3d-face multibiometrics for face recognition...agenda • multibiometrics in general •...

Multibiometrics for Face Recognition

3D Face Project – End User Meeting

2007-03-22 / DarmstadtVolker Kempert

(Cognitec Systems GmbH)

Agenda• Multibiometrics in general • Multibiometrics related to the face• Biometric Face Identifiers • Capturing Biometric Face Identifiers• Fusion related to the face• Conclusions

27.03.2007 23dface - End User Meeting

Multibiometrics

27.03.2007 3dface - End User Meeting 3

BiometricAlgorithmBiometric

Algorithm

BiometricAlgorithmBiometric

Algorithm

FusionEngine

Biometric SampleBiometric

Sample

Biometric SampleBiometric Sample

BiometricAlgorithm Score

ScoreScore

ScoreScore

Score

Biometric Identifier

Biometric Identifier

•Multiple biometric identifier•Multiple biometric samples

•at different sample qualities•Multiple biometric algorithms•One combined score

Why Multibiometrics?• Compared to single biometrics identifier

– Higher Accuracy More secure – More robust– Higher fraud resistance

• Disadvantages– More complex biometric capturing processes– More complex devices and algorithms

Higher operational costs


October 23, 2006 FaceVACS Algorithms: An Overview

Multi-Modal Biometrics

Example for worse performance (John Daugman, 1999):

• Biometric system A: EER = 1%• Biometric system B: EER = 0.1%

Have A and B operate at their EERs; conduct 100,000 verification attempts with impostors, 100,000 with authentics; then:

• A alone: 2000 errors; B alone: 200 errors• “AND” rule: 1099 FR’s, 1 FA 1100 errors• “OR” rule: 1099 FA’s, 1 FR 1100 errors

Multibiometrics using the face


Intensity Algorithm

ShapeAlgorithm

FusionEngine

Intensity data

Shape data

CombinedAlgorithm Score

Score

Score

Score

Score

Intensity Data

Shape Data

Skin Texture

Data

Skin Texture

Algorithm

Alligned biometric samples•Intensity data•Shape data

July 2006 http://www.cognitec.com 7

Sample FRGC imagesControlled

Uncontrolled


Sample frontal Yale images

Varying lighting conditions


Example: shape data

vertex data is subject to noise

Monocular, fixed view sensors produce

occlusions

Example: shape data


Gap Filling /plane surface patches

some vertices have large deviations

Example: skin texture


Skin texture ananlysis indicates the degree to which two surfaces are the same if the blocks match in an orderly fashion

Simultanious Capture


Intensity Data and

Skin Texture sample

Shape Datasample

Intensity identifier

Shapeidentifier

Skin Texture

identifier

Combined Sensor

High Resolution Camera

3d Shape Sensor

Real Person withFacial characteristics

Digital representation of biometric samples


Fusion Architecture for Face

Fusion EngineUse dependency between the scores to produce a fused

one

Comparison EngineIntensity Images

Comparison EngineShape Images

Comparison EngineIntensity and Shape Images

Compensate intensity image using parameters of the shape

Shape EstimatorEstimates shape from single or multiple

intensity images

scor

e

scor

e

Scor

e

final score

14

Promising Results

Equal Error Rates•Using Intensity Images („2d“): 3.14%•Using Shape Data („3d“) 2.54%•Using both (2d+3d): 1.01%

Promising Results (2)


Conclusions

• Multi (modal) biometrics help to improve accuracy compared to single biometrics

• Face is an object that allows the simultaneous capturing of multi-biometrics identifiers

• Multi-biometrics systems are more difficult to outsmart


3d-face multibiometrics for face recognition...agenda • multibiometrics in general •...

Documents