cs 691 - team 5 alex wong raheel khan rumeiz hasseem swati bharati biometric authentication system

CS 691 - Team 5CS 691 - Team 5Alex WongRaheel KhanRumeiz HasseemSwati Bharati

Biometric Biometric Authentication Authentication

SystemSystem

Project ObjectivesProject Objectives Develop a biometric authentication systemDevelop a biometric authentication system Application coded in JavaApplication coded in Java Determine the feasibility of the Dichotomy Determine the feasibility of the Dichotomy

Model Model Report results using standard authentication Report results using standard authentication

system performance statistics system performance statistics

Biometric Biometric Authentication SystemAuthentication System

CS 691 - CS 691 - Team 5Team 5

Dichotomy ModelDichotomy Model

A statistically inferable approach to establishing the individuality of a biometric

Classifies two biometric samples as Classifies two biometric samples as coming either from the same person coming either from the same person (intra-variation) or from two different (intra-variation) or from two different people (inter-variation) people (inter-variation)

Uses distance measure between two samples of the same class and between those of two different classes



Objective of Dichotomy Objective of Dichotomy ModelModel

Validation of individuality of biometric data statistically

Not the detection of differences of specific instances

Find the individuality of the entire population based on the individuality of a sample of n people, where n is much less than the population.

Allows inferential classification of individuals where large classes are involved and the whole population is not available for sampling



Dichotomy vs. Dichotomy vs. PolychotomyPolychotomy

Binary decision, yes/no Authentication or

Verification process A user is verified as

being the person s/he claims to be

More suitable for establishing individuality of a person, where number of classes is too large to completely sample, eg. population of an entire nation.

One-of-many decision Identification process A user is identified

from within a population of n users One-of-n responseOne-of-n response



Original Feature Vector Original Feature Vector Data FileData File



Dichotomy Converted Dichotomy Converted FileFile



Dichotomy Conversion Dichotomy Conversion ExampleExample First row :First row :

SAME , 254 , |0.11431427822210534 - 0.0|,..SAME , 254 , |0.11431427822210534 - 0.0|,..

Fifth row :Fifth row : DIFF, 254, |0.11431427822210534 - 0.32848686484618|,..DIFF, 254, |0.11431427822210534 - 0.32848686484618|,..

Total number of Total number of Intra (SAME) class data samples :Intra (SAME) class data samples :

m * (m-1) * n /2 m * (m-1) * n /2 Inter (DIFF) class data samples :Inter (DIFF) class data samples :

m * m * n * (n-1) /2m * m * n * (n-1) /2 Where Where

n = number of subjectsn = number of subjects m = number samples from each subjectm = number samples from each subject

For the given example : For the given example : Intra-class size = 40; Inter-class size = 150; Intra-class size = 40; Inter-class size = 150;

n=4, m=5n=4, m=5 Biometric Biometric Authentication SystemAuthentication System


Polychotomy to Polychotomy to Dichotomy ConversionDichotomy Conversion

Reference:http://www.icgst.com/gvip/v5/P1150511001.pdf Biometric Biometric

Authentication SystemAuthentication SystemCS 691 - CS 691 - Team 5Team 5

System EvaluationSystem Evaluation

FRR (False Reject Rate)FRR (False Reject Rate) Same person’s biometric data identified Same person’s biometric data identified

as coming from two different peopleas coming from two different people FAR (False Accept Rate)FAR (False Accept Rate)

Biometric data provided by two Biometric data provided by two different people are classified as coming different people are classified as coming from the same personfrom the same person

System PerformanceSystem Performance Biometric data correctly classifiedBiometric data correctly classified



Project SpecificationsProject Specifications

Convert training and testing files of Convert training and testing files of n-class feature data into files of 2-n-class feature data into files of 2-class (inter and intra-class) class (inter and intra-class) dichotomy-model feature datadichotomy-model feature data

Prepare sets of inter and intra-class Prepare sets of inter and intra-class data for training and testingdata for training and testing

Implement the nearest-neighbor Implement the nearest-neighbor technique to obtain accuracy results technique to obtain accuracy results on the data (Euclidean distance)on the data (Euclidean distance)



Application Design Application Design DecisionsDecisions

Allows for users to save Test Dichotomy Allows for users to save Test Dichotomy Data both intra and inter class data setsData both intra and inter class data sets

