1. Facial Recognition in Biometrics
Susan Simmons
University of North Carolina Wilmington

2. Biometrics
Biometrics (wikipedia) -- Biometrics are used to identify the identity of an input sample when compared to a template, used in cases to identify specific people by certain characteristics.
possession-based: using one specific "token" such as a security tag or a card
knowledge-based: the use of a code or password.
Biometric – physical or behavioral characteristic
Biometrics (questbiometric.com) -- The word "biometrics" is derived from the Greek words 'bios' and 'metric' ; which means life and measurement respectively. This directly translates into "life measurement”. General science has included biometrics as a field of statistical development since the early twentieth century.
Biometrics technologies measure a particular set of a person's vital statistics in order to determine identity. Biometrics in the high technology sector refers to a particular class of identification technologies. These technologies use an individual's unique biological traits to determine one's identity or to verify one’s identity.The traits that are considered include fingerprints, retina and iris patterns, facial characteristics and many more.

3. A little history

4. Examples

5. We see biometrics in many different places today.
Voter Registration
Driver Licensing
Border Control
Passport / VISA
Criminal ID / Wanted Persons Lookup
Airports / Frequent Traveler / Passenger Tracking

6. Facial recognition Some examples Can be used for surveillance
Find criminals, terrorists, missing children
Involves non-invasive, contact-free process
Can be integrated with existing surveillance systems
Marketing
Identify demographics interested in products
Some examples

7. Problems in Facial recognition
Privacy issues facial recognition
Violation of people’s privacy?
Right to search database for match of images captured in public surveillance cameras?
Uncontrolled background (including lighting, shadows, glares)
Camera angle
Image resolution
Part of face hidden (sunglasses, hat, profile, etc)

8. Identify facial images
Need to identify facial images from a video or picture

9. Eigenfaces Take an N x N image and convert it to an N2 x 1 vector
Use a subset of the face images as a training set (each face must be centered and of the same size)
Calculate the eigenvectors of the covariance matrix of the images, keeping on K eigenvectors (corresponding to the larges K eigenvalues)
Uses of eigenfaces
After centering new images, calculate the distance between the new image and all images in database. If distance is less than a set cutoff, then the picture is recognized as that face.
Can also do this same exercise to determine if an image is a face.

10. Active Appearance Models
The following approach works well with facial images that are forwarding facing and not much facial expression.
The shape of a face is found using Active Shape Models (ASM). Identifies the outline of the face as well as important landmarks on the face.

11. Sean Connery (1959)

12. Modeling shape in AAMs
The algorithm uses a subset of images to train the model.
Points are aligned into a common co-ordinate frame and represented by a vector x (x = (x1, x2,x3…, y1, y2, y3,…)T).
Principal Component Analysis (PCA) is then applied to the data

13. Modeling grey-level appearance
Each image is warped so that its control points match the mean shape (using a triangulation algorithm)
To minimize the effect of global lighting variation, we normalize the example samples by applying scaling and offset
PCA is applied to the normalized data

14. Using the models to vary parameters

15. Movie
Varying the first parameter
Varying the seventh parameter

16. Age estimation
Random Forest (by Leo Brieman) uses decision trees to estimate age.
Support Vector Regression – (input info about support vector regression)
MAE % error w/+5
Random Forest
Support Vector Machine

17. Aging faces

18. Current methodology
Use Monte Carlo simulation to simulate potential b vectors
Use a classification method from the model to estimate age (for example, support vector regression or random forest, etc)
Create a look-up table for the average b-vectors
Use the look-up values to age individuals (use their b-vectors)

19. An example
A hypothetical “b-vector look-up table with the dimension of b = 4
AGE
A new individual’s “b-vector” at age 20
To age this individual to 30, we need to shift their b-values
( ) =
( ) =
( – ( )) =
( ) =

20. Some examples

21. Conclusions Biometrics continuously receives more attention
Need to create database
Much more work to be done by MANY different fields

22. Special Thanks to Ms. Amrutha Sethuram
Drs. Karl Ricanek (Computer Science), Yishi Wang (Statistics)
Mr. Fernando Schiefelbein (graduate student), Mr. Philip Whisenhurst (undergraduate student)