Biometrics with Topics in Face Recognition - People Server at UNCW

Biometrics with Topics in Face
Recognition

Dr. Karl Ricanek, Jr.

Assistant Professor

Computer Science Dept

University of North Carolina, Wilmington

Discussion Overview


Biometrics


Definition/History


Technologies


Face Recognition


History/Issues


Research Focus


Questions and Answers

Biometrics Definition


(Merriam
-
Webster online): the statistical analysis
of biological observations and phenomena.


Biometrics are automated methods of
recognizing a person based on a physiological
or behavioral characteristic.
(
http://www.biometrics.org
)


Phenotypic biometric
–

based upon features or
behaviors that are acquired through experience and
development.


Genotypic biometric
–

based upon genetic
characteristics or traits.

Biometrics History


First documented example: Egypt several
thousand years ago.
(Biometrics: Advanced Identity Verification
the complete guide, Julian Ashbourn)


Khasekem, assistant to chief administrator,
used phenotypic biometrics for identification of
food provisions.


Notes were kept on every worker (100,000 or more)
detailing physical characteristics (eg. age, height,
weight, deformities) and behavioral characteristics
(eg. General disposition, lisp/slurs in speech, etc.)

Biometrics History


Biblical Reference


Judges 12:5
-
6: “Then said the men of Gilead
unto him, Say now Shibboleth: and he said
Sibboleth: for he could not frame to pronounce
it right. Then they took him, and slew him at the
passages of the Jordan: and there fell at that
time of the Ephraimites forty and two thousand.”


Phenotypic biometric, in particular, voice, was
used to identify Ephraimites, the enemy of the
Gileadites.


Ephraimites pronounced “Sh” as “S”

Biometrics History


Modern


Belgian mathematician and astronomer Adolphe
Quetelet ushered in the modern use of biometrics with
his treatise of 1871, “
L’anthropometrie ou mesuare des
diffenretes facultes de l’homme
”


Frenchman Alphonse Bertillon, applied Quetelet work to
develop a system to identify criminals based on
anatomical measures.


Argentinean police officer Juan Vucetich was the first to
use dactyloscopy in 1888. Dactyloscopy is the taking of
fingerprints using ink.

Biometric Technologies: Selected


Fingerprint


Voice


Iris/retina


Gait


Face Recognition

Biometric Technologies


Fingerprint


Pros:


Years of research and
understanding


Security community
comfortable with technology


Innately distinctive feature


Cons:


Can be altered/worn over
time


Some ethnic groups exhibit
poor discrimination of finger
prints


Automatic techniques not
trusted


Biometric Technologies


Voice


Pros


Non
-
invasive


Distinctive w.r.t. vocal
chords, vocal tract,
patalte, sinuses, and
tissue w/in mouth


Cons


Easily corrupted with
noise


High false rates (positive
and negative) w.r.t.
physical ailments (colds,
sinus drains, etc.)


Biometric Technologies


Iris/Retina


Pros


Innately unique


No change over time
(static)


Left and right within
themselves


Genetic inheritance
(Genotypic)


Cons


Acquiring image

•
Alignment/position

•
Pupil size change




Biometric Technologies


Gait


Pros


Non
-
invasive


Discriminate under
various conditions (eg,
walking, jogging,
running)


Promising research


Cons


Can be altered


Too early in research

Biometric Technologies: Face
Recognition


History


1888 Galton

Profile Id

Kanade 1977,

Kaya 1972,

Bledsoe 1964

Feature Metric

Turk 1991

Hong 1991

Shirovich 1987

Statistical

Akamtsu 1991

Brunelli 1992

Neural Network

Psychophysic

neuroscience

approaches

Ricanek 1999

Variable Lateral

Pose Recognition

Ricanek, Patterson & Albert 200X

Craniofacial Morphology:

Models for Face Aging

(Research in progress)

Face Recognition Techniques


Image Based


Statistical based on O(2
nd
)


PCA/Eigenfaces (dominant)


Fisherfaces (LDA)


Etc.


Template matching


Spectral analysis


Gabor filtering


Etc.


Feature Based


Geometric


Feature metrics (spatial
relationships)


Morphable models
(shape/texture)

FRT Diagram

Probe

Gallery (DB)

Face Recognition

System



Rank ordered lists

from gallery set with

confidence factor

Preprocessing

Preprocessing

Face Recognition Technologies: Field
Reports


ACLU Press Release:
Data on
Face
-
Recognition Test at Palm
Beach Airport Further
Demonstrates Systems' Fatal
Flaws
. May 14, 2002.


ACLU press release:
Drawing
a blank: Tampa police records
reveal poor performance of
face
-
recognition technology:
Tampa officials have
suspended use of the system
.
Jan. 3, 2002.


Etc.


Reports that system in real
world app was effective 53% of
the time




“System logs obtained by the
ACLU through Florida's open
-
records law show that the
system never identified even a
single individual contained in
the department’s database of
photographs.”

Face Recognition Technologies:
Problems


Resolution/Quality


Orientation


Scale


Disguise


Lighting


Image Currency


Physiologic changes
due to growth


Physiologic changes
due to aging

My Research Niche: Age Progression


Age Progression


Growth
–

from infancy
to full maturation (~18)


Maturation
–

from full
maturation to
senescence (elderly
years)


My Research Niche: Age Progression


Maturation Age Progression


Face undergoes significant changes during the
adult age progression which dramatically
impacts face recognition technologies.


Loss of epidermis elasticity causes the formation of
rhytides and ptosis.


Elasticity loss is caused primarily by photoaging but
contributory factors include smoking, alcohol
consumption, drug use, and some prescribed
medications.


Skin texture changes occur also, rougher skin,
blotchiness/discoloration, hanging skin, etc.

My Research Niche: Age Progression

Face Recognition Rates (offline)


Probe
-
Gallery (temporally current)


Image based: mid 90%


Feature based: mid 90%


Probe
-
Gallery (temporally displaced)


Image based: 80% (1yr)
–

50% (5yr)


Feature based: unknown

Face Recognition Rank Curve: Normal

0
5
10
15
20
25
30
35
40
45
50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
Face Recognition Rank Curve: Age
Progression

0
5
10
15
20
25
30
35
40
45
50
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
Team’s Research


Constructing the first craniofacial database where each subject
contains multiple images that span from late adolescences
through senescence.


Formulate understanding of the mechanisms of morphological
changes in the human face as it ages from late adolescence (i.e.,
ages 18
-
21 years) to senescence (i.e., ages 60+ years).


Which features fundamentally change with age?


Which features DO NOT change with age?


Develop models based on analysis of features for consistent
patterns versus idiosyncratic variations of craniofacial change due
to aging. Develop soft tissue texture map models that simulate
aging of skin.


Detailed evaluation of FRT against the database.


How and why does the FRT algorithm fail?


Develop FRT algorithm that is robust against aging.


Develop face detection and tracking techniques.


Questions and Answers