Allows for users to also save the Train Allows for users to also save the Train Dichotomy Data both intra and inter class Dichotomy Data both intra and inter class data setsdata sets

Users are able to view a log file of what Users are able to view a log file of what action is currently being executedaction is currently being executed

Results can be saved as a .html file to easily Results can be saved as a .html file to easily save and distribute themsave and distribute them

GUI is simple, clear and easy to useGUI is simple, clear and easy to useBiometric Biometric


Application Application DemonstrationDemonstration




Biometric Authentication System Biometric Authentication System TutorialTutorial

Experimental ResultsExperimental Results

Experiments Performed on data Experiments Performed on data obtained from obtained from Mouse Movement biometric systemMouse Movement biometric system Stylometry biometric systemStylometry biometric system Keystroke biometric systemKeystroke biometric system

Results showResults show Overall System Performance %Overall System Performance % FRR (False Reject Rate) %FRR (False Reject Rate) % FAR (False Accept Rate) %FAR (False Accept Rate) %

Mouse Movement Mouse Movement ResultsResults

Different subjects same conditions Different subjects same conditions

Intra-Inter class SizesFRR (%) FAR (%) Performance (%)

Train Test

1455-5100 1005-3000 64.37 22.83 66.74

1005-3000 1455-5100 58.48 23.64 68.61



Training set : 115 samples from 5 subjects

30 samples each from 3 subjects, 15 samples from 1 subject, 10 samples from 1 subject

Testing set : 90 samples from other 5 subjects

10 samples from 3 subjects, 30 samples each from 2 subjects

Mouse Movement Mouse Movement ResultsResults

Using all subjects; train and test sets captured Using all subjects; train and test sets captured 3 weeks apart3 weeks apart


Train Test

225-1000 225-1000 72.00 16.30 73.46

225-1000 225-1000 78.66 16.30 72.24



Training set : 50 samples from all 5 subjects

10 samples from each 5 subjects

Testing set : 50 samples from all 5 subjects

10 samples from each 5 subjects ; approximately 3 week interval

Stylometry Results Stylometry Results Different subjects same conditionsDifferent subjects same conditions


Train Test

270-1500 270-1500 91.11 10.80 76.94

270-1500 270-1500 73.70 24.86 67.68







Stylometry ResultsStylometry ResultsTrain and test set on all subjects by dividing Train and test set on all subjects by dividing

the samplesthe samples


Train Test

120-1650 120-1650 93.33 5.27 88.75

120-1650 120-1650 85.83 10.12 84.74



Training set : 60 samples from all 12 subjects


Testing set : 60 samples from all 12 subjects


Keystroke ResultsKeystroke ResultsDifferent Subjects Same Conditions

Conditions

Intra-Inter Class Sizes FRR

(%)FAR (%) Performance (%)

Train Test

Desktop/Copy

180-3825 180-3825 11.11 6.01 93.75

Laptop/Copy

180-3825 180-3825 7.77 4.36 95.48

Desktop/Free

171-3570 176-3740 28.40 1.39 97.39

Laptop/Free

180-3825 180-3825 15.55 3.73 95.73






5 samples from each 18 subjects; all intra-inter data used

Keystroke ResultsKeystroke ResultsDifferent Subjects Same Conditions – Using a

randomized set of 500 inter-class data

Conditions

Intra-Inter Class Sizes FRR (%)

FAR (%)

Performance (%)Train Test

Desktop/Copy

180-500 180-500 10.00 13.40 87.50

Laptop/Copy

180-500 180-500 1.66 10.20 92.05

Desktop/Free

171-500 176-500 18.75 5.00 91.42

Laptop/Free

180-500 180-500 9.44 10.80 89.55






5 samples from each 18 subjects; 500 intra-inter sets used

Keystroke ResultsKeystroke ResultsTest results for “old” keystroke data (180

samples : 36 subjects 5 samples each) on same subjects and different conditions.

