Abstract
Faces are recognized more easily when upright than inverted. Contrary to many theories, recent studies suggest that the inversion effect is not due to subjects using different spatial or spatial frequency information. So the question remains: What causes the inversion effect? Here we examine whether sensitivity to information at different orientations may account for the face inversion effect. Face identity is conveyed primarily by information in the horizontal band (Dakin & Watt, J Vis 2009), and observers are more sensitive to this information for identification of upright faces than inverted faces (Goffaux & Dakin, Front Psychology 2010; Pachai et al., VSS 2010). To determine whether these sensitivity differences are directly associated with face identification, we assessed orientation tuning of upright and inverted face identification using noise masking. Stimuli were masked with Gaussian noise filtered to contain information in one of 8 orientation bands (bandwidth = 23deg) ranging from −90 (vertical), through 0 (horizontal) to 67.5 degrees. We measured 10-AFC identification thresholds in 16 subjects in each orientation condition, as well as a white noise baseline condition, with upright and inverted faces. On average, we found strong masking centred on the horizontal band for upright faces and significantly weaker masking for inverted faces. Furthermore, the degree of horizontal tuning was strongly correlated with baseline identification performance for upright, but not inverted, stimuli. Finally, the change in horizontal tuning following inversion and the size of the inversion effect in the baseline condition were strongly correlated (r = 0.564, p = 0.023). Together, these results show that sensitivity to horizontal information in the face is associated with face identification performance, and supports the idea that a loss of this sensitivity following inversion underlies the face inversion effect.
NSERC, Canada Research Chairs.