Abstract
Visual perceptual learning (VPL) of orientation occurred as a result of exposure to task-irrelevant supra-threshold natural stimuli (NS), but not to artificial stimuli (AS) that consisted of the identical primitive statistics such as the luminance, orientation, and spatial frequency distributions to those of the NS (Shibata et al., VSS, 2020). Here we examined what factor in NS allows task-irrelevant VPL to occur. Given that the above-mentioned primitive statistics information in AS that was extracted from NS was not sufficient for task-irrelevant VPL, one possibility is that information about spatially regular patterns in NS, which cannot be synthesized by the primitive statistics, is necessary for task-irrelevant VPL. To test this possibility, we used Portilla & Simoncelli stimuli (PSS), which include correlational structures among signals from different positions, orientations, and scales and have the spatially regular pattern information of NS (Portilla & Simoncelli, Int J Comput Vis, 2000). For 10 days, three different groups of participants (n=12 for each) were repeatedly exposed to a set of supra-threshold NS, PSS, or AS with a dominant orientation (trained orientation) while performing a rapid serial visual presentation (RSVP) task. Before and after the exposure, orientation discrimination performance was measured on both the trained and untrained orthogonal orientations. For the trained orientation, significant performance improvement was found in the PSS condition, although it was not as great as in the NS condition. However, no significant improvement was obtained in the AS condition. For the untrained orientation, none of the conditions showed significant improvement. We also found that performance on the RSVP task was significantly lower in the NS and PSS conditions than in the AS condition. These results suggest that the higher-order statistics in the PSS are not suppressed even if the stimuli are task-irrelevant and partially contribute to VPL of the orientation in the stimuli.