Visual attention: the past 25 years

Abstract

This review focuses on covert attention and how it alters early vision. I explain why attention is considered a selective process, the constructs of covert attention, spatial endogenous and exogenous attention, and feature-based attention. I explain how in the last 25 years research on attention has characterized the effects of covert attention on spatial filters and how attention influences the selection of stimuli of interest. This review includes the effects of spatial attention on discriminability and appearance in tasks mediated by contrast sensitivity and spatial resolution; the effects of feature-based attention on basic visual processes, and a comparison of the effects of spatial and feature-based attention. The emphasis of this review is on psychophysical studies, but relevant electrophysiological and neuroimaging studies and models regarding how and where neuronal responses are modulated are also discussed.

Fig. 1

Number of articles on visual attention published as a function of year in (a) all scientific journals and (b) in Vision Research.

Fig. 2

Sequence of events in a trial (after the classic Posner paradigm). Observers perform a two-alternative forced-choice (2AFC) orientation discrimination task on a tilted target Gabor patch, which appears at one of four isoeccentric locations. The target is preceded by a central cue (instructing observers to deploy their endogenous attention to the upcoming target location), a peripheral cue (reflexively capturing attention to the upcoming target location), or a neutral cue (baseline). The cue in this example is valid. The timing (precue and interstimulus interval) for endogenous and exogenous conditions differs (along with their respective neutral conditions), in order to maximize the effectiveness of the cues.

Fig. 3

(a) Stimulus enhancement affects both the signal and the external noise in the input stimulus in the same way; thus, there is no retuning of the perceptual template entering the decision. (b) Stimulus enhancement has an effect at low- but not high-levels of external noise. (c) External noise exclusion reduces the effects of external noise through filtering, or retuning of the perceptual template that enters in the decision. (d) External noise exclusion improves performance in the region of high external noise, where there is external noise to exclude. (e) Internal noise reduction. Multiplicative noise increases with increasing contrast in the stimulus display. (f) Performance improves across all levels of external noise. [Reprinted from Lu et al., 2009].

Fig. 4

(a) Trial sequence illustrating an invalid cue: a central cue indicates a location; the stimulus and the response cue appear simultaneously indicating to the observer the target location. (b) Contrast threshold vs. external noise contrast data from a sample observer. Endogenous attention reduces contrast thresholds at high noise levels but not at low noise levels, suggesting tuning of the perceptual template to exclude external noise. [Reprinted from Dosher & Lu, 2000a, 2000b].

Fig. 5

(a) A trial sequence. Following a fixation point, a cue appears either above one of the two Gabor locations (peripheral cue) or at fixation (neutral cue). After an ISI, two Gabor stimuli are simultaneously presented (randomly oriented to the left or to the right) on the horizontal meridian. Then a response cue appeared at fixation to indicate the target Gabor for which the observer had to report the orientation. On one third of the trials the response cue points to a precued Gabor. On another third of the trials it points to the Gabor that was not precued. In the remaining trials the precue was presented in the center of the screen and the response cue was equally likely to indicate the Gabor to the right or to the left of fixation. (b) Examples of types of trials. In a valid trial the locations indicated by the peripheral cue and by the response cue match. In an invalid trial the locations indicated by the peripheral cue and by the response cue do not match. In a neutral trial the cue is presented at fixation and the response cue indicates the left Gabor in half of the trials and the right Gabor in the other half.

Fig. 6

Increment thresholds vs. pedestal contrast for two observers (rows). Filled symbols are data from the unattended condition; unfilled symbols are data from the attended condition. For judgments on luminance gratings (a and b) an attentional effect is visible from medium to high pedestal contrasts. For judgments on chromatic gratings (c and d), an attentional effect is visible only at high contrasts. These data are consistent with a response gain mechanism. [Reprinted from Morrone et al., 2004].

