Comparing explicit and implicit ensemble perception: 3 stimulus variables and 3 presentation modes

Abstract

Visual scenes are too complex for one to immediately perceive all their details. As suggested by Gestalt psychologists, grouping similar scene elements and perceiving their summary statistics provides one shortcut for evaluating scene gist. Perceiving ensemble statistics overcomes processing, attention, and memory limits, facilitating higher-order scene understanding. Ensemble perception spans simple/complex dimensions (circle size, face emotion), including various statistics (mean, range), and inherently spans space and/or time, when sets are presented scattered across the visual scene, and/or sequentially in rapid series. Furthermore, ensemble perception occurs explicitly, when observers are asked to judge set mean, and also automatically/implicitly, when observers are engaged in an orthogonal task. We now study relationships among these ensemble-perception phenomena, testing explicit and implicit ensemble perception; for sets varying in circle size, line orientation, or disc brightness; and with spatial, temporal or spatio-temporal presentation. Following ensemble set presentation, observers were asked if a test image, or which of two test images, had been present in the set. Confirming previous results, responses reflected implicit mean perception, depending on test image distance from the mean, and on its being within or outside ensemble range. Subsequent experiments asked the same observers to explicitly judge whether test images were larger, more clockwise, or brighter than the set mean, or which of two test images was closer to the mean. Comparing implicit and explicit mean perception, we find that explicit ensemble averaging is more precise than implicit mean perception-for each ensemble variable and presentation mode. Implications are discussed regarding possible separate mechanisms for explicit versus implicit ensemble perception.

Keywords: Dual-task performance; Visual awareness; Visual perception.

Fig. 1

Top: Implicit and explicit ensemble perception tests. Rapid Serial Visual Presentation (RSVP) of a sequence of images differing in A circle size, B line orientation, or C disc brightness is followed by presentation of two test images (a). To test implicit ensemble perception, observers are asked which image was present in the sequence. Their responding according to the shorter distance from the sequence mean indicates implicit set mean perception. Explicit mean perception is tested by directly asking which of the two test images is closer to the mean. An alternative testing method (b) presents only a single test image and asks if it was present in the sequence (implicit mean perception) or if it is larger, more clockwise, or brighter than the set mean (explicit mean perception). Bottom: illustration of the other presentation modes—spatial (e.g., circle size) and spatiotemporal (e.g., line orientation). These were followed by the same 2 or 1 test image(s) as in (a). See text

Fig. 2

Examples of different implicit-2-test-images trial subtypes. In each case, the 8 sequence members are indicated, as well as their mean (M) and the two test images (SEEN and NEW). From the top: SEEN test image = sequence mean, expecting observers to correctly choose it; NEW = mean, expecting incorrect choice of NEW image; neither = mean, SEEN and NEW equidistant from mean, expecting ~50% chance performance; SEEN closer, better than 50%; NEW closer, less than 50%; NEW out of range, expecting easy rejection of NEW, and choice of SEEN, whether = mean or not. Note that performance accuracy is measured by choice of the SEEN test image and not by of the test image closer to the mean, though our central interest is the effect of mean perception on this choice

Fig. 3

Experiment 1—implicit, 2-test image paradigm—Membership task performance as a function of trial subtype for three testing variables (columns: circle size, line orientation and disc brightness) and for three presentation modes (rows: temporal, spatio-temporal, and spatial). In every case, accuracy (proportion reporting that the SEEN image was present in the set) was greater for trials where SEEN = mean than those where NEW = mean, with the baseline subtype (neither = mean) between them, close to 50% chance performance. Best membership task performance was for NEW outside the sequence range and easily rejected. Error bars are standard error of the mean (SEM). (Color figure online)

Fig. 4

Experiment 1—implicit, 2-test image paradigm—Performance for individual participants as a function of trial subtype for 3 presentation modes, averaging over test variables (top), for 3 test variables, averaging over presentation modes (middle), and averaging over all cases (bottom). Despite considerable scatter among participants, average membership task performance is clearly, and significantly dependent on trial subtype. Each circle corresponds to a single participant’s performance; horizontal lines correspond to the average performance over participants; error bars are SEM. (Color figure online)

Fig. 5

Experiment 1—implicit, 2-test image paradigm—Performance for individual participants as a function of which test image was closer to the mean, SEEN or NEW (excluding data for NEW out of set range). Performance, fraction choosing the SEEN image, was superior for almost all participants, in all conditions, when the SEEN image was closer to the mean, despite considerable scatter among participants. Each circle and line connecting performance for the two conditions, correspond to a single participant’s performance

Fig. 6

Experiment 1—implicit, 2-test image paradigm—Membership task performance as a function of parameter Δ for three test variables and three presentation modes. Graphs show data and best-fit sigmoid function, including data for NEW out of set range. Δ is the difference between absolute distances of NEW and SEEN images from the mean, with SEEN closer to mean on the right side of each graph, NEW closer on the left. Choice of test image closer to the mean reflects implicit ensemble perception. Red, orange, blue and gray data points reflect average performance for trials where, respectively, SEEN = mean, NEW = mean, neither = mean (baseline), and NEW is outside the ensemble range. Green curves are integral of data from Fig. 7. (Color figure online)

Fig. 7

Experiment 1—implicit, 2-test image paradigm—Fraction responding “member of set” as a function of distance of chosen test image, whether SEEN or NEW, from set mean. Data for 3 variables and 3 presentation modes, and averages over variables, presentations, and both. Each graph shows data and best-fit Gaussian function, including data of trials with NEW out of set range. Choice of the test image closer to the mean reflects implicit ensemble perception. Framed and nonframed circles correspond respectively to fraction of selecting the NEW or the SEEN image. Blue curves are derivative of black curves of Fig. 6, where x-axis is not distance from mean but difference of distances from mean (Δ). (Color figure online)

