Project 4 - Eigenfaces
Alexander Faucher 0321059
Testing
recognition with cropped class images
10
9
8
7
6
5
4
3
2
1
Average
base.user
10 Eigenfaces
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
| 531911 |
400873 |
501975 |
545528 |
704872 |
268648 |
1810730 |
237490 |
368143 |
664958 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
| 529120 |
531710 |
995626 |
348898 |
847846 |
285296 |
350836 |
378429 |
1150370 |
1026070 |
 |
 |
 |
 |
 |
 |
 |
|
|
|
|
|
 |
 |
 |
|
 |
|
|
|
| 929694 |
326076 |
225774 |
265869 |
523381 |
650852 |
279168 |
|
|
|
Total 20/27
Questions
1. It should be clear increasing the number doesn't
improve results at all past a certain point. Using more then this
# results in slower execution, larger storage, and no fewer errors.
11 eigen faces yeilded the best result in this case.
2. See above for misclassifed results.
Cropping and finding faces
The cropped result isn't informative. This is only marginly
better. It didn't work nor could I see a way to rectify it.
Elf
Min 0.45, Max 0.55, Step 0.01, Mark 10
Portrait
Min 0.07, Max 0.12, Step 0.01, Mark 1
The elf image returned 2 false postitives and 2 false negatives.
I don't know why it thought the chair/wall looked like people.
I also implemented verifyFace