Fig. 7

Possible effects of attention on the contrast response function. (a) The left panel depicts a contrast gain model for attention. Contrast gain predicts an increase in sensitivity that is a function of stimulus intensity, and is characterized by a leftward threshold (C50) shift in the contrast response function. The dashed curve represents the signature curve shift brought about by attentional contrast gain; the shape of the function does not change but rather shifts leftward – boosting the effective contrast of the stimulus. (b) In the right panel, the dashed curve (attended) represents the effects of attention according to response gain models. Response gain predicts an increase in firing rate, which is characterized by a change in the shape of the curve – in slope and asymptote (R_max). C50, threshold; R_max, asymptote; n, slope; C, contrast level; N, attentional modulation; and M, response at lowest stimulus intensity.

Fig. 8

Response of a sample neuron from area V4 as a function of attention and stimulus contrast. (a) The contrast of the stimulus in the receptive filed increased from 5% (bottom panel) to 10% (middle panel) to 80% (top panel). The monkey had to detect a grating at the attended location. On each trial, attention was directed to either the location of the stimulus inside of the receptive field (solid line) or a location far away from the receptive field (dotted line). Attention reduced the contrast threshold to elicit a response (middle panel), but did not affect the response at saturation contrast (top panel). (b) Averages responses of V4 neurons while the monkey attends to the location (thick line) or away (thin line) of the receptive field (thin line). The horizontal line depicts the five different contrast values of the gratings presented inside the RF, which spanned the dynamic range of the neuron. The dashed and dotted lines show percentage and absolute difference in firing rate, respectively, across the two attention conditions, as a function of contrast. [Adapted from Reynolds et al., 2000].

Fig. 9

Effects of exogenous and endogenous attention on performance (d′) as a function of contrast. (a and b) Large stimulus with small attention field. (c and d) Small stimulus with large attention field. Exogenous attention is shown in (a and c). Endogenous attention is shown in b and d. Shown are plots of psychometric functions for each attentional condition (valid, neutral and invalid pre-cues) and parameter estimates (c50, contrast yielding half-maximum performance; d_max ′, asymptotic performance at high contrast). Each data point represents the mean across 4 observers. Error bars on data points are ±i.e. Error bars on parameter estimates are confidence intervals obtained by bootstrapping. [Reprinted from Herrmann et al., 2010].

Fig. 10

(a) Average gap-size thresholds (75% localization accuracy) for both exogenous (top-left panel) and endogenous (top-right panel) attention for the cued, neutral, and uncued conditions. (b) The bottom panels depict the average percent change in acuity thresholds at cued and uncued locations as compared to the neutral condition for exogenous (left) and endogenous (right) attention. Negative values indicate a cost in acuity, whereas positive values indicate a benefit. Error bars show ±1 SE. [Reprinted from Montagna et al., 2009].

Fig. 11

Observers’ performance as a function of target eccentricity and cueing condition for the three viewing distances. Viewing distance is indicated at the top of each panel. Because viewing distance varied, the eccentricity values (abscissa) differ in the three panels [Adapted from Yeshurun & Carrasco, 1998].

Fig. 12

Proportion correct as a function of the target-flankers distance. The vertical lines indicate the critical distance (90% of asymptotic value) for cued and neutral conditions [Adapted from Yeshurun & Rashal, 2010].

Fig. 13

Size changes of the receptive field center with attention. (a) Receptive field maps of a representative cell when a task was done at the fixation point (left graph; fixation is indicated by the white square) and when a position inside the RF was attended (right graph; the attended location is indicated by the white circle). The RF area, outlined in white, is clearly reduced with attention inside the RF. Colors represent spiking activity in Hz. (b) Plots the distribution of RF size changes (in% of the RF diameter in the fixation condition) for all 142 cases in which attention was directed into the RF [Adapted from Anton-Erxleben et al., 2009].

Fig. 14

Predictions of the ARF model. A broken line may be perceived as solid when unattended (large ARFs—left panel), but as broken when attended (small ARFs—right panel). [Reprinted from Shalev & Tsal, 2002].

