How reliable is the result?

paraderollΤεχνίτη Νοημοσύνη και Ρομποτική

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

192 εμφανίσεις

How reliable is the result?


Who are these people?


Are you sure?

1

2

3

4

1

Defining Reliability

Ground truth

Accurate landmark

Inaccurate landmark

0
)
(
ˆ
1
)
(
ˆ
)
(
)
(


j
j
l
r
l
r
Estimation based on intensity model of
each landmark

Intrinsic

precision

1

2

3

4

2

A. Martinez (2002)

IEEE Transactions on Pattern Analysis and
Machine Intelligence,
, 24(6):748

763,

Mutual Information


Intrinsic precision



Reliability



Reliability estimate



Maximize information given by the estimate:

1

2

3

4

3

Examples from IOF
-
ASM

Reliable


Unreliable

1

2

3

4

4

Examples from IOF
-
ASM

Reliable


Unreliable

1

2

3

4

5

Examples from IOF
-
ASM


Reliable


Unreliable

1

2

3

4

6

Examples from IOF
-
ASM

1

2

3

4

7

Reliability of a shape

1

2

3

4

Intrinsic precision
(average error)

Outlier threshold

(unacceptable error)

8

Incremental accumulation of evidence

1

2

3

4

9

Incremental accumulation of evidence

1

2

3

4

Reliable

Unreliable

Undefined

10

Segmentation results: IOF
-
ASM

2214
images

1.92 (
±
0.01)
pix

avg

146
images

3.67 (
±
0.18)
pix

avg

2360
images

XM2VTS
database


Avg

p2c error =


2.03 (
±
0.02)
pix

1

2

3

4

11

Segmentation results:
ASM

2141
images

2.69 (
±
0.04)
pix

avg

219
images

6.80 (
±
0.72)
pix

avg

2360
images

XM2VTS
database


Avg

p2c error =


3.06 (
±
0.08)
pix

1

2

3

4

12

Application I: Automatic model selection

?

1

2

3

4

13

Application I: Automatic model selection

Accuracy: 89.6 %

Accuracy: 82.1 %

1

2

3

4

14

Confusion Matrices (
color
-
coded)

Application II: Reliable Identification

1

2

3

4

XM2VTS Database

w/ BAD initialization

15

Application II: Reliable Identification

1

2

3

4

16

Conclusions on Reliability Estimation

1

2

3

4


High correlation of proposed measure with accuracy



Generic approach for ASM methods


Only requirement is a local metric for each landmark


Does not introduce changes in the algorithms



Very low false positives rate



Useful to provide robustness to biometric systems


Based on consistency with training data


17

Journal publications


F.M.
Sukno
, S.
Ordas
, C. Butakoff, S. Cruz, and A.F. Frangi.
Active shape models with
invariant optimal features: Application to facial analysis
.
IEEE Transactions on Pattern
Analysis and Machine Intelligence, 29(7):1105
-
1117, 2007.


F.M.
Sukno

and A.F. Frangi.
Reliability Estimation for Statistical Shape Models
.
Conditionally accepted for publication in IEEE Transactions on Image Processing, pending
minor
revision


F.M.
Sukno
, J.J. Guerrero and A.F. Frangi.
Projective Active Shape Models for
Posevariant

Image Analysis of Quasi
-
Planar Objects
: Application to Facial Analysis.
Submitted
for publication



C. Hoogendoorn, F.M.
Sukno
, S.
Ordas
, and A.F. Frangi.
Bilinear Models for
Spatiotemporal
Point Distribution Analysis: Application to Extrapolation of Left
Ventricular, Biventricular and Whole Heart Cardiac Dynamics
. Submitted for
publication

18

Conferences


A. Ortega, F.M.
Sukno
, E.
Lleida
, A.F. Frangi, A. Miguel, L.
Buera
, and E.
Zacur.AV@CAR
:
A
spanish

multichannel multimodal corpus for in
-
vehicle automatic audiovisual
speech recognition
. In
Proc. 4th Int. Conf.
on Language Resources and Evaluation,
Lisbon, Portugal, volume 3, pages 763
-
767. (www.cilab.upf.edu/ac), 2004.


