Vinod Biometric

7/31/2019 Vinod Biometric

1/50

1

Biometric Identification

Assoc. Professor Vinod ChandranSchool of Engineering Systems

QUT


2/50

2

Speech, Audio, Image and VideoTechnologies

Academic staff:Prof Sridha SridharanA/Prof. Vinod Chandran, Prof. M. Moody, A/Prof. W. Boles

Postdoctoral ResearchersDr. Michael Mason - Research FellowDr. Clinton Fooks Research Fellow

Dr. David Cole - Research Fellow

Postgraduate students: 19 PhD


3/50

3

Track record of Speech, Audio, Image andVideo Technologies Group

1992-20041. Graduated 18 PhD students and 12 Mastersby research students.

2. Currently supervising 19 PhD students.3. Over 200 refereed journal and conference

publications4. Working with 15 different industries and

government organisation.5. Average external funding of

$300,000/annum.


4/50

4

Technology Transfer and Commercialisation

Codan Pty Ltd : Speech enhancement system for NewGeneration HF Transceiver Australia Post : Voice Controlled Parcel Sorting Telstra : Automatic Speech Quality Measurement for

Mobile Communication Systems Queensland Police: Covert Speech Enhancement and

Suspect Identification by Voice

(Name withheld) : Intelligent Multi-Microphone SpeechEnhancement System and Covert Listening PostDesign.


5/50

5

Technology Transfer and Commercialisation

Motorola Australian Research Centre: Multi-Microphone Based Speech Recognition in Cars

Boeing : Digital HF radio design Harris corporation, USA: Analog Speech Encryption

Systems.

Genista Corporation, Japan: Perceived Audio QualityMeasurement: Commercial Monitors: Automatic Audio Segmentation

and Recognition for Broadcast Monitoring.

Avaya (Lucent Technology): Speech qualitymeasurement for internet telephony. Edcare: Automated English pronunciation training

system.


6/50

6

Research areas we are working on Speaker Recognition Language recognition Word Spotting Speech recognition in PDAs, mobile phones and wearable computers. Speech recognition for broadcast transcription Face Recognition

Iris Recognition Palm Recognition Finger Print Recognition Gait Recognition Motion Detection

Person tracking and human activity detection Gesture and facial expression recognition Multi-modal Recognition Hand Written Signature Recognition Document recognition


7/507

Fingerprint

Iris & Retina scan

Handwriting

Voiceprint

Facial Geometry

DNA Typing Style

Other biometricssuch as ear shapepalm print, hand-shape, vein shapehave also beenused.

Our main focus ison voice and facerecognition.

Introduction - Biometrics


8/508

Basis for secure access

What we know Password (can forget)

What we possess Secret key on disk, card (can be stolen)

What we are Biometric


9/509

Requirements of a goodbiometric

Universality Everyone should have it

Distinctiveness It should not be the same for two persons

Permanence It should be unchanged for reasonable

period of time Collectability It should be possible to acquire it


10/5010

For real-life applications

Good performance (accuracy, speed,resource requirements)

Acceptability (harmless, preferably non-intrusive, easy to work with)

Circumvention (robust againstimpersonation attacks and fraud)


11/5011

What is biometrics?

Automated processing (with digitalcomputers usually) of biometric data foridentifying or verifying the identity ofliving human individuals.


12/5012

Potential applications

Crew/passenger verification Secure access to premises Criminal investigation Surveillance and counter terrorist

measures

Authenticated access to servers Authenticated electronic commerce andbanking etc.


13/5013

Recognition


14/5014

Verification


15/5015

Watch List


16/5016

Performance of differentbiometrics


17/5017

Generic architectureAcquisition

Detection and pre-processing

Feature Extraction

Matching

Identification/Verification

TrackingSegmentation

Fiducial points (eg eyes)Normalization

morphable models

PCA (eigenspace)LDA (Fisherspace)

2D Fourier spectrumCorrelation filtersGabor wavelets

Bispectral integrals

Statistical classifiersStructural methods

ANNsSVMs


18/5018

Biometrics for Internet security

Encryption keys may be stored on smartcard

Biometric to access the keys

Cancellable biometrics one reservedbiometric or key, others encryptedbefore providing to third parties


19/5019

Fingerprint

Around 2% FR at 0.1% FA (FVC20022% EER best in open category)

Sensors chip and optical

Contact imaging no need for scalenormalization

Sensor cost low, around $25

Suitable for smart card implementation Susceptible to fraudulent copying

need liveness tests


20/5020

Fingerprint processing

Consist of ridges and burrows Ridge endings and bifurcations

Minutiae are important Minutiae extracted with tuned Gabor

filters and morphological opertions Minutiae points represented as a graph Graph matching after morphing for

plastic deformations of the skin


21/5021

Fingerprint usage

Already in use at some airports Large databases available with FBI and

Police in many countries

Has potential for secure internettransaction implementations (recentpapers on secret keys stored in smartcards accessed with fingerprints)

