1 Binocular rivalry and perceptual ambiguity David Alais 1 & Randolph Blake 2 To appear in: Oxford Handbook of Perceptual Organization Oxford University Press Edited by Johan Wagemans 1. School of Psychology, University of Sydney, Sydney, New South Wales, Australia 2. Department of Psychology, Vanderbilt University, Nashville, Tennessee, USA 1. Introduction and background Humans possess the impressive ability to achieve coherent and reliable perception of the external world. Remarkably, this achievement is realized despite the relatively low resolution of the retinal images, images that are inherently two-dimensional and often under-represent what one is actually looking at. Consequently, many important aspects of objects and scenes are fundamentally ambiguous at the input stage to vision, including size, distance, depth ordering, shape and color. The general reliability of visual perception is striking given that not all pieces of the puzzle are present in the retinal input. To overcome this limitation, perception relies on perceptual organization (Wertheimer 1923) and knowledge about the likely properties of the external world acquired through evolution or learned from experience to make “unconscious inferences” (von Helmholtz 1925) about the world we live in. Thanks to these processes, we are generally able to construct a plausible interpretation of the world from the ambiguous and incomplete retinal image. Circumstances may arise, however, that defeat the brain’s ability to infer a single coherent percept (Leopold and Logothetis 1999). In cases where more than one plausible percept is possible, the competing perceptual interpretations alternate over time in an irregular fashion each second or so, as the reader can experience by viewing a well-known ambiguous figure known as the Necker cube (Figure 1a). This class of phenomenon, generally labelled bistable perception, reveals the competition or ‘rivalry’ that occurs when the perceptual system is confronted with ambiguous visual information (e.g., Blake and Logothetis 2002). As well as competition, bistable perception also reveals a key role for inhibition, as the competing percepts are mutually exclusive: only one interpretation is visible at a time, with the other being suppressed from perceptual awareness.
27
Embed
1 Binocular rivalry and perceptual ambiguity David Alais ...
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
1
Binocular rivalry and perceptual ambiguity
David Alais1 & Randolph Blake
2
To appear in:
Oxford Handbook of Perceptual Organization
Oxford University Press
Edited by Johan Wagemans
1. School of Psychology, University of Sydney, Sydney, New South Wales, Australia
2. Department of Psychology, Vanderbilt University, Nashville, Tennessee, USA
1. Introduction and background
Humans possess the impressive ability to achieve coherent and reliable perception of the
external world. Remarkably, this achievement is realized despite the relatively low resolution of
the retinal images, images that are inherently two-dimensional and often under-represent what
one is actually looking at. Consequently, many important aspects of objects and scenes are
fundamentally ambiguous at the input stage to vision, including size, distance, depth ordering,
shape and color. The general reliability of visual perception is striking given that not all pieces of
the puzzle are present in the retinal input. To overcome this limitation, perception relies on
perceptual organization (Wertheimer 1923) and knowledge about the likely properties of the
external world acquired through evolution or learned from experience to make “unconscious
inferences” (von Helmholtz 1925) about the world we live in. Thanks to these processes, we are
generally able to construct a plausible interpretation of the world from the ambiguous and
incomplete retinal image. Circumstances may arise, however, that defeat the brain’s ability to
infer a single coherent percept (Leopold and Logothetis 1999). In cases where more than one
plausible percept is possible, the competing perceptual interpretations alternate over time in an
irregular fashion each second or so, as the reader can experience by viewing a well-known
ambiguous figure known as the Necker cube (Figure 1a). This class of phenomenon, generally
labelled bistable perception, reveals the competition or ‘rivalry’ that occurs when the perceptual
system is confronted with ambiguous visual information (e.g., Blake and Logothetis 2002). As
well as competition, bistable perception also reveals a key role for inhibition, as the competing
percepts are mutually exclusive: only one interpretation is visible at a time, with the other being
suppressed from perceptual awareness.
2
Figure 1. a) Three well-known examples of ambiguous visual stimuli giving rise to
bistable perception. The Necker cube, Schroeder’s stairs and Rubin’s face/vase
illusion. Inspecting any of these for a minute or so is sufficient to demonstrate the
basics of bistable perception: i) when stimuli support two plausible
interpretations, perception will alternate between the two, ii) the alternations are
irregular in duration, and iii) it is not possible to stop the alternation process by
willful attention. b) Binocular rivalry is a commonly studied form of bistable
perception. It is triggered when incompatible images are presented to the eyes
and produces perceptual alternations between one eye’s image and the other’s.
As with any bistable stimulus, the alternations are irregular and form a skewed
distribution, such as a Gamma or log-normal distribution.
Examples of bistable perception are found in many areas of vision including 3D perspective,
figure/ground organization, binocular rivalry (Wheatstone 1838), and new varieties discovered
in motion (e.g., Hupe and Rubin 2003), perception of human action (Vanrie, Dekeyser et al.
2004) and stereo-depth organization (van Ee, Adams et al. 2003). Other modalities, too, must
deal with stimulus uncertainty. Conflicting dichoptic auditory messages also compete for
dominance, creating binaural rivalry (Brancucci and Tommasi 2011). Tone sequences that can be
perceptually grouped into two distinct patterns produce auditory bistability (e.g., Pressnitzer
and Hupe 2006). In the tactile domain, rivalry occurs when vibrotactile sequences supporting
two interpretations are applied to a fingertip (Carter, Konkle et al. 2008). See chapters by
Denham and Winkler (this volume) and Kappers and Bergmann Tiest (this volume) for further
discussion of perceptual ambiguity in the auditory and tactile domains, respectively. In general,
fluctuations in perception seem to be the rule when sensory input is ambiguous. The
phenomenology of all forms of bistable perception is broadly similar in that all involve exclusive
3
alternations between the competing perceptual interpretations. One common hallmark is the
apparent randomness of the alternations between competing interpretations, as evidenced by
the Gamma-like, skewed normal frequency histograms of dominance durations (Fox and
Herrmann 1967) (see Figure 1b). Several studies have shown that diverse instances of
perceptual rivalry all exhibit this pattern of temporal dynamics (Carter and Pettigrew 2003; Long
and Toppino 2004; Brascamp, van Ee et al. 2005; van Ee 2005; O'Shea, Parker et al. 2009),
suggesting that it may be a general characteristic of bistable perception.
