BIOMERTICS

Lec1
What is Biometrics?
-Biometrics is a science that applies statistical and mathematical methods to data
analysis problems in the biological sciences
-Biometrics is a technique that identifies people based on their unique physical
characteristics or behavioral traits
-Biometrics is a field of science that uses computer technology to identify people
based on physical or behavioral characteristics, such as fingerprints or voice
-Biometrics is a science, which deals with the automated recognition of
individuals (or plants/animals) based on biological and behavioral characteristics
Types of Biometrics:
-Physiological : are related to the shape of the body.
Examples: fingerprint, face recognition, DNA, hand and palm geometry, iris recognition, and retina.

-Behavioral : are related to the behavior of a person.

Examples: gait, voice, signature, and keystroke .

Differenece between Biometrics , Biometry , Biometric system :

Biometrics : science, which deals with the automated recognition of individuals (or
plants/animals) based on biological and behavioral characteristics .

Biometry : mathematical and statistical analysis of biological data.

Biometric system: a pattern recognition system that recognizes a person by determining the
authenticity of a specific biological and/or behavioral characteristic (biometric).

Lec2
factors determining trait used in biometric application:
-Universality: Every individual accessing the application should possess the trait.
-Uniqueness: The given trait should be sufficiently different across individuals comprising the
population.

-Permanence : The biometric trait of individual should be sufficiently invariant over a period of time.
Measurability: possible to acquire the biometric trait using suitable devices.

-Performance : accuracy, speed.

Acceptability: Individuals in the target population that will utilize the application should be willing to
present their biometric trait to the system.

-Circumvention: Ease with which the trait of an individual can be imitated using artifacts (e.g., fake
fingers).

Examples of physical biometrics:

-Bertillonage: measuring body lengths (no longer used).
-Fingerprint : analyzing fingertip patterns.

-Facial Recognition : measuring facial characteristics.

- Iris Scan : analyzing features of colored ring of the eye.

-Retinal Scan : analyzing blood vessels in the eye.

-Vascular Patterns : analyzing vein patterns.

-DNA : analyzing genetic makeup.

Examples of behavioral biometrics:

-Speaker Recognition : analyzing vocal behavior.

-Signature : analyzing signature dynamics.

-Keystroke : measuring the time spacing of typed words .

Biometrics Applications:
- Biometric Time Clocks or Biometric time and attendance systems: used in organizations to
control employee timekeeping.

-Biometric safes and biometric locks: security to the homeowners.

- Biometric access control systems: security at entrances.

-Biometric systems : are developed for securing access to pc's.

-Biometrics technology : in identifying DNA patterns for identifying criminals.
-Biometrics airport security devices to enhance the security standards .

Identification sciences(Forensics Science):

- Forensics Science : is the application of the natural and physical sciences to questions of
legal or public concern.

- The most common application is the analysis of evidence, such as blood, hairs, fibers, and
fingerprints, from criminal cases .

NOTES:
BIOMETRICS IS PRE-EVENT.

FORENSIC SCIENSE IS POST-EVENT.

Lec3
Identification:
-compares that sample against every template in the database .

-called a “one-to-many” search (1:N).

-low security.

-consumes a lot of time and high memory.

Two types of identification:

-positive identification

-Negative identification

positive identification:
user enrolled in data base.

-Negative identification:
User is not enrolled in database.

Matching process of two phases:

-Acquisition and Comparison of biometric sample.

-Decision as to whether to accept or reject the input as authentic.

Enrollment:
-Use a password with permission to associate a biometric with their log -in information.

-Store a template in database .

Mistake during Enrollment led to Failure to Enroll (FTE).

-TO ACCESS COMPUTER:

- Correct User is known as genuine match

- If similarity score is high, user is allowed to proceed.

-If similarity score is low, user is not allowed to proceed.

WHY CORRECT USER IS REJECTED?

- Fingerprint sample is poor quality led to false reject rate (FRR)
- fingerprint quality was bad led to failure to acquire (FTA).

- Different biometric characteristics lead to impostor match.

- Weakness of matching algorithm.

WHY INCORRECT USER IS ACCEPTED?

-impostor fingerprint look similar to enrolled one lead to False Accept Rate (FAR).

– Spoofing.
- Weakness of matching algorithm.

Score Histogram: it represents probability distribution of the scores.

-It shows frequency of scores for both genuine and impostor matches over full range of possible
scores.

Standard graphs: are used in biometric testing as interpreting biometric evaluation results.
EQUAL ERROR GRAPH:
-allows reading trade-off between making system easy to use (few false rejections) and
secure (few false acceptances).

Lec4
MULTIMODAL SYSTEMS:
-Use more than one source of information for biometric recognition .

- For example, a multimodal biometric system may use both iris recognition and fingerprint
recognition to confirm the identity of a user.
Unimodal is (monomodal) systems.
-Advantages of Multimodal :
- Increase accuracy

- few enrollment problems.

- enhanced security.

-DISADVANTAGES OF MULTIMODAL SYSTEMS:

-Expensive and complicated.

- Systems may be difficult to use for user.

