Biometrics

Individual physiological and/or behavioural characteristics

Two types of biometrics

  • Physical biometrics (ex: DNA, fingerprints, iris, palmprint, etc.)
  • Behavioural biometrics (ex: gait, keystrokes, signatures, multi-touch gestures, etc.)

Advantages of biometrics

  • All the users of the system have an equal level of security (There's no password choice and doesn't require you to remember anything)

Disadvantages of biometrics

  • Speed is perceived as the biggest problem
  • FMR will increase when scaling up an identification application
  • Irrevocability (Unlike passwords, you can't change leaked biometric data)

Biometric Hypothesis Testing

Hypothesis testing
  • H0 (null hypothesis): The inputted biometric measurement doesn't match the biometric data template of the person that they're trying to authenticate themselves as
  • H1 (alternative hypothesis): The inputted biometric measurement matches the biometric data template of the person that they're trying to authenticate themselves as
Decisions
  • D0 (null decision): The user is not who they claim to be
  • D1 (alternative decision): The user is who they claim to be
Hypothesis testing formulation errors types
  • Type a - False Match (D1 when H0)
  • Type b - False Non-Match (D0 when H1)

Biometric models

Unimodal biometrical systems
  • A biometric system that uses a single biometric measurement to authenticate users
Multimodal biometrical systems

  • The fusion of multiple biometric features together to build a more complex template and make a more accurate and efficient biometric system

  • Data fusion levels \n

    • Fusion at the sensor level \n

    • Fusion at the feature level

    • This combination strategy is usually done by a concatenation of the feature vectors extracted by each feature extractor

    • This yields an extended-size vector set

      • Two drawbacks
      • There is little control over the contribution of each vector component to the result
      • Both feature extractors should provide identical vector rates

    • Although it is a common belief that the earlier the combination is done, the better result is achieved, state-of-the-art data fusion relies mainly on the opinion and decision level

       Fusion at the feature level

  • Fusion at the opinion level

    • The score must be adjusted first (Normalization must be done):

      • The similarity measures must be converted into distance measures
      • The score generated by each classifier must have the same range

    • The combination strategies can be classified into three main groups

      • Fixed rules / equal weight
      • Trained rules / unequal weight
      • Adaptive rules / adaptive weight

         Fusion at the opinion level

  • Fusion at the decision level

    • The Borda count method can be used for combining the classifiers' outputs (This approach overcomes the scores normalization that was mandatory for the opinion fusion level)

    • One problem that appears with decision-level fusion is the possibility of ties

    • For verification applications, at least three classifiers are needed

    • An important combination scheme at the decision level is the serial and parallel combination (AND and OR combination)

      • the AND combination improves the False Acceptance Ratio
      • the OR combination improves the False Rejection Ratio

         Fusion at the decision level

Biometric system

A biometric system is essentially a pattern recognition system

What biological measurements/features qualify to be a biometric?

  • Universality (Everyone must possess one)
  • Distinctiveness (It should be only limited to you)
  • Permanence (Should be the same all the time)
  • Collectability (It should be collectable)

Other issues to consider when thinking of a biometric measurement/feature to use in your biometric system

  • Performance (How fast is your system? etc.)
  • Acceptability (How acceptable is your biometric? etc.)
  • Circumvention (How circumventable is your system? etc.)

A biometric system is designed using four main modules

  • Sensor module
  • Feature module
  • Matcher module
  • System database module

A biometric system may operate either in

  • Verification mode (Does this biometric data belong to this person?, etc.)
  • Identification mode (Whose biometric data is this?, etc.)

Two types of biometric systems errors

  • False Match - Mistaking biometric measurements from two different people to be from the same person
  • False Non-Match - Mistaking two biometric measurements from the same person to be from two different people

Important specifications in a biometric system

  • False Match Rate (FMR)
  • False Non-Match Rate (FNMR)
  • Failure to Capture (FTC) ex: faint fingerprints, etc.
  • Failure to Enroll (FTE) ex: not long enough contact with the sensor, etc.
  • etc.

Biometric protection templates

Template protection

  • protects the privacy and security of biometric features
  • revokes and re-issues biometric templates if any leaks were to occur
  • prevents linking across databases
  • allows matching in an encoded space
  • etc.
Three categories of template protection
  • Straight feature protection
    • Protecting the original biometric features by using some one-way transformation to encrypt the data
    • The matching is done in an encoded space

 Biometric template protection: Straight feature protection

  • Key-generating
    • The biometric data goes into a key generation or hashing algorithm and then outputs a key
    • Very difficult to do since the same biometric must output the same key, meaning that you have to ensure that your biometric systems are exact

     Biometric template protection: Key-generating

  • Key-binding
    • You generate a random key using an algorithm and you mix that key with your biometric
    • The database stores a mixture of the template and the randomly generated key
    • To authenticate yourself, you simply provide the biometric data that you used and the algorithm removes it from the data set, which should leave you with a key, and if you get the correct key, then you get authenticated and so on
    • Irreversible

 Biometric template protection: Key-binding

Template protection schemes
  • Robust hashing
    • A robust hash function maps two similar inputs to the same hash values whereas inputs that are significantly apart hash to unpredictable hash values

     Biometric protection template scheme: Robust hashing

  • Cancelable biometrics \n

   Biometric protection template scheme: Cancelable biometrics

  • Fuzzy vaults
    • Not specific to biometric data, but typically applied to minutiae-based fingerprint matches as a key binding biometric cryptosystem

     

  • Fuzzy vaults vulnerabilities

    • Chaff Point Identification
    • Improved brute force attack
    • Correlation attack / Key attack
    • Hill Climbing
    • etc.

  • Fuzzy commitments (Another well-known key binding approach)

    • Enrollment
    • Commit a codeword (C), that'll act as the key, of an error-correcting code using a fixed-length biometric feature vector (X) as a witness
    • Store a hash (h) of C as "helper data"
    • Fuzzy commitment requires a fixed-length feature vector representation of a biometric modality

       

  • Fuzzy extractors (A key binding cryptosystem)
    • The goal is to extract a uniformly random string (R) from its input (w) in a noise-tolerant way (If the input changes in some way, but remains close, the string (R) can still be reproduced exactly)
    • An attractive proposition, but difficult due to intra-user variability

     \n