Subjects

Ten observers (5 females, ages 25–50) with normal or corrected-to-normal vision including two of the authors (YSB, THD) participated in the experiments. Most participants were experienced observers from previous MIB experiments.

The study and informed consent procedure were approved by the Weizmann Institute of Science Human Research Ethics Committee and a written informed consent was obtained from all participants.

Apparatus

Stimuli were displayed on a 19” CRT monitor controlled by dedicated software developed by the first author running on a Windows PC. The video format was true color (RGB), 100-Hz refresh rate, 1280×1024 pixels resolution, and 25°×20° field of view. Luminance values were gamma-corrected to obtain linear luminance response with a range of 6.6–110 cd/m². The viewing distance was 70 cm. All experiments were conducted in the dark.

Stimuli

Stimuli consisted of a single Gaussian luminance patch (the target, σ = 0.3^o) and a rotating mask (see example in Figure 1). The target was presented in the upper left quadrant (5.45^o of eccentricity) on a gray background (30 cd/m²) with one of seven different luminance contrasts, starting with 25% above background (peak luminance of 39.1 cd/m²) with increments of 0.15 log units. We use the term “contrast” to denote “luminance contrast” throughout the paper. The mask (when present) consisted of a square grid of 11×11 gray ‘+’ patterns (luminance of 13.3 cd/m², width of 0.8^o, spacing of 1.5 widths, total grid width of 12^o). The mask rotated at one of 3 speeds: static (0), slow (0.023 cycle/sec equal to 43s cycles) and fast (0.37 cycle/sec equal to 2.7 s cycles). Both speeds were chosen with pilot experiments by the authors that verified high percentage of disappearance with the fast speed mask, similar to previous studies, and a clearly visible movement of the slow mask. There was a small black circle (0.27^o diameter) as a fixation point. Both the fixation point and the target were surrounded by a “protection zone” of background space (diameter:1.8^o) to avoid local interaction with the mask.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Figure 1. Example of stimuli used to measure MIB for different luminance contrast levels of a peripheral Gaussian target.

Observers reported disappearance and reappearance of the Gaussian target by depressing buttons in 1 minute trials.

https://doi.org/10.1371/journal.pone.0092894.g001

Procedure

All experiments consisted of several trials (duration: 1 min) involving continuous viewing of one of the above-mentioned stimulus configurations, which remained constant throughout the trial. Observers were instructed to fixate the central dot and depress and release a button to report disappearance and reappearance respectively of the target. Trials were initiated by the observer with a minimum of 5 sec period between trials. Stimuli varied between trials in target contrast and mask type (different speeds and no mask).

We performed two experiments. In experiment 1, five observers were tested only on the comparison between MIB (at fixed mask speed) and Troxler (no mask), with various different target contrasts. Each session (lasting for about half an hour) consisted of 2 repetitions of each of 28 trial types (7 contrast levels×2 stimulus conditions [MIB, Troxler]). The trial types were presented in random order. Each observer performed 2–4 sessions on different days, making each result the average of 4–8 samples per observer per trial type.

In experiment 2, the other five observers were tested in two further stimulus conditions in addition to MIB and Troxler. The four stimulus conditions were: fast mask (MIB), slow mask, static mask, and no-mask (Troxler), which were tested in separate blocks of trials in random order. Each block had the same 7 levels of contrast randomly presented in 1 minute trials, making 28 trials per session (about 30 minutes). Observers participated in 2–8 sessions (average of 5 per observer, only one observer with 2 sessions); to avoid sampling bias, results were first averaged across sessions individually for each observer, and then averaged across observers with equal weighting. The two sets of experiments differed in the order of presentation (mixed vs. blocked) and in the two additional stimulus conditions (static and slow mask).

Data Analysis

Four parameters were computed for each trial: (1) percentage of total viewing time that the target was invisible (henceforth termed “percentage of time invisible”); (2) the average duration of target-invisible states (henceforth termed “mean invisible period”), (3) the average duration of target-visible states (henceforth termed “mean visible period”); (4) the number of disappearance events per minute (henceforth termed “transition rate”). These four parameters are inter-dependent, with two free parameters (i.e., two of the parameters are sufficient to derive the others, but one is insufficient). For example, the same percentage of time invisible can be produced by a few long invisible periods (at a low rate) or by many (high rate) short invisible periods.

Each of the four parameters was first averaged across trials, individually for each observer. A baseline value for each parameter and each observer was computed by averaging across contrast levels. The baseline values were subtracted from the values for each contrast to normalize the single-observer parameters. The normalized parameter values were then averaged across observers, and adjusted by adding the average (across observers) of the baseline values.

