Eigenfaces for Familiarity

A previous experiment tested subjects’ new/old judgments of previously-studied faces, distractors, and morphs betw e n pairs of studied parents. We examine the extent to which models based on principal component analysis (eigenfaces) can predict human recognition of studied faces and false alarms to the distractors and morphs. We also compare eigenface model s to the predictions of previous models based on the positions of faces in a multidimensional “face space” derived from a multidimensional scaling (MDS) of human similarity ratings. W e find that the error in reconstructing a test face from its posi tion in an “eigenface space” provides a good overall predict ion of human familiarity ratings. However, the model has difficulty accounting for the fact that humans false alarm to morp hs with similar parents more frequently than they false alarm t o morphs with dissimilar parents. We ascribe this to the limit ations of the simple reconstruction error-based model. We th en outline preliminary work to improve the fine-grained fit with in the eigenface-based modeling framework, and discuss the re sults’ implications for exemplarand face space-based mod els of face processing.