Object perception: When our brain is impressed but we do not notice it

Abstract

Although our eyes receive incomplete and ambiguous information, our perceptual system is usually able to successfully construct a stable representation of the world. In the case of ambiguous figures, however, perception is unstable, spontaneously alternating between equally possible outcomes. The present study compared EEG responses to ambiguous figures and their unambiguous variants. We found that slight figural changes, which turn ambiguous figures into unambiguous ones, lead to a dramatic difference in an ERP (“event-related potential”) component at around 400 ms. This result was obtained across two different categories of figures, namely the geometric Necker cube stimulus and the semantic Old/Young Woman face stimulus. Our results fit well into the Bayesian inference concept, which models the evaluation of a perceptual interpretation's reliability for subsequent action planning. This process seems to be unconscious and the late EEG signature may be a correlate of the outcome.

Introduction

What do you see in Figure 1: An old or a young woman? Prolonged inspection suggests both—but not simultaneously. When we observe an ambiguous figure, like the “Old/Young Woman” (Boring, 1930), our percept is unstable, changing spontaneously between two or more possible interpretations. The information that enters via our eyes is very often ambiguous, not least due to the projection of the 3D world onto our 2D retinae. Visual perception can thus be viewed as an incessant attempt to solve the visual “inverse problem” of how several perceptual interpretations can result from one and the same retinal image. Given this uncertainty, how is the brain at all able to construct a stable percept of the world, reliable enough for us to act successfully in the world and survive in the long run? There is a general consensus regarding the earliest steps of visual perception, such as the visual areas' retinotopic and columnar organization (Holmes & Lister, 1916; Schira, Wade, & Tyler, 2007), but a sizable gap remains if we approach object representation and consciousness. It is vividly discussed whether an object is represented by a small sample of grandmother-like neurons (e.g., Barlow, 1972; Quiroga, Reddy, Kreiman, Koch, & Fried, 2005) or by population codes of distributed neural assemblies (Pouget, Dayan, & Zemel, 2000). Ambiguous figures, like Boring's Old/Young Woman (Figure 1, Boring, 1930) or the Necker cube (Figure 2b, Necker, 1832) are found in the chapters on conscious perception of nearly every neuroscience textbook. Many authors regard those figures as key stimuli with which to address the topics of object representation and conscious perception, since they enable us to separate neural activity related to the visual input from activity related to the perceptual outcome (e.g., Blake & Logothetis, 2002). 

Figure 1

 

Old/Young Woman (Boring, 1930): This face profile can be seen as an old or a young woman looking to the right. The pronounced nose of the old woman can also be interpreted as the chin of a young girl. The wart on top of the old woman's nose can be seen as the nose of the young girl. And the old woman's eye can be interpreted as the young woman's right ear.

However, after research over more than 200 years (Long & Toppino, 2004), the mechanisms underlying spontaneous perceptual reversals are still poorly understood (Blake & Logothetis, 2002). 

Recent physiological studies concentrated on the question of when and how perceptual reversals take place. Several authors thus investigated the time intervals, close to perceptual reversals (e.g., Başar-Eroglu, Strüber, Schürmann, Stadler, & Başar, 1996; Kaernbach, Schröger, Jacobsen, & Roeber, 1999; Kornmeier & Bach, 2004, 2006; O'Donnell, Hendler, & Squires, 1988; Pitts, Nerger, & Davis, 2007; Roeber et al., 2008; Strüber & Herrmann, 2002). In the present study, however, we examined how the processing of visual objects, resulting in enduring stable percepts, differs in general from object processing, which in turn leads to unstable percepts that spontaneously break down time and time again. What is the difference between the neural processing and representation of ambiguous and unambiguous stimuli? One surprising result is this: Although the figural changes rendering ambiguous figures unambiguous (Figure 2b) and enabling observers to identify them with >94% accuracy are tiny (Figures 2c and 2d), certain EEG correlates of the perception of these stimuli differ dramatically, by up to factor 4. Apparently the huge differences in neural activity underlying these EEG results are not consciously noticed. This finding occurred with stimuli from two completely different categories, namely the geometric Necker cube and the semantic Old/Young Woman. 

