Conceived and designed the experiments: AN BU MT MN JW AY. Performed the experiments: AN MT. Analyzed the data: AN BU MT. Wrote the paper: AN BU.
The authors have declared that no competing interests exist.
Observers misperceive the location of points within a scene as compressed towards the goal of a saccade. However, recent studies suggest that saccadic compression does not occur for discrete elements such as dots when they are perceived as unified objects like a rectangle.
We investigated the magnitude of horizontal vs. vertical compression for Kanizsa figure (a collection of discrete elements unified into single perceptual objects by illusory contours) and control rectangle figures. Participants were presented with Kanizsa and control figures and had to decide whether the horizontal or vertical length of stimulus was longer using the two-alternative force choice method. Our findings show that large but not small Kanizsa figures are perceived as compressed, that such compression is large in the horizontal dimension and small or nil in the vertical dimension. In contrast to recent findings, we found no saccadic compression for control rectangles.
Our data suggest that compression of Kanizsa figure has been overestimated in previous research due to methodological artifacts, and highlight the importance of studying perceptual phenomena by multiple methods.
Several lines of evidence demonstrate that observers misperceive the location of points within a visual scene presented at about the time of a saccade
Contrary to Morrone et al.'s
However, we can induce perception of discrete elements as a single object not only through element proximity but also by other means. To illustrate, the Gestalt principle of closure suggests that we perceive familiar figures with gaps (i.e., constellation of discrete elements) as complete figures or objects. Perhaps one of the best-known illustrations of closure is the Kanizsa figure
Sogo and Osaka
Bars indicated the averaged magnitude of the saccadic compression for all conditions: Disks, Packman, Illusory contour, Real contour, and Filled figures. Error bars indicated standard deviation of the compression among their three participants. The figure highlights that Sogo and Osaka observed substantial horizontal saccadic compression even for the control figures: the real contour and filled contour conditions.
We used Kanizsa figure to gain insight into the nature of processes involved in saccadic compression. More specifically, our aims were to obtain unequivocal evidence, free of interpretive difficulties, showing whether saccadic compression occurs with Kanizsa figure stimuli, to determine at which stage of processing such compression occurs, to investigate whether saccadic compression is dependent on proximity or closure. As a starting point of our design and based on extensive prior evidence, we took for granted that visual space is not extended just prior to saccadic movement, neither in the horizontal nor vertical dimension (e.g.,
In Experiment 1a, we investigated horizontal saccadic compression for Kanizsa and control rectangle figures presented in two different sizes (small, large). Participants were presented with Kanizsa and control rectangles and had to decide whether the horizontal length of each stimulus was longer using the two-alternative force choice method. To control for any possible perceptual biases and to assess participants' ability to estimate relative length of horizontal vs. vertical dimension of critical figures (Kanizsa, rectangle), participants were also required to make the two-alternative force choice judgments about the stimuli while their eyes remained stationary (fixation condition). Relative to the fixation condition, Experiment 1a revealed horizontal saccadic compression for large Kanizsa figures but not for large rectangle stimuli, small Kanizsa stimuli, or small rectangle stimuli.
To find out whether the illusory contours are responsible for preventing saccadic compression of the small Kanizsa figure, we repeated Experiment 1a with three different small figures: rectangle, Kanizsa, and disk-rectangle. The disk-rectangle was similar to the Kanizsa figure except that the Pac-men – the illusory shape inducers – were replaced by disks (filled circles). If illusory contours are critical for prevention of saccadic compression, then saccadic compression would occur in the disk-rectangle but not in the Kanizsa conditions. In contrast, if proximity of the figure elements is critical, then saccadic compression would not occur in either condition. Experiment 1b revealed no horizontal saccadic compression for any of the small figures.
As noted in the introduction, Kaiser and Lappe
Four women and nine men (mean age 24.3 years, range 21 to 24 years) participated in Experiment 1a. One participant was an author of the present study but other participants did not know the purpose of the experiment. The design had three within-subject factors: stimulus type (Kanizsa, rectangle), size (small, large), and eye movement condition (saccade, fixation). The experiment was approved by Research Ethics Committee of the Kwansei Gakuin University and all participants signed written informed consent form prior to participating in the experiment.
