Figure 2
From: Integration or separation in the processing of facial properties - a computational view

Face space for invariant facial properties.
Fisherface projections of the first components. Each panel (A–E) shows the 3D projection of the first three components (left) and enlargement of the 2D projection of the first two components (upper right panel, magnifier plots) and a histogram of Euclidean distances of each sample to each other sample in the face space average according to class labels (‘within’: same class; ‘across’: difference classes). Figure titles indicate training and testing samples (e.g. training on identity and testing on race (ID:RA)). Colors in the 3D and 2D panels indicate classes. Note that in panel (A) due to the great number of classes (here identities), two different classes (identities) can have relatively similar colors but are located apart from each other in the face space. In panel (A) full circles indicate testing samples; outline circles indicate training samples. In (B–E) light colors indicate training samples.