A topology preserving mapping for face recognition

We review a form of topology preserving mapping which uses the same underlying structure as the generative topographic mapping (GTM) but organises the projections of the latent points into data space based on the method of harmonic K-means. We show that projections of the Olivetti face database onto this latent space show good performance in terms of identifying all images of a particular individual as lying in the same section of the latent space.