Figure 2

 

(a) Paradigm: EEG was recorded while participants compared successive briefly presented stimuli (S _i) and indicated in a go/nogo task in one condition perceptual changes and in a separate condition perceptual stability (two consecutive stimuli perceived as identical). In the case of ambiguous stimuli, perceptual changes had an endogenous origin inside the brain, while in the case of the unambiguous variants changes were induced exogenously. Lattices and faces were presented in separate experiments with different participants. In each experiment ambiguous stimuli (L1 or F1) and unambiguous variants (L2 and L3 or F2 and F3) were presented in separate experimental blocks. (b) Stimuli: L1: Necker lattice; L2, L3: Unambiguous variants. F1: Ambiguous Old/Young Woman; F2, F3: Unambiguous variants. (c) Normalized magnitude plots (spectra) of a 2-dimensional Fourier analysis of the stimulus pictures directly above in row (b). To enhance small details, the gray level represents the square root of the magnitude in spatial frequency space. The spectra of the lattice and the face clearly differ. For instance in the lattice the regular grid leads to equidistant-spaced harmonics in the spectrum. The difference between ambiguous and non-ambiguous, while obvious in row (b) is hard to detect in the spectra. This suggests that the ERP findings cannot be explained by low-level image statistics. (d) Total duration that the specific stimulus from above (b) was perceived relative to the total presentation time of that stimulus. The relative percentages indicate that the ambiguous versions are close at the 50:50% ratio and that small figural changes render the stimuli almost completely unambiguous.

Methods

Participants

Nine women and five men aged from 20 to 27 with a median age of 24 years took part in the lattice experiment. Eight women and five men aged from 21 to 30 with a median age of 25 years took part in the face experiment. All participants were naive as to the specific experimental question and gave their informed written consent. All of them had normal or corrected-to-normal visual acuity. The study was performed in accordance with the ethical standards laid down in the Declaration of Helsinki (World Medical Association, 2000) and was approved by the local ethics review board. 

Stimuli

The geometric stimuli were a “Necker lattice” ( Figure 2b, L1), consisting of 3 × 3 Necker cubes (Kornmeier & Bach, 2004) and two unambiguous variants with depth cues (shading, central projection, and aerial perspective, OpenGL lighting model, Woo, Neider, & Davis, 1998). The semantic stimuli were a line drawing of an ambiguous face profile (Figure 2b, F1; Gale & Findlay, 1983), a variant of Borings “Old/Young Woman” (Boring, 1930). We added two unambiguous variants ( Figure 2b, F2 and F3). The lattice stimuli were presented at a viewing angle of 7.5° × 7.5°. The face stimuli were presented at a viewing angle of 7.5° × 12.5°. The stimuli were created with a Macintosh G4 computer and displayed on a Philips GD 402 monochrome CRT screen at a refresh rate of 75 Hz. 

Procedure

The stimuli were presented discontinuously with 800-ms presentation time ± 91-ms uniformly distributed random time jitter to prevent potential habituation effects ( Figure 2a). Successive stimulus presentations were intermitted by 27-ms inter-stimulus intervals with a blank screen. EEG was recorded while the participants compared successive stimuli (S _i). They indicated in go/nogo-tasks in the reversal condition perceived orientation reversals in the two opposite directions (e.g., from lattice front side bottom right to front side top left or vice versa) with different hands pressing different keys. In a separate stability condition they indicated stability of each of the two possible percepts (either two consecutive lattices perceived with their front side as bottom right or as top left) with different hands pressing different keys. In the case of ambiguous stimuli, perceptual changes had an endogenous origin inside the brain, while in the cases of the unambiguous variants the changes were induced exogenously in a random order with a probability of P = 0.5. 

The geometric lattice stimuli ( Figure 2b, L1–L3) and the face stimuli ( Figure 2b, F1–F3) were presented in separate experiments with the same experimental protocol but with different participants. Each experiment (lattice and face) consisted thus of four conditions, namely unambiguous stimulus/indicate reversal (“unambiguous/reversal”), unambiguous/stability, ambiguous/reversal, and ambiguous/stability (see Table 1). Each experiment consisted of 24 experimental runs with six repetitions of each of the four conditions in an ABCDDCBA sequence. Each experimental run stopped after participants had responded 16 times at the latest after 7 minutes. 

Table 1

 

Overview of the experimental structures.

