Keystroke Biometric ROC Experiments Team Abhishek Kanchan Priyanka
Keystroke Biometric : ROC Experiments Team Abhishek Kanchan Priyanka Ranadive Sagar Desai OC Pooja Malhotra R : c i tr e. Ning Wang m o i B e ok ts r t n s y e e K im r e Exp Keystroke Biometric: ROC Experiments
WHAT IS KEYSTROKE BIOMETRIC ? • The keystroke biometric is one of the lessstudied behavioral biometrics. • Keystroke biometric systems measure typing OC characteristics believed to be unique to an R : c i tr e individual and difficult to duplicate. m o i B e ok ts • Used for Identification r t n s y e e K im r e • Used for Authentication Exp • Developed over the past 6+ years Keystroke Biometric: ROC Experiments
Introduction to ROC Curves Used for binary decisions Signal detection – signal / no signal Medical diagnosis – disease / no disease Biometric authentication – you are the OC R : c person you claim to be / you are not i etr m o i B e ok ts r t n s y e e K im r e Exp Keystroke Biometric: ROC Experiments
Introduction to ROC Curves In biometrics the ROC curve varies from FAR=1 & FRR=0 at one end to FAR=0 & FRR=1 at other FAR = False Accept Rate – the rate an imposter is falsely accepted FRR = False Reject Rate – the rate the correct OC R : c person is falsely rejected i tr e m io B e k o s r t t n s ROC Charts are expressed in terms of percentages (0 Key erime 100%) or probabilities (0 -1). These are used Exp interchangeably. Keystroke Biometric: ROC Experiments
ROC Authentication Analogy • Supreme Court – nine judges – Usual procedure – majority required to make decision – Like 9 NN needing majority to authenticate a user • ROC Curve – effectively creates many potential procedures and provides FAR/FRR C tradeoff for each O R : (here is the m-k. NN method) ic r t e m – Need 9 votes to make decision (very conservative) io B e k o s – Need 8, 7, 6 votes to make decision (conservative) r t t n s ey rime K – Need 5 votes to make decision (majority) e p x E – Need 4, 3, 2 votes to make decision (liberal) – Need 1 or even 0 votes to make decision (very liberal) Keystroke Biometric: ROC Experiments
ROC EXPERIMENTS • Derived from four nonparametric techniques. • ‘Weak' and ‘Strong' training experiments. – Weak Enrollment data, only non-testsubject data is used to train the system. OC R : c i tr e – Strong enrollment uses test-subject data to m o i B e train the system, and then uses ok ts r t n s y e e K im r independent (different) test-subject data to e Exp test the system. • Large Data Experiments Keystroke Biometric: ROC Experiments
SYSTEM OVERVIEW O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp C Keystroke Biometric: ROC Experiments
Parametric Procedures Ø Parametric techniques are well studied. Ø Data follows a normal or Gaussian distribution. OC R : c i Ø Vary a threshold to tr e m o i obtain the tradeoff B e ok ts between FAR/FRR. r t n s y e e K im Ø Probability density r e Exp functions can be calculated without estimation. Keystroke Biometric: ROC Experiments Parametric ROC - Probability Density Function - Adapted from Cha, et al (2009)
Cha Dichotomy Model Ø Simplifies complexity Ø Transforms a feature space into a OC R : c i distance tr e vector space. B i o m e k o Ø Uses y s t r e n t s Ke erim distance xp E measures. Keystroke Biometric: ROC Experiments Multi-class to two Class Transformation Process, Adapted from Yoon et al (2005)
Pure Rank Method – m-k. NN Ø Pure Rank Method. Ø Evaluate the top 7 NN. Ø Q is authenticated if # within-class OC R : c i matches is >= tr e m o decision threshold i B e of 4 NN. s t r o k n t s y e e m K i r Ø Unweighted. All e p Ex W’s are equal in weight. Keystroke Biometric: ROC Experiments
Rank Method Weighted by Rank Order wm-k. NN Ø Authenticate if W choices are > weighted match (m) Ø Score varies from 0 to =k(k+1)/2 C O R Ø For every m, : c i FAR/FRR pair or tr e m o ROC point. i B e ok ts Ø If m=0, FAR=1, r t n s y e e FAR=0 …All users K im r e accepted. Exp Ø If m=15, FAR=small, FRR=large, few Q’s accepted. Keystroke Biometric: ROC Experiments
m-k. NN and wm-k. NN ROC’s O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp Lap. Free – Weak Training C Keystroke Biometric: ROC Experiments
Distance Threshold Method t-k. NN Ø A positive vote is within a distance threshold from the user’s sample. Ø Uses feature vector space distances only. OC R : Ø At 0, no distance c i tr e vectors are m o i B authenticated. e ok ts r t FAR=0, FRR=100%. At n s y e e K im t=100, all distance r e vectors are Exp authenticated. FAR=100, FRR=0. Keystroke Biometric: ROC Experiments
Threshold (t-k. NN) Method O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp Desk. Free (left) and Lap. Free (right) Data C Keystroke Biometric: ROC Experiments
Threshold (ht-k. NN) Method Ø Weighted vote based on distances to the k. NN. Ø Hybrid of rank method and vector C O R space distances. : c i tr e Ø For each test m o i B sample, the withine k o s r t t class weight (WCW) n s ey rime K is calculated based e p x E on the distance vectors. Desk. Free (left) and Lap. Free (right) Data Keystroke Biometric: ROC Experiments
Weak & Strong Training Percent Accuracy k. NN Performance 100. 00% 99. 00% 98. 00% 97. 00% 96. 00% 95. 00% 94. 00% 93. 00% 92. 00% 91. 00% 90. 00% O R : c i tr e m o i B e ok ts r t n s y e e K im r e p 1 E x 3 5 7 9 11 Desk. Free (WT) C 13 Nearest Neighbor Lap. Free (WT) Desk. Free (ST 18) 15 Desk. Free (ST 36) 17 19 21 Desk. Free (ST 54) Keystroke Biometric: ROC Experiments
DELIVERABLE • Deliverable 5 – Authentication Experiments – Ideal Conditions/ Weak Enrollment Part I Status – Completed • Deliverable 6 - Authentication Experiments – Ideal Conditions/ OC R : Weak Enrollment Part II t r i c e m io Status – Completed B e k o s r t t n s Key erime • Deliverable 7 – Enhance and Correct Refactor-BAS. jar ROC Exp interface Status - Completed Keystroke Biometric: ROC Experiments
DELIVERABLE 7 • Implement Perl ROC with threshold logic in JAVA. • Unify the code in Java which was supported by a Perl program earlier for calculating ROC OC R : c i tr threshold Values. e m o i B e • Maintain the performance of Perl code in Java. ok ts r t n s y e e K im r e • Some changes in User Interface of ROC Exp program. Keystroke Biometric: ROC Experiments
UI CHANGES O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp C Keystroke Biometric: ROC Experiments
O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp C Keystroke Biometric: ROC Experiments
O R : c i tr e m o i B e ok ts r t n s y e e K im r e Exp C Keystroke Biometric: ROC Experiments
TEAM COMMUNICATION • Google Group for information sharing and discussion • Skype Meetings • Emails • • OC R : c i Personal Meetings tr e m o i B e Documented Minutes of Meeting ok ts r t n s y e e K im r e Team Website status updates Exp Assigned Task progress check by team leader Keystroke Biometric: ROC Experiments
Questions? OC R : m o i B e ok ts r t n s y e e K im r e Exp ic r t e Keystroke Biometric: ROC Experiments
- Slides: 23