Abstract
Dakin and Watt (J Vis., 2009, 9(4):2, 1-10) suggested that face identity is conveyed primarily by the horizontal structure in a face. We evaluated this hypothesis using upright and inverted faces masked with orientation filtered Gaussian noise. Observers completed a 10-AFC identification task that used faces that varied slightly in viewpoint. Face stimuli were presented in horizontal and vertical noise, and in a noiseless baseline condition. Both face and noise orientation were varied within subjects, with face orientation blocked and counter-balanced across two sessions and noise orientation varying within each session. We measured 71% correct RMS contrast thresholds for each condition and then converted the thresholds into masking ratios defined as the logarithm of the ratio of the masked and unmasked thresholds. There was a significant effect of noise orientation for upright faces (F(1,11)=5.162, p<0.05), with horizontal noise producing more masking than vertical. However, this effect did not appear for inverted faces. In a second experiment, we found that the pattern of masking did not change significantly with the RMS contrast of the masking noise (F(2,4)=1.013, P>0.4). Finally, we simulated the performance of Dakin and Watt's so-called barcode observer for our experimental conditions, and found that the predictions of the model were consistent with the masking data obtained with upright faces. Together, these data suggest that observers may indeed identify faces preferentially using the horizontal structure in the stimulus.