Cancelable Biometrics - World Customs Organization

IBM Research

© 2005 IBM Corporation

Nalini K. Ratha*Exploratory Computer Vision GroupIBM T. J. Watson Research CenterHawthorne, NY [email protected]

Cancelable Biometrics

*inputs from J. Connell, R. Bolle, and S. Chikkerur

mailto:[email protected]

2

IBM Research


Introduction

� Privacy issues in biometrics

� How can privacy be enhanced

Survey of existing methods

Cancelable biometrics

� Operational issues

� Sample transforms Conclusions

Revocable/Rescindable/Anonymous/Cancelable biometrics

3

IBM Research


Large Scale Biometrics Identification

Biometrics identification has become a �must have� tool in homeland security and the next generation intelligent infrastructure� Government: Passports/Visas, Citizen identification, Employee

identification� Financial Services: Consumer point-of-sale ID, Confirmation of

financial transactions These new uses bring new challenges

� Meeting expectations for accuracy (false negative/false positive)

� Supporting transaction response rates where identification or authentication are involved

� Achieving the scale required by emerging applications� Understanding and handling privacy issues

4

IBM Research


Large scale and Cancelable are not different �

� Two sides of the same coin

Large collection leads to privacy issues

5

IBM Research


Attack model

Formidable adversaries: Thieves

Hackers

Users

Customers

Employees

Merchants

Competitors

Competitors� governments

6

IBM Research


Attack Methods

Hardware/Software/Database Attacks

Trojan horse for feature extractor Trojan horse for matcher Overriding templates Feature-based dictionary attack

Other Attacks

Phishing Farming Hill climbing attack Swamping attack Piggy-back attack Spoofing the sensor Collusion at the enrollment process

Channel Attacks

Override result Replay attack Channel attack between matcher and template DB Channel attack at the enrollment time

7

IBM Research


Biometrics vs. Passwords

Always the sameConstantly varyingData input

Yes (easily)No Revocability

NoYes (mostly)Non-Repudiation

Exact, 100%Inexact, fuzzy, Never 100%

Match algorithm

Typically 6-8 alphanumeric characters

Usually about 100 bytes or more

Size

Hash of the password string

Features (constant size features vs. variable size features from signal)

Internal representation

PasswordsBiometrics

8

IBM Research


Biometrics and public perception

In a 2002 poll commissioned by SEARCH (funded by US Bureau of Justice Statistics)

- 88% were concerned about possible misuse of their biometrics data

- 80% were comfortable with the use of biometrics �as a means of helping prevent crimes�

9

IBM Research


Issues

You give up part of yourself that is unique to you

The use of biometrics introduces a problem: biometrics cannot be replaced� biometrics is not a secret � once compromised, compromised forever

What if a biometric is compared: cross matching? � Biometrics collected for one application can be shared to retrieve other

private information (health care, law-enforcement, financial background)

Can we find a function which permits us to safely replace biometrics just like stolen credit cards...

10

IBM Research


Hashing as a solution

Privacy:The original biometric is not stored

Each application uses a different transformation function Security

It is computationally hard to recover B given T(B)

One way hashT()

One way hashT()

DBT(B)

B

B�Match

T(B),T(B�)

T(B�)

Matching

Enrollment

11

IBM Research


Hash Functions : Ideal for passwords and text

33B21856A91D2FBB5BC4144C69B23F85

FIRE ALL LINUX

PROGRAMMERS

43C08679B2FD54C65467DDCC9C00AD49

1 character difference

65 bitsdifference !!

MD5

HIRE ALL LINUX

PROGRAMMERS

MD5

Can we simply hash a fingerprint?!

12

IBM Research


Hashing : Doesn�t work for biometrics

26 pointsmatch

Don�t match at ALL !!F313C86188DDE96b

D48AD58CDECDB9E8

MD5

80BC979099C2FA643E4C5432A03E01B8

MD5

15 pointsdon�t match

OK

13

IBM Research


Solutions?

