Machine Recognition in Complex Domain - High Dimensional Neurocomputing: Growth, Appraisal and Applications - page 150

Fig. 7.3 Eigenvectors corresponding to the smallest eigenvalues, derived from the ORL face data-
base

Fig. 7.4 ICA basis vectors derived from the ORL face database

of eigenvectors of R PCA. Therefore with 48 subspace dimensions C PCA yields
better accuracy in recognition as well as better class distinctiveness in comparison
to R PCA.