Purchase this article with an account.
Harry Griffin, Alan Johnston; Reconfigurable face space for the perception of inter-gender facial resemblance. Journal of Vision 2010;10(7):705. https://doi.org/10.1167/10.7.705.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Psychophysical and neurophysiological evidence suggests that faces are represented in a mean-centered multi-dimensional face space. However, the organization of face space is poorly understood. We used a novel markerless morph-vectorization technique, based on the multi-channel gradient model, to investigate the organization and rapid reconfiguration of face space. Face spaces for male and female faces were created via principal component analysis (PCA). Faces' shape and texture were described as vector deviations from populations' mean faces. Novel faces were synthesized by translating these vectors within and between male and female face spaces and then reconstructing to image form. The mathematical basis of perceptual similarity between male and female face spaces was investigated by showing subjects cross-gender pairs of faces which had either similar, unrelated or opposite vector deviations from their population mean. Subjects perceived faces with similar vector deviations from their respective means (sibling-faces) as most similar and faces with opposite vector deviations as least similar. Facial identity aftereffects also transferred between male and female face spaces. Adaptation to a male face yielded a shift in perceived identity of female faces toward the mathematically opposite female face. The perceptual similarity of synthesized sibling-faces indicates that face space can be dynamically partitioned into mean-centered subspaces e.g., male and female. This ability may underpin the perception of “family resemblances” in disparate groups of faces with widely varying underlying image statistics. Cross-sibling adaptation indicates the existence of relational as well as absolute coding in face space.
This PDF is available to Subscribers Only