password


Something you are (Biometrics)


Fingerprint

Security Threats


We now live in a global society of increasingly
desperate and dangerous people who cannot be
trusted based on
identification documents
.


The nineteen 9/11 terrorist
-
hijackers had a total of
63 valid driver licenses.

Acts of one person
deliberately impersonating
another with the objective
of personal gain.


Identity thieves steal PIN
(date of birth) to withdraw
money from accounts.


3.3 million identity thefts in U.S. in 2002

6.7 million victims of credit card fraud




Identity Theft

Unencrypted personal information including names and Social
Security numbers of 600,000 employees of Time Warner that was
backed up on tapes
was lost during transportation by a truck to
the storage location.

NY Times, May 8, 2005

Misplaced Identity

Many Types of Measurements

Fingerprint Acquisition

Computer Login using USB

Swipe Sensor

Building Access

PDAs and Cellphones

Fingerprints


Advantages


Fingerprints are unique (even between twins)


Fingerprint sensors are cheap


Latent fingerprints are easy to detect and are
commonly used in criminal cases


Disadvantages


Almost 10% of the population do not have fingerprints
that can be detected by typical scanners


Most fingerprint readers do not work well when wet
or dirty


Many people do not like to use fingerprint technology

Fingerprint Features


Fingerprint Style


Whorl


Loop


Arch


Most detect “minutia”


Core


Bifurcation


Delta


Ending

Core

Bifurcation

Ending

Delta

QUERY

TEMPLATE

30%

65%

5%

Image from
-

http://encarta.msu.com/

Biometric Applications


Verification



Compare a
measurement to a single
claimed identity



Identification



Compare
a measurement to an
entire database and find
the best match


Face Recognition


Advantages


Face is the most common biometric used by humans


Identification at a distance


Easy to capture from low
-
cost cameras


Non
-
contact data acquisition (free from contagious disease)


Covert data acquisition (ubiquitous surveillance cameras)


Legacy database (passport, visa and driver license)


Disadvantages


Faces change easily


Faces have the same salient features

13

Anchor Point Based Systems


Kanade, T. (1973).
Picture Processing System by Computer Complex and Recognition of
Human Faces
. Ph.D. Thesis Kyoto University, Japan.


Brunelli, R. and T. Poggio (1993). "Face Recognition: Features versus Templates."
IEEE
Transactions on PAMI

15
(10): 1042
-
1052.


14

Anchor Point Detection


Xiao, J., S. Baker, et al. (2004).
Real
-
Time Combined 2D+3D Active Appearance
Models
. IEEE Conference on Computer Vision and Pattern Recognition, Washington
D.C.



15

Global Features


Turk, M. and A.
Pentland

(1991). "
Eigenfaces

for
Recognition."

Journal of Cognitive Neuroscience
3(1).


16

Feature Regions


Pentland, A., B. Moghaddam, et al. (1994).
View
-
Based and Modular
eigenspaces for face recognition
. Proceedings of the IEEE Computer Society
Conference on Computer Vision and Pattern Recognition.


Bartlett, M. S., H. M. Lades, et al. (1998).
Independent component
representations for face recognition.

Proceedings of the SPIE, Conference on
Human Vision and Electronic Imaging III.

17

Model Fitting


Blanz, V. and T. Vetter (2003). "Face Recognition Based on Fitting a 3D
Morphable Model."
IEEE Trans. PAMI

25
(9): 1063
-
1074.


Measurement Difficulties


Large Intra
-
class variation





Small Inter
-
class variation


news.bbc.co.uk/hi/english/in_depth/americas/2
000/us_elections

Spoofing


Trying to trick a system (Secret Agent Stuff)








Live
-
ness Detection

Faces Can Lie

21

Cooperative vs. Uncooperative
Subjects


Uncooperative
Variations


Large pose changes


Large expression
changes


Intentional moving
during scan process


Cooperative Variations


Growing / changing
hairstyles (including
beards)


Wearing makeup


Blinking



Environment Variations


Changes in lighting


Changes in background


Which Algorithm is the Best?


We need numbers that say method ‘X’ is much
better than method ‘Y’


Equal Error Rate


Rank One matching


We need to test different algorithms on the
same database


FRGC


FRVT


FacePix

Types of Matching


True Accept



I claim to be me and the
computer agrees


True Reject



Someone else claims to be
me but the computer disagrees


False Reject


I claim to be me but the
computer disagrees


False Accept


Someone else claims to be
me but the computer agrees

24

3D Scanners

Laser

Color Camera

3D Face Recognition


Advantages


Hard to Spoof


Easy to find Anchor
Points


Pose tolerant


Lighting tolerant


Disadvantages


Expensive


Slow

27

Frontal Anchor Point Detection
Demo

FRGC 1.0 Data Collection


275 subjects, 943 subject sessions


Between 1 and 8 acquisitions per subject

943 x 943 Similarity Matrix

87
94
96
78
89
97
98
68
81
91
97
76
0
20
40
60
80
100
0.01
0.1
1
10
False accept rate (%)
Verification rate (%)
FRGC 1.0 Results

At CUbiC


Social Interaction Assistant


Camera Motion


Lighting Changes


Expression Changes


Distance Changes


Pose Changes

Multiple Biometric Grand Challenge


Low quality still images


High and low quality video
imagery


Face and iris images taken
under varying illumination
conditions


Off
-
angle or occluded
images