CSE 455 Project 4: Face Recognition

By Andy Dumovic

As can be seen from the above graph, the benefit from adding more EigenFaces reaches its peak at 11 faces and after that never reaches a better value. So as you can see, the only thing that will happen if you use more than 11 EigenFaces is that your code will run slower.

Program in Action

Here is an error that I was unable to correct; I always got this patch of wall above his head to be my first result. Unfortunately I was never able to correct this particular problem.

Here is the cropped version of the below image L :

Here is an image of me and the result of cropping it;

Cropped image of me: .15 to .22 in .01 :

IMAGE_0031: .52 to .54 in .01

Image of us Guys: 4 to .5 in .02

While working on the above images I also got a few errors that don’t seem very reasonable, like the image below. I can’t understand why the really dark region near the middle fellow’s…zipper would match up well, but I will assume it is because there is some very dark but high contrast features that could be mistaken as a combination of the faces. Also, in the image below I get a false positive on his arm, but there is more texture there and especially with major features in the middle (like mouths or noses) I can see how it might be possible to construct this as a combination of EigenFaces.