5% of the population do not have legiblefingerprints


22/5022

Face recognition

Mature commercial implementations 10% FR for 1% FA with indoor images Within class variations pose, lighting,

expression, facial hair etc. Acceptablity is quite high but standing in

front of a booth is time-consuming Many algorithms and extensive

benchmarking efforts


23/5023

Still Face Recognition : OpenIssues

Addressing the issue of recognition being toosensitive to inaccurate facial feature localization

Robustness Small and/or noisy images Images acquired years apart Outdoor acquisition:

lighting and pose

Scaling well to larger databases optimally arbitrating and combining local and

global features


24/5024

Face Recognition -Commercial Systems


25/5025

Face Recognition usage

Natural for passport, drivers licence Easier for untrained human checking of automated

result(s) Already in use at airports and other premises User cooperation not necessary can be used with

surveillance Higher computational, storage and transmission

requirements may be a hurdle to smart cardimplementation

Potential for continued authentication of internettransactions such as an online examination or anonline chess game with biometric verificationinformation embedded in packets at presenatiationlayer.


26/5026

Face Recognition -Performance Evaluations

FERET 93-97 FRVT 2000 FRVT2002

http://www.frvt.org M2VTS XM2VTS BANCA

http://www.ee.surrey.ac.uk/Research/

VSSP/xm2vtsdb Colorade State University Web Site

http://www.cs.colostate.edu/evalfacerec


27/50

27

Still Face RecognitionSystems

Holistic matching methods (Classification using whole face region) Principal component analysis (PCA) Eigenface* Linear discriminant analysis (LDA) Fisherface and subspace LDA (FLDA)*

Feature-based (structural) matching methods (Structural classification usinglocal features) Pure geometry methods Graph matching methods* Gabor wavelets & image graphs

Hybrid methods (Using local features and whole face region) Eigenface & Eigenmodules Local & global feature methods Face region and components * Top 3 in FERET tests


28/50

28

3-D face recognition usingdepth information

Figure shows a 3-Dreconstruction of aface using depth

information acquiredusing a stereocamera system.


29/50

29

Basic premise for Super resolution


30/50

30

HR Original


31/50

31

Reference : Super -Resolution Optical Flow, Technical Report CMU -RI-TR-99-36,Carnige Mellon University, USA


32/50

32

Background removal

Segment moving objects from a staticbackground. Person Tracking, Face detection in video

Background changes with time of day Algorithm works by clustering and

modelling background pixels

Simple background subtractionineffective, need to adapt to lightingchanges, object movements etc.


33/50

33

Comparison of methods

Original Truth VAR GMM1 GMM2 NHD


34/50

34

Face Recognition research atQUT

Use of 3D data and stereo images Face tracking using colour, depth as

well Super-resolved faces from surveillance

video By-passing depth estimation and

extracting depth-dependent features Hybrid 2D-3D methods and fusion


35/50

35

Voice as a biometric Speakerverification

Can be quite accurate and reliable Text-dependent and Text-independent

systems Can be low-cost (microphones, sound cards) Sensitive to audio noise, acoustic channel

changes

Natural for telephone based applications Could become important with multimedia 3G

mobile services


36/50

36

Voice based systems

Most are Mel-scale cepstral coefficientand Gaussian Mixture model based

NIST evaluations technology quitemature

QUT systems have been placed no. 1 insmall vocabulary (and language id)categories


37/50

37

MFCC features

Windowed overlapping frames DFT each frame Log in cepstrum converts multiplicative

effects such as channel transfer function toremovable additive bias

Frequency scale warped (linear up to 1000Hz, factor of 1.1 thereafter) to correspond to

human perception. Called Mel-scale. DCT of log of spectral energies averagedover Mel-scale bands


38/50

38

Speaker Modelling and ID

Speakers are modelled using GMMs (means, covariance matrix

The speaker model, k , that maximisesthe likelihood of the given test speech(or observation), X , is identified, i.e.,

where S is the registered # of speakers.)|(maxarg

1 k Sk X pS


39/50

39

Feature clusters and GMMs

HOS d MFCC i


40/50

40

HOS and MFCC comparisonwith noisy speech


41/50

41

Handwritten Signatures

Behavioural biometric Pen tablet systems cost only

around a hundred dollars Spatial coordinates, pen

pressure and pen angles canbe captured

Dynamics are difficult toforge acceptability is high Reliability is poorer than iris,

fingerprint, voice or facebecause of large intra-classvariations

Even with relatively lowEER, savings are potentiallyhuge with credit card fraudreduction


