Random and systematic effects of neural noise on low-level and high-level pattern vision

Abstract

The visual system never perfectly represents the physical world. Rather, neural noise sometimes perturbs perception of simple shapes, especially when they are viewed briefly. Neural noise can randomly influence perception, but may also systematically distort perception, as in feature exaggeration. We examined these random and systematic influences of neural noise on curvature perception when short contour segments were presented alone (primarily involving low-level processing) and when they were embedded as the mouths within schematic faces (involving high-level object processing). Inverted-and-scrambled versions of the faces were used to control for a potential effect of crowding. To detect perceptual noise with high sensitivity, we used a procedure developed by Baldassi, Megna, and Burr (2006) in which brief presentations were comprised of an array of straight segments with either one curved target or no curved targets. The task was to report the curvature of the segment that appeared most curved using a magnitude-estimation procedure. Random noise was estimated as the variance of the reported curvatures for displays where curved targets were absent. Curvature exaggeration was measured as the average difference between the reported curvature and the actual curvature of the target. Random noise in curvature perception was greater when segments were presented within faces or inverted-scrambled faces compared to when they were presented alone, suggesting that crowding increased random perceptual noise. Curvature exaggeration, in contrast, was substantial when segments were presented alone or within inverted-scrambled faces, whereas exaggeration was weakest (reduced by ∼40%) when segments were presented within faces. This pattern of curvature exaggeration may reflect the possibility that systematic exaggeration is beneficial in low-level processing because a unique feature may become more salient, but could be harmful in high-level processing because perceptual exaggeration of facial expressions may lead to socially inappropriate responses.