Testing Recognition with Cropped Class Images

Eigen Faces:

Average Face:

Program output | .bat | .face | .user 

1. My program found 19 faces correctly and 3 faces incorrectly.

2. The average position for the correct answer when the program was wrong is 4th place.

Incorrect Matching:

For the first place guess was with actual face placed second.

For the first place guess was with actual face placed seventh.

For the first place guess was with actual face placed fourth.

3.  I do not see a correlation between recognizing a female face and having the runners up also be female.  It does, however, look like the second runner up for males is most likely male.   

female faces 

m,f,m alissah
f,f,m esp
f,f,m jwkim
m,f,m margaux
f,m,m melissa
f,f,f mhl
m,m,f squinn
m,f,m tamoore
m,m,f tshail

male faces

m,f,f adeakin
m,f,m amiratuw
m,m,f crosetti
f,m,m ddewey
m,m,f djj
m,f,m galen
m,f,f jaydang
m,m,f mbixby
m,f,f merlin
m,f,m paullarp
m,m,m rhinnig
m,f,m seitz
m,f,f tanderl

4. 

I would not use the entire user set to computer the face space.  This would add to computation time and I think it would not affect the outcome all that much.  You get a good approximation of a face with a small amount of faces. 

5.

It would be better to use a face set independent of the user set to computer the eigenfaces so that the faces used do not have a bias, that is, the faces in the face set and are used for eigenfaces are more likely picked.