42/50

42

Potential Applications

credit card transaction verification secure access to computers secure access to databases passport and customs checks Identity checking at examinations

Identity confirmation when voting


43/50

43

An example stroke and x,y,pressure and corresponding

derivatives

Sig t ifi ti


44/50

44

Signature verificationcommercial products

PenOp (Peripheral Vision, New York) for access to systems Sign-On (Peripheral Vision, New York)

claims 2.5% EER, built into software instead of password

Signer Confidence (Peripheral Vision, New York) static, used for signature verification on cheques

Cadix ID-007 (verification in 1 second) CounterMatch (AEA Technology, UK) Kappa (British Technology Group, UK) ApproveIT (Silanis Technology, Canada)


45/50


46/50

46

Signature Verification Benchmarks(SVC 2004) skilled forgeries


47/50

47

QUT System

On line signatures X, Y, Pressure (pen angles can be used*) overlapping frames bispectral invariant phases and other features Gaussian mixture models (with some

temporal order information as in HMMs*) Language independent Handwriting sensitive *not in demo


48/50

48

Performance (in house data)


49/50

49

Advantages

No need to normalize and align, warp ormorph

Works with any language signature

Robust to intrapersonal variations,scaling Fast verification

Low memory requirements uncompressed data in a few KB permodel


50/50

References1. J. L. Wayman, Digital Signal Processing in biometric identification: A review, Proc. of ICIP -2002, vol. 1, pp. 22-25, Sept,

2002.2. K. Delac and M. Grgit, A survey of biometric recognition methods, Proc. of 46th Intl. Symposium on Electronics in marine

Elmar 2004, pp. 184-193, June 2004.3. http://www.biometrics.org 4. http://www.nist.gov 5. P.J. Phillips, P. Grother, R.J. Michaels, D.M. Blackburn, E. Tabbassi and M. Bone, Face Recognition Vendor Test 2002:

Overview and Summary, NIST Technical Report, March 2003. 6. M. Yang, D.J. Kriedman and N. Ahuja, Detecting faces in images: a survey, IEEE Trans. on Pattern Analysis and Machine

Intelligence, vol. 24, no. 1, pp. 34-58, Jan. 2002.7. R. Chellappa, C.L. Wilson and S. Sirohey, Human and Machine Recognition of Faces: A Survey, Proc. of the IEEE, vol. 83,

no. 5, pp. 705-740, May 1995.

8. K.W. Bowyer, K. Chang and P. Flynn, A survey of approaches to three -dimensional face recognition, Proc. of 17th Intl.Conf. on Pattern Recognition (ICPR), vol. 1, pp. 358-361, Aug. 2004.9. D. A. Reynolds and R. C. Rose, Robust Text -Independent Speaker Identification using Gaussian Mixture Speaker Models,

IEEE Trans. On Speech and Audio Processing, vol. 3. no. 1, pp. 72-83, Jan. 1995.10. V. Chandran, D. Ning and S. Sridharan, Speaker Identification using Higher Order Spectral Phase Features and their

Effectiveness vis--vis Mel - Cepstral Features, Proc. of the International Conference on Biometric Authentication (ICBA -2004).

11. G. Gupta and A. McCabe, A Review of Dynamic Handwritten Signature Verification, Technical Report, James CookUniversity, Australia, 1997.

12. R. Plamondon, Looking at Handwriting Generation from a Velocity Control Perspective, Acta Psychologica, vol. 82, pp. 89 -101, 1993.

13. M.S. Hwang and L.H. Li, A new remote user authentication scheme using smart cards, IEEE Trans. on Consumer Electronics, vol. 46, pp. 28-30, 2000.

14. J.K. Lee, S.R. Ryu and K.Y. Yoo, Fingerprint based remote user authentication scheme using smart cards, IEE ElectronicsLetters, vol. 38, no. 12, pp. 554-555, 2002.

15. U. Uludag and A.K. Jain, Multimedia content protection via biometrics -based encryption, Proc. of Intl. Conf. on Multimediaand Expo (ICME03), vol. 3, pp. 237 -240, July 2003.

16. B.T. Tsieh, H.T. Yeh, H.M. Sun and C.T. Lin, Cryptanalysis of a Fingerprint -based Remote User Authentication SchemeUsing Smart Cards, Proc. of 37th Annual Intl. Carnahan Conf. on Security Technology, pp. 349 -350, Oct. 2003.
http://www.biometrics.org/http://www.nist.gov/http://www.nist.gov/http://www.nist.gov/http://www.nist.gov/http://www.biometrics.org/http://www.biometrics.org/

Vinod Biometric

Documents