“Averaging” Dominates Feature Integration in Low-Contrast Crowding

Ruijie Jiang,Nihong Chen

DOI: https://doi.org/10.1167/jov.22.14.3534

2022-01-01

Journal of Vision

Abstract:Visual crowding, the inability to identify objects in clutter, is considered as a fundamental limit in peripheral vision. There is a growing consensus that crowding originates from inappropriate feature combination. An "averaging" strategy was first proposed explaining the integration of orientation signals in a crowded display, and later a "signed-max" rule was believed to contribute to the nonlinearity in the perceptual results. To investigate the feature integration mechanism, we presented an array of Gabor patches at low contrast (Michelson contrast: 10.5% ± 6.4%, determined at 75% correct threshold for a single patch in a 2AFC spatial task). The array consisted of a central target at 5° horizontal eccentricity and eight surrounding flankers. Some flankers were tilted as the target, and the others were vertical. The stimuli varied in the number of tilts (1/2/3/7) and the tilted angle (1°/2°/4°/8°/16°, clockwise or counterclockwise). The tilting flankers were arranged either along the radial axis (the axis connecting the target to fovea) or along the tangential axis. Subjects’ perceptual distributions were measured using a magnitude matching method, in which they reported the direction and the magnitude of the tilts. At the end of each trial, we asked subjects to rate their confidence. Low-confidence responses were filtered out in subsequent analyses. We compared the performance of an "averaging" model and a "signed-max" model by calculating correlation coefficients between the perceptual distributions from model predictions and measurements. We found that the "averaging" model fits the data better than the "signed-max" model. A spatial anisotropy was also observed: when incongruent flankers were arranged radially, the “averaging” model fits the data better than when incongruent flankers were arranged tangentially. These findings indicate a dominant role of "averaging" strategy in feature integration in low-contrast crowding.