Scribd
Upload
13 views

Biometrics-2022

Uploaded by

shawaqi96
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
13 views

Biometrics-2022

Uploaded by

shawaqi96
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 32

Biometrics

ISEC 413 Security Architecture and Mechanism


Dr. Ahmed Al Faresi
Issues with Biometric
Authentication
➢ Matches and probabilities

➢ Higher storage requirements

➢ Devices usually required

➢ People preferences
Matches and probabilities
➢ Identifying information is not typed in, but obtained
by a device

➢ Characteristics mapped from analog to digital and


Loading…
not all of the original information is retained

➢ Devices for most common biometrics may not


produce identical results or even identically
repeatable results
● Ex: fingerprint readers depend on environmental
factors such as the positioning of the finger, the
“moisture” of the hand, oils, and occupational issues
which may cause a print to be roughened over time
Higher
➢ “Size” ofStorage
the templateRequirements
as stored can be quite large in
comparison with a password and is not necessarily
directly tied to the accuracy

➢ Some typical template sizes:


● Finger-scan: 250 - 1000 bytes
● Hand geometry: 9-20 bytes
● Iris: 512 bytes
● Retina: 96 bytes
Devices Usually
The device collectingRequired
the data is often proprietary
and/or uses proprietary algorithms

Patents protect much of the technology


Loading…
There may be considerable computation involved
in computing a “validator” or template for storage
(far beyond the Unix validator)

Sometimes the biometric requires local installation


of a specialized reader device (such as for
fingerprints, but not for voice)
5
People preferences
According to the International Biometric Group's
Consumer Response to Biometrics, people did not
like facial scans as much as fingerprints as a
substitute for a PIN in ATM, but both technologies
rated between “somewhat comfortable” and “neutral”
Reasons seemed to be:
● People do not like to look at their own images in low
resolution
● People do not like their picture taken
● People do not recognize “facial id” as an authenticator
in the same way they recognize fingerprints
● Facial scans do not require consent (with hidden
cameras)
6
Finger scanning most commercial
Fingerprints
UK police uses for forensics, US for identifying
arrested persons
Automatic recognition has equal error rate of 1-2%
Widely used in 1990s in welfare / pensions
Banking: India, other LDCs
Since 9/11: US-VISIT
Forensic use: 16-point match taken as gospel until
the McKie case

:
The McKie Case
➢ Identifying people from ‘16
points’ thought infallible
➢ Error rate “2.5x10-10”
➢ Shirley McKie was
prosecuted; acquitted
➢ Police panic: they then
prosecuted the experts!
Actual McKie Case Photos

Even harder, isn’t it?


And what about the odds now we have
computers?
Iris

➢ John Daugman’s idea:


irises
Loading…
➢ Work fine in some apps
– equal error rate very
low
➢ Unattended operation?
Iris Scan
➢ See http://www.cl.cam.ac.uk/users/jgd1000 for
details
➢ Gather data by a camera within 3 feet of eye
➢ 512 byte IrisCode represents the visible
characteristics of the eye
● It describes the phase characteristics of the iris in a
polar coordinate system.
➢ It is claimed that the odds of the same IrisCode
being returned by two different people is less than
1 in 1052

12
Iris Scan

➢ IrisCode includes “266 spots” to distinguish


between irises (claim is most other biometrics
have between 10-60 distinguishing spots)

➢ IrisCode may vary by as much as 25% for a given


eye … but the odds of two different eyes being
75% similar is said to be 1 in 1016, so this seems
acceptable

13
Face Recognition
➢ Automatically locate the face

➢ Recognize the face under different illuminations

➢ Facial expression

➢ Aging effects
Face recognition
Step 1: Face detection
● Faces are taken under
varying lighting
conditions

● Face variations in
color, scale, position,
orientation, 3D pose,
and facial expression
Face recognition
➢ Step 2: Face detections procedure

Light compensation

Skin detection

Variance segmentation

Eyes detection

Mouth detection

Face boundary detection


18
Authentication vs. Identification
➢ Face authentication and verification (1:1 matching)

➢ Face identification/recognition (1:N


Matching)
Challenges in face recognition

➢ Identify similar face (inter-class similarity)

Twins Father and son


Challenges in face recognition
➢ Accommodate (intra-class variability)
● Head pose

➢ Illumination condition on facial expression


Challenges in face recognition
➢ Accommodate (intra-class variability)
● Expression

● Facial accessories

22
Challenges in face recognition
Accommodate (intra-class variability)
Aging effects

Cartoon faces
Challenges in face recognition
Accurate automated facial recognition requires
taking 3D images of the subject in varying lighting
conditions. This is not ideal as in identification we
may have low quality images or low quality
recording in security video of a moving subject.

Experiments done by the UK national physical


laboratory found face recognition to be the worst
Its single attempt equal error rate was 10%
Challenges in face recognition
Tests done in the UK office trial in 2005 found that
FR recognized 69% of users. It was still adopted in
Airport control.

Since 2012 with the neural network revolution the


FR performance has improved drastically. The use of
convolutional neural networks (cnn) the miss rate is
0.1%

FR algorithms using cnn where close to experts


results
Challenges in face recognition
In 2018 some algorithms have doubled in accuracy !
Thanks to deep learning

CCTV now can change from a tool for crime scene


forensics into tools for real time person recognition
and tracking. China and Russia pre and post the
pandemic.
Voice Biometrics
Voice recognition is the problem of identifying a
speaker from a short utterance
Issues: noise, text-dependency and real time
recognition
Like fingerprints it is used for identification and
forensics. Mainly used for telephone authentication
such as phone banking and military.
Voice and Video Morphing
AI has helped create algorithms that morphs voice
and video data into unreal illusions.

DEEP FAKE

https://www.youtube.com/watch?v=Wrald_EZgDQ
DNA
DNA has become a valuable tool for crime scene
forensics and determining parenthood in child
support cases.
Loading…
Genotype/Phenotype (twins)
Sloppy data and sequencing expenses
Privacy issues
Comparison of Biometric Solutions
Characteristics Weighting

Effortless Iris > Face > Finger > voice

Non-intrusive Iris > Voice > Face > Finger

Accurate Iris > Finger > Face ~ Voice

Inexpensive Voice > Finger > Face > Iris


Attacks in Biometrics
● Compulsion like in china
● Assumption that combining biometric solutions
will give a lower equal error rate. It will improve
either the flase accept rate or the false reject rate
and make the other worse.
● Hill climbng attack. Templates were considered
not private information.

You might also like