lecture 12: introduction to biometricsmp/lectures/lect12.pdf · • adaptive fuzzy segmentation...
TRANSCRIPT
CSE489-02 & CSE 589-02 Multimedia Processing
Lecture 12: Introduction to Biometrics
Spring 2009
New Mexico Tech
* Most slides are extracted from Prof. IoannisPavlidis’s lectures5/3/2009
Personal Identification Associating an identity with an individual
is called personal identification. Verification refers to the problem of confirming
or denying a person’s claimed identity. Recognition refers to the problem of
establishing a subject’s identity.
#25/3/2009
Applications Accurate identification of a person could
deter crime and fraud, streamline business processes, and save critical resources. MasterCard estimates the credit card fraud at
$450 million per annum, which includes charges made on lost and stolen credit cards.
ATM related fraud is worth approximately $3 billion annually.
#35/3/2009
Identification Methods Traditional
A person’s possession (e.g., key or card) A person’s knowledge of a piece of information
(e.g., user id and password)
Biometrics A person’s physical characteristics (e.g.,
fingerprints, face, and iris)
#45/3/2009
Biometrics Any human physiological or behavioral
characteristic could be a biometric provided it has the following desirable characteristics: Universality: every person should have the
characteristic. Uniqueness: no two persons should be the same in
terms of the characteristic. Permanence: the characteristic should be invariant with
time. Collectibility: the characteristic can be measured
quantitatively. Performance: the resource requirements to achieve an
acceptable identification accuracy. Acceptability: to what extent people are willing to accept
the biometric system. Circumvention: how easy is to fool the system by
fraudulent techniques.#55/3/2009
Biometrics Technologies Voice Infrared Facial and Hand Vein Thermograms Fingerprints Face Iris Ear Gait Keystroke Dynamics DNA Signature and Acoustic Emissions Odor Retinal Scan Hand and Finger Geometry
#65/3/2009
Voice – Pros and Cons Pros
Voice capture is unobtrusive and voice print is an acceptable biometric in almost all societies.
Some applications entail authentication of identity over telephone. In such situations, voice may be the only feasible biometric.
Cons Voice is not expected to be sufficiently unique to permit
recognition. A voice signal available for authentication is typically
degraded in quality by the microphone, communication channel, and digitizer characteristics.
Voice is a behavioral characteristic and is affected by a person’s health (e.g., cold), stress, and emotions.
Some people are very good at mimicking the voice of others.
#75/3/2009
Voice - Categories Text-dependent speaker verification
authenticates the identity of a subject based on a fixed predetermined phrase.
Text-independent speaker verification is more difficult and verifies a speaker identity independent of the phrase.
Language-independent speaker verification verifies a speaker identity irrespective of the language of the uttered phrase and is even more challenging.
#85/3/2009
Voice - Methodology The amplitude of the input signal
may be normalized and decomposed into several band-pass frequency channels.
The features extracted from each band may be either time-domain or frequency domain features. Log of the Fourier Transform
The matching strategy may typically employ approaches based on hidden Markov model, vector quantization, or dynamic time warping.
#95/3/2009
Infrared Facial and Hand Vein Thermograms –Pros and Cons
Pros Human body radiates heat and
the pattern of heat radiation is a characteristic of each individual body.
Infrared Imaging is unobtrusive.
Cons The absolute values of the heat
radiation are dependent upon many extraneous factors.
The technology is expensive.
#105/3/2009
Fingerprints – Pros and Cons Pros
It is one of the most mature biometric technologies.
Fingerprints are believed to be unique to each person.
Cons It has a stigma of criminality.
#115/3/2009
Fingerprints - Methodology Four basic approaches to identification
based on fingerprint are prevalent: Invariant properties of the gray scale profiles
of the fingerprint image or part thereof. Global ridge patterns, also known as
fingerprint classes. Ridge patterns of fingerprints. Fingerprint minutiae – the features resulting
mainly from ridge endings and bifurcations.
#125/3/2009
Face – Pros and Cons Pros
Face is one of the most acceptable biometrics because it is one of the most common method of identification that humans use.
It is non-intrusive. Cons
Difficult to develop techniques to tolerate the effects of aging, facial expressions, slight variations in the imaging environment, and variations in the pose of face with respect to camera (2D and 3D rotations).
Facial disguise is of concern in unattended applications.
#135/3/2009
Face - Methodology Two primary approaches for face
recognition are: Transform Approach: the universe of face
image domain is represented using a set of orthonormal basis vectors. Currently, the most popular basis vectors are eigenfaces.
Attribute-Based Approach: facial attributes like nose and eyes are extracted from the face image and the invariance of geometric properties among the face landmark features is used for recognition.
#145/3/2009
Iris – Pros and Cons Pros
It is unique for each person and each eye.
