Semin Oh¹ (), Fangfang Hong², Derek Nankivil³, Edda B. Haggerty⁴, Robert Stauble⁵, Nicolas P. Cottaris², John Buch³, Geoffrey K. Aguirre², David H. Brainard²; ¹Justus Liebig University Giessen, ²University of Pennsylvania, ³Johnson & Johnson, ⁴Growth Minded Co., Wayne, US, ⁵SeeSharp, Philadelphia, US

The chromatic spatial contrast sensitivity function (csCSF) provides a fundamental characterization of pattern vision. Models of the csCSF often focus on group-averaged data, potentially obscuring meaningful individual variation. Here, we explored individual differences in the csCSF by presenting participants (N = 32) with L-M cone-contrast Gabor images (0.75° SD; stimulus size = 7°) at five different spatial frequencies (3, 6, 9, 12, 18 cpd) using a custom Maxwellian-view hyperspectral display. To support our scientific goals, we did not adjust the stimuli for individual isoluminance. On each trial participants reported the stimulus orientation (tilted +/- 45° relative to vertical). Contrast was varied to determine threshold. While the group-averaged csCSF was low-pass, individual csCSFs had diverse shapes; many exhibited a notch at intermediate spatial frequencies. We modeled the data using a computational simulation. Retinal images were obtained by convolving each stimulus with 79 different polychromatic (i.e. including chromatic aberration) optical point spread functions (PSFs), derived from published wavefront-aberration measurements in a different set of participants. For each PSF, we found a computational-observer csCSF by i) computing the Poisson-distributed cone excitations of a simulated cone mosaic for each retinal image; ii) training a linear classifier to discriminate the stimuli; iii) evaluating classifier performance across contrast. For each of our participants' csCSFs, we found the best-fitting of the 79 computational csCSFs, with one free gain parameter for overall sensitivity. This model captured individual csCSF shapes (R2 = 0.831), outperforming an individually-scaled group-average csCSF model (R2 = 0.685), and approaching the measurement-noise-limited performance of a model that fits CS at each spatial frequency for each participant separately (R2 = 0.890). These findings indicate that optical factors (including chromatic aberration) coupled with sampling by the cone mosaic account for individual variation of the L-M csCSF, and should be incorporated into models of human pattern vision.