Statistical significance was verified by a two-way, repeated-measures within-observer analysis of variance (ANOVA). The two factors were contrast (7 levels) and mask condition (2 types in experiment 1, and 4 in experiment 2), both measured in the same observers.

Invisible and visible periods vs. transition rate

The effects of contrast on mean invisible period (Figure 2c), mean visible period (Figure 2a), and transition rate (Figure 2b) are related as expected by their definition. For example, if the typical visible period does not change with contrast while the typical invisible period decreases, then the rate must increase.

The contrast dependence of the mean invisible period was similar for MIB and Troxler (Figure 2c), suggesting that the time it takes for a target to reappear is determined by processes not affected by the moving mask, thus common to MIB and Troxler. This can be (at least in part) a consequence of reappearances being triggered by small fixational eye movements (e.g., microsaccades), which do not depend on mask presence and speed [9], [26].

Mean visible periods were roughly constant with contrast for MIB (Figure 2a) but increased for Troxler, suggesting that MIB may involve an additional process that is largely independent of target contrast. The constant level of visible periods found in MIB could not reflect a “floor effect”, since measures much below this level were obtained in the same experiment for the invisible duration, and since the perceptual effect itself did not saturate as reflected in the changes of rate and invisible periods. The additional contrast-invariant component in MIB might be related to competition between the neuronal populations in visual cortex processing the target and the moving mask [35], [36].

Relation to previous studies of MIB

In a previous study of MIB [1] the percentage of time invisible was found to increase with the luminance contrast of the target. Our current results (Figure 2d) show a different pattern, with a small decrease in the percentage of time invisible rather than increase. Among the main methodological differences we note the small eccentricity (1 deg) in the original study, which was not possible to repeat here to compare with a measurable Troxler effect (at 5.45^o). Other differences include the presentation of 3 targets instead of one, and smaller target patterns with sharp edges rather than the Gaussian blobs used here. We suspect that the difference stems primarily from the difference in eccentricity which modulates the relative weight of the two suggested components involved in the process: contrast adaptation and neural competition. This results in a difference in the effect of contrast with eccentricity.

Another study which tested the effect of contrast on MIB [20], compared MIB with perceptual filling-in of the type first reported by Ramachandran and Gregory [4]. For MIB, the percentage of time invisible was similar in range to the current results (40–60%), for a similar range of target luminance contrast levels (10-fold and more). In comparison to the moderate decrease of the percentage of time invisible with contrast we observed (Figure 2d), they found a slight increase for brighter targets as well as for very dim targets (1 cd/m² on a black background), possibly due to a difficulty in judging their visibility. In addition, they found the initial time-to-fade in MIB to be largely invariant to contrast (except from the very low luminance target) at about 3 sec, as compared to our moderate increase around 5 sec (Figure 4a). These moderate differences could be accounted for by our larger (∼3-fold) target size at roughly the same eccentricity, the different mask we used, as well as the larger variability in the Hsu et al data: standard errors 10-fold larger than our data, possibly due to a smaller sample (N = 5, half our sample size) and a larger inter-subject variability, which was minimized in our study by normalization (i.e., subtracting a baseline for each observer before averaging across observers), and by including observers experienced in MIB (see Methods, last paragraph).

More significant and striking is the discrepancy between the two studies in the static mask condition. While our static mask results were almost identical to the Troxler (no-mask) condition, showing decrease from 50% to almost zero invisibility with contrast, as well as increase from 5 sec to 20 sec in the initial time to fade, the Hsu et al study found around 30% invisibility and around 5 sec time-to-fade, largely independent of contrast. These differences are most likely due to the smaller target and the random dot mask in the Hsu et al study, which adds an additional component of perceptual filling-in. Hsu et al. [20] proposed that the non-intuitive contrast dependence (more MIB with higher contrast) is due to the effect of target-mask grouping: the more dissimilar the target is to the mask (e.g. differs in brightness), the less it groups with the mask, and the more it disappears. This explanation cannot account for the opposite pattern of results we obtained with a static mask and MIB (Figure 3b), and is inconsistent with the Bonneh et al. [1] results, where high luminance mask induced the strongest invisibility effect with high luminance targets. Given these discrepancies, the similarities between the effects of contrast in MIB and perceptual filling-in [20] cannot be used to make a strong claim about common mechanisms causing both effects. Instead, we propose that filling-in by a surrounding pattern (e.g. a proper static mask, but not Troxler) is an additional mechanism that affects disappearance. Taken together, the discrepancies between the current results and those from previous studies do not undermine our conclusions. The stimuli in the previous studies appear to have induced additional interactions beyond those processes revealed by our study.