In this chapter we focus on the most widely studied form of bistable perception, binocular
rivalry (Blake 2001; Tong 2001; Blake and Logothetis 2002; Alais and Blake 2005). We begin by
describing the basic properties of binocular rivalry, and then review work on rivalry relating to
perceptual organisation, including figure/ground segregation and perceptual grouping. The
second half of the chapter broadens the scope by discussing the role of attention in binocular
rivalry and considering the impact of top-down and contextual influences. Broader still, the final
section examines recent work studying binocular rivalry in a multisensory context.
2. Binocular rivalry
Binocular rivalry is a compelling bistable phenomenon first systematically studied by
Wheatstone (1838) following his invention of the mirror stereoscope. Binocular rivalry occurs
when each eye views incompatible images at the same retinal location, where ‘incompatible’
means stimuli sufficiently different to prevent a binocular match. This can be easily achieved in
the laboratory using a mirror stereoscope to present a different image to each eye, as shown in
Figure 1b. Perceptually, binocular rivalry is experienced as seemingly random fluctuations in
dominance between one image and the other that continue as long as the dissimilar images are
viewed. For stimuli of similar salience, these stochastic fluctuations tend to even out over time
so that each image is seen equally often during extended viewing. Stimulus salience in binocular
rivalry is largely governed by low-level stimulus properties, such as contrast, luminance, and
orientation, with a relatively small but demonstrable role for high-level stimulus factors such as
attention and context (reviewed later in the chapter). Generally, while one image is dominant,
little or no trace of the other image is perceived. Interest in binocular rivalry has increased in
recent decades, in part because rivalry allows systematic examination of processes governing
perceptual competition, neural dynamics and selection of the contents of visual awareness.
Although binocular rivalry has much in common with other forms of bistable perception, some
very important differences set binocular rivalry apart. First, binocular rivalry is unique in
presenting a different stimulus to each eye, whereas other bistable examples involve a single
stimulus viewed binocularly. This interocular conflict disrupts normal binocular vision and
triggers binocular rivalry, in part because the conflict interferes with the establishment of
binocular correspondence necessary for stereomatching. Second, the alternations in binocular
rivalry are generally mutually exclusive, such that when one image is perceived the other is
completely suppressed. Other forms of bistable perception involve a single stimulus that
supports two interpretations, and it is those interpretations that alternate over time while the
stimulus itself remains visible. The Necker cube, for example, elicits bistable alternations of
4
perceived perspective without any part of cube disappearing from visual awareness. Third,
binocular rivalry has a strong local component, as revealed by the phenomenon of piecemeal
rivalry in which large images tend to alternate as a patchwork (O'Shea, Sims et al. 1997). By
contrast, other bistable stimuli tend to alternate globally and do not exhibit obvious ‘piecemeal’
states. There are, however, conditions under which rivalry behaves globally, and this makes it
useful as a tool for studying perceptual organization. Accordingly, the following sections review
basic features of binocular rivalry that illustrate its links to the principles of perceptual
organization.
3. Gestalt organizing principles in binocular rivalry
3.1. Figure/ground segregation and binocular rivalry
One of the primary processes in perceptual organization is figure/ground segregation, the
process by which some regions within the visual image merge perceptually to form objects while
remaining regions are treated as the background against which those objects appear. The
relationship between figure and ground is one of occluder and occluded because the figure, in
terms of depth ordering, must be nearer than the background. Surprisingly little work in
binocular rivalry has examined figure/ground organization directly, although it has been widely
studied in other contexts (see Kogo & Van Ee, this volume). In one old study, Alexander (1951)
attempted to weaken the strength of rivaling figures by using dashed lines instead of continuous
contours to portray shapes and by reducing the lines’ contrast by printing them on gray paper.
The rationale was that these manipulations would reduce ‘figural strength’ and make vigorous
rivalry less likely, because figural strength entails resistance to distortion, impressiveness,
internal articulation, density of energy and symmetry (Koffka 1935). In fact, Alexander did find
reduced alterations rates for the weak figures, but a contemporary interpretation of that finding
would focus simply on the accompanying variations in stimulus contrast: stimuli higher in
contrast and greater in contour strength produce more vigorous rivalry (Levelt 1965),
presumably because of contrast-dependent responses in early cortical areas tuned to
orientation. Still, it could be argued that those response properties in turn contribute to
figure/ground relationships.
One reasonable hypothesis arising from figure/ground organization is that stimulus regions
defined as figure should engage more vigorously in rivalry than regions deemed to be
background. This is in line with traditional thinking on figure/ground classification and also
squares with modern thinking about visual processing in which visual objects are extracted from
the visual image and compete for visual attention (Desimone and Duncan 1995), although there
is no direct test of this notion in the published literature on rivalry. A simple test would be to
present dichoptic displays consisting of a small figure region (e.g., red horizontal lines) within a
surrounding background region (e.g., green vertical lines), with the reverse pattern in the other
eye, as shown in (Figure 2a). More vigorous rivalry for the figure region could be demonstrated
in two ways, by showing that rivalry alternations were faster in the figure region, consistent with
the figure having greater stimulus strength, or by measuring contrast sensitivity to probe stimuli
– a common method for measuring rivalry suppression strength (Fox and Check 1968; Nguyen,
Freeman et al. 2003; Alais and Melcher 2007). The prediction would be that probes presented in
5
the figure region would show greater threshold elevation during rivalry suppression than probes
presented in the background region.