The identification error rate using iris technology is believed to be extremely small and the method is very fast.
Cons It requires cooperation from
the user.
#155/3/2009
Ear – Pros and Cons Pros
It is known that the shape of the ear and the structure of the cartilegenous tissue of the pinna are distinctive.
Cons The features of an ear are
not expected to be unique to each individual (good for authentication only).
No commercial systems are available yet.
#165/3/2009
Gait – Pros and Cons Pros
Non-obtrusive.
Cons Gait is not supposed to be unique to each
individual, but is sufficiently characteristic to allow identity authentication.
Gait is a behavioral characteristic and may not stay invariant over a large period of time (e.g., fluctuations of body weight or injuries involving joints or brain).
#175/3/2009
Keystroke Dynamics – Pros and Cons Pros
It is unobtrusive Cons
This behavioral biometric is not expected to be unique to each individual but it offers sufficient discriminatory information to permit identity authentication.
One may expect to see large intra-individual variability.
#185/3/2009
Keystroke Dynamics - Methodology Keystroke dynamic features are based on:
Time durations between the keystrokes Dwell times – how long a person holds the
key.
Typical matching approaches use a neural network architecture.
#195/3/2009
#20
DNA DNA (DeoxyriboNucleic Acid) is the 1D
ultimate unique code for one’s individuality.
Identification for forensic applications only.
Three factors limit the utility of this biometric for other applications Contamination and sensitivity Automatic real-time identification issues Privacy issues
#21
Signature and Acoustic Emissions The way a person signs her name is known to
be a characteristic of that individual.
A related technology is authentication of an identity based on the characteristics of the acoustic emissions emitted during a signature scribble.
Pros Acceptable practice in many transactions
Cons They require contact and effort They evolve over time and are influenced
by physical and emotional conditions of the signatories.
Professional forgers can reproduce signatures to fool the unskilled eye.
#22
Odor It is known that each object exudes an odor that
is characteristic of its chemical composition and could be used for distinguishing various objects.
Methodology A whiff of air surrounding an object is blown
over an array of chemical sensors, each sensitive to a certain group of aromatic compounds.
The feature vector consists of the signature comprising of the normalized measurements from each sensor.
After each act of sensing, the sensors need to be initialized by a flux of clean air.
#23
Retinal Scan The retinal vasculature is rich in
structure and is supposed to be characteristic of each individual and each eye.
Pros It is supposed to be the most
secure biometric since it is not easy to change or replicate the retinal vasculature.
Cons The image acquisition involves
cooperation of the subject, entails contact with the eyepiece, and requires a conscious effort on the part of the user.
#24
Hand and Finger Geometry Some features related to a human hand,
e.g., length of fingers, are relatively invariant and peculiar (although, not unique) to each individual. Finger geometry is a variant of hand geometry and is a relatively new technology, which relies only on geometrical invariants of fingers (index and middle).
Pros The representational requirements of
the hand are very small (9 bytes). Cons
Suitable for verification only. It requires cooperation from the
subject. The registration of the palm is accomplished by requiring the subject’s fingers to be aligned with a system of pegs on the panel, which is not convenient for subjects suffering from arthritis.
#25
Biometrics Technologies: A Comparison In the context of biometrics-based identification
(authentication) systems, an application is characterized by the following properties: Does the application need identification or
authentication? Is it attended (semi-automatic) or unattended
(completely automatic)? Are the users habituated (or willing to be habituated) in
the given biometric? Is the application overt or covert? Are the subjects cooperative or non-cooperative? What are the storage requirement constraints? How stringent are the performance requirement
constraints? What types of biometrics are acceptable to the users?
#26
Why Infrared?
• Visible light has no effect on images taken in the thermal infrared spectrum.
• Even images taken in total darkness are clear in the thermal infrared.
#27
Why Infrared? (Contd..)
Illumination Invariance Major problem in visible domain.
Uniqueness and Repeatability Sense thermal patterns of blood vessels
under the skin, which transport warm blood throughout the body.
Remain relatively unaffected by aging. Even identical twins have different
thermograms.
Immune from Forgery Disguises can be easily detected.
#28
Eigenfaces Approach - Training Training set of images represented by
1,2,3,…,M
The average training set is defined by Ψ = (1/M) ∑M
i=1 i
Each face differs from the average by vector Φi = Γi – Ψ
A covariance matrix is constructed as: C = AAT, where A=[Φ1,…,ΦM]
Finding eigenvectors of N2 x N2
matrix is intractable. Hence, find only M meaningful eigenvectors. M is typically the size of the database.
