close

Вход

Забыли?

вход по аккаунту

код для вставкиСкачать
Biometrics II
CUBS, University at Buffalo
http://www.cubs.buffalo.edu
http://www.cedar.buffalo.edu/~govind/CSE717
[email protected]
Maintaining Biometric Databases
 Enrollment
 Base truth established through seed documents
 Positive Enrollment
 Negative Enrollment
 Problems in enrollment
 Biometric System is only as secure as its enrollment
 Fake Identities
 Duplicate Identities
 Quality
 The biometric zoo
 Sheeps,Goats,Wolves,Lambs
Large Scale Databases




How large is ‘large scale’?
Application
 US-VISIT
 National ID – Thailand, Hong Kong
 Voter Registration – Peru
Accuracy issue
 FARN= N*FAR
 FRRN = FRR
 Number of False accepts scales as O(N2)
Scalability and retrieval
 Indexing
 Filtering
 Binning
Indexing Problem
 Biometric template have no natural order
 Natural groups cannot be inferred from statistical descriptors
Indexing Problem: Solutions
Binning and pruning
Natural feature representation
Combination of biometric matchers
•Combination of the matching results of different biometric features
provides higher accuracy.
Fingerprint
matching
Signature
matching
Hand geometry
matching
Alice 26
Bob 12
:
:
Alice 0.31
Bob 0.45
:
:
Alice 5.54
Bob 7.81
:
:
Combination
algorithm
Alice 0.95
Bob 0.11
:
:
Sequential combination of matchers
Fingerprint
matching
Combination
algorithm 1
No
Signature
matching
Yes
Combination
algorithm 2
No
Hand geometry
matching
Desired confidence
achieved?
Desired confidence
achieved?
Combination
algorithm 3
Yes
Alice 0.95
Bob 0.11
:
:
Information Fusion






Multiple sensors
 Compensates environmental variation
Multiple samples
 Allows interaction induced variation
Multiple instances of the same biometric
Multiple matchers
 May use multiple representations
 Serial and hierarchical combination
Multi-modal biometrics
 Feature level Fusion
 Score level Fusion
 Decision level Fusion
 Loosly coupled integration
Combination of biometrics and tokens
Multimodal biometrics:Fusion approaches
Security of biometric data


Issues in biometrics
 Biometrics is secure but not secret
 Permanently associated with user
 Used across multiple applications
 Can be covertly captured
Types of circumvention
 Denial of service attacks
 Fake biometrics attack
 Replay and Spoof attacks
 Trojan horse attacks
 Back end attacks
 Collusion
 Coercion
Attacks on a Biometric System
Biometric
Sensor
Feature Extraction
Database
Enrollment
Biometric
Sensor
Feature Extraction
Matching
ID : 8809
Authentication
Result
Gummy finger
Fake biometrics
Synthetic fingerprint
Securing biometric templates






Liveness testing
Cancelable biometrics
 Signal domain distortion
 Feature domain distortion
Hashing
 Uses non invertible transform
Watermarking and steganography
 Ensures that sensor input can be trusted
Encryption and digital signature
 Combines security of encryption with non-repudiation of
bometrics
Challenge response systems
 Conversational biometrics
 CAPTCHAs
Securing password information
Hashing
Cancelable biometrics
Privacy, social and ethical concerns




Aspects of privacy
 Secrecy
 Solitude
 Anonymity
Unintended functional scope
 Retinal pattern is capable of revealing diabetes
 Hand irregularities can be correlated with genetic defect
 This information may be used for discrimination
Unintended application scope
 Can compromise anonymity of an individual
 Biometric can be used for covert surveillance
 Can be used to track purchases leading to spam
Social acceptance
 Stigma associated with biometrics
 Reduces level of expected privacy
Avenues for research





Face recognition
Palm verification
Key stroke dynamics
Speaker recognition
Soft biometrics





Multimodal biometrics
Classifier combination
Binning and Indexing
Security of biometrics
Interchange standards
and API
Summary








Enrollment into biometric databases
Large scale databases
Filtering, binning and indexing biometric templates
Combination of classifiers
Multimodal biometrics
Attacks on Biometric System
Securing Biometric Templates
Privacy Social and Ethical Concerns
Thank You
[email protected]
1/--страниц
Пожаловаться на содержимое документа