Fig. 8

Results for Experiment 1—Explicit, 2-test image paradigm. Sigmoid curves of mean estimation accuracy performance as function of the relative distance of the test images (target and distractor, closer and further) from the trial mean (i.e., Δmean). A gradual increase in task accuracy is seen as a function of the difference of the test image distances from the mean, for all stimulus variables and presentation modes

Fig. 9

Experiment 1—2-test images paradigm—Comparing implicit and explicit ensemble perception. Normalized data for each set variable and presentation mode, and their averages. Implicit test data (black) from Fig. 6: participants asked which test image was a member of the previously presented set; Δ is the difference between distances of NEW and SEEN images from the mean. Explicit test data (blue): participants asked to explicitly estimate set mean and judge which of 2 test images is closer to the set mean; normalized data from Fig. 8, showing normalized fraction of responding to the closer test image as a function of variable Δ, the difference in distances of the test images from the mean. The explicit (blue) sigmoid has a sharper slope, compared to that of the implicit (black) sigmoid. (Color figure online)

Fig. 10

Experiment 1—comparing implicit and explicit ensemble perception, 2-test images paradigm. Normalized data for each set variable and presentation mode, and their averages. Implicit test data (black) from graphs of Fig. 7: fraction of choosing the test image as a function of its distance from the mean. Explicit test data (blue): Gaussian curve derived by taking the derivative of the (blue) sigmoid curve of Fig. 9 as a function of the difference, Δ, of the test image distances from the mean. Note the narrower Gaussian explicit (blue) curve, compared to the width of the implicit (black) Gaussian. (Color figure online)

Fig. 11

Experiment 2—implicit, 1-test image paradigm—Fraction responding “member of set” as a function of distance of single test image from set mean. Columns: data for 3 variables: circle size, line orientation, disc brightness, and their average; Rows: data for 3 presentation modes, temporal, spatio-temporal spatial, and their average. Graphs show data and best-fit Gaussian function (black), including data for test images included (red) or excluded (orange) from the set, or out of set range (gray). Attributing membership on basis of test image proximity to set mean reflects implicit ensemble perception. Blue curves are derivatives of corresponding sigmoid curves (of the explicit task; Fig. 12). (Color figure online)

Fig. 12

Experiment 2—explicit, 1-test image paradigm—Participants were explicitly asked to estimate set mean and compare it to the test image. Fraction of responding “greater” (larger, more clockwise, brighter) as a function of test image size/orientation/brightness compared to set mean. Columns: data for 3 variables. Rows: data 3 presentation modes. Graphs show data and best-fit sigmoid (black), including data for test images included (red) or excluded (orange) from the set, or out of set range (gray). Green curves are integrals of corresponding Gaussian curves of the implicit task (Fig. 11). (Color figure online)

Fig. 13

Results for Experiment 2—implicit and explicit, 1-test image paradigm, temporal presentation, comparing explicit and implicit ensemble perception, and introducing results for naïve participants who performed the explicit tests without prior experience with the implicit tests. Plot shows performance dependence on distance of the test image from the set mean, for the 3 variables together. The 3 curves are the best fits for the explicit test, as in Fig. 12 (red), the explicit test of the naïve participants (orange) and the sigmoid curves (of Fig. 12) derived from the implicit data of Fig. 11 (green). The orange and red explicit curves are very similar; the green implicit curve is significantly shallower, confirming that the difference between explicit and implicit ensemble performance does not derive from prior experience with the implicit tests. (Color figure online)

Fig. 14

Summary of comparisons of results for different tests and measures, comparing Top: direct-result implicit Gaussian (purple) with derivative of either implicit (dashed purple) or explicit (orange) sigmoid; Bottom: direct-result implicit (purple) or explicit (orange) sigmoid curves, with curves derived by integral of implicit Gaussian (dashed purple). Averaged data over presentation modes and set variables, all normalized for comparison. Left: Experiment 1: 2-test image paradigm—Comparing different tests of implicit ensemble perception. Top: Gaussian curves from Fig. 7; Bottom: sigmoid curves from Fig. 6. Note good coincidence of original and derived curves. Center (Experiment 2: 1-test image paradigm) and Right (Experiment 1: 2-test image paradigm): Comparing implicit and explicit ensemble perception. Top: Implicit test: Gaussian curves (purple), fraction responding “member of set” as function of test image distance from set mean. Explicit test: Gaussian curves (orange) derived from sigmoid curves; from Figs. 10 and 11; Z-score normalization was done for all data. Bottom: Center: Experiment 2: 1-test image paradigm—Explicit test: sigmoid curve (orange), fraction responding “greater” (larger, more clockwise, brighter) as function of test image size/orientation/brightness compared to sequence mean; participants asked to compare test image to explicitly estimated set mean. Implicit test: sigmoid curve (purple) derived from Gaussian curve of implicit test; from Fig. 12. Right: Experiment 1: 2-test image paradigm—Explicit test: sigmoid curve (orange), fraction explicitly choosing test image closer to mean as function of difference of test images’ distances from sequence mean. Implicit test: sigmoid curve (purple) of implicit test; from Fig. 9. Note lack of coincidence of original and derived curves. Narrower Gaussians and steeper sigmoid slopes for explicit data indicate explicit ensemble perception is more precise. (Color figure online)