-TYPES OF MULTIMODAL SYSTEMS :

-sources of biometric information

-Mode of operation

1-SOURCES OF INFORMATION:
-Multiple modalities: different biometric modalities are used.
-Multiple characteristics: use of different instances of same biometric.

-One characteristic, multiple sensors: uses multiple captures of same biometric from different
sensor types.

-One sample, multiple algorithms : combination of multiple algorithms used to match same
sample

- Multiple impressions : uses multiple impressions of same biometric characteristic.

- Soft biometrics Multimodal systems : use information from “soft” biometrics traits like
height, weight, and eye color.

2. MODES OF OPERATION:
-Multimodal system operate :

-in parallel: all process at the same time.

-in serial: many users will only need to use one sensor.

-BIOMETRIC FUSION TECHNIQUES:

1.feature level fusion.
2. score level fusion.

3. decision level fusion

-FEATURE LEVEL FUSION (low level fusion) :

-better performance than the two others.

-Reason: most information is available, which may be lost when fusion is conducted at higher
levels.

-Difficulties: feature sets for different information sources may be different.

-SCORE LEVEL FUSION (match score level fusion) (similarity score level fusion):
-Fusion at match score level.

-score level fusion achieves better results than decision level fusion.
-Reason: useful information is lost when match score is disregarded.

-APPROACHES OF COMBINATION AT SIMILARITY SCORE LEVEL:

1-classification : consider verification as classification problem with two classes: ‘Accept’ and ‘Reject’.

2. score combination: Apply formula to several score to combine them in single score.
-Example: adding scores together, take average, or select minimum or maximum score.

-decision level fusion:

-Decision level fusion is the highest level combination.

-Common approach: majority voting scheme

Lec5
-Evaluation of system performance:
-Accuracy and Security.
-Calculated results from trial provide accurate reflection of performance when system is in
operation.
-Evaluation Types:
1-Technology evaluation:

Recognition algorithm.

Face Recognition Vendor Tests (FRVT). Fingerprint Verification Tests.

2-Scenario evaluation: Undertake testing in simulated, controlled environment close to
operational , given cost and time considerations.

3-Operational evaluation: Ensure that system is operating as expected .

-Elements of Evaluation:
1-primary test elements:
-Type and quality of data .

-Demographics and user behavior.

-Data must be sampled from same sensor types.

2-secondary test elements: -Vulnerability -Fraud -Open-set -Blindness -Validation -Metadata

-Publicity -Repeatability -Scalability -Storage artifacts -Subject behavior -Reliability

-Test plan:
-Identifying test target.

-Determining data to be collected.

-Specifying how data is matched.

-Describing how results will be presented.

-TEST SET:
1-Enough biometric samples.

2-Similar to population and environment of target system.

-Errors in Test Set:

To minimize errors in test set: either before or after data collection.

-Before Data Collection:

Sources of error: 1. test subject forgets their unique ID, or enters wrong one during
enrollment or verification. Solution: use of tokens.

2. Unfamiliar people with using technology.

Solution: habituation test subjects receive training before test.

-After Data Collection:

Sources of errors:

1. genuine match labeled as impostor have unusual high score.

2. impostor match labeled as genuine have unusual low score.

-Data Collection:
-Test subjects interact correctly with the acquisition system.

-Recording quality issues.

-Measuring the relevant environmental parameters.

-Labeling the ground truth data correctly.

-Handling the data securely.

-Separating systems and data.

-Storing all the data acquired in a format easy to report.

-Matching:
-Storing the top results.

-Storing the impostors and genuine matches above a threshold.

-Storing only results in the target range.

Lec6
-Verification:
The key feature of verification systems is that they involve a claim of identity.

-Hypothesis Testing:
Verification systems make decision, accept or reject.

How likely is the verification system to make an incorrect decision?

-Problem Formulation:
Hypothesis testing begins by formulating a null hypothesis (H0 ), which is presumed true
unless there is evidence to the contrary.

If there is evidence to the contrary of the H0 , an alternative hypothesis H1 is accepted.

Case of biometric verification, hypotheses refer to user’s claim of identity.

-Decision Errors:
Type I error (false non-match) : occurs when null hypothesis is true, but is rejected.

Type II error (false match) : occurs when null hypothesis is false, but is accepted.

Ex- 1-FNMR 2- FMR

-Performance Rates:
Representing Match Score Distributions
-test or z-test.

-Performance Graphs:
Limitations of FMR and FNMR as performance measures:

1. performance is of interest over a range of thresholds.

2. Threshold values are relevant to specific algorithm, and cannot be compared between
systems due to different ranges.
Two standard plots: ROC curve and DET curve .

Lec7
-Identification vs Verification:
-Identification: One-to-many match.

-Verification: One-to-One match.

-Verification: can be viewed as special case of identification in which system only has single.
enrollment.
-Probe: is biometric sample from user who’s identity is unknown.

-Database, gallery or watchlist : are set of people enrolled in the system.

Operator : Identification systems involve operator who manually results .

open-set identification: there is no guarantee that the probe even belongs to gallery.
candidate list : list of potential matches returned
-Number of Enrollments:
- performance is dependence of identification systems on the number of people enrolled in
the system.

