First we took a training set of face images. We reduced the
resolution and converted them to black and white. From there each image
could be thought of as a vector each pixel being a dimension. We
reduced this space using PCA.
I subtracted the average face from the face then found the covariant
matrix of these points. I then computed the eigenvectors of the matrix
(actually this we done indirectly, see extra credit). I then took the
top n eigen vectors to be used for the rest of the project. Below are
those vectors reinterpreted as faces.
Using a 25 x 25 resolution and the neutral class pictures.
Eigen Vector | Eigen Value |
1 | 12321061 |
2 | 8632369 |
3 | 5174586 |
4 | 3351286 |
5 | 2359568 |
6 | 1868694 |
7 | 1431015 |
8 | 1191470 |
9 | 1102703 |