#29
Eigenfaces Approach - Training Consider eigenvectors vi of ATA such that
ATAvi = μivi
Pre-multiplying by A, AAT(Avi) = μi(Avi)
The eigenfaces are ui = Avi
A face image can be projected into this face space by
Ωk = UT(Γk – Ψ); k=1,…,M
#30
Eigenfaces Approach - Testing The test image, Γ, is projected into the face space to
obtain a vector, Ω:Ω = UT(Γ – Ψ)
The distance of Ω to each face class is defined by Єk
2 = ||Ω-Ωk||2; k = 1,…,M
A distance threshold,Өc, is half the largest distance between any two face classes:
Өc = ½ maxj,k ||Ωj-Ωk||; j,k = 1,…,M
#31
Eigenfaces Approach - Testing Find the distance, Є , between the original image, Γ,
and its reconstructed image from the eigenface space, Γf,
Є2 = || Γ – Γf ||2 , where Γf = U * Ω + Ψ
Recognition process: IF Є≥Өc
then input image is not a face image; IF Є<Өc AND Єk≥Өc for all k
then input image contains an unknown face; IF Є<Өc AND Єk*=mink Єk < Өc
then input image contains the face of individual k*
#32
Limitations of Eigenfaces Approach
Variations in lighting conditions Different lighting conditions for enrolment and query. Bright light causing image saturation.
Differences in pose – Head orientation 2D feature distances appear to distort.
Expression Change in feature location and shape.
#33
Segmentation Noise in the background may
effect the performance of a face recognition system.
Remove the background.
Use thermal information on face to compute the features.
• Adaptive Fuzzy Segmentation (kakadiaris02)– Fuzzy affinity is assigned to spels w.r.t. target object spel.– Affinity is computed as weighted sum of the temperature and the
temperature gradient in the neighborhood of the target spel.– Minimal user interaction because of dynamically assigned weights.
#34
Problem with Single Seed Temperatures on face are
different at different regions.
• If a single seed is chosen in a particular region, then the connectivity stretches only along this region and the segmentation goes wrong.
#35
Multiple Seeds Solution to this problem is to
choose multiple seeds in different regions on face and merge the resulting segmented parts .
• Choose a seed pixel on face wherever there is sharp change in gradient.
• Works well even when the subject is wearing glasses.
• Robust to variation of poses.
#37
Assumptions Merge all resultant segmented
regions to form final image.
ASSUMPTIONS
• The center of the image contains the pixel from facial region.• The temperatures at all pixels are mapped between 0 and 255.
– If this mapped temperature at a pixel is between 175 - 200, it is classified to be in blue region.
– If this mapped temperature at a pixel is between 200 - 225, it is classified to be in pink region.
– If this mapped temperature at a pixel is between 225 - 255, it is classified to be in cyan region.
#38
Feature Extraction
The Gabor filter bank is given by:
• The segmented facial image is divided into its spectral components using Gabor filters.
• The resultant Gabor filtered images are modeled using Bessel models.
#41
Bessel Parameters
The filtered images are modeled using Bessel parameters:
SK – Sample Kurtosis
SV – Sample Variance
• Each segmented image in training set is convolved with the filters in Gabor filter bank to obtain Gabor filtered images.
#42
Sample Variance and Kurtosis
Sample Variance is the measure of the “spread” of the distribution.
• Sample Kurtosis is the measure of the “peakedness” or “flatness”.
Sample Kurtosis,
#43
Bessel Model Using the bessel parameters p and c, the filtered
image I(j)(x,y) is modeled as:
(p) is gamma function Iv(z) is modified bessel function of first kind given by:
#45
Performance of Bessel K Forms
Kullback-Leiber divergence:
KL div=0.0013 KL div=0.0027 KL div=0.0055 KL div=0.0058
– observed marginal density– Estimated Bessel Form
#46
Comparing IR Images Images modeled into Bessel parameters can be
compared by:
• L2-metric between two Bessel forms f(x;p1,c1) and f(x;p2,c2) in D:
#47
Hypothesis Pruning Applying a high-level classifier on entire
database is computationally very expensive.
Pruning of hypotheses can be achieved by using Bessel parameters (anuj01).
Helps in short listing best matches.
Bessel parameters for images in database can be computed offline which helps in saving a lot of computation time.
#48
Hypothesis Pruning (Contd..) Define a probability mass function on the database
A:
(p(j)obs,c(j)
obs) – observed Bessel parameters for test image I(j)
(p(j),s,c(j)
,s) – estimated Bessel parameters which can be computed offline
• Images in database A with P1(|I) greater than a specific threshold value are short listed as best matches.
(D=0.3 for Equinox dataset)
#49
Hypothesis Pruning (Contd..) Shortlist the subjects of A with P1(/I) greater than a
specific threshold:
#52
Results – ROC Curves
Correct Positive : Test image is in the database and is correctly recognized.
False Positive : Test image is not in the database, but is recognized to be an image of the database
Negatives : Test images that are not in the database.