6 Recognition , Classification and Inversion of Faces in the Multidimensional Space

1.1 The multidimensional space One of the leading models of face recognition is the multidimensional space (MDS) model proposed by Valentine (1991a) which suggests that faces are encoded as points in a metaphoric multidimensional space, and that any characteristic which differentiates among faces can serve as a dimension in that space (Bruce, Burton, & Dench, 1994; Valentine, 1991a; Valentine & Endo, 1992). According to the model, faces are normally distributed along each of the dimensions which share a common center (Bruce et al., 1994; Johnston, Milne, Williams, & Hosie, 1997; Lewis & Johnston, 1999b). Most faces that one encounters are typical, and as such are distributed around the center of the MDS. Distinctive faces, on the other hand, are located far from the center. Therefore, the more typical a face is on the various dimensions, the closer it is to the center of the MDS and to other faces, and consequently the smaller is its representation space (Lewis & Johnston, 1999a; Tanaka, Giles, Kremen, & Simon, 1998). Thus, the greater the similarity among the faces, the more densely they are located and hence their recognition is more difficult (Busey, 1998; Johnston, Milne et al., 1997; Lewis & Johnston, 1999a; Tanaka et al, 1998; Valentine, 1991a,b, 2001; Valentine & Bruce, 1986a,c). As the MDS model suggests, distinctive faces are indeed recognized faster and more accurately than typical ones (Lewis & Johnston, 1997; Valentine, 1991a, 2001; Valentine & Bruce, 1986a,c; Valentine & Ferrara, 1991; Wickham, Morris, & Fritz, 2000).