Panel A shows examples of Kanizsa and rectangle stimuli. Panel B shows sequence of events during each trial.
Participants were seated with their heads stabilized by a chin rest in a dimmed room. All stimuli were presented on a 21-inch CRT monitor (Calix, TOTOKU) using a 1152×864 resolution and 120 Hz refresh rate color mode. The distance between participants' eyes and the surface of the monitor was fixed to 32 cm. Eyelink eye tracker (SR research Ltd.) sampled horizontal dominant eye movement at 250 Hz with a resolution of less than 0.5 degree. Participants pressed one of the two keys on a keyboard placed in front of them to register their responses. Presentation of stimuli, collection of eye movements and registration of participants' responses was controlled by a Macintosh G4 (Apple) computer running customized Matlab based software (Mathworks) and Psychophysics and Eyelink toolboxes (
On each trial of the saccade condition (see
To assess participants' ability to estimate the relative length of horizontal vs. vertical dimension of critical figures (rectangle, Kanizsa) while their eyes remained stationary, participants were also given fixation trials where they were only required to fixate on the fixation point without making a saccade. While they fixated on the fixation point, a critical figure was flashed on the screen at the same location as that used for the saccade condition. All other aspects of the fixation and saccade trials were the same.
If the participants' eye position was out of the 2-degrees square window surrounding the fixation point before the fixation disappeared, a warning beep was sounded to alert participants that they failed to maintain fixation; the fixation point was replaced by a black screen, the trial was discarded, and the next trial commenced. Discarded trials were re-inserted into a random position within the remaining sequence of trials. For saccade trials, we collected data until there were at least 12 valid trials for each condition (i.e., figure type x figure size x width level) that met the following criteria: saccade latency over 130 ms; saccade peak velocity less than 800 ms/s; saccade amplitude more than 13 degrees; and stimulus onset from saccade onset between −50 to −10 ms. We, therefore, avoided influences of the phosphor persistence on the monitor because even though the stimulus was presented at −10 ms before the saccade, the phosphor persistence finished before retinal images smeared by the saccade. For fixation only trials, we collected data until there were at least 12 valid trials for each condition.
The saccade and fixation trials were blocked and the order of the blocks was counterbalanced across participants. Within the blocks, the sequence of stimuli (i.e., type of figure, width level) was randomized individually for each participant and participants were given a chance to rest after each 48 trials (both valid and invalid). The experiment lasted about 3 hours including rests.
Eye movements were analyzed on-line to determine the saccade onset, saccade velocity, saccade amplitude, timing of the critical stimuli relative to the saccade onset (determined by the average saccade onset during the preceding 5 trials). The saccade onset was defined as the time when the angular velocity of the horizontal saccade exceeded 40 deg/s.
We used bootstrapping to fit psychometric curves to each participant's data and to determine individual 50% perceptual subjective equality or PSE thresholds with 95% confidence interval limits and we used maximum likelihood estimation (MLE) to fit psychometric curves to data averaged across all participants
Three women and eight men (mean age 22.4 years, range 21 to 24 years) with normal or corrected-to-normal visual acuity participated in Experiment 1b. The experiment design, materials, and procedure were the same as in Experiment 1a with the following exceptions: three figures – rectangle, Kanizsa, and disk-rectangle – were used as critical stimuli and all were presented in small size only. The diameter of the disks was the same as that of the Kanizsa figure inducers. The experiment was approved by Research Ethics Committee of the Kwansei Gakuin University and all participants signed written informed consent form prior to participating in the experiment.
