Project 4: Eigenfaces

Landon Todd Detwiler
CSEP 576 Winter 2005

Let's play, who in my class looks the most like ...?
(I apologize in advance to those who were selected by my program)

Who looks the most like Britney Spears?

Who looks the most like Brad Pitt?

OK, seriously though, below are the links to my results from the 2 main experiment types:

• Testing recognition with cropped class images
• Cropping and finding faces

For extra credit on this assignment I implemented the "speedup". This extra makes calculating the eigenfaces much faster by reducing the dimensions of the covariance matrix (instead of n squared by n squared where n is the height or width of a square image, it is reduced to m by m where m is the number of images). The observation that makes this speedup possible is that if we multiply the eigenvectors of aTa (where a is an array and T means transpose) by a, we get back vectors that are equivalent (possibly with a change in sign) to the eigenvectors of aaT.