Featural representation and internal noise around the visual field

Abstract

In human adults, visual performance varies systematically around the visual field. It is higher along the horizontal than the vertical meridian (horizontal–vertical anisotropy, HVA) and higher at the lower than the upper vertical meridian (vertical–meridian asymmetry, VMA). Although these robust performance fields have been linked to non-uniform neural resources, the system-level computations that translate neural constraints into perceptual asymmetries remain largely unexplored. Here, we used reverse correlation to characterize feature weighting and internal noise during peripheral orientation detection. Reverse correlation revealed non-ideal feature weighting in the joint orientation–spatial-frequency space, which was incorporated into a noisy-observer model jointly constrained by trial-wise detection responses and double-pass consistency. Across observers, the magnitude of the HVA in contrast sensitivity was correlated with individual asymmetries in orientation sensitivity and additive internal noise. In contrast, we found limited evidence that any tested representational or noise components reliably accounted for individual differences in VMA magnitude. Spatial-frequency tuning exhibited substantial individual variability, often peaking below the signal’s spatial frequency, but did not vary systematically across locations or explain performance asymmetries. These findings suggest that the HVA reflects systematic variation in the feature weighting of task-relevant orientations and internal noise, while constraining which computational components provide robust explanations of polar-angle asymmetries. Moreover, our framework provides a principled approach that links this prevalent perceptual asymmetry to the system-level computations that transform sensory information into perceptual decisions. Author summary Human vision is surprisingly uneven across our field of view. At the same distance from where we look, vision is better along the horizontal axis than the vertical axis, and better in the lower than the upper half of the vertical axis. These differences are well established, but little is known about their computational basis. To investigate, we combined visual detection tasks with computational modeling. By analyzing how observers detected faint patterns within noisy images, we measured how they process distinct features—like patterns with different orientations and spatial frequencies at different visual field locations. We then fit a mathematical model to separate distinct sources of internal noise. We found that differences in how the brain processes line orientations predicted the magnitude of the horizontal-vertical asymmetry across individuals, whereas distinct components of internal noise predicted these asymmetries in different ways. Our results show that visual field asymmetries are not driven by a single visual bottleneck, but rather by location-specific combinations of how the brain encodes relevant feature information and neural noise. This study helps explain why human vision is fundamentally uneven across our field of view.