Probabilisticsubspacemixturemodels,asproposedoverthelastfewyears,areinterestingmethodsforlearningimagemanifolds,i.e.nonlinearsubspacesofspacesinwhichimagesarerepresentedasvectorsbytheirgrey-values.Theirlackofaglobalmappingcanberemediedbyarecentlydevelopedmethodbasedonlocallylinearem