Assignment 2
Topic: EigenFaces and FisherFaces
Due on or before:
25 March (Monday) 2024, 12:00 noon
Maximum Marks: 8
East is East, and West is West...The Twain Shall Never Meet
One normally does not mix, or compare unsupervised learning
methods, and supervised ones. This is an exception.
Please also submit an ASCII .txt file, which contains a
description of the specific parameters used.
The recommended programming language is C/C++/Java/Python, and the
recommended software environment is
OpenCV. No MATLAB for this
assignment, please.
This assignment aims at getting experience with Machine
Learning algorithms which can easily be coded up. (There are many
which cannot, and may be used as black boxes. Hence, please feel
free to use routines such as the SVD, and methods to compute
eigenvalues and eigenvectors. Please do not try to code these
for this course!).
This assignment has been kept open-ended for a reason:
to play around with parameters, and try to find the
physical significance of the operations involved. Thsi will
encompass most of what you have learnt in this course with regard
to eigenspaces in unsupervised learning, and Fisher's linear
discriminant, in supervised learning. Start from a 2-class
problem (different views of faces of two separate individuals)
For the supervised case, try to find the best separating
hyper-plane. For the unsupervised case, try to find the
corresponding decision boundary. Look at cluster centres, and
distances. This assignment will also require you to divide the
dataset into training and testing sets, for instance. Please feel
free to use any face recognition database. You can find an
extremely comprehensive list at the
Face Recognition Homepage's databases section
Try extending the 2-class cases to multiple classes.
Please do not copy/cut-and-paste
code from the Internet: this will be penalised, as mentioned on
the course webpage (the front page). You may read up different
implementations, but please code from scratch.
Demo Schedule:
(To be announced)
Venue:
Demos:
The schedule has been put up on the Google Drive page
associated with the contact information
Sumantra Dutta Roy,
Department of Electrical Engineering, IIT Delhi, Hauz Khas,
New Delhi - 110 016, INDIA.
sumantra@ee.iitd.ac.in