Crypto community:

� Reduce uncertainty of the biometric - quantization

� Borrow randomness from key to compensate for lost entropy

� Approaches

� Biometric Hardening (Goh et al �03, Teoh et al �04, Soutar et. Al �98)� Biometric Keying (Davida et al. �98, Monrose �99, Monrose �01)� Fuzzy techniques (Juels & Watenberg �98, Juels & Sudan 02, Dodis 04,

Tuyls 04)

Biometric community:

� Mask the original biometric � preserves entropy (CMU)� Cancelable biometrics (IBM)

14

IBM Research


Biometric Hardening Template is combined with user specific random information This is similar to �salting� of passwords before hashing

Feature

ExtractionFeature Salting

Error Tolerant

Discretization

Goh and Ngo, 2003

�Face Biometrics

�`Eigen faces� features

Soutar et al,1998

�Fingerprints

�Fourier transform features

�Features are projected

on to user specific

orthogonal random

vectors

�Fourier features are

multiplied with user

specific random phase

array

�Binary values are

derived using quantization

�The key acts as a Shamir

secret key share



�Key is embedded using a

redundant lookup table

High uncertainty Zero uncertainty

15

IBM Research


Biometric Keying

The binary key is directly derived from the biometric template The transformation has to be error tolerant More scalable than �biometric hardening� methods

Feature

ExtractionBinarization

Error tolerant

Representation

Davida et al., 1998

�Iris Biometric

�Iris code features

Monrose et al., 1999

�Key stroke dynamics

�Key duration and latency

time features

�Monrose et al., 2001

�Speech biometric

�Cepstral features

�Features are already

binary

�Binarization is done by

comparing feature value

with a global threshold �T�

�User specific hamming

codes are used to correct

errors caused by offsets

�Consistency of each

feature is learned over

time for each user

�The inconsistent features

are discarded

16

IBM Research


Biometric Hardening Template is combined with user specific random information. This is similar to `salting` of passwords before hashing

Feature

ExtractionFeature Salting

Error Tolerant

Discretization

Goh and Ngo, 2003

�Face Biometrics

�`Eigen faces� features

Soutar et al,1998

�Fingerprints

�Fourier transform features

�Features are projected

on to user specific

orthogonal random

vectors

�Fourier features are

multiplied with user

specific random phase

array



�The key acts as a Shamir

secret key share



�Key is embedded using a

redundant lookup table

High uncertainty Zero uncertainty

17

IBM Research


Cancelable signal transform (CMU)

Requires the use of the MACE correlation engine

18

IBM Research


IBM Solution: Cancelable Biometrics

Intentional repeatabledistortion� alters signal but still in

correct format

� generates a similar signal each time

Compromised scenario:� a new distortion

creates a new biometrics

Comparison scenario:� different distortions for

different accounts© New Yorker Magazine (Charles Addams)

19

IBM Research


Cancelable Biometrics: Example

Two images of the same face

repeatable distortion

DON�T

MATCH

DON�T

MATCH

MATCH

MATCH

20

IBM Research


Operational Issues

Application: Must be applied directly at the sensor

There should be no scope for the original signal to leave the sensor

The transform can be applied at

� signal level� feature level

Registration: For repeatability, often we have to register (align) before applying

any distortion transform

Use invariant points to align two patterns

� core and delta in fingerprint images� nose and mouth in face images

21

IBM Research


Cancelable Biometrics vs. Biometric Cryptography

NONOYESYESYESRevocable

YesNoNoYESYesRetains entropy?

NONONOYESYES/NO

Preservesrepresentation?

YES(Juels et al,Uludag et. al)

NONOYESYESApplicable forfingerprints(minutiae)?

FuzzyTechniques

BiometricKeying

BiometricHardening

CancelableBiometrics

Ideal

Made in IBM!

22

IBM Research


Real example: two images of the same face

23

IBM Research


Registration and Distortion

24

IBM Research


Images look similar, but not like the original

25

IBM Research


Fingerprint example: two impressions

Registration based on �core� and �delta�

Original 1 Original 2

26

IBM Research


Distorted versions still appear similar

Distorted 1 Distorted 2

27

IBM Research


Minutiae of distortions match, but not to original

Original 1 Distorted 1 Distorted 2

no match match

28

IBM Research


Conclusions

Privacy issues in biometrics databases need to be addressed for acceptable mass deployment

Privacy enhancement for biometrics requires both information security and biometrics experts to contribute

Our initial experimental results are extremely encouraging

Cancelable Biometrics - World Customs Organization

Documents