Abstract
Humans and other species use saccadic eye movements to redirect the high-acuity fovea toward salient locations in a visual scene. In the course of shifting gaze, saccades necessarily introduce abrupt changes in the luminance signals impinging onto retinal receptors. Much work has been devoted to understanding how the visual system deals with the potentially negative consequences of saccade transients. However, it has also been argued that the saccade-induced reformatting of luminance patterns into temporal modulations is actually beneficial for visual encoding [Boi et al., 2013]. To explore this idea, we examined the frequency content of the visual signals delivered by saccades to the retina. We first recorded the eye movements of human observers at high resolution during free-viewing of natural scenes. We then reconstructed the visual input to the retina around the time of saccades and estimated its power in the joint space-time frequency domain. We show that the power redistribution resulting from saccades consists of two regimes. Below a critical spatial frequency, dynamic power (the power at non-zero temporal frequencies) impinging on the retina does not depend on spatial frequency. This happens because in this range, the saccade-induced conversion of spatial patterns into temporal modulations counterbalances the spectral density of natural scenes. Above the critical spatial frequency, the dynamic power follows the spectral density of natural scenes. Although this critical spatial frequency depends on saccade amplitude, there is a broad range of spatial frequencies for which the resulting spectrum is independent of saccade amplitude, thus providing a veridical representation of the visual scene independent of the specific saccade performed. We elucidate the origin of these effects, model their impact on the responses of neurons in the early visual system, and discuss their possible consequences for the establishment of visual representations during the saccade-fixation cycle.
Meeting abstract presented at VSS 2018