Fig. 15

(a) Time course of the event-related magnetic field (ERMF, difference between corresponding efflux- and influx-field maxima) response for each probe distance; PD0: location was probed; PD1: location adjacent to the target was probed; PD2-PD4: non-adjacent locations to the target were probed (b) Mean size of the probe-related response between 130 and 150 ms, collapsed across corresponding probe-distance conditions. The size of the effect represents the average of the ERMF difference between the observers’ individual field maxima and minima. [Adapted from Hopf et al., 2006].

Fig. 16

Motion aftereffect directions obtained from pairing eight attentional motion directions with the adapting motion. Data from active (attended) and passive trials are plotted separately. The amplitudes of the best-fitting sinusoids are 24.27° and 4.63°, respectively. Data points show the averages for four observers; error bars indicate ±1 SEM [Adapted from Alais & Blake, 1999].

Fig. 17

Feature-based attention in MT. (a) Schematic representation of tasks used to assess the effects of attention to direction of motion. Two patches of random dots were presented, one within the RF of the neurons being record (broken white line). On some trials (top), a cue at the beginning instructed the animal to pay attention to the motion of the patch outside the receptive field to detect a change in that motion. On other trials (bottom) the attention of the animal (gray arrows) was directed to the fixation spot to detect a change in luminance. (b) Responses of a representative MT neuron to different directions of motion during the two attentional conditions. Attention to the preferred direction of motion increased the response of the neuron, but attention to the null direction of motion decreased its response. [Reproduced from Martinez-Trujillo & Treue, 2004, as printed in Maunsell & Treue, 2006].

Fig. 18

(a) Classification accuracy based on responses in regions-of-interest (ROIs) contralateral to the focus of spatial attention (spatial attention inside RF). (b) Classification accuracy based on responses to an ignored stimulus (spatial attention outside RF). (c) Classification accuracy based on the responses to an unstimulated region of space (baseline activity). [Adapted from Serences & Boynton, 2007].

Fig. 19

(a) Attention induced MAE magnitude for each adapter-test pair. The icons below the x-axis depict the locations of the adapter and test stimuli (solid circles: adapter, empty circles: test). There were 4 possible locations on the intercardinal axes at 10 deg of eccentricity. Dashed vertical line separates results when the adapter was in the upper right quadrant and in the lower right quadrant. (b) Attention induced MAE magnitude averaged across adapter locations, near SH: near location, same hemifield (1st and 4th columns in a), near DH: near location, different hemifield (2nd and 5th columns in a), far DH: far location, different hemifield (3rd and 6th columns in a). [Adapted from Liu & Mance, 2011].

Fig. 20

Psychophysical data revealing how subjects deploy attentional gain (a) Normalized contrast detection thresholds when observers were engaged in a fine discrimination task. The x-axis labels refer to orientation offset of the to-be-detected Gabor from the target orientation. Positive values along the x-axis refer to rotation in the direction indicated by the color of the cue, and negative values refer to rotation in the direction opposite of that indicated by the cue. For example, if a red cue indicated that targets were rotated clockwise with respect to distractors, then all distractors rotated clockwise from the target would be denoted with a positive value and all distractors rotated counterclockwise would be denoted with a negative value. Note that since there is only one distractor orientation, positive rotational offsets denote exaggerated target features and negative offsets denote the distractor feature (−5° from the target) and exaggerated distractor features. (b) Normalized contrast detection thresholds when observers were engaged in a coarse discrimination task. All error bars are ±1 SEM. [Adapted from Scolari & Serences, 2009].

Fig. 21

The effect of gain and tuning on neural population responses and equivalent-noise curves. (a) A hypothetical population response to an attended upwards-moving stimulus. Dotted lines correspond to changes with attention. A gain model proposes an overall multiplicative increase in the population response to a stimulus (left panel). This amplified response would only lead to a benefit in discriminability at low levels of external noise (middle panel). This is the pattern of responses obtained for spatial attention (right panel). (b) A tuning model proposes a sharpening of the population response around the attended stimulus feature (left panel). This narrowed response would only lead to a benefit in discriminability at high levels of external noise, when there is noise to suppress (middle panel). Feature based attention leads to noise reduction across all external noise levels. Indicating both gain and tuning [Adapted from Ling et al., 2009].