Face Biometric Systems
For general background information see Overview of Biometric Projects.
Many face biometric techniques have been explored for both authentication and identification applications.
Although the current trend focuses on three-dimensional techniques, we will implement simpler two-dimensional techniques.
Many algorithms exist to perform facial recognition.
These fall into the broad categories of feature point,
which use facial landmarks such as the eye, mouth or nose and appearance based,
which uses color and other recognition techniques to identify a subject.
Some use still images and others video clips.
Challenges exist to recognize a subject who may be in a variety of poses or facial expressions.
Some of the experiments will be designed to test the accuracy of the application in dealing with these challenges.
Purchase or Find Software
We can spend $300-400 on face recognition software,
but for reimbursement purposes the actual order and payment must be made by your instructor.
Possible purchases include:
It may be possible to find implementations of face biometric techniques on the internet,
and a thorough search should be made.
It may also be possible to request copies of code from authors of papers.
Implement or Modify Existing Software
Eigenface TechniqueThe Eigenface algorithm, although rather mathematical, is popular and reasonably
simple to implement, see for example the following references:
The Eigenface technique was implemented in an earlier project in 2003.
At that time the algorithm was coded using Matlab software, and we should be able to find that code.
Because that code may be difficult to work with, however, it
might be easier to find code available on the internet.
The objective is to get some version of the Eigenface algorithm up and running.
Neural Network Technique
An algorithm that offers the potential to work with the limited processing capabilities
(such as the PDA application of the companion project)
is a neural network back propagation algorithm.
This algorithm is of particular interest as it can compare a variable posed image against a stored image
by estimation techniques, and deals with multiple facial expressions.
This is beneficial as the pose or expression of the user is variable and the time of image capture.
Security Classroom Face Recognition Demonstration - Dr. Tappert
Here we create an interesting demonstration that can be performed using any available face recognition software.
This is also a simple test of any face recognition software.
The interesting aspect of the demonstration is to determine which students look similar to the system,
and depending on the system the results can sometimes be surprising.
Possible demonstration scenarios are as follows:
We will focus on the first scenario, and your primary task is to make the demonstration easy to conduct.
Therefore, you should choose software that is easy to use,
and provide easy-to-follow instructions for conducting the demonstration.
For appropriate results it is advisable to take all student photos with one digital camera,
with the same lighting and background, and with each face centered and sized identically.
- One photo of each student in the classroom is taken and each photo is compared to all the other photos
to obtain distance (or similarity) scores, and a confusion matrix is obtained that shows which students look most alike.
- Photos of several (ideally all) students in a classroom are taken put into
the system's database during an enrollment (training) phase,
and each training photo is labeled with the name of the student.
Then, during the recognition (testing) phase, we take and enter a new photo
(unknown to the system, not one used for training)
of one of the students and the system finds the best five matches among the training photos
and returns these photos and the names of the students.
The system should correctly recognize the student within the top five choices, and ideally on its first choice.
It is best to show the resulting matching matrix by using the photos as follows.
Each row contains a student's photo (alphabetical by student's last name),
followed the best matching photos (best, second best, etc.) of other students (with matching score under the photo).
If you obtain (or develop) several face recognition systems,
it would be interesting to compare the outputs of the systems using this demonstration.
First Milestone: run a prototype demonstration at our second classroom meeting (middle of the semester).