Abstract
Images of androgynous faces typically are generated by morphing masculine and feminine faces, based on an assumption that androgynous faces are equally masculine and feminine. Our past work (VSS 2020) challenged that assumption, finding that androgynous faces could be perceived simultaneously as both androgynous and strongly gendered. The current study uses a reverse correlation technique (Dotsch & Todorov, 2011) to address directly the question of what makes faces appear more or less androgynous. On each of 600 trials, an observer viewed a pair of male or female faces embedded in Gaussian white noise and chose the face that appeared more androgynous. The two noise fields varied across trials, but were anti-correlated within each trial. Noise fields were sorted based on observer responses, and averaged to create a Classification Image (CI) and an antiCI. Initial results from two female observers showed that when the CI and antiCI were added to base faces with the same gender as their original base, resulting images clearly differed, with base-face+CI less strongly gendered (more androgynous) than base-face+antiCI. Interestingly, adding the CI and antiCI to the opposite gender face had the opposite effect: a CI obtained from a female base-face made a male base-face appear more masculine, whereas the antiCI made the male face appear more androgynous. We currently are conducting statistical analyses and behavioural experiments to examine i) how the spatial structure in the CI varies across base face images; and ii) the consistency of CIs across observers. We also plan to repeat the study with more faces, tasks, and observers to determine the limits of generalizability, and to determine more precisely the relationship between androgynous CIs and gendered CIs. The results of these studies will shed much needed light on our understanding of how the visual system processes face gender information.