Purchase this article with an account.
Daniel Fiset, Caroline Blais, Frédéric Gosselin, Philippe Schyns; Effective frequency tuning of three face categorization tasks. Journal of Vision 2006;6(6):273. doi: https://doi.org/10.1167/6.6.273.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Using a new application of the Bubbles method (Gosselin & Schyns, 2001), we estimated the frequency tuning of three face categorization tasks (Identity, Gender and Expressive or not (Exnex)) applied to 20 face pictures (5 males and 5 females identities, smiling and neutral). On each trial, we exposed 4 observers to a continuous sample of Spatial Frequency (SF) information. The continuous sample of spatial frequencies involved several computations. For each trial, we first produced a white noise curve of random numbers (between 0 and 1) convolved with a Gaussian function (std = .07 of the Nyquist). We also computed the complement curve (1 - white noise curve). We Fourier transformed a randomly chosen face and multiplied the amplitude of Fourier coefficients with the same random curve, for each orientation of the SF in Fourier space. The complement curve similarly multiplied the Fourier coefficients of face noise. The full spectrum stimulus summed the resulting random sample of the Fourier coefficients of the face and complement face noise. In each task, we maintained observers' accuracies halfway between chance and ceiling (75% for gender and Exnex, and 55% for identity) by adjusting on each trial the sampled amplitude of face frequencies. Multiple linear regressions on response accuracy and sampling noise revealed task-dependent frequency tuning: In Gender, a narrow band peaked at 4 cycles per face (cpf); in Exnex, a broader band covered 5 to 12 cpf. In Identity, two bands (one peaking at 5 cpf, the other at 12 cpf) characterized performance.
This PDF is available to Subscribers Only