Figure 2. Two examples of perceptual organisation in binocular rivalry. a) A rivalry
stimulus with well-defined figure and ground regions. Viewing this stimulus
dichoptically tends to produce two independent rivalry processes, one for the
figure region, and one for the background region. Rivalry in the figure region
tends to be more vigorous and alternate faster, implying greater salience for
figure relative to background. One prediction that has never been tested is that
rivalry suppression should be stronger in the figure than in the background region
(see text). b) Dichoptically viewing a ground plane (left-hand side) and a ceiling
plane (right-hand side) produces rivalry (Ozkan and Braunstein 2009). The two
planes are identical apart from a 180° rotation, yet the ground plane tends to
dominate the ceiling plane, highlighting the importance of ground planes in
perceptually organising surface layout and spatial relations (see text).
Although there is little work directly examining the impact of figure/ground organization on
binocular rivalry, several studies have looked at other aspects of visual scene organization. One
examined the salience of different regions of a visual scene by inducing rivalry between a
simulated ground plane and a simulated ceiling plane (Ozkan and Braunstein 2009). The ground
plane was a receding checker board appearing to incline towards the horizon while the ceiling
plane was a receding checker board appearing to decline towards the horizon (Figure 2b). Thus,
the two stimuli were identical except for one being a rotated version of the other, and yet the
ground plane tended to predominate over the ceiling plane. Moreover, the ground plane, when
suppressed, returned more quickly to dominance than did the ceiling plane. Other studies have
highlighted the relevance of surface layout, finding that it influences the dynamics of rivalry
alternations by inhibiting false matches between the eyes according to ecological constraints.
Other aspects of surface properties such as natural boundary contours (Ooi and He 2006) and
the coherence of surfaces (Ooi and He 2003) influence dynamics and dominance durations in
6
rivalry. As an example, continuous or homogenous surfaces tend to dominate over
discontinuous images (Ooi and He 2003).
3.2. Perceptual grouping in binocular rivalry
Another fundamental process in perceptual organization is grouping. Unlike the paucity of work
on figure/ground classification in rivalry, a good deal of research has been done on perceptual
grouping. For example, Whittle et al. (1968) demonstrated grouping by similarity in showing
robust configural effects among multiple, small contour segments when each engaged in rivalry.
Observers tended to see simultaneous dominance of segments that formed an extended line,
even when those segments were presented to different eyes. More dramatic versions of figural
grouping encouraging globally synchronized dominance have been reported by Dorrenhaus
(Dorrenhaus 1975), Kovacs et al. (1996) and Alais et al. (2000) which suggest that grouping in
rivalry is possible at a binocular level (Figure 3). In a similar vein, Van Lier and De Weert (2003)
showed grouping by colour in binocular rivalry: in a multi-element display, similarly coloured
features tended to dominate together. Kim and Blake (2007) showed this also occurs with
illusory colors experienced by color-graphemic synesthetes. In the domain of motion
perception, spatially distributed dots that move in the manner of a human figure (so-called
point-light animations) remain dominant as an entire figure more often during rivalry than does
the same configuration when inverted to form an upside down figure, or when distributed
between the eyes (Watson, Pearson et al. 2004). Evidently, conjoint dominance of individual
dots is promoted when they form a dynamic and globally coherent human figure.
Figure 3. Examples of high-level perceptual organisation in binocular rivalry. Most
models of rivalry focus on competition between competing monocular
mechanisms, with one image at a time being suppressed and the other allowed to
7
dominate. These examples, however, demonstrate that a degree of higher-level
processing must be involved. a) The left- and right-eye images are incompatible in
terms of form (lines vs. circles) and colour (red vs. green) and trigger vigorous
rivalry between each eye’s image (Alais et al., 2000). However, extended viewing
of this image also elicits periods where alternations occur between entirely green
circles and entirely red lines – percepts that are only possible by combining across
both the left- and right-eye images simultaneously. b) Dichoptically viewing the
upper pair of images produces rivalrous alternations between the left- and right-
eye stimuli. The lower pair also produce left- vs. right-eye rivalry, but in addition
produces periods of rivalry between the coherent images (the monkey face vs. the
page of text) which requires grouping elements from each image simultaneously
across the eyes (Kovacs et al. 1996). These demonstrations show that coherent
perceptual organisation can be imposed on conflicting monocular images when
strong Gestalts are present. Because this requires interocular grouping, it implies
a binocular process over-riding earlier interocular suppression.
The findings summarized above pertain to perceptual grouping among multiple, spatially
distributed elements each engaged in rivalry. Grouping can also occur within a single large-field
stimulus, especially when they contain meaningful spatial structure (Lee and Blake 2004; Alais
and Melcher 2007), although before reviewing this work it is necessary to describe the
phenomenon of ‘piecemeal rivalry’. When two small stimuli engage in binocular rivalry, they will
usually produce coherent fluctuations in perception so that either one image or the other
dominates entirely. This is generally true for stimuli subtending a degree or two of visual angle.
Rivalry between larger stimuli, however, tends to fragment into a patchwork of local
alternations, with the local patches appearing to alternate between the left and right eyes’
images independently of each other. This mosaic of independent local rivalry zones is commonly
referred to as ‘piecemeal’ rivalry and is very common when large images engage in rivalry.
Piecemeal rivalry points to the local nature of rivalry, yet there are also occasions when large
stimuli appear to alternate in a coherent or synchronized manner. Clearly some cooperative
grouping process is at work in coordinating these otherwise independent local processes.
