Secure graphical password system for high traffic public areas Bogdan Hoanca and Kenrick Mock University of Alaska Anchorage
Outline Shoulder surfing as security threat in information systems Eye tracking based authentication Error rates of eye tracking hardware Error-aware eye tracking Systematic errors Random errors Conclusions March 27, 2006 Hoanca/Mock ETRA 2006
Shoulder surfing Stealing authentication information Critical threat for mobile users or in public places Safest bet: assume “naked” user under constant surveillance March 27, 2006 Hoanca/Mock ETRA 2006
Defending against shoulder surfing Screen filters Challenge-response schemes Physical key schemes Biometric schemes March 27, 2006 Hoanca/Mock ETRA 2006
Eye tracking based authentication Use the eye tracker without on-screen feedback to select on-screen objects Ideally, transparent for the user Secure from shoulder surfing Slower than typing Still vulnerable to key logger and screen capture programs March 27, 2006 Hoanca/Mock ETRA 2006
Graphical Password Entry via Eye Tracking March 27, 2006 Hoanca/Mock ETRA 2006
Sample authentication log March 27, 2006 Hoanca/Mock ETRA 2006 Image size 700x482
High error rates due to hardware limitations “Low error” user Mean D = 12 pixels “High error” user Mean D = 30 pixels Plots of actual gaze location as compared with intended target (red); black is the center of gravity Distances are in pixels and scale is -40…40 in both X and Y March 27, 2006 Hoanca/Mock ETRA 2006
Error rates of eye tracking hardware Using the ERICA system from Eye Response Technologies Error types Systematic errors Due to head tilt Slowly varying with time Dependent on screen geometry and location Random errors Highly user dependent March 27, 2006 Hoanca/Mock ETRA 2006
Handling random errors Loss of cryptographic complexity depends on how much error is acceptable Success rate (%, 0…100) vs. distance in pixels (1…1000, log scale) March 27, 2006 Hoanca/Mock ETRA 2006
Handling systematic errors Red – raw data Green -- corrected Success rate (%, 0…100) vs. distance in pixels (1…1000, log scale) one user 25 sessions “low error” Average over multiple attempts or over multiple users Loss of cryptographic complexity – equivalent to one click less March 27, 2006 Hoanca/Mock ETRA 2006
Handling systematic errors (continued) Success rate (%, 0…100) vs. distance in pixels (1…1000, log scale) one user 25 sessions “high error” Limited usefulness for high error users Red – raw data Green -- corrected March 27, 2006 Hoanca/Mock ETRA 2006
Summary Eye tracking is a promising technology for authenticating from public places with reduced danger of shoulder surfing Wide acceptance will require eye tracking technologies that are More stable and accurate Ideally, head tracking-capable and calibration-free Much lower in price March 27, 2006 Hoanca/Mock ETRA 2006