To generate a set of eigenfaces, a large set of digitized images of human faces, taken under the same lighting conditions, are normalized to line up the eyes and mouths. They are then all resampled at the same pixel resolution (say m×n), and then treated as mn-dimensional vectors whose components are the values of their pixels. The eigenvectors of the covariance matrix of the statistical distribution of face image vectors are then extracted.
Since the eigenvectors belong to the same vector space as face images, they can be viewed as if they were m×n pixel face images: hence the name eigenfaces.
Viewed in this way, the principal eigenface looks like a bland androgynous average human face. Some subsequent eigenfaces can be seen to correspond to generalized features such as left-right and top-bottom asymmetry, or the presence or lack of absence of a beard. Other eigenfaces are hard to categorize, and look rather strange.
When properly weighted, eigenfaces can be summed together into a gray-scale rendering of a human face. Remarkably few eigenvector terms are needed to give a likeness of most of most people's faces, so eigenfaces provide a means of applying data compression to faces for identification purposes.