Poster Session II: Computational analysis of the effect of cone temporal filtering on detection threshold with and without retinal motion.

Abstract

Retinal stimulus motion can increase visual acuity, but recent experimental evidence indicates that for briefly presented stimuli this benefit does not always occur (Braun et al, VSS 2023). To understand this effect, we modeled how the temporally low-pass filtering that occurs within each cone is expected to impact visual performance. We simulated a grating (10 cpd Gabor) detection task in which the stimulus was present for two 15 msec frames. We used the ISETBio software to compute cone excitations and cone photocurrent responses to the stimulus, for different contrasts and retinal positions, and used linear SVM classifiers to estimate computational-observer detection thresholds. We examined three retinal-motion conditions: 1) stimulus stabilized on the retina; 2) stimulus shifted 0.5 cycle orthogonal to grating orientation across the two frames; 3) stimulus shifted 1 cycle. Across all three conditions, detection threshold estimated on the basis of cone excitations varied little. Threshold estimated on the basis of photocurrent, however, was highest for the 0.5 cycle shift, and about the same for the no shift and 1 cycle shift conditions. Qualitatively, this result recapitulates the findings of Braun et al. and suggests that those findings may be understood as a consequence of the initial visual encoding. We note, however, that we analyzed detection rather than acuity; planned work will attempt to bring the computations into more direct contact with the experimental results.