-
A matrix of test images and training images were obtained.
-
Both the matrices were normalized by subtracting the mean.
-
The training images were decomposed using the SVD into eigenfaces(eignevectors).
-
Adequate number of principle components(k) were determined amongst the largest eigenvalues.
-
Each training image was then represented as a linear combination of the k-eigenfaces (eigenvectors).
-
The testing images were then projected on the the eigenspace and the minimum distance of the testing image with a training image.
-
The least distant image was recognized and compared with the correct label(f(x)).
-
If h(x) = f(x), then a correct face was recognized.
