BIOMETRICS IN ABC: COUNTER-SPOOFING RESEARCH Hong Wei, Lulu Chenand James Ferryman Computational Vision Group School of Systems EngineeringUniversity of Reading, UK11th October 2013
Outline of the presentation • Introduction • Research on counter spoofing attacks • for face • for fingerprint • for iris • Challenges and future research
Introduction • ABC requires: fast, secure identity verification. • EU FastPass project (www.fastpass-project.eu): harmonised, modular reference system for future European ABC. • Biometrics: face, fingerprint and iris – important for ABC. • Potential vulnerability: sensor-level attacks – “spoofing”. • Counter-spoofing measure: crucial for FastPass. FastPass Heavy traffic and pressure at current border control – how to ease these?
Face counter spoofing approaches • Three categories in developing face anti-spoofing algorithms • Motion analysis: make use of significant difference between motions of planar objects and real face (3D) in optical flow fields. • Texture analysis: extract image texture features which reflects difference between real face and printed or replayed faces. • Liveness detection: detection of life signs such as eye blinking, lip movement, etc.
Relevant competitions • Competitions on counter measures to 2D facial spoofing attacks • First with IJCB 2011: 6 teams [Chakka, et al 2011] . • Second with ICB 2013: 8 teams [Chingovska, et al 2013] • Participants: academic teams • Replay-Attack face spoofing database: • Printed photographs • Photographs on a mobile device (iPhone / iPad) • Videos replayed on a mobile device
Summary of the 2013 competition Methods used in algorithms development • Database: Replay-Attack Face Spoofing Database • Teams 1&4 achieved 100% accuracy in both development and test. • All 8 teams developed highly sophisticated methods, and some introduced methods beyond the three categories, such as human pulse.
Open issues and challenges • Facial features: • Change over time • Similarity between family members: e.g. twins, father and son. • Capture: • Moving subject • Efficient sensor-level fusion (VR+NIR, stereo vision) • 2D face only: 3D attacks should be added, e.g. 3D face masks.
Fingerprint sensing technologies • Optical sensors • Relatively big in size • In border control, biometric enrolment • Solid state fingerprint sensors • Different types, most common: capacitive • Can be compact • Sensitive to electrostatic discharges • Multiple spectral imaging sensor • 3D scanner • Can be touchless
Fingerprint counter-spoofing methods • Solutions: Hardware and Software based • Hardware: • Pulse oximetry, smell, temperature, blood pressure, heart beat etc. • Requires additional hardware • Software: • Perspiration pattern • Skin distortion • Pores • Image quality measure
Software based Pore distribution Perspiration patterns Skin distortion Image quality measure Real Fake
Fingerprint counter-spoofing competitions • LivDet: Fingerprint Liveness Detection Competition • Since 2009, every two years • Software based methods • Multiple sensors: • Optical and swipe • Various reproduction materials: • Gelatin • Silicon • Play-doh • Latex
Fingerprint LivDet • Ferrfake: a false acceptance of a spoof image. • Ferrlive: a false rejection of a live subject.
Open issues and challenges • Various spoofing materials • Recent approaches: • Handles certain types well • Poor performance on other materials • New methods should: • Detect and handle all types of spoofing materials • Combine successful algorithms • Enable more balanced ferrfake and ferrlive rates • Touchless fingerprint scan
Iris counter-spoofing approaches • Spectrographics based: optical properties • Purkinje reflection • Retina light reflection: ‘Red eye’ effect • Image quality measure • Behaviour based: dynamic properties • Eye hippus • Pupil and iris constriction and dilation • Eyelid blinking • Other: • 3D structure
Iris counter spoofing competitions • LivDet-Iris: First Liveness Detection-Iris Competition 2013 • Held by IEEE BTAS Conference (September 29 – October 2, 2013)
Open issues and challenges • FastPass: • Data capturing: involves additional noise • Small target • Long-distance • On-the move • Illumination, focus • Research: • Data collection: lack of data for training • Lack of standardisation on spoofing experimental datasets
Challenges • Current biometric systems: • Have counter-spoofing mechanism integrated. • However, research show vulnerability from spoofing attacks (e.g. TABULA RASA Spoofing Challenge 2013). • Arms race: between spoofing and counter-spoofing. • Spoofing using new technologies, e.g. Hand-held mobile devices. • Increased throughput and intuitive user-friendly devices. • Lack of training data. • Data interchange format. • E-Passport storage capacity.
Future of counter spoofing • Multimodal: • Increase difficulty to spoof multiple traits. • Not every person has all biometric features. • Multi-sensor system: • Visible range and Near IR • 2D and 3D • Combined approaches: • Handle different types of materials. • Tackle continuously updated and practised spoofing attacks. • Increase difficulty for replication process.
Future research • Explore and develop new algorithms on counter spoofing measures and attacks in ABC applications. • Assemble ABC-specific databases with more realistic spoofing attacks. • Create a framework for counter spoofing measures. • Analyse data fusion effects at a quantitative level: • At the feature level for a single trait; • At the decision (score) level for multiple traits; • Combination of feature fusion and trait fusion; • Adaptive fusion schemes taking additional information (quality, thresholds) into account.
Summary • ABC demands harmonized capturing devices adopting international standards • non intrusive and enable remote biometric capture • faster, more secure, efficient, seamless • robust counter spoofing detection Subject to appropriate legal and ethical controls
Thank you.& Questions www.fastpass-project.eu