Keystroke Biometric System Client: Dr. Mary Villani Instructor: Dr. Charles Tappert Team 4 Members:...

Preview:

Citation preview

Keystroke Biometric SystemClient: Dr. Mary VillaniInstructor: Dr. Charles Tappert

Team 4 Members: Michael Wuench ; Mingfei Bi ; Evelin Urbaez ; Shaji Mary Varghese ; Michael Tevnan

Contents

1. Introduction1. Introduction

2. New System2. New System

3. Experimental Results 3. Experimental Results

4. Conclusion4. Conclusion

5. Future Studies5. Future Studies

6. Demo6. Demo

Previous Work

Pace University has 5+ years in keystroke biometrics research.

Built a complex system of interworking JAVA and PHP programs to support academic research in biometrics.

System can successfully identify and authenticate individuals with a relatively high degree of accuracy especially during same time periods.

Introduction

Objectives

• Modifies the existing systems toward practical usage.

• Attempts to verify users taking an online test based on the characteristics of their typing.

• Analyze results on new input data.

• Present possible methods for determining instant authentication.

Current Study

Results

New System Overview

SubjectSubjectRegistrationRegistration

SubjectSubjectRegistrationRegistration

Classifier (BAS)Classifier (BAS)Classifier (BAS)Classifier (BAS)

Test Taker AppletTest Taker AppletSubjectSubject

DemographicDemographic

KeystrokeKeystrokeEntryEntry

KeystrokeKeystrokeEntryEntry

Raw FileRaw File

Actual Text Actual Text FileFile

Client Client Reviews TestReviews Test

Feature Extractor (BioFeature)Feature Extractor (BioFeature)Feature Extractor (BioFeature)Feature Extractor (BioFeature)

New System

New System consists of the following:

1. A PHP Website registers the user.

2. A modified Java applet captures 300 keystrokes and produces two files: a raw data file and a text file.

3. A Java program, BioFeature, extracts 239 feature measurements.

4. A Java program, Biometric Authentication System (BAS), performs authentication tests.

Test-Taker Authentication System

Feature Extraction – BioFeature Extracts 239 features from raw data collected

from applets Ex. Features file

Authentication Classifier Uses 2 features files One is trained-on and the other is tested-on. Returns False Acceptance Rate (FAR), False

Rejection Rate (FRR) & combined performance Ex. BAS results (html file)

Authentication Classifier

Authentication Transformation

feature space (left) feature distance space (right)

Experimental Design: Data Sets

1. Team member Data: 5 samples of free text using enrollment applet

(~650 keystrokes) 5 samples of free text data using test-taker

applet (~300 keystrokes)2. Outsider Sample:

5 samples of free text data using test-taker applet (~300 keystrokes)

3. Original 36 subjects from 2006 Study: 5 samples of laptop free text using

enrollment applet (~650 keystrokes)

Test | Train

5 | 5

5 | 5

6 | 5

5 | 6

FRR

0.0%(0/50)

6.0%(3/50)

0.0%(0/60)

12.0%(6/50)

FAR

4.8%(12/250)

1.2%(3/250)

11.2%(42/375)

0.4%(1/250)

Performance

96.0%(288/300)

98.0%(294/300)

90.3%(393/435)

97.7%(293/300)

1

2

3

4

Experimental Results

Biometric Authentication System (BAS) results using test and enrollment samples (5 per subject) collected in the fall of 2008

Study 1: Fall 2008 Team with Outsider

Performance is high (at least 95%) with same subject testing

No significant difference in results due to keystroke length (300 vs 650)

Immediate drop in performance when an subject that is not enrolled is used.

Increased number of subjects is recommended

Study 1 Conclusions

Study 2 (partial)

Test | Train FRR FAR Performance

10 | 5 0.0% (0/225) 27.7% (277/1000) 77.4% (948/1225)

10 | 10 8.4% (19/225) 8.5% (85/1000) 91.5% (1121/1225)

10 | 15 7.1% (16/225) 5.9% (59/1000) 93.9% (1150/1225)

10 | 20 25.3% (57/225) 1.8% (18/1000) 93.9% (1150/1225)

10 | 25 29.3% (66/225) 1.4% (14/1000) 93.5% (1145/1225)

10 | 30 44.0% (99/225) 0.9% (9/1000) 91.2% (1117/1225)

10 | 36 39.6% (89/225) 0.6% (6/1000) 92.2% (1130/1225)

Study 2: Original-36 Training-on Tests

Testing on combined fall 2008 enrollment and test-taker samples (10 per subject) and training on original-36 subject samples (5 per subject).

Study 2 Conclusions

Again, keystroke length has little effect on results.

When the number of subject is large (30+), it produces a very low FAR (should be as low as possible for maximum security).

Performance increases (above 90%), FAR decreases, and FRR increases as # of subject is 10 or more.

Study 3 (partial)

Test | Train FRR FAR Performance

5 | 10 10.0% (5/50) 11.2% (28/250) 89.0% (267/300)

10 | 10 3.0% (3/100) 23.2% (232/1000) 78.6% (865/1100)

15 | 10 2.7% (4/150) 21.6% (216/1000) 80.8% (930/1150)

20 | 10 5.0% (10/200) 57.5% (575/1000) 51.3% (615/1200)

25 | 10 2.0% (5/250) 68.1% (681/1000) 45.1% (564/1250)

30 | 10 1.3% (4/300) 72.0% (720/1000) 44.3% (576/1300)

36 | 10 0.3% (1/360) 78.5% (785/1000) 42.2% (574/1360)

Study 3: Original-36 Testing-on Tests

Testing on original 36 subject samples (5 per subject) and training on combined fall 2008 enrollment and test-taker samples (10 per subject).

Study 3 Conclusions

Yet again, keystroke length has little effect on results.

Overall poor performance indicates that system requires adequate training data.

FAR increases substantially, and FRR decreases as # of subject is 10 or more.

Study 2 and Study 3 hint that 30 or more subjects will yield a more reliable authentication.

Future Studies

Convert the current Java programs to web applications using J2EE or PHP.

Further testing should be done with at least 30 enrolled subjects.

Use the test taker applet as individual samples to test against a large enrollment database.

Continue to modify the Authentication Classifier (BAS) to implement the proposed k-nearest-neighbor procedure. (NEXT)

Simple k-nearest-neighbor procedure

W

4

8

7

19>=Accept

B

6

2

3

11

1

2

3

Totals

Proposed authentication using k-nearest-neighbor procedure.

Matching sets:

[W B B W B B W W B B]

[W W B W W W W W B W]

[W B W W B B W W W W]

K = 10 (the 10 nearest neighbors) W = within (accept) class B = between (reject)

Demo: Test-Taker Authentication System

How to access the system (Taking the Test ):

User must first enroll into the system.

Click on the Web Link (http://utopia.csis.pace.edu/cs691/2008-2009/team4/testtakersite).

The enter same name you enrolled with and take the test in the applet

Demo: Taking the Test

Presented with five questions and must provide five answers.

Answer should be more than 50 words, based on the assumption that words are approximately 6 keystrokes in length.

Once completed, click on “Submit”

If 50 words are not meet, user will be presented with an error.

Demo: New Online Test System

Logging In

New Online Test System

Test Applet

New Online Test System

Test Applet (continued)

New Online Test System

User reached at least 300 keystrokes

Demo: Test Applet

User did reach or surpass the 300 keystrokes

Q&A Thank You!Pace University

Recommended