Conditions Intra-Inter Class Sizes FRR (%)

FAR (%)

Performance (%)Train Test Train Test

Desktop/Copy

Desktop/Free

360-500 347-500 8.06 17.80 86.18

Desktop/Free

Desktop/Copy

347-500 360-500 3.33 13.00 91.04

Laptop/Copy

Laptop/Free

360-500 360-500 3.61 40.40 75.00Laptop/

FreeLaptop/Copy

360-500 360-500 5.83 3.40 95.58

Desktop/Copy

Laptop/Copy

360-500 360-500 5.27 6.80 93.83

Laptop/Copy

Desktop/Copy

360-500 360-500 4.72 18.00 87.55

Desktop/Free

Laptop/Free

347-500 360-500 3.05 38.80 76.16

Laptop/Free

Desktop/Free

360-500 347-500 8.93 6.80 92.32

Desktop/Copy

Laptop/Free

360-500 360-500 5.83 22.20 84.65

Laptop/Free

Desktop/Copy

360-500 360-500 5.27 8.79 92.67

Desktop/Free

Laptop/Copy

347-500 360-500 1.66 14.39 90.93

Laptop/Copy

Desktop/Free

360-500 347-500 3.17 27.60 82.40Biometric Biometric


Keystroke ResultsKeystroke ResultsLongitudinal authentication test results on

same subjects and conditions but at two-week data collection interval.

Condition

Intra-Inter Class Sizes FRR (%) FAR (%) Performance (%)

Train Test

Desktop/Copy

40-150 40-150 2.50 4.66 95.78

Laptop/Copy

40-150 40-150 2.50 10.00 91.57

Desktop/Free

40-150 40-150 0.00 4.66 96.31

Laptop/Free

40-150 40-150 0.00 10.00 92.10



Training set (baseline) : 20 samples from 4 subjects


Testing set (2-week interval): 20 samples from 4 subjects


Keystroke ResultsKeystroke ResultsLongitudinal authentication test results on

same subjects and conditions but at four-week data collection interval.

Condition

Intra-Inter class Sizes FRR (%) FAR (%) Performance (%)

Train Test

Desktop/Copy

40-150 40-150 2.50 12.66 89.47

Laptop/Copy

40-150 40-150 0.00 0.00 100.00

Desktop/Free

40-150 40-150 2.50 1.33 98.42

Laptop/Free

40-150 40-150 0.00 8.00 93.68



Training set (baseline) : 20 samples from 4 subjects


Testing set (4-week interval): 20 samples from 4 subjects


Project AchievementsProject Achievements

Utilized the dichotomy model in the authentication of biometric data obtained from the Keystroke, Stylometry and Mouse Movement biometric systems.

Sought to establish that the dichotomy model is the preferred model over the polychotomy model when dealing with an enormous number of classes where the whole population is not available for sampling, that it is the statistically inferable approach.



Summary of ResultsSummary of Results For the mouse movement and stylometry For the mouse movement and stylometry

biometric data – small number of users biometric data – small number of users (classes) (classes) System performance : between 66% and 76% System performance : between 66% and 76% FAR and FRR : highFAR and FRR : high

For the keystroke biometric data - large For the keystroke biometric data - large number of users (classes)number of users (classes) System performance : above 90% in most System performance : above 90% in most

casescases FAR : less than 15% in most cases FAR : less than 15% in most cases FRR : almost always less than 10%.FRR : almost always less than 10%.



ConclusionConclusion

The results on the keystroke The results on the keystroke biometric data are encouraging and biometric data are encouraging and indicate that the dichotomy model indicate that the dichotomy model may be a feasible solution to the may be a feasible solution to the authentication problem when a large authentication problem when a large number of classes are involved. number of classes are involved.



Future WorkFuture Work

Comparative analysis of the Comparative analysis of the dichotomy authentication results dichotomy authentication results with polychotomy authentication with polychotomy authentication results obtained on the same results obtained on the same keystroke biometric data. keystroke biometric data.

Study to see whether the results for Study to see whether the results for the mouse movement and stylometry the mouse movement and stylometry data improved significantly as the data improved significantly as the sample sizes increased.sample sizes increased.



Biometric Biometric


Please Visit Our Please Visit Our WebsiteWebsite

ahttp://utopia.csis.pace.edu/cs691/2007-2008/team5/index.html

To obtain the latest downloads

and information please visit us online.

Thank youThank you



cs 691 - team 5 alex wong raheel khan rumeiz hasseem swati bharati biometric authentication system

Documents