Purchase this article with an account.
Philip M. Bronstad, Judith H. Langlois; Image warping does not model variation in facial masculinity. Journal of Vision 2005;5(8):978. doi: 10.1167/5.8.978.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Differences in facial appearance between women and men (facial masculinity/femininity) have been proposed to account for variance in perceived attractiveness of faces. Studies of men's facial attractiveness show that, in many cases, masculine faces are preferred to feminine faces. However, some studies show the opposite, particularly studies that used image warping to masculinize or feminize images of men's faces. We propose the terms masculinity (variation in unaltered faces) and masculinization; (images warped to be more masculine-appearing) to differentiate the methods.
We altered men's images with image warping and found masculinity to be positively - but masculinization negatively - associated with judgments of men's attractiveness. To understand how masculinity and masculinization differ, we used a computational model to determine whether they are subsets of the appearance differences between men and women. Once trained to differentiate images of men and women, the model moderately predicted masculinity judgments of unaltered men's face images. The model did not predict masculinity judgments of masculinized images of men. We also demonstrate that the ways in which appearance variation in unaltered and warped images predicts masculinity judgments are remarkably distinct.
Thus, warping produces images that do not parallel natural variation in masculinity. At least two reasons explain the incongruity. First, growth and differentiation are unlike a single deformation. Second, warping methods generally don't model the pigmentation differences between men and women.
Warping is useful to isolate sources of facial variation for experimentation, so we suggest a guideline to its use. If warping is used to produce representations of the results of “real-world” processes such as growth, the burden of proof is on the experimenter to carefully demonstrate that the altered images represent valid variants.
This PDF is available to Subscribers Only