Eigen Faces:
Average Face:
Program output for 05 eigen faces | 10 eigen faces | 12 eigen faces | 15 eigen faces | 20 eigen faces | .bat | .face | .user
It is best to use as few eigenfaces as possible while still getting good results because the more eigenfaces you have the more computation there is and therefore the longer it will take to run the tests.
1.
05 - 2 right, 20 wrong
10 - 3 right, 19 wrong
12 - 4 right, 18 wrong
15 - 4 right, 18 wrong
20 - 4 right, 18 wrong
2. I think the incorrect identifications is reasonable. They do look similar to the actual person. I see similar facial features in the images.
input: 
output: 
correct output: 
input: 
output: 
correct output: none
3. The program performs more poorly in this recognition task than the previous one because:
a. Not all of the class has ugrad pictures, and therefore cannot get the correct match up.
b. The ugrad pictures were taken a year or more ago, and therefore the people could look older or have grown facial hair or have glasses. All of witch will make them look different.
c. The lighting is different. This makes the skin look like a different color and the shadows could be in a different place. I also do not normalize the faces when I make the eigenfaces and or later when I am comparing. This will exaggerate the problem.
4. With a very small number of eigenfaces the result were worse. There became a point when increasing the number of eigenfaces did not increase the number of correctly guessed faces. The optimal number of eigenfaces is 12.