Osman Kavcar¹ (benkavcar@gmail.com), Michael E. Rudd^1,2, Christabel Arthur¹, Michael A. Crognale^1,2; ¹University of Nevada, Reno, ²Center for Integrative Neuroscience

According to Rudd’s edge-integration model with contrast gain control, the achromatic appearance of a match disk varies with the target disk’s luminance and its contrast polarity relative to the annulus in a simple disk / disk–annulus display. This dependence appears in the coefficients of the achromatic color-appearance matching functions (ACAMFs), obtained by fitting a second-order regression model to the match-disk settings at each sampled annulus luminance. Thus, the brightness-matching behavior in a disk–annulus stimulus should carry a hidden imprint of the target disk luminance under which the match was made, even though that luminance is not explicitly encoded in the match settings themselves. We tested whether this imprint is recoverable from behavior alone by training a smooth forward model exclusively on data collected at a single target disk luminance (T18). For each contrast-polarity condition, a Gaussian Process Regression (GPR) model was fit using annulus luminance, annulus radius, and (in a separate version) participant identity as predictors of log match-disk settings. These models captured the characteristic induction function for T18 and were subsequently frozen as fixed reference manifolds. New datasets collected at T11, T14, T22, and T28 were then passed through the T18 models, and trial-wise residuals were computed and plotted as a function of log annulus luminance. Across all polarity conditions, residual curves corresponding to different target luminances formed orderly, non-overlapping “fans,” maintaining a consistent vertical ordering (T11 < T14 < T22 < T28) and preserving curvature within polarity. Including participant identity did not alter this structure. These properties demonstrate that target disk luminance systematically shifts the induction function in a stable, recoverable manner. Thus, brightness-matching behavior embeds a discernible imprint of the disk luminance under which observers perform the task, even when the model has no explicit access to that variable.

Acknowledgements: Research reported in this presentation was supported by National Institute of General Medical Sciences of the National Institutes of Health under grant numbers [P30 GM145646 and/or P20 GM103650]