Homework #2
EECS-841
Due: Monday, Oct 6 (midnight)
Total Points: 8
Celebrity Look-Alikes
Although the technology for facial recognition is still a quickly evolving work in progress, even simple
facial recognition approaches are sufficiently powerful for another application: identifying which celebrity
looks most like you. In this assignment, we will use the eigenfaces technique discussed in class to write a
matlab program that identifies which celebrity you most resemble.
1. Offline Processing: Computing the Eigenfaces (3 pts)
Write a matlab routine
function [mean_face, eigenfaces, singularvalues, celebrity_PCA_coefs] =
ComputeEigenfaces(celebrity_directory, binary_mask)
where celebrity_directory is a directory containing the celebrity photo database available from the
class website. Each of these images is a 330x280x3 color image, and the scale, rotation, and offset of
each image has been normalized so that the eyes of each celebrity are in the same location. Use
dir([celebrity_directory, '*.jpg'])
to get a list of files in the celebrity directory.
binary_mask is the boolean 330x280x3 array that is found at the class website. This boolean mask
zeros-out parts of the images that typically lie outside of faces, so that the image background does not
affect the results.
ComputeEigenfaces first computes the mean celebrity face, and then "centers" the celebrity images by
subtracting out this mean face. mean_face should be a 330x280x3 array.
Next, ComputeEigenfaces should apply the binary mask to each image, to zero-out the background.
Next, compute the eigenfaces. To do this, you will first want to construct an array that contains each
celebrity image. It is convenient here to keep only those pixels that are not masked out. You can do this
by:
unmasked_pixels = find(binary_mask);
im_vector = im(unmasked_pixels);
Later, when you need obtain the masked image from im_vector, use:
full_im = zeros(size(binary_mask));
full_im(unmasked_pixels) = im_vector;