Experiments	Blocks	Conditions
Lattice experiment	Necker lattices (ambiguous)	ambiguous/reversal
		ambiguous/stability
	Unambiguous lattices	unambiguous/reversal
		unambiguous/stability
Face experiment	Ambiguous Old/Young Woman	ambiguous/change
		ambiguous/stability
	Unambigous face variants	unambiguous/change
		unambiguous/stability

EEG recordings and data analysis

EEG was recorded from nine gold-cup scalp electrodes at Oz, P3, P4, Pz, T7, T8, Cz, Fz, Fpz (American Clinical Neurophysiology Society, 2006) with a linked ear reference. Horizontal and vertical EOG detected blinks and eye movements. Impedance at each electrode location was kept below 10 kΩ. Signals were band-passed at 0.1–70 Hz, digitized at a sampling rate of 500 Hz, and written to disk. In the case of artefacts from eye movements and amplitude excursions exceeding ±100 μV, the EEG sweeps were automatically rejected. They were averaged per condition and digitally filtered with a latency-neutral low-pass at 20 Hz. Peak amplitude was measured relative to baseline, which was defined as the average from 80 ms before to 20 ms after stimulus onset. Four types of EEG data were sorted according to stimulus type (ambiguous and unambiguous), task (“response” and “nonresponse,” collapsing over reversal and stability), and electrode position (“channel”) and averaged separately into ERPs (“event-related potentials”, Bach, 1998; Luck, 2005) with respect to stimulus onset as time reference. 

To provide an analysis of the ERP data with high temporal resolution and concurrently avoiding an inflation of first order errors due to multiple testing we used a method introduced by Blair and Karniski (1993). It is based on randomization tests (Edgington, 1995), which generate their reference distribution by permuting the pooled original data. Randomization tests require fewer assumptions than, e.g., a student t-test. Blair and Karniski (1993) extended the principle of randomization tests to EEG data. In the following we denote this method as “BK-statistics.” For each stimulus type (geometric and semantic) a subsequent repeated-measurement MANOVA (in the following denoted as “rm-MANOVA”) analyzed the resultant ERP components from the difference traces (“dERPs”: unambiguous minus ambiguous) with the factors task (response and nonresponse) and channel (=electrode positions, including only those electrodes at which the BK-statistics had indicated significance) regarding the variables amplitude and latency (time interval from stimulus onset to the extremum of the regarded ERP component). The amplitudes from individual participants were defined as the extrema from the ERP-difference traces in specific time windows (100–200 ms after stimulus onset for the early ERP components and 200–600 ms for the later P400 ERP component; see below). 

The total durations of stable percepts of each stimulus interpretation was reconstructed from the participants' responses and averaged across participants. 

Results

Lattices: Unambiguous versus ambiguous stimuli

Figure 3a shows the ERP traces for the lattice stimuli of both the response and nonresponse trials. The BK-statistics identified a small negativity at around 180 ms after stimulus onset at the frontopolar electrode in the case of the unambiguous lattices. It is absent in the case of the ambiguous Necker lattices. This early negativity is marked with a black arrow in the difference traces (dERPs: unambiguous minus ambiguous lattices, averaged across participants, Figure 3c, P < 0.05, BK-statistics). 

Figure 3

 

ERP traces to the geometric lattice stimuli (a) and to the semantic face stimuli (b) on schematic scalps. The difference traces ± SEMs (unambiguous minus ambiguous; c, d) show the strong deviations from the null hypothesis of flat traces and the similarity of P400 patterns for the two stimulus types. The arrows indicate the earliest significant dERP deflections.

Twenty milliseconds later, at 200 ms, a further occipital negativity appears. Due to its high variability, it was not significant based on the BK-statistics. Further, a huge positive deflection can be observed at around 400 ms after stimulus onset (labeled here as “P400”) at all electrode positions, maximal at Pz and Cz, in the case of the unambiguous lattices. 

This P400 is almost absent in the case of the ambiguous lattices ( Figure 3c, P < 0.001, BK-statistics). The P400's latency decreases monotonously from occipital over central to frontal and frontopolar positions ( Figure 4a, F(10,120) = 21.9, P < 0.001 for the factor channel concerning the variable latency, rm-MANOVA). 

Figure 4

 

Mean latencies (a, b) and amplitudes (c, d) ± SEMs of the dERP P400 component (unambiguous minus ambiguous) across central electrode positions for the geometric lattice stimuli (a, c) and the semantic face stimuli (b, d).