The existence of piecemeal rivalry prompts two fundamental questions. First, what determines
the size of local rivalry zones, and second, what are the cooperative processes that promote
interactions among these local zones? Regarding the first question, there is good evidence that
the spatial extent of local rivalry zones is governed by the size of receptive fields in early visual
cortex. In central vision, rivalry zones are typically about a degree or so in diameter, however
their size increases with eccentricity at a similar rate to the expanding size of V1/V2 receptive
fields with eccentricity (O'Shea, Sims et al. 1997). This implies that rivalry has a spatial extent
governed by the sizes of receptive fields in early visual cortex and that rivalry alternations are
more likely to be piecemeal when stimuli activate neurons spanning multiple receptive fields.
The link with receptive field size also relates to another interesting observation, namely that
rivalry appears to have a minimum size. It has been shown that even when the interocular
conflict is limited to a single point, as when two thin orthogonal lines are viewed dichoptically,
there exists a zone of suppression that extends around that point (Kaufman 1963), with the size
of the suppression zone depending on eccentricity. Rivalry therefore appears to be a process
8
that operates locally over an extent determined by receptive field sizes in early cortex. One
advantage of rivalry being local is that suppression is localized and allows binocular vision to
operate normally in any binocularly congruent regions outside the region of interocular conflict.
The second question prompted by piecemeal rivalry is why independent local rivalry zones
sometimes appear to function synchronously to form global alternations. One study examined
this question by presenting two adjacent gratings to one eye, rivaling with corresponding noise
patches in the other eye (Alais and Blake 1999). Observers tracked rivalry alternations at the two
grating locations and the orientations of the gratings were manipulated over blocks to be either
collinear, orthogonal or parallel. The perceptual fluctuations reported in the orthogonal
condition were independent, meaning that both gratings occasionally were visible at the same
time but not more often than would be expected by chance alone. In the collinear condition,
however, the gratings were often jointly dominant, significantly more than predicted by
independence (Figure 4). This grouping tendency was very strong when the two pairs of rivaling
stimuli were adjacent in the same hemifield (therefore projecting to adjacent columns in the
same cortical hemisphere), and was still quite strong when the rivaling stimuli were placed on
either side of fixation. The fact that grouping was still observed for grating patches placed on
either side of fixation suggests that callosal connections between hemispheres are able to
establish the adjacency of the grating patches in the visual field as well as their orientation
relationship. Consistent with this suggestion, a study of binocular rivalry in a split-brain observer
found that coordinated dominance between rivalry patches did not occur when those patches
were located either side of the midline (O'Shea and Corballis 2005). The corpus callosum does
indeed seem critical for perceptual grouping across the vertical midline.
9
Figure 4. Spatial grouping in binocular rivalry. Binocular rivalry has a strong local
component, as revealed by the tendency for large stimuli to break up into
independent local zones of rivalry (‘piecemeal’ rivalry). Despite this, cooperative
processes between local elements can produce spatial grouping and coherent
alternations. Motivated by contour interactions research (see ‘Contour
Integration’ chapter, Hess et al, this volume), binocular rivalry studies have shown
that local elements engaged in rivalry tend to coordinate and become jointly
dominant if they are close and collinear, but less so as separation or relative angle
increases, consistent with the “association field”. Subsequent work applied this
approach over larger stimulus regions to study “traveling waves” of dominance
and mapped this phenomenon onto early visual cortex using neuroimaging
methods (see text).
Binocular rivalry is therefore a process occurring in local zones, but these can group together
into pairs or larger ensembles (Bonneh and Sagi 1999) according to the principle of the
“association field” (Field, Hayes et al. 1993). This notion (see Figure 4) is similar to the Gestalt
principle of common fate or good continuation and posits that collinear orientations will tend to
associate more strongly than oblique contours (Alais, Lorenceau et al. 2006), and that the
strength of association declines with distance. The association field is thought to have a basis in
10
the long-range horizontal connections in V1 which are known to be longer and stronger for
collinear orientations and to fall off monotonically with angular difference (Kapadia, Ito et al.
1995). Related work shows that spatial interactions influencing rivalry can arise outside regions
of the visual field within which rivalry is occurring. For instance, the predominance and strength
of suppression of a patch of grating engaged in rivalry are influenced by a surrounding grating
that is not engaged in rivalry (Paffen, te Pas et al. 2004; Paffen, Alais et al. 2005). This interaction
is thought to have a neural basis in center-surround interactions between classical and extended
receptive fields (e.g., Blakemore and Tobin 1972; Fitzpatrick 2000).
Another line of work pointing to local grouping between rivalry zones comes from studies of
‘traveling waves’ of rivalry dominance (Wilson, Blake et al. 2001; Kang, Lee et al. 2010). These
studies examined the often noted observation that when a large rivalry stimulus is suppressed,
dominance will often breakthrough in a single small region and then spread like a wave,
sweeping across the entire stimulus until it is fully visible. Psychophysical observations have
shown that traveling waves tend to travel faster and further along collinear contours than non-
collinear contours (see Figure 4), in keeping with the association field hypothesis (Wilson, Blake
et al. 2001; Kang, Lee et al. 2010). An fMRI study (Lee, Blake et al. 2005) has shown that when a
traveling wave is experienced in rivalry it produces a concomitant wave of changing BOLD
activity across the occipital cortex that is correlated spatially and temporally with the perceived
traveling wave. The speed of the wave in perception, in other words, is tightly correlated with
the spreading wave within neural tissue, as is the spatial movement of the wave in the visual
field and in retinotopic cortical areas (Lee, Blake et al. 2007).
Taken together, these findings are consistent with binocular rivalry being a local process with
lateral interactions capable of coordinating rivalry states across adjacent locations, thereby
allowing coherent states to emerge through perceptual grouping and synchronized transitions.