-Score Normalization :
- Another difference between verification and identification is that score normalization can
be used for identification.

-When members of gallery are known in advance, score normalization can be used to better
distinguish between subjects.

-one approach is transformation to feature space that maximizes inter-person distances and
minimizes intra-person distances.

-enrollments belonging to the same person are brought closer together, while those
belonging to different people are distanced from each other.

-the algorithm to better discriminate between two people who were originally very close in
the feature space.

-Identification Performance :
1-False-negative identification error : When user enrolled in system does not have true
template returned in the candidate list after an identification attempt.

2-False-positive identification error : When candidate list template is impostor. each

template returned by identification query that does not belong to the user being identified
3-False-negative identification error rate: proportion of identification attempts in which user
enrolled in system does not have at least one of templates returned among candidate list.

4-False-positive identification error rate: proportion of identification attempts in which non-

empty candidate list is returned, which does not include any genuine matches .
5-Correct Detect and Identify : requirement for this is that genuine match is both in the
candidate list, and ranked first.

There are two different types of identification systems: closedset and open-set:
-Closed-set identification : Every input is known to have enrollment in database .

-Closed-set identification : it is known that enrollment exists, and the question is “how
many of the top ranked matches need to be examined before finding the correct match?
-Open-set identification : sample submitted to system may or may not have an enrollment
in database. Open-set identification is also known as watchlist problem.

-Open-set identification : it is not known if the enrollment exists in the database or not.
returning more matches does not necessarily increase chance of correct identification.

Lec8
-Curves to evaluate identification system :
1. CMC Curve use to Closed-set system ( used to top x match).

2. Alarm Graph use to open set system ( used to threshold).

-Alarm Graph(watchlist ROC curve plots):

detection rate against the false alarm rate over a range of alarm thresholds.
Alarm curve is a function of watchlist size

-There are two performance rates :

1. detection rate : probability that correct match, when it exists, is included as part of the
candidate list.

-This is known as detection rate inclusion in candidate list is based only on the alarm
threshold, this rate does not depend on database size.
-If a genuine match scores above the threshold it is included in the candidate list,
regardless of ranking and how many impostor matches are above threshold

2. false alarm rate : which is the proportion of identification queries that raise false

alarms false
- alarm rate depends on database size.

-more people are enrolled, more impostor matches that are conducted, and greater
probability that there will be at least one above threshold and included in candidate list.

-Dealing with Uncertainty :

Uncertainty is important topic for biometrics, for designing test plans and reporting
results.

- Confidence intervals : are used to present the range of uncertainty

-There are two primary sources of errors that prevent one from knowing
“true” performance statistics:
1. systematic errors

2. sampling errors

1. Systematic Errors :
-result from environmental factors or test procedures that bias the result in any direction .
example of systematic error for biometric system is when the demographics of the test
subject population does not match that of the target population.

- Behavioral aspects can also introduce systematic errors.

- The best approach for dealing with systematic errors is through careful experiment
design

2. Sampling Errors :
-errors are consequence of using a finite set of subjects to represent larger target
population

- Sampling errors are always present, even when the subjects are selected randomly.
-unique individuals tend to perform differently within a biometric system.

-For example, some people may have trouble authenticating, while others may be prone
to being falsely accepted. exact set of people selected in trial will have some influence on
collective error rates.

-Confidence interval:
-is an interval estimate [L,U] of an observed error rate, along with an associated
probability p that the true error rate is between L and U.

-Confidence intervals are especially useful when comparing results.

- if confidence intervals of the two measurements do not overlap, there is a statistically

significant difference between the values.

-Boxplots :
-Another method for illustrating statistical uncertainty is by using boxplots.

- Boxplots are similar to confidence intervals, except they contain more information , as
opposed to simply upper and lower bound.

- boxplots are generated using a sampling technique. test set is partitioned into smaller
sets, and the performance rates are computed for each subset.

- Boxplots are non-parametric way.

 Lec9
-ENROLLMENT AND ACQUISITION ERRORS:
-Failure to enroll(FTE)

- Failure to acquire (FTA)

-SYSTEM DECISION POLICY : The protocol defined for enrollment and verification or
identification.

-FALSE ACCEPT : A false accept occurs when a verification transaction confirms the identity
of an impostor.

-FALSE REJECT : A false reject is a genuine verification that is incorrectly denied.

Lec10
-DESIRABLE BIOMETRIC ATTRIBUTES :
-Distinctiveness

-Stability

-Scalability

-Usability

-Insensitivity
-Privacy

-Maintenance

-Health

-Quality
- Integration

- Cost

-BIOMETRIC DATA:
1. RAW DATA:
- The raw biometric information (known as the biometric sample) is data gathered

- directly from the sensor before any processing has been carried out.
2. TOKEN DATA :
-A token is representation of the raw data that has had some minimal amount of processing
applied.

3. TEMPLATE DATA :
-A template is the refined, processed and stored representation of the distinguishing
characteristics of a particular individual.