one is a Mahalanobis-like distance between the new image and the prototype centroid, defined within the subspace spanned by the 75 largest eigenvectors of the prototype cluster, while the other is the Euclidean distance from the new image to its projection in the subspace.