Finding flicker: critical differences in temporal frequency capture attention

Abstract

Rapid visual flicker is known to capture attention. Here we show slow flicker can also capture attention under reciprocal temporal conditions. Observers searched for a target line (vertical or horizontal) among tilted distractors. Distractor lines were surrounded by luminance modulating annuli, all flickering sinusoidally at 1.3 or 12.1 Hz, while the target's annulus flickered at frequencies within this range. Search times improved with increasing target/distractor frequency differences. For target-distractor frequency separations >5 Hz reaction times were minimal with high-frequency targets correctly identified more rapidly than low frequency targets (~400 ms). Critically, however, at these optimal frequency separations search times for low and high-frequency targets were unaffected by set size (slow flicker popped out from high flicker, and vice versa), indicating parallel and symmetric search performance when searching for high or low frequency targets. In a "cost" experiment using 1.3 and 12.1 Hz flicker, the unique flickering annulus sometimes surrounded a distractor and, on other trials, surrounded the target. When centered on a distractor, the unique frequency produced a clear and symmetrical search cost. Together, these symmetric pop-out and search costs demonstrate that temporal frequency is a pre-attentive visual feature capable of capturing attention, and that it is relative rather than absolute frequencies that are critical. The shape of the search functions strongly suggest that early visual temporal frequency filters underlie these effects.

Keywords: attention; capture; cost; flicker; pop-out; temporal frequency; visual channels; visual search.

Figure 1

Illustration of an example search display. In Experiment 1 all distractor annuli modulated with an identical frequency within a given trial (1.3 or 12.1 Hz) whilst target annuli modulated at either 1.3, 2.7, 4.0, 6.7, 8.1, or 12.1 Hz.

Figure 2

Results from Experiment 1. Data points show reaction times for correct target identification plotted as a function of target frequency, for three different set sizes. Continuous lines show the best-fitting Gaussian functions (see Eq. 1). Error bars for individual subjects depict SE of all correct trials. Average data points represent between-subject means. Error bars for averaged data represent SE of individual means. Gaussian fits of averaged data were calculated independently of fits for individual subjects.

Figure 3

Comparison of psychophysical temporal frequency channels derived using a classic overlay masking paradigm (dashed curves) and Gaussian-fitted average search times (Experiment 1: set size = 11 items) resulting from differences in target–distractor flicker rate (solid curves).

Figure 4

Mean reaction times for correctly identifying the orientation of the target element (horizontal vs. vertical) when presented under the various temporal contexts in Experiment 2 (set size = 7 items). Error bars represent SE of inter-subject means (n = 4).