System-Level Computations Underlie Visual Field Heterogeneity

Abstract

Human visual perception for basic dimensions varies with eccentricity and polar angle, influencing daily activities such as reading, searching and scene perception. We investigated whether and how system-level computations that transform visual input into perception underlie these heterogeneities. Using the equivalent noise method and perceptual template model, we estimated gain, internal noise, and nonlinearity for orientation discrimination across eccentricity (fovea, parafovea and perifovea) and around polar angle. Participants discriminated the orientation of Gabors embedded in dynamic white noise and showed the expected variations across eccentricity and around polar angle. Importantly, visual performance declined with eccentricity due to decreased gain and nonlinearity and increased internal noise. Observers with stronger eccentricity effects showed greater gain decrease. Only gain varied with polar angle--higher along the horizontal than vertical meridian, and lower than upper vertical meridian--paralleling performance asymmetries. This dissociation aligns with known variations in neuronal count and tuning, suggesting that neural correlations and neural noise contribute to these system-level computations. By revealing distinct system-level computations underlying the eccentricity effect and polar angle asymmetries, our findings link perceptual heterogeneity across the visual field and neural architecture and provide insights into how the human brain encodes information under neural constraints. Significance StatementHuman visual performance varies across eccentricity-distance from gaze-and around polar angle -circular dimension. Retinal factors and cortical surface area account for eccentricity, but only partially for polar angle variations. Here, we show that system-level computations--how signal are amplified and how noisy the system is--distinctly underlie these perceptual variations. Decline across eccentricity stems from both reduced gain and nonlinearity and elevated internal noise, whereas variation around polar angle arises solely from gain difference. This dissociation implies fundamental differences in information processing along two axes of the visual field. Our results provide a computational link between behavior and its neural bases and highlight the importance of considering both eccentricity and polar angle in modeling and understanding human perception.