Project EESM5547: Face recognition and average face!!
• Prerequisites: Gather a face database of the class (Let’s
do it within the next week)!
• At least, each student takes a picture with front and
centred view of your face!
• Each student provides his face image from different
orientations and variation of expression and wearings
etc.!
• The pictures can be taken with slightly variant
illuminance conditions or blurring as well!
• Each student provides 15 portraits!
• Send it to the TA by next weekend!
• You can do the project in group of two or per-student. In
your final report and presentation, the contributions need
to be specified to individual. The two-student projects
need to show significantly larger work package (e.g.
include other methods) or much deeper analysis to
the result. Grades will be provided per individual. If you
are a UG, you can choose to do it with others in a group.!
!
Project definition
Face detection and recognition with comparative
studies!
!
GENERAL REQUIREMENT:!
From this project, the following key techniques should be
included in the procedure:!
* Image enhancement in either spatial or frequent domain!
* Restore distortion when necessary!
* Face detection / recognition!
* Feature Recognition!
General Guidelines and minimum work packages
!
MODELLING!
Use the Course-dataset (preferred, since you worked on it
together) or the one in http://www.cl.cam.ac.uk/research/dtg/
attarchive/facedatabase.html to train your model.
Remember to divide them into training and testing dataset.
Consider the following questions to help:!
(a) How should you represent you image as inputs for PCA?
How about using other methods, such as keypoint-based
methods?!
(b) How do the (leading) eigenfaces look like as an image
(show some examples in your report)? What do they mean?!
(c) How does the importance of the eigenfaces decrease?!
(d) When keypoint-based methods are used, what are the
limitations (e.g. repetitive features)? How to reduce the side-
effect caused by them? !
!
PREPROCESSING!
Before doing anything else, try to enhance the visibility of
the datasets using learnt techniques. Use examples to
illustrate the improvement.!
!
RECONSTRUCTION!
Calculate the most average look of the class. (In order to get
a perfect results, probably you want to rotate and shift the
image manually a bit to align the eye positions. These
operations are not required in your results, but it would be
nice to have.) Consider the following questions to help your
project:!
(a) Observe the difference between reconstructed and
original images, as the number of eigenfaces used in
reconstruction increases.!
(b) How many eigenfaces are required to recover an original
face with reasonable errors?!
(c) Does the number of needed eigenfaces change from
person to person, or not?!
(d) How do you select the co-efficiencies to get an
“average”-look?!
!
RECOGNITION!
Use the images from the testing dataset to demonstrate
your face recognition statistically. As comparison, please
take 20 additional arbitrary images (not face, or faces of
animals or cartoon), show the recognition results
comparatively. Also, please discussion over the results and
give your justification.!
!
IDENTIFICATION!
Using another subset of your testing data, or take several
other faces from you and friends to identify who is the guy in
the picture. Show your results statistically. (remember to use
false samples as well.) Consider the following hints:!
(a) Use the same training and testing sets as above!
(b) Develop your method base on what you learn from
image features. If you method is able to tell that a new face
is known, how does it continue to tell which face in the
training set it corresponds to?!
(c) The following plot shows a general flow. You can use
your own as well.!
!
Grading
– Content: 50%!
– Report: 30%!
– Presentation: 20%!
Extension of work packages will be rewarded and strongly
encouraged!
Coding language other than Matlab is encouraged!
!