Imprecision in Smooth Pursuit Facilitates Visual Processing

Humans use smooth pursuit eye movements to inspect non-stationary objects of interest. These movements fail to perfectly stabilize the object at the center of gaze and are often interspersed by small saccades, resulting in considerable motion of the stimulus on the retina. This behavior is commonly regarded as the unavoidable consequence of limits in oculomotor control. However, it may be perceptually beneficial, as it yields luminance transients on the retina that are likely to strongly stimulate neurons in the visual system. A similar oculomotor behavior during fixation on stationary objects has been shown to play a fundamental role in visual processing by reformatting spatial patterns into an efficient temporal code. Here we examine whether “errors” in smooth pursuit could also serve similar functions. Emmetropic observers (N=9) were asked to discriminate the orientation (±45°) of a Gabor (1 or 10 cycles per degree) embedded in a circular uniform patch (4° diameter), which either remained stationary on the display or translated at a constant speed (4°/s). The Gabor contrast increased gradually (1.5 s ramp) to minimize transients not caused by eye movements. Eye movements were continually measured at high resolution via a dDPI eye-tracker (Wu et al, 2023) and used to reconstruct the spatiotemporal stimulus on the retina. Our results show that, like during fixation, luminance modulations from imperfect tracking counterbalance (whiten) the power spectra of natural scenes over a broad range of spatial frequencies, an operation that attenuates redundant correlations and improves coding efficiency. The frequency band of whitening is narrower than for fixation, but contains more power. Perceptual performance closely followed the structure of temporal modulations, with better discrimination at low and high spatial frequency during pursuit and fixation, respectively. These results suggest that imperfect tracking, which maintains retinal image motion of a tracked object, facilitates visual encoding.