Eigen-faces
Testing recognition with cropped class images, procedure 1. Use the cropped, non-smiling faces (24 in total) to compute 10 eigenfaces. The face size I used is 40 by 40, and the average face and eigenfaces shown below are upsampled to 120 by 120 in Photoshop.
Average face:
The top 10 eigen faces are:
![]() |
![]() |
![]() |
![]() |
![]() |
1 | 2 | 3 | 4 | 5 |
![]() |
![]() |
![]() |
![]() |
![]() |
6 | 7 | 8 | 9 | 10 |
Extra Credits Item:
I implemented the trick for speeding up calculating eigenvectors, so that the covariance matrix is always 24 by 24 (24 is the total number of training images), and independent on the face image size.