Lattices: Response versus nonresponse conditions

The early frontopolar negativity is present in the response condition and absent in the nonresponse condition ( Figure 3c, P < 0.05). The amplitude of the subsequent P400 is twice as high in the response trials as in the nonresponse trials at the parietal and central electrode positions ( F(1,12) = 22.1, P < 0.001 for the factor task and F(10,120) = 13.7, P < 0.001 for the interaction channel × task regarding the variable amplitude, rm-MANOVA). 

Faces: Unambiguous versus ambiguous stimuli

Figure 3b depicts the ERP traces for the ambiguous face and the unambiguous variants for the response and the nonresponse trials. The BK-statistics identified higher amplitudes for the unambiguous stimuli compared to the ambiguous stimuli ( P < 0.001) as early as 130 ms after stimulus onset. 

This amplitude difference can be observed more easily in the dERPs (unambiguous minus ambiguous; averaged across participants) as positive deflections at parietal, temporal, central, and frontal electrode positions ( Figure 3d, arrows). Further, a huge positivity can be observed at all electrodes with maxima at parietal and central positions (P400). Its amplitude is about four times as high as with the ambiguous stimuli ( P < 0.001, Figure 3d). 

Faces: Response versus nonresponse trials

The early positivity is indicated as significant (BK-statistics) for the parietal, temporal, central, and frontal electrodes in the response trials but only at the parietal electrode in the nonresponse trials, where it is much weaker ( F(1,12) = 22.9, P < 0.001 for the factor task, F(4,48) = 19, P < 0.001 for the factor channel, and F(4,48) = 7.5, P < 0.001 for the interaction task × channel regarding the variable amplitude, rm-MANOVA). 

The P400 amplitude is higher in the response trials than in the nonresponse trials ( F(1,12) = 123.3, P < 0.001 for the factor task, and F(10,120) = 8.3, P < 0.001 for the interaction channel × task regarding the variable amplitude, rm-MANOVA). Further, the latency decreases monotonously from occipital over central to frontal and frontopolar electrode positions ( Figure 4b, P < 0.001 for the factor channel, regarding the variable latency, rm-MANOVA). 

The first dERP difference between unambiguous and ambiguous face stimuli occurs 50 ms earlier (130 ms), at different locations, it is more prominent and has a different sign compared to the lattice stimuli. 

After 250 ms the results for the face stimuli are remarkably similar to those for the lattice stimuli: Although the pictorial differences between ambiguous and unambiguous stimuli are very small ( Figure 2c), there is a pronounced difference in amplitude of the P400 dERP-component (this can be seen in the difference traces in Figures 3c and 3d that would be flat in case of no difference). And although the lattice and face stimuli belong to different categories with substantial pictorial differences (comparing Figures 2b and 2c, right with left), the spatial and temporal pattern of the P400 is very similar (comparing Figures 3c with 3d and Figures 4a and 4c with Figures 4b and 4d shows very similar patterns concerning P400 differences). 

Discussion

We analyzed the ERP differences (dERPs) between ambiguous stimuli and unambiguous stimulus variants from two different stimulus categories: Geometric wireframe cube lattices (lattice stimuli) and simplified face drawings (face stimuli). Participants observed the ambiguous and unambiguous stimuli from each stimulus category in different experimental runs and indicated in separate conditions in a go/nogo-task perceptual changes between or perceptual stability across successively presented stimuli. 

Altogether, four different types of dERP traces were analyzed, according to stimulus category (lattice and face stimuli) and task (response and nonresponse), which differ markedly depending on the time window: 

Up to 250 ms after stimulus onset the results for the face stimuli are different from those for the lattice stimuli: The first dERP difference between unambiguous and ambiguous lattice stimuli occurs at 180 ms after stimulus onset, is small, and restricted to the frontopolar position. The diverging early dERP results for the lattice and the face stimuli ( Table 2) may represent differences in perceptual processing of the very different object categories. This is in accordance with recent evidence of differences in neural processing of animate objects (e.g., faces) and nonanimate objects (e.g., buildings, Kanwisher & Yovel, 2006). Task-related attentional effects (Reynolds & Chelazzi, 2004) may explain the response-related dERP enhancements. 

Table 2

 

ERP differences (dERPs) “unambiguous minus ambiguous.”

