Peripheral spatial frequency processing affects timing and metrics of saccades Laubrock, J., Cajar, A., & Engbert, R. University of Potsdam, Germany
[email protected]
The existence of a visual acuity gradient, cortical magnification, and the two-streams hypothesis suggest a specialization of foveal vision for highacuity analysis, and of peripheral vision for orienting and target selection. How does this affect eye movements? Are different spatial frequency bands in foveal vs. peripheral vision effective in guiding saccade target selection and timing? We used Gaussian-envelope gaze-contingent filtering of spatial frequencies in pictures of natural scenes to investigate influences of level-of-detail on saccade amplitudes and fixation durations in free-viewing (E1) and search (E2) tasks.
Fig. 1. Large foveal high-pass, small peripheral high-pass, small foveal low-pass, large peripheral lowpass. Gaze is near the center, target (E2 only) in the green circle (not visible to participants).
Method • Subjects: E1: N=11 (mean age: 24 yrs), E2: N=26 (24 yrs) • Task: E1: free-viewing (15 s) in preparation for memory test; E2: search (max 20 s) for semi-transparent Gabor patch. • Material: 120 (E1), 144 (E2) photographs of natural scenes • Design: Filter Location (periphery, fovea) × Filter Type (low-pass, high-pass) × Window Size (E1: constant, 5.6 deg, FWHM; E2: small, 5.6 deg vs. large, 12 deg FWHM) + Control (no filter) • Filter: Laplacian, low-pass (E1), high-/low-pass filtering by Fourier analysis with cutoff frequency 3.2 cpd (E2). Attenuation by more than 10 dB of frequencies greater than 1.4 cpd for low-pass or smaller than 0.8 cpd for Laplacian filtering. • Procedure: Gaze-contingent filtering using the moving-window and -mask techniques (McConkie & Rayner, 1975; Rayner & Bert‐ era, 1979), window or mask always centered on current gaze po‐ sition. Precomputed foreground and background images were composited using alpha-blending with a Gaussian profile. • EyeLink 1000 (SR Research, Ontario, Canada) @ 1000 Hz
The spatial distribution of the first few fixations emphasizes the im‐ portance of peripheral vision for saccade target selection.
Spatial results reveal a preference for selecting upcoming fixation locations in the region that was unfiltered during the current fixation, and an additional influence of filter type.
Surprisingly, temporal results indicate that peripheral information does play a role in controlling fixation duration.
Filter Type control low-pass high-pass
7.5
6.5
6.0
Fixation no. in trial
4-9
6-11
Fig. 2. Smoothed spatial distribution for strong mask conditions (E2), averaged across all images and participants, and across six fixations (2-7, …).
260
Filter Type control low-pass high-pass
Filter Type control low-pass high-pass
255
250
310 305
245
300
6.5
Lowpass
14-19
315
7.0
Highpass
10-15
Fixation Duration [ms]
Filter Type control low-pass high-pass
Lowpass
Fovea (large)
2-7
320
Saccade Amplitude [deg]
Saccade Amplitude [deg]
7.0
Highpass
Periphery (small)
Fixation Duration [ms]
8.0
Control
240
295
5.5 6.0 control
periphery
Filter Region
fovea
control
periphery large
periphery small
fovea large
fovea small
control
periphery
Filter Region
fovea
control
periphery large
periphery small
fovea large
fovea small
Filter Region and Size
Filter Region and Size
Fig. 3. Saccade amplitudes by filter type and filter region in E1 (left panel), and additionally by window size in E2 (right panel). Tunnel vision evoked by peripheral masks shortens saccades, and artificial scotoma evoked by foveal masks lengthen saccades. These effects are differentially modulated by spatial frequency, suggesting higher importance of low frequencies in the periphery and of high frequencies in the fovea.
Fig. 4. Fixation durations by filter type and filter region in E1 (left panel), and additionally by window size in E2 (right panel). If matching information is severely attenuated (low-pass fovea, high-pass periphery), then default timing seems to take over. If matching information is still mostly available (high-pass fovea, low-pass periphery), then fixation duration increases.
We conclude that both foveal and peripheral vision contribute to where and when we move our eyes.This generalizes across task sets, as shown by comparable results in free-viewing and visual search tasks. An influence of peripheral vision on the control of viewing duration needs to be incorporated into models of eye movements during scene perception, such as CRISP (Nuthmann et al. 2010). Supported by DFG LA-2887/1