The face inversion effect is nothing “spatial”

Abstract

Face identification accuracy declines and response time increases when stimuli are presented upside-down compared to when they are presented upright (Yin, 1969). This face inversion effect (FIE) is one of the most robust findings in the face literature. Here, we investigated whether inversion leads to qualitative changes in spatial frequency use. Participants were instructed to identify 10 individuals from 20 grayscale face photos. Stimuli were constructed by randomly sampling the spatial frequency information in the images using 45 Gaussian “bubbles” with a standard deviation of 1.5 octaves applied to the logarithm of spatial frequencies (Gosselin & Schyns, 2001; Fiset et al., 2006, VSS). By adding white Gaussian noise, performance for upright faces was adjusted to about 80% and 92% correct (mean response time of 1496 ms), in the accuracy and the response time versions of the task, respectively. The same amount of Gaussian noise was applied to the inverted faces; this led to a performance of 52% and 72% correct (with a mean response time of 1935 ms), in the accuracy and response time versions, respectively, thus demonstrating a clear FIE. Our results show the same spatial frequency band peaking at approximately 9 cycles per face width (octave width of about 2) in all conditions. The present findings thus do not support that inversion leads to qualitative changes at the spatial frequency level. They rather suggest that the same spatial frequencies are processed more efficiently during upright face identification.