Time window	Features	Lattice	Face
0–250 ms (earliest peak)	Latency:	180 ms	130 ms
	Size:	−0.9 μV	1.6 μV
	Polarity:	minus	plus
	Location:	Fpz	Pz, Cz, T7, T8, Fz
>250 ms (P400)	Latency:	280–410 ms	340–440 ms
	Size:	2.6–12.5 μV	1.5–6.7 μV
	Polarity:	plus	plus
	Location:	max. at Pz and Cz	max. at Pz and Cz

In the present study we were interested in physiological differences between stimuli with different degrees of ambiguity across two different stimulus categories. We were less interested in physiological differences between representations of different stimulus categories or states of attention. We thus concentrate our discussion on the large difference of P400 amplitude between ambiguous and unambiguous stimuli, which occurs for both the geometric (lattices) and the semantic (faces) category. 

The difference in P400 amplitude may be explained by a temporal jitter of perceptual processing from trial to trial (with respect to stimulus onset) in the case of the ambiguous visual information. This could result in reduced (or nearly absent) mean P400 amplitudes for the ambiguous stimuli from both stimulus categories. However the following observations speak against this explanation. Figure 5a shows the present data selectively averaged across trials with change indicated and separately across trials with stability indicated. Clear differences between change and stability can be observed. However, these aspects are not in the focus of the present study, they are discussed elsewhere (Kornmeier & Bach, 2004, 2005, 2006; Kornmeier, Ehm, Bigalke, & Bach, 2007; O'Donnell et al., 1988; Strüber & Herrmann, 2002). For the present discussion it is important that the temporal jitter with ambiguous stimuli should be reflected in longer median reaction times and broader interquartile ranges (Sommer, Leuthold, & Hermanutz, 1993; Verleger, 1997). However, as can be seen in Figure 5, the differences in reaction time variability cannot predict the differences in P400 amplitudes between ambiguous and unambiguous stimuli. 

Figure 5

 

(a) EEG data selectively averaged across trials with change indicated and separately across trials with stability indicated. Grand means for the Pz electrode across participants. Left: Lattice experiment. Right: Face experiment with different participants. Stimulus onset at 0 s. Clear differences between reversal and stability can be observed, whereas the huge P400 difference between ambiguous and unambiguous stimuli is common to both reversal and stability. (b) Median reaction times across participants with respect to stimulus onset after the blank period. The boxes indicate the ±25 percentiles, the lines indicate the ±95 percentiles. UC: unambiguous change; US: unambiguous stability; AC: ambiguous change; AS: ambiguous stability.

In a very similar experimental paradigm Kormeier and Bach (2004, 2005) recently found that endogenous perceptual changes of Necker lattices are synchronized with stimulus onset with a precision of ±30 ms or less. 

One might interpret the P400 as a correlate of object representation but this would not be compatible with the sparse coding hypothesis (e.g., Barlow, 1972), which assumes that the information carried by neurons is represented by relatively few active neurons within a largely populated neuronal network: Amplitudes of over 10 μV imply the synchronous activity of many neurons (reviewed in Bach, 1998). The activation strength hypothesis (e.g., Zeki, 2005), which assumes perceptual awareness of an object as a function of the strength of neural activation, would likewise be ruled out: The P400 absence in the case of the ambiguous stimuli would imply low neural activity and thus no conscious perception of those stimuli. 

The P400 amplitude, its distribution, and its latency decrease from posterior to anterior positions ( Figures 4a and 4b) may indicate a superposition of two well-known “cognitive” ERP components, namely the P3a and the P3b (e.g., Polich, 2004). 

The P3a, or “novelty P300,” typically occurs with a distractor stimulus that appears randomly between target and standard stimuli in “three-stimulus” paradigms. It has a more anterior central/frontal maximum, shorter latencies, and rapidly habituates in contrast to the P3b (Polich, 2004). Since the present paradigm has no third distractor stimulus, P3a can be ruled out as contributing to our P400. The P3b is typically induced in oddball paradigms with a randomly occurring low-probability target stimulus. Its amplitude is known to be maximal at parietal electrodes. In his very influential paper, Johnson ( 1986) suggested a “triarchic model” integrating most of the experimental findings regarding the modulation of the P3b amplitude. He assumes two additively related factors, namely ‘subjective probability’ (global probability and sequential expectancies) and ‘stimulus meaning’ (task complexity, stimulus complexity, and stimulus value). Both factors are scaled multiplicatively by a third factor, which he called ‘information transmission’ (equivocation and inattention). 