Rivalry thus exhibits spatial grouping over space and time. This grouping is made possible by
cooperation along collinear or near-collinear orientations and is likely mediated by lateral
cortico-cortical networks (Kapadia, Ito et al. 1995; Angelucci, Levitt et al. 2002). For a full review
of contour interactions, see Hess et al. (this volume). Consistent with this reasoning, natural
images – which contain locally correlated orientations across spatial scales – tend to resist
breaking into piecemeal zones and will remain coherent at much larger image sizes than
gratings will (Alais and Melcher 2007). Natural images will also tend to predominate over non-
natural images when the two are pitted against one another in rivalry (Baker and Graf 2009).
4. Dynamics of binocular rivalry
One of the striking features of binocular rivalry is that the competition between conflicting
monocular inputs never seems to be resolved. Alternations in dominance between dissimilar
monocular patterns persist for as long as those patterns are viewed, although the incidence of
mixed dominance tends to increase when one views rivalry for very long periods of time (Klink,
Brascamp et al. 2010). What underlies the temporal dynamics of binocular rivalry? This section
will review the factors governing rivalry dynamics, and in doing so will lay the groundwork for
the subsequent sections discussing top-down and contextual influences on binocular rivalry.
11
Levelt (1965), one of the first to examine rivalry dynamics in detail, borrowed the idea of
reciprocal inhibition from early neurophysiologists. He contended that when conflicting rival
images first activate respective neural populations, reciprocal inhibition would inevitably cause
one response to dominate the other. The reason is that a stronger response in one population –
even a slight one – leads to greater inhibition over the other population. Any degree of
advantage less inhibition is exerted back by the weaker population, freeing the stronger
population to respond even more strongly (and exert still further inhibition over the other). This
process rapidly leads to one population completely inhibiting the other so that only one image is
visible. Most subsequent models of binocular rivalry have employed reciprocal inhibition to
account for rivalry suppression (Lehky 1988; Blake 1989; Mueller 1990; Laing and Chow 2002;
Freeman 2005).
Reciprocal inhibition offers an explanation of the suppression of one image at rivalry onset, but
how does it explain the ensuing alternation of perceptual dominance? Simply adding neural
adaptation to the reciprocal inhibition process is sufficient to account for ongoing fluctuations in
dominance because it reverses the process. Adaptation gradually attenuates the responses
within the dominant population, progressively weakening its inhibitory hold over the
suppressed population. Concurrent with weakening inhibition, the suppressed neurons are also
recovering from adaptation incurred in their previous dominance phase and are thus gaining
strength. Over time, responses in the two populations converge towards a balance point where
any minor change in response can trigger a flip in perceptual dominance. The adapting
reciprocal inhibition model of binocular rivalry is sufficient to explain both suppression and
alternation dynamics. Importantly, the tipping point is somewhat variable, as it is influenced by
external factors such as eye movements or blinks, or by internal factors such as attentional
shifts or neuronal noise in response levels (Kim, Grabowecky et al. 2006; Lankheet 2006;
Moreno-Bote, Rinzel et al. 2007). These potential tipping factors assume increasing significance
as the tipping point approaches and can trigger perceptual shifts at irregular times, consistent
with the fundamentally stochastic nature of rivalry dynamics (Brascamp, van Ee et al. 2006;
Shpiro, Moreno-Bote et al. 2009).
The adapting reciprocal inhibition model of rivalry predicts that suppression strength should
weaken over a dominance period, reaching a minimum level just prior to a dominance switch.
Two studies testing this prediction found sensitivity for detecting probes in the suppressed eye
late in a suppression period were not better than early in the period (Fox and Check 1968;
Norman, Norman et al. 2000), implying that inhibition was not weakening over time. However,
two limitations may explain their null finding. First, both studies used gratings as rival stimuli but
measured sensitivity using completely different probes (letters or small spots of light) that
would not tap into the same neurons signaling (and adapting to) the suppressed grating.
Second, the ‘late’ probes in these studies were presented at the median dominance duration so
that no genuinely late probes were measured. Recently, a new approach solved these problems
(Alais, Cass et al. 2010). First, the probe was a contrast increment of the suppressed stimulus
itself, meaning it directly probed contrast sensitivity of the neurons encoding the suppressed
stimulus. Second, in a new ‘reverse correlation’ approach, hundreds of probes were presented
at random times and their timing relative to suppression onset was later mapped onto
12
observers’ rivalry alternation data. In this design, probes could fall early or late in a rivalry phase
with equal probability. Plotting probe sensitivity within rivalry phases showed a striking
reciprocity: dominance performance was initially stable but declined late in the period, and
suppression performance was initially stable but improved in a complementary fashion late in
the period (Figure 5). The complementarity of these curves confirms the reciprocity of the
model, and their convergence late in the period confirms the role of adaptation in rivalry
dynamics.
Figure 5. Data showing a reciprocal change in contrast sensitivity for dominance
and suppression states during the course of a single (normalised) rivalry period. A
new ‘reverse correlation’ technique (Alais, Cass et al. 2010) allowed a critical
prediction of the reciprocal inhibition model of rivalry to be tested. Because of
neural adaptation, the neurons encoding the dominant stimulus should weaken
over the course of a single rivalry phase, producing a decline in contrast
sensitivity. At the same time, a weakening response in the dominant neurons
should increasingly free the suppressed neurons from inhibition and improve
contrast sensitivity in the suppressed eye towards the end of a rivalry phase. The
reciprocal pattern of contrast sensitivity changes observed in this study confirmed
for the first time the predictions of the reciprocal inhibition model of binocular
rivalry.