The idea of two components involved in MIB was previously suggested by Gorea and Caetta [13] and is further supported by a detailed study of contrast detection and discrimination under invisible periods in MIB [37]. This study found both a sensitivity reduction indicative of inhibition and decisional criterion elevation indicative of an additional factor operating in MIB during invisibility [37]. In a following study, Gorea and Caetta [13] further specified these two components in terms of two gain modulation mechanisms: adaptation (of the neural population responding to the target), which causes a known response-gain change over time during continuous stimulation, and transient-to-sustained inhibition (of the response to the static target by the response to the moving mask), which causes a contrast-gain. These two components, together, can cause responses to the target to fall below perceptual threshold.

As discussed in the introduction, a simple implementation of this proposal implies a symmetric change of visible and invisible periods with changes in stimulus strength that alter these gains, for both MIB and Troxler fading. Our findings well match this pattern of results for Troxler but strongly deviate from it in MIB. This deviation is not related to adaptation over time, as we obtained a similar pattern of results when discarding the initial part of inspection (see Results), and thus should be explained by the dynamical properties of the proposed inhibition mechanism. In order to produce time-to-fade (or visibility period), which is (almost) invariant to contrast, one should assume inhibition which is proportional to contrast with a time course (to reach full inhibition effect) which is largely independent of contrast. Physiological evidence for contrast-dependent inhibition exists [38]. An alternative explanation, which is indicated by the similarity of the results to other types of bi-stable phenomena is related to competition and is discussed below.

Relation between MIB and other bi-stable visual phenomena

Our findings link MIB (and Troxler fading) to two other bi-stable visual phenomena, for which parametric manipulations of the strength of the competing stimulus components produced analogous effects on the dynamics of perception: binocular rivalry (BR), [28] and a rotating structure-from-motion (SFM) sphere [29]. The perceptual dynamics in both phenomena are thought to emerge from the interplay between local adaptation of stimulus-specific neuronal populations, competition between them, and noise [28], [30], [31],with additional local spatial interactions in BR, not present in MIB due to the “protection zones” surrounding the target (see Methods). The analogy can be summarized in terms of Levelt's propositions for binocular rivalry (BR) and the deviations from these propositions at extreme values of stimulus strength [28], [29]. Unlike in BR, where stimulus dominance is quantified as visibility of one of the two stimuli, in MIB and Troxler we quantify the two (asymmetric) dominance states as target visibility and invisibility.

For MIB, the visible and invisible periods match Levelt's proposition 2: strengthening only one stimulus (target contrast) does not affect the duration of the corresponding percept (target visible period), but shortens the dominance duration of the alternative percept (target invisible period). By contrast, Troxler fading does not match that proposition. Exactly the same deviation as in Troxler also occurs in BR, when the contrast of the other (fixed-contrast) image is close to detection threshold [28]. Further, the results of our experiment 2 (manipulation of the “strength”, i.e., speed, of the mask) revealed that Levelt's proposition 2 also holds for the mask in MIB. Within the target contrast range (twofold increase), the mask speed does not affect the duration of the corresponding percept (invisible period), but shortens the duration of the alternative percept (visible period). Again, this symmetry of the effects of target and mask strength is analogous to BR [28] and SFM [29]. Finally, the effects of mask speed and target strength on the transition rate are analogous to SFM [28], [29]. This is true with respect to both matches and mismatches with Levelt's proposition 4: strengthening both stimuli will increase the alternation rate. In particular, when the strength of one of the two stimulus components is close to zero (i.e., static or absent mask: Troxler fading), the transition rate decreases with the strength of the other stimulus component (i.e., target contrast, Figure 2a, Troxler fading). The same is evident in SFM (Figure 5 of Klink et al., 2008).

Taken together, this comparison of results establishes common principles underlying MIB on the one hand and BR and SFM on the other hand. Such common principles are often suggested (e.g., [2], but direct evidence is scarce. Most authors have pointed to the similar (asymmetric) shapes of the distributions of the perceptual state durations in these phenomena (e.g., [2]. Measuring the perceptual dynamics under a full parametric manipulation of the relevant stimulus parameters, like the ones performed here, is more sensitive to establish such links, and our results are consistent with those for BR and SFM.

Conclusion

In a detailed comparison of MIB and Troxler fading with respect to target contrast we found a lawful pattern of results with marked qualitative differences. Increasing target contrast increased (doubled) the rate of disappearance events in MIB, but shortened it to half in Troxler. On the other hand, MIB and Troxler did not differ in the mean invisibility periods which decreased with contrast. We interpret the results in terms of two processes involved in perceptual suppression: contrast adaptation, which is common to both MIB and Troxler, and neuronal competition, which is dominant in MIB, as well as in other bistable phenomena such as binocular rivalry.