The present paradigm can be regarded as a typical go/nogo paradigm, where the P400 occurs in both go and nogo trials. In the case of the unambiguous stimuli both the go and the nogo trials occur with equal probability. Further ambiguous and unambiguous stimuli are presented in different experimental runs and have thus both a probability of P = 1. Probability can thus be ruled out as an explanation. 

The “stimulus meaning” (Johnson, 1986), which may be higher in the go than the nogo trials (due to “task complexity” and/or “stimulus value”) can only explain a small part of the whole effect. The major P400 difference occurs between ambiguous and unambiguous stimuli in both the go (higher value) and the nogo trails (lower value). 

This major part of the difference may be explained by Johnson's ‘equivocation’ factor. Sutton, Braren, Zubin, and John (1965) already proposed that the P3 is a correlate of the resolution of prior uncertainty. Hillyard, Squires, Bauer, and Lindsay ( 1971) concluded that their P3 amplitude reflects the level of confidence in the detection of a signal. Johnson regarded equivocation as the amount of information loss during stimulus processing, expressed as subject's a posteriori uncertainty about their percept and reflected in a decrease of P3b amplitude. The present P400 difference may thus indicate different degrees of information transmitted by ambiguous compared to unambiguous stimuli. 

These considerations remind on the “Bayesian Inference concept,” a model of how the perceptual system resolves the visual inverse problem (e.g., Kersten, Mamassian, & Yuille, 2004): The probability P(S∣I) of a scene S, given the retinal image I, is proportional to the product of the probability P(S) of the scene prior to the retinal image information based on prior knowledge, and the probability P(I∣S) of the retinal image, given the scene S: 

$P(S|I)\propto \mathrm{P}\left(S\right)P(I|S).$

(1)

 

The core idea is that the part of the information, which is unavailable to our eyes but necessary to construct a stable percept, can be estimated by combining the available retinal information with prior perceptual experiences stored in our memory. 

This Bayesian probability P( I∣ S) can be regarded as a reliability indicator expressing “how much an evaluation instance is impressed by the visual input.” The P400 may be a direct correlate of this indicator, reflecting the quality of the perceptual outcome: the more ambiguous the visual information is, the less certain the inferences drawn by the evaluation instance, the smaller the P400 amplitude, the less stable the percept, and the more probable a spontaneous perceptual change. We assume this reliability estimation to be unconscious for the following reasons:

1.  
   After the experiment several participants expressed their disbelief about the purely endogenous nature of perceived changes. They assumed subtle stimulus manipulations as a cause even though the ambiguity of these stimuli was explained at the beginning. This corresponds well with numerous reports of surprise of naive observers when they experience sudden perceptual changes and speaks against a general conscious uncertainty regarding the ambiguous stimuli, which should be also felt during stability periods.
2.  
   Verleger, Jaskowski, and Wascher (2005) assume the P3b to be a correlate of a conscious perceptual decision process, dependent on certainty. Higher ambiguity of the stimulus should thus cause higher uncertainty in the decision process, which should be reflected in longer median reaction times and higher variability. This, however, is not the case in the present data ( Figure 5b, AC/AS vs. UC/US).

Conclusions

We must constantly construct percepts from limited sensory information. It may thus have been an evolutionary key advantage for survival to rely our (re)actions on the result of the perceptual outcome's reliability. The more reliable our percept is rated, the faster, more appropriately and more goal-directed we can prepare and execute our actions. We assume the P400 as a correlate of an evaluation instance's reliability estimation. The less reliable one interpretation is, the better it would be to come up with an alternative interpretation or concept about the environment. Such an evaluation instance—from the visual or other sensory domains—may thus be a basic precondition for invention and cognition. The late positive deflection at 400 ms may be a correlate of its outcome. 

Acknowledgments

Support by the Deutsche Forschungsgemeinschaft (BA 877-16-2) is gratefully acknowledged. We thank Christine M. Hein for data collection. 

Commercial relationships: none. 

Corresponding author: Jürgen Kornmeier. 

Email: [e-mail]. 

Address: Uni-Augenklinik, Sektion Funktionelle Sehforschung, 79106 Freiburg, Germany; Institut für Grenzgebiete der Psychologie, 79098 Freiburg, Germany.