Three women and five men (mean age 22.9 years, range 21 to 25 years) with normal or corrected-to-normal visual acuity participated in Experiment 2. The experimental design, materials, and procedure were the same as in Experiment 1a with the following exceptions: in the small size condition, the width of stimuli was fixed to 7 degrees and height varied from 5.0 to 12.5 degrees in six levels (5.0 6.5 8.0 9.0 11.0 12.5) whereas in the large size condition, the width of stimuli was fixed to 10 degrees and the height varied from 7.5 to 16.0 degrees in six levels (7.5 9.0 10.5 12.0 14.0 16.0). The experiment was approved by Research Ethics Committee of the Kwansei Gakuin University and all participants signed written informed consent form prior to participating in the experiment.
Vertical bars indicate SEMs.
Vertical bars indicate 95% within subject confidence intervals
The ANOVA over individual 50% threshold data with stimulus type (rectangle, Kanizsa), stimulus size (small, large), and eye movement (saccadic, fixation) as within subjects factors revealed significant main effects of stimulus type, F (1,12) = 53.78, MSe = 1.18, p<0.001, stimulus size, F(1,12) = 782.78, MSe = 0.49, p<0.001, eye movement, F(1,12) = 7.71, MSe = 0.99, p = 0.017, significant eye movement x stimulus size interaction, F(1,12) = 9.24, MSe = 0.26, p = 0.010, and significant three way eye movement x stimulus size x stimulus type interaction, F(1,12) = 7.69, MSe = 0.39, p = 017. No other affects approached significance. Follow up ANOVAs for small stimuli with stimulus type and eye movement as withing subjects factor revealed no significant effects or interaction except significant effect of stimulus type, F(1,12) = 66.97, MSe = 0.38, p<0.001. In contrast, Follow up ANOVAs for large stimuli with stimulus type and eye movement as withing subjects factor revealed significant stimulus type effect, F(1,12) = 30.93, MSe = 1.26, p<0.001, eye movement effect, F(1,12) = 16.56, MSe = 0.57, p = 0.002, and stimulus type by eye movement interaction, F(1,12) = 6.69, MSe = 0.71, p = 0.024. Whereas eye movement effect (saccade vs. fixation) was significant for Kanizsa figure, t(12) = 3.88, p<0.01, it was not significant for rectangle figure, t(12) = 1.03, p = 0.32.
To determine whether the magnitude of saccadic compression increases when the Kanizsa figure is presented closer to the saccade onset, we examined the relationship between the probability that the width of the stimulus was perceived as longer than its height as a function of time between the figure onset and the saccade onset. For this purpose, we classified all trials in the saccade conditions into one of the 3 time bins – −60 to −40 ms, −40 to −20 ms, and −20 to 0 ms – depending on time elapsed between the figure and saccade onsets, calculated means for each participant and each time bin, and then calculated mean values across all participants for each time bin. Next, we used MLE methods to to find out the amount of compression in degrees for each time bin for all conditions (large Kanizsa, large rectangle, small Kanizsa, small rectangle) in the saccade condition relative to the fixation condition.
Several alternatives can be offered to explain why saccadic compression occurred only for the large Kanizsa figure. According to one class of explanations, the illusory contour induced by Kanizsa figure may prevent saccadic compression but the speed of illusory contour completion depends on the distance between the inducers allowing the re-mapping processes, running in parallel, to remap the visual space in the absence of illusory contours
Vertical bars indicate SEMs.
Vertical bars indicate 95% within subject confidence intervals
The finding that saccadic compression did not occur with either small Kanizsa or with small disk-rectangle stimuli suggests that the absence of saccadic compression for small Kanizsa figure found in Experiment 1a was not due to illusory contours but simply due to the close proximity of individual elements forming the figure. In turn, these findings suggest that element proximity determines whether saccadic compression is observed.
Vertical bars indicate SEMs.
Vertical bars indicate 95% within subject confidence intervals
The ANOVA over individual 50% threshold data with stimulus type (rectangle, Kanizsa), stimulus size (small, large), and eye movement (saccadic, fixation) as within subjects factors. The ANOVA revealed no significant effects nor interaction except the main effect of stimulus size (large, small), F(1,12) = 1069.01, MSe = 0.18, p<0.001.