A. Ortega, F.M.
Sukno
, E.
Lleida
, A.F. Frangi, A. Miguel, L.
Buera
, and E.
Zacur
.
Base
de datos audiovisual y
multicanal en castellano para reconocimiento automático del habla multimodal en el automóvil
. In
III Jornadas
en Tecnologías del Habla,
pages

125
-
130,
(www.cilab.upf.edu/ac), 2004.


F.M.
Sukno
, S.
Ordas
, C. Butakoff, S. Cruz, and A.F. Frangi.
Active shape models with invariant optimal
features (IOF
-
ASMs).
In
Proc. 5th Int. Conf. on Audio
-

and Video
-
Based Biometric Person Authentication, New York,
NY, USA. Lecture Notes in Computer Science
vol. 3546, pages 365
-
375, 2005.


F.M.
Sukno
, J.J. Guerrero and A.F. Frangi.
Homographic active shape models for
viewindependent

facial
analysis
. In
Proc. SPIE Biometric Technologies for Human Identification,
Orlando
, FL, USA, volume 5779, pages 152
-
163, 2005.


F.M.
Sukno

and A.F. Frangi.
Exploring reliability for automatic identity verification
with statistical shape models
. In
Proc. IEEE Workshop on Automatic Identification Advanced
Technologies,
Alguero
,
Italy
, pages 80
-
86, 2007.


D.
González
-
Jiménez
, F.M.
Sukno
, J.L. Alba
-
Castro and A.F. Frangi.
Automatic pose
correction for local
feature
-
based face authentication.

In
Proc. 4th IEEE Workshop on Motion of Non
-
Rigid and Articulated Objects,
Mallorca, Spain. Lecture Notes in Computer
Science vol. 4069, pages 356
-
365, 2006.


C. Hoogendoorn, F.M.
Sukno
, S.
Ordas
, and A.F. Frangi.
Bilinear models for spatiotemporal
point distribution
analysis: Application to extrapolation of whole heart cardiac dynamics
. In
Proc. IEEE ICCV 2007 8th Int.
Workshop on Mathematical Methods
in Biomedical Image Analysis, Rio de Janeiro, Brazil,, 2007.

19

Projects


BIOSECURE
: Biometrics for Secure Authentication (IST
-
2002
-
507534) European Excellence
Network of FP6/2002/IST/1.
Comisión Europea.
www.biosecure.info



iE
-
VULTUS
: Desarrollo de un sistema centralizado de biometría facial de tercera generación
para el control de acceso y seguridad en entornos inteligentes (Proyecto Coordinado TIC2002
-
04495
-
C02). Ministerio de Ciencia y Tecnología


HERMES:
Análisis biométrico de actividades óculo
-
faciales con técnicas de modelado
estadístico robusto para sistemas de asistencia a la conducción segura de vehículos, (Plan
Nacional de
I+D+i
, Proyectos de Investigación Aplicada TEC2006
-
03617/TCM). Ministerio de
Educación y Ciencia.


iEYE
: (
en conjunto con
Scati

Labs
) Definición de un sistema de tercera generación para
seguridad en entornos inteligentes mediante técnicas de visión por ordenador (Programa de
Fomento de la Investigación Tecnológica PROFIT FIT
-
070000
-
2002
-
935, FIT
-
070200
-
2003
-
112,
FIT
-
390000
-
2004
-
30, Proyecto
Iberoeka

IBK 02
-
263). Ministerio de Industria, Turismo y
Comercio.


eMedusa
:
(
en conjunto con
Scati

Labs
). Estrategias de adquisición, análisis, vi
-
sualización

y
fusión de información y su integración en un sistema avanzado de seguridad para entornos
complejos. Programa de Fomento de la Investigación Tecnológica (PROFIT/
Iberoeka

FIT
-
360000
-
2006
-
55, FIT
-
390000
-
2007
-
30
).
Ministerio de Industria, Turismo y Comercio.

20

Automatic Face Recognition demo

21

Automatic Face Recognition demo

22

Conclusions


IOF
-
ASM demonstrated consistently superior to ASM


Different databases with frontal images (30% more accurate)


Multi
-
view databases (70% more accurate)



The coplanar face model w/ PASM


Adds robustness to head rotations


Requires stronger image intensity models



Average performance of ASM methods is acceptable
-

Adding
reliability estimates:


Helps to automatically discards outliers


Allows for model selection and convergence assessment


23

Acknowledgements

24

THE END

25