A study by van Ee (van Ee 2009) explored the role of noise in rivalry dynamics using a
computational model. A comparison was made between adding noise to the adapting
representation of the dominant stimulus or to the cross-inhibited neural activity. The intention
was to clarify whether the mutual inhibition process adapts, as has been suggested (Klink,
Brascamp et al. 2010), or whether it is the response to the dominant stimulus. Results showed
that adding noise to the cross-inhibition process did not produce typical rivalry dynamics, but
adding noise to the dominant response did. They suggest this reflects differing time scales.
13
Cross-inhibition is a fast process (millisecond scale) and no amount of noise perturbation
produces significant variations in dominance durations (typically lasting a second or so).
However, noise added to the adaptation of the dominant stimulus does produce typical rivalry
dynamics, showing that noisy adaptation within a reciprocal inhibition framework can account
for stochastic rivalry dynamics. This and related work by others has seen noise and adaptation
become key, interacting features in recent rivalry models (Brascamp, van Ee et al. 2006; Kim,
Grabowecky et al. 2006; Moreno-Bote, Rinzel et al. 2007; Kang and Blake 2011; Seely and Chow
2011; Roumani and Moutoussis 2012).
Another key characteristic of rivalry dynamics is that phase durations are significantly affected
by stimulus contrast (Mueller and Blake 1989; Lankheet 2006). Rivalry alternation rate reliably
increases as the contrast of both stimuli increases, with each stimulus perceived for shorter
periods on average. Within the reciprocal inhibition model, this is attributed to faster adaptation
arising from stronger neural responses to high-contrast stimuli. Interestingly, increasing the
contrast of only one stimulus will also increase alternation rates but in a curious way: increasing
one image’s contrast can slightly increase its dominance duration, but the main consequence is
to decrease the dominance duration of the other image (Levelt 1965; Mueller and Blake 1989;
Bossink, Stalmeier et al. 1993). This counterintuitive relationship is easily explained within the
framework of reciprocal inhibition where a given stimulus generates not an isolated response
but one linked to the response generated by the other, competing stimulus.
This underscores the distinction between overall rivalry alternation rate and the relative
durations of the dominance and suppression phases making up a rivalry cycle, which is referred
to as ‘predominance’. Rivalry predominance is measured by tracking rivalry alternations and
then calculating the proportion of time each image was visible. Alternation rate relates to the
period of a full rivalry cycle (i.e., dominance plus suppression duration), whereas predominance
effectively measures the duty cycle (the proportion of each phase relative to the cycle period).
Both measures are important, as a change in predominance of one stimulus over the other (e.g.,
from 50:50 to 70:30) could go unnoticed if only alternation rate were measured. This is an
important point for the following sections where we discuss how perceptual organization, as
manifest through a variety of contextual and top-down effects, influences rivalry dynamics. By
way of preview, contextual and top-down effects in rivalry generally affect the duration that a
given rival target is dominant, but less often when it is suppressed. This implies that perceptual
organization’s influence during rivalry operates primarily on the rival pattern already selected
for conscious awareness.
5. Top-down and contextual influences on binocular rivalry
5.1. Attention in binocular rivalry
The first top-down influence on rivalry we consider is attention, a concept closely linked to
rivalry over the years because both can be thought of as acts of selection. Attention involves
selecting among competing objects and rivalry could be interpreted as perceptual selection
between competing images. The role of attention in binocular rivalry has been debated since
the beginnings of experimental psychology. Von Helmholtz thought attention played a key role
14
and that rivalry alternations were under volitional control and easily manipulated by will. Hering
adopted a contrary position and considered rivalry to be driven by physiological processes
related to the stimuli. More than a century later, both positions have support. There is ample
evidence supporting Hering’s contention that basic stimulus properties such as contrast and
spatial frequency are important determinants of rivalry. In support of von Helmholtz, it is also
clear that attention can modulate aspects of rivalry such as alternation dynamics, dominance
durations, and selection of initial perceptual dominance. The key point, however, is that no act
of attention or will-power can arrest the alternations of rivalry so that a single image remains
dominant, undermining the notion that rivalry is completely synonymous with attentional
selection.
In more recent times, Lack was the first to systematically examine the role of attention in
binocular rivalry (Lack 1978). Lack found that attentional control over rivalry was generally
limited, although with training observers were better able to select and hold one stimulus. This
led to extended dominance durations (by about 20%) relative to a baseline condition, showing a
degree of endogenous or volitional control over rivalry (although much less than von Helmholtz
had suggested). In other experiments, Lack used spatial cueing to draw attention to the
dominant image, which extended its dominance duration, or to cue the suppressed stimulus,
which increased the likelihood of it becoming dominant. This established that exogenous
attention could also influence binocular rivalry. Other papers have confirmed that voluntary and
involuntary attention affect binocular rivalry. Ooi and He (1999) presented four targets to the
dominant eye and asked observers to attend to one. A transient signal in the suppressed eye,
which would normally trigger a dominance switch, was less likely to cause a switch when it
occurred at the attended location, compared to the three unattended locations. Voluntary
attention can therefore help maintain the ‘selected’ image despite transient exogenous stimuli.
These authors also used a monocular pop-out cue flanking a suppressed image to show that
involuntary attention directed to a suppressed stimulus could cause it to become dominant. In
related work, Paffen & Van der Stigchel (2010) presented rivalry at two locations and added an
exogenous cue around one of them, finding that alternations occurred earlier and more
frequently at the cued location, linking rivalry dynamics to the spatio-temporal properties of
visual attention. In other words, drawing attention to a spatial location increases the rate of
perceptual alternation at that location.
Object-based attention can also bias which image dominates in binocular rivalry. In one study
(Mitchell, Stoner et al. 2004), observers were first presented with two objects superimposed in
transparency that were binocularly viewed for a brief period before shutter glasses activated
and streamed them separately to the two eyes to trigger rivalry. Just before the rivalry stage,
one object was exogenously cued with a transient movement. This caused the cued object to
achieve perceptual dominance at rivalry onset and showed that an object selection made during
normal binocular viewing is maintained despite a change to rivalrous dichoptic viewing. A
subsequent study using different techniques drew the same conclusions (Chong and Blake
2006). Endogenous cuing, too, has been shown to produce a similar effect (Chong, Tadin et al.
