Purchase this article with an account.
Danelle A. Wilbraham, Aleix M. Martinez, James T. Todd; Exploring the nature of the multidimensional face space. Journal of Vision 2009;9(8):521. doi: https://doi.org/10.1167/9.8.521.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Many researchers agree that faces are perceptually represented in a multidimensional space; however, the nature of the constituent dimensions of this space remains unclear [Wilbraham et al. (2008), JOV 8(15):5]. The research described here further explored the nature of this face space. In Experiment 1, a variety of homeomorphic image transformations, each at several magnitudes, were applied to images of faces. Some of these transformations approximated naturally-occurring craniofacial changes (e.g., growth), while others did not. On each experimental trial, observers indicated which of two transformed images depicted the same individual as an untransformed sample image. The difference in image structure produced by each transformation was indexed with several metrics involving either pixel intensities or wavelet outputs, enabling performance comparisons across transformations. The results suggested that observers' judgments may have been based on configural relations among facial features that remain relatively invariant over some types of transformations, but not others.
In Wilbraham et al. (2008), we observed that randomizing the amplitude spectrum produced reasonably recognizable images, while randomizing the phase information did not. Based on this observation, Experiment 2 was designed to examine face matching performance when only phase information is preserved. Observers matched unaltered sample images to three types of comparison images: unaltered images, images with randomized amplitude spectra, and transformed images from Experiment 1 that produced changes in image structure similar in magnitude to that resulting from amplitude randomization. The amplitude randomization had little effect on performance with respect to baseline performance on unaltered images, unlike the effects of other variations of similar magnitude. From these results we concluded that the information specified in the phase spectrum is sufficient for face recognition, while the amplitude information is relatively unimportant, suggesting that phase information likely plays a significant role in the ability to recognize faces.
This PDF is available to Subscribers Only