Biometrics Angela Sasse – Dept of Computer Science

Goals of this lecture What are biometrics? How they are applied Usability and security issues

See http://www.bromba.com for a good FAQ on Biometric jargon biometric = biological or behavioural property of an individual that can be measured and from which distinguishing, repeatable biometric features can be extracted for the purpose of automated recognition of individuals biometric sample = analog or digital representation of biometric characteristics prior to biometric feature extraction process and obtained from a biometric capture device or biometric capture subsystem (raw data) biometric template = stored biometric features, applied to the biometric features of a recognition biometric sample during a comparison to give a comparison result. See http://www.bromba.com for a good FAQ on Biometric jargon

Some basics Verification using ID + biometric, or Enrolment = capture of biometric feature and generation of biometric sample and/or template Full images or templates templates are more efficient Images can be used to reverse-id/create new templates Verification using ID + biometric, or identification (biometric compared to database

Physical/behavioural Fingerprint Finger / Palm Vein Hand geometry Face recognition Iris Retina Earshape Behavioural Voice print Dynamic Signature Recognition (DSR) Typing pattern Gait recognition Heart rate analysis

Enrolment Crucial for security and subsequent performance In some context, identity of enrolee needs to be checked Biometrics enrolled need to be genuine (see attacks) good enough quality to work Enrolment procedure needs to be formalised Staff need to be trained Staff need to be trustworthy or closely checked Time taken to carry out enrolment often under-estimated

FTE FTE (failure to enrol) rate = proportion of people who fail to be enrolled successfully FTAs: users can be enrolled but biometric sample too poor quality to verify Reasons for FTE/FTA Biometric not present or temporarily inaccessible Biometric not sufficiently prominent or stable Problem for Universal Access – may exclude Older users Disabled Equipment may be too difficult to use

FTE in UKPS enrolment trial Face Iris Finger Quota 0.15% 12.30% 0.69% Disabled 2.73% 39% 3.91% UKPS (UK Passport Service) enrolment trial 2004

FAR & FRR FAR (False Acceptance Rate) accepting user who is not registered mistaking one registered user for another High security: FAR of .01% acceptable FRR (False Rejection Rate) – rejecting legitimate user High FRRs reduce usability, high FARs reduce security customer-based applications tend to raise FAR

Performance User performance depends on frequency of use: Frequent users complete faster and with fewer errors, infrequent users need step-by-step guidance and detailed feedback Degree of cooperation Total usage time (not just for matching) Quality of enrolled and presented samples has key impact (e.g. fingerprints 1 or 10 at a time?) Different performance for identification and verification (1-1 verification or 1-many identification)

"We were aiming for it to scan 12 pupils a minute, but it was only managing 5 so has been temporarily suspended as we do not want pupils' meals getting cold while they wait in the queue." Careful balancing of business process requirements and security requirements needed

Average 12-20 seconds, longer with infrequent users Total Usage Process Time quoted by suppliers often only refer to capture of live image & matching Walk up to machine Put down bags, remove hats, etc. Find token (if used) Put on token (if used) Read token Wait for live image to be captured & matched Walk away & free machine for next user Plus average number of rejections & re-tries Average 12-20 seconds, longer with infrequent users

FRR in UKPS enrolment trial Face Iris Finger Quota Time: 30.82% 39 sec 1.75% 58 sec 11.70% 1 min 13 sec Disabled 51.57% 1 min 3 sec 8.22% 1 min 18 sec 16.35% 1 min 20 sec

Performance: Smartgate Sydney Airport Problem: speedy & secure immigration Technology: Face recognition system Users: Quantas air crew (2000) Performance: FAR “less than 1%” FRR 2% “could be faster” (average 12 secs) Several re-designs necessary, including updating of image templates

Example: BKA face recognition trial Railway station with 20,000 passengers/day 2 month trial of 3 systems 200 people on watch list, who passed through every day, making no effort to conceal their identity FAR fixed at .1% (= 23 false alarms/day) Best performing system at under most favourable detected caught 60% (down to 20%)

Usability Issues: Finger Which finger? How to position Where on sensor? Which part of finger? Straight or sideways? Problems: arthritis, long fingernails, handcreme, circulation problems

Which finger?

Finger position?

Usability Issues: Iris What is it – iris or face? One or both eyes? One eye: how to focus? Distance adjustment Positioning “rocking” or “swaying” Glasses and contact lenses about half of population wear them Target area difficult to see when glasses are removed Example: Project IRIS at Heathrow

Focussing

Height adjustment Often not sufficient for very short (under 1.55 m) or very tall (over 2.10) people, or wheelchair users Need to use hand to adjust If card needs to be held, other things users carry or hold need to be put down

Height adjustment

… but users may not realise this … or be reluctant to touch equipment, or think it takes too long

Usability Issues: Face What is it? Where do I stand? Where do I look/what am I looking at? Standing straight, keeping still “Neutral expression” Hats, changes in (facial) hair, makeup

Distance

“Neutral expression”

User Acceptance Issues –Finger Hygiene, Hygiene, Hygiene Association with forensics/criminals Finger chopped off

Liveness detection Detects movement, pulse, blood flow Fitted to several systems, but tends to increase FRR Users: fine, but do the criminals know about it?

User Acceptance Issues - Iris Risk to health (e.g. damage to eyes, triggering epilepsy) Covert medical diagnosis Illnesses (iridology) Pregnancy Drugs “Minority Report” attacks

User Acceptance Issues - Face Covert identification Surveillance/tracking Direct marketing

User Acceptance – General Issues Data protection – threat to  privacy Abuse by employer, commercial organisations, state, or malicious individuals Mission creep Increasing capability of technology – e.g. iris recognition at a distance Integration with other technologies – e.g. RFID Doubts about reliability Sophisticated attackers Can government really keep systems secure? Cheap systems and successful attacks erode confidence

Attacks - Finger Simple Activate latent prints: breathing, bag with warm water Sophisticated Lift print with tape or photograph Gelatine print (gummy bear attack) – lasts 1x Silicone print

CCC strikes again Pay-by-touch system in German supermarket chain Superglue Plastic bottle cap Digital camera PC with laser printer Plastic foil Wood glue Published fingerprint of German Home Secretary

Attacks - Iris Simple Picture of eye stuck on glasses Sophisticated Coloured contact

Attacks - Face Simple Replay attack (Photo or video of person) Glasses with strong frames Sophisticated Mask (Mission Impossible attack) http://www.heise.de/ct/english/02/11/114/bild7.jpg