2005; Klink, van Ee et al. 2008), although in both cases the cue’s influence in determining image
dominance is restricted to the early phase of rivalry, after which normal alternation dynamics
are observed. Studies with other kinds of perceptually bistable stimuli show similar modulatory
15
effects of attention (Struber and Stadler 1999; van Ee 2005) in that attention can bias which
percept tends to dominate, although several studies have found that attentional control over
rivalry is generally weaker than control over other forms of bistability (Meng and Tong 2004; van
Ee, van Dam et al. 2005).
These studies manipulated attention by selecting one of the perceptual alternatives, either
endogenously or exogenously. An alternative approach involves directing attention away from
the rival stimuli towards a peripheral secondary task. Paffen et al. used this method to show
that removing attention from the stimuli causes rival alternations to slow. The slowing effect
was graded, being stronger for a more difficult secondary task (Paffen, Alais et al. 2006), with
some evidence that alternations cease altogether when attention is completely removed from
rival stimuli (Brascamp and Blake, 2012). A similar paradigm was used to show that perceptual
alternations in bistable motion perception are also slowed by a difficult attentional distractor
(Pastukhov and Braun 2007). In a neuroimaging study examining the withdrawal of attention,
Lee et al. (2007) investigated rivalry between large images designed to produce a travelling
wave of dominance following a path of ‘good continuation’ along locally similar orientations.
With attention directed to the rival images, the traveling waves of perceptual dominance
produced corresponding waves of activity sweeping across retinotopic areas V1, V2 and V3.
However, when attention was diverted to a letter monitoring task at the center of the display,
activity in V2 and V3 no longer indicated a travelling wave and rivalry-related activity was
restricted to V1.
5.2. Interpretation and affect influence rivalry dynamics
As noted already, there is abundant evidence that low-level visual attributes impact on
binocular rivalry dynamics. Indeed, most reciprocal inhibition models described earlier assume
that rivalry transpires early in visual processing where inhibitory competition occurs between
local features signaled by monocular neurons. Several lines of evidence, however, have emerged
to show that seemingly “high-level” influences can govern the occurrence and dynamics of
rivalry, as can feedback from mid-level vision (Alais and Blake 1998; Watson, Pearson et al.
2004; Pearson and Clifford 2005; van Boxtel, Alais et al. 2008). Top-down approaches to rivalry,
in focusing on interpretation of ambiguous retinal input, broaden the scope of potential
influences on rivalry. We will focus here on results implicating high-level influences operating
during rivalry, for those results bear on the role of perceptual organization in governing rivalry
dynamics. We start by summarizing findings from a growing list of studies showing that the
meaning or emotional content of rivalry stimuli can influence rivalry dynamics.
The question of cognitive and motivational influences on rivalry goes back to the middle of the
previous century (reviewed by Walker 1978). In early studies, rival stimuli with conflicting
emotional or symbolic content were presented to different groups and predominance was
measured. When Jewish and Catholic observers viewed the star of David versus a Christian
cross, Jewish observers tended to see the star more than the cross, and vice versa for Catholic
observers (Losciuto and Hartley 1963). In a similar vein, figures a person had seen before tended
to predominate in rivalry over figures never seen before (Goryo 1969). These results were
interpreted to mean that non-visual factors such as affective content and familiarity influence
the resolution of stimulus conflict during binocular rivalry (Walker 1978). Recently, interest in
16
this question has returned with several new papers addressing this topic (reviewed by (Blake
2013)). For example, studies report that emotionally arousing pictures – whether positive or
negative – produce longer dominance durations than non-arousing pictures, even when both
images have comparable low-level image properties (Sheth and Pham 2008). Dominance
durations are also longer for emotional faces rivalling against neutral faces. An emotional face is
also more likely to dominate first at rivalry onset (Alpers and Pauli 2006). More remarkably,
neutral looking faces dominate significantly longer if they have previously been associated with
negative social behaviors through conditioning (“threw a chair at a classmate”), relative to faces
associated with positive or neutral behaviors (Anderson, Siegel et al. 2011). Even the simple act
of imagining a given stimulus can subsequently boost its dominance in rivalry, implying a boost
in stimulus strength from the act of imagining (Pearson, Clifford et al. 2008).
Top-down influences such as these are not too surprising given our knowledge that attention
can modulate rivalry durations (Lack 1978; Paffen, Alais et al. 2006): familiar, imagined or
emotionally charged stimuli may command greater attention and, hence, receive a boost in
rivalry. Accordingly, enhanced rivalry predominance could arise from lengthened dominance
durations, for it is presumably the dominant stimulus that receives attention during rivalry. Is
that the sole basis of context’s modulation of rivalry? To answer this, we turn to recent work
using a new procedure that isolates context’s influence on suppression durations. These new
studies all employ continuous flash suppression (CFS: Figure 6), a robust form of binocular
rivalry produced when one eye views a rapidly changing array of densely overlaid, high-contrast
shapes (the CFS inducer) and the other eye views a more conventional, static rival image
(Tsuchiya and Koch 2005). Because of the broadband spatio-temporal energy spectrum of the
CFS inducer (Yang and Blake 2012), it is always the initially dominant stimulus at rivalry onset,
and it remains dominant for an unusually long duration compared to rivalry produced by
conventional rival stimuli.