We also directly compared the size of the saccadic compression for the large Kanizsa figures in Experiment 1 vs. 2 and expected to obtain significant 3-way interaction between the experiment, figure, and saccadic condition. Consistent with this expectation, the ANOVA over individual 50% threshold data with the experiment (1A and 2) as between subjects factor and the figure (rectangle, Kanizsa) and saccadic condition (fixation, saccade) revealed a significant 3-way interaction between the experiment, figure, and saccadic condition, F(1,22) = 5.72, Mse = 2.65, p<0.05.
The finding that participants perceived the vertical sizes of figures as larger in Experiment 1 but not in Experiment 2 in both the fixation and saccade conditions is consistent with prior research
The null effect of the fixation vs. saccade condition could be due to (1) no horizontal and no vertical compression or (2) both vertical and horizontal compression occuring simultaneously. However, the data from Experiment 1 make the second interpretation extremely unlikely. Experiment 1 revealed that the saccadic compression occurred only for large Kanizsa (10 degrees in height) and that the size of the compression increased with the horizontal width, that is, the compression increased as the horizontal distance between the inducers increased from 5.5 to 16 degrees. Specifically, no horizontal compression occurred when the horizontal width was smaller than 10 to 11 degrees and the horizontal compression was the largest when the horizontal distance between the inducers was large, 14 to 16 degrees. Given that the horizontal distance between the inducers in Experiment 2 was only 10 degrees, we can assume that no horizontal compression occurred in Experiment 2. In turn, given that there were no differences between the fixation and saccade conditions and the vertical distances between the inducers ranged up to 16 degrees in Experiment 2, the data indicate that the absence of the null effect of the fixation vs. saccade condition indicates that there was no vertical compression. Moreover, this interpretation is consistent with the findings of Kaiser and Lappe
In combination, these results indicate that the vertical compression does not occur even with the presentation of the complex stimuli such as rectangles and Kanizsa figures.
Our findings unequivocally show that large but not small Kanizsa figures are perceived as compressed in the horizontal dimension and that such compression is stronger when the Kanizsa figure is presented closer to the saccade onset. In contrast, small Kanizsa figures and rectangle figures, regardless of size, are perceived as undistorted with no horizontal compression. Finally, we found no evidence of saccadic compression for our control stimuli, rectangles, in either the horizontal or vertical dimension.
Thus, using different methodology, we have obtained results consistent with Sogo and Osaka's
In our experiments, participants make decisions about the ratio of horizontal vs vertical dimensions in Experiment 1 and the ratio of vertical vs. horizontal dimensions in Experiment 2, while we varied the horizontal and vertical dimensions of stimuli in Experiments 1 and 2, respectively. Accordingly, someone may argue that our finding of horizontal compression in Experiment 1 but not in Experiment 2 could be due to a bias effect rather than to saccadic compression. In any case, more studies are needed to examine whether horizontal vs. vertical compression occurs for different stimuli and/or stimuli with different ranges of dimensions (i.e., larger or smaller than in our experiments). However, we believe our findings reflect true saccadic compression in Experiment 1 for the following reasons. First, any perceptual biases (for example, horizontal vs. vertical illusion effects) not related to saccadic movement are controlled by our fixation conditions. Second, to explain the findings, any such saccade related bias would need to operate only on large Kanizsa figures but not on control rectangle figures or small Kanizsa figures that were not subject to any saccadic compression. Third, one may argue that the results could be due to vertical extension rather than horizontal compression of Kanizsa figure. While this is a possibility, it is unlikely because no one has ever reported vertical nor horizontal saccadic extension in prior research. Fourth, our findings are consistent with Sogo and Osaka's
Our findings showing no horizontal compression for our control figures – rectangles – is consistent with the finding by Matsumiya and Uchikawa
One explanation for these contrasting findings is that the design features of Sogo and Osaka's method are responsible for the unexpected horizontal compression of the control rectangle figures in their experiment. First, as pointed out in the introduction, Sogo and Osaka's data may underestimate or overestimate the degree of horizontal compression depending on the particular task strategy participants choose on any given trial – adjusting the length of the probe to the
We thank Amy L. Siegenthaler for comments on the manuscript.