Figure 6. A schematic diagram of continuous flash suppression. A new approach to
producing strong interocular suppression, known as continuous flash suppression
(CFS), was developed by Tsuchiya and Koch (2005). As in binocular rivalry, CFS
involves presenting incompatible images to each eye, with the difference that in
CFS one of the eyes receives a dynamic sequence of random ‘Mondrian’ patterns
17
presented at a rate of ~10 Hz. The advantage of CFS is that it produces binocular
rivalry-like interocular suppression, but does so much more strongly and for far
longer periods than is typical in binocular rivalry. When the dynamic Mondrian
pattern is pitted against a static image, as shown in this figure, the dynamic
pattern may dominate for several 10s of seconds before the suppressed image
becomes visible.
Exploiting the robustness of CFS, recent studies have used a variant whereby the CFS inducer is
initially presented to one eye and a probe stimulus is presented to the other eye shortly after.
The predominance of CFS at onset prevents observers from seeing the probe at first, but probe
contrast is steadily increased until eventually the observer can indicate in which of four display
quadrants the probe appeared. In some cases, contrast of the CFS inducer is also gradually
decreased, to ensure the probe will eventually be perceived. The dependent measure is the
duration of suppression, the period from probe onset until successful reporting of the probe’s
location. Using this approach, several recent studies have asked what stimulus properties
empower an initially suppressed probe to overcome the potent suppression from the CFS
inducer. Whatever those properties turn out to be, they cannot be due to a boost from
attention because the identity and location of the suppressed probe remains unknown to the
observer until it emerges from suppression. Some examples of findings from these studies are:
• Upright faces emerge from suppression more quickly than inverted faces, as do words
printed in familiar script that can be read by an observer compared to words in
unfamiliar script (Jiang, Costello et al. 2007).
• Angry faces escape suppression faster than neutral or happy faces (Yang, Zald et al.
2007; Tsuchiya, Moradi et al. 2009).
• Faces implying direct eye contact break suppression faster than the same faces with
gaze slightly diverted (Stein, Senju et al. 2011).
• Scenes containing an object (e.g., a watermelon) in a bizarre context (a basketball game)
are freed from suppression faster than the same scenes with a contextually appropriate
object (e.g., a basketball) (Mudrik, Deouell et al. 2011).
Based on this kind of speeded emergence from suppression, most (but not all) of these studies
conclude that meaning, affective connotation and contextual relevance of suppressed stimuli
are still registered, despite being completely absent from visual awareness. At first glance, these
kinds of findings seem to rule out attention as the modulating factor in enhanced predominance
of certain stimuli engaged in rivalry. However, there are some reasons to take that conclusion
with a grain of salt. Two papers that used CFS together with emotional faces adopted a more
cautious tone by pointing to actual feature differences between faces that break suppression
early and those that do not (Yang, Zald et al. 2007; Gray, Adams et al. 2013). Also, the
investigators that documented gaze direction’s effect on dominance (Stein, Senju et al. 2011)
expressed in another paper doubt about the adequacy of control measures typically employed
to rule out alternative explanations (Stein, Hebart et al. 2011).
5.3. Rivalry in a multisensory context
Next we turn to studies that have asked whether sensory inputs from modalities other than
vision can influence binocular rivalry dynamics. As we live in a multisensory world, there are
18
many occasions when visual signals from the external environment are accompanied by
auditory or tactile signals (see chapter by Spence, this volume, for multisensory processing,
including a section on multisensory bistability). Psychophysical and neurophysiological evidence
shows the brain combines information across senses if it is likely to refer to the same stimulus
event (see recent reviews: Alais, Newell et al. 2010; Spence 2011). This helps achieve a more
veridical and less ambiguous percept, one of the main functions of cross-modal interactions
(Ernst and Bulthoff 2004). Recent results suggest multisensory signal combination can
significantly modulate rivalry dynamics. Specifically, a sound congruent with one of the rival
stimuli biases perceptual dominance towards that stimulus (Kang and Blake 2005; van Ee, van
Boxtel et al. 2009; Conrad, Bartels et al. 2010; Chen, Yeh et al. 2011), and rubbing a finger back
and forth over a tactile grating promotes dominance of a visual grating of matched orientation
(Lunghi, Binda et al. 2010; Lunghi and Alais 2013). Even smelling a distinctive odor while
experiencing binocular rivalry can bias dominance in favor of a congruent visual rival target
(Zhou, Jiang et al. 2010). The motor system, too, can influence binocular rivalry dynamics, as
evidenced by increased predominance when the motion of a rival stimulus is controlled by the
observer’s self-generated actions (Maruya, Yang et al. 2007). More broadly, motor and non-
visual sensory signals can bias other forms of visual bistability, including ambiguous motion
(Sekuler, Sekuler et al. 1997) and ambiguous depth perspective (Blake, Sobel et al. 2004).
One way that multisensory interactions can influence binocular rivalry is by boosting the degree
of attentional control over perceptual alternations. A recent multisensory study added two
different auditory signals to the rivalry stimulus, with one signal being congruent with one of the
visual stimuli (van Ee, van Boxtel et al. 2009). It was found that attentional control over rivalry
was augmented by a congruent auditory signal, relative to the non-congurent signal. The boost
to attentional control over rivalry was also shown with a congruent tactile signal. In a trimodal
experiment, a combination of both auditory and tactile congruency afforded even more
attentional control over binocular rivalry than either modality alone. This study shows that the
attentional resources involved in exerting voluntary control over binocular rivalry are central or
‘supramodal’, and squares with another study showing that attending to an auditory distractor
task slows binocular rivalry (Alais, van Boxtel et al. 2010), in the same way that attending to a
visual distractor slows rivalry (Paffen, Alais et al. 2006).
These multisensory influences in binocular rivalry demonstrate perceptual organization in
its full breadth, as information from all available sensory modalities is used in pursuit of a
coherent, disambiguated interpretation of the external world.