Content uploaded by Mark James Fenske
Author content
All content in this area was uploaded by Mark James Fenske on Mar 22, 2018
Content may be subject to copyright.
Martinez-Conde, Macknik, Martinez, Alonso & Tse (Eds.)
Progress in Brain Research, Vol. 155
ISSN 0079-6123
Copyright r2006 Elsevier B.V. All rights reserved
CHAPTER 1
Top-down facilitation of visual object recognition:
object-based and context-based contributions
Mark J. Fenske, Elissa Aminoff, Nurit Gronau and Moshe Bar
!
MGH Martinos Center for Biomedical Imaging, Harvard Medical School, 149 Thirteenth Street, Charlestown, MA 02129,
USA
Abstract: The neural mechanisms subserving visual recognition are traditionally described in terms of
bottom-up analysis, whereby increasingly complex aspects of the visual input are processed along a hi-
erarchical progression of cortical regions. However, the importance of top-down facilitation in successful
recognition has been emphasized in recent models and research findings. Here we consider evidence for top-
down facilitation of recognition that is triggered by early information about an object, as well as by
contextual associations between an object and other objects with which it typically appears. The object-
based mechanism is proposed to trigger top-down facilitation of visual recognition rapidly, using a partially
analyzed version of the input image (i.e., a blurred image) that is projected from early visual areas directly
to the prefrontal cortex (PFC). This coarse representation activates in the PFC information that is back-
projected as ‘‘initial guesses’’ to the temporal cortex where it presensitizes the most likely interpretations of
the input object. In addition to this object-based facilitation, a context-based mechanism is proposed to
trigger top-down facilitation through contextual associations between objects in scenes. These contextual
associations activate predictive information about which objects are likely to appear together, and can
influence the ‘‘initial guesses’’ about an object’s identity. We have shown that contextual associations are
analyzed by a network that includes the parahippocampal cortex and the retrosplenial complex. The
integrated proposal described here is that object- and context-based top-down influences operate together,
promoting efficient recognition by framing early information about an object within the constraints pro-
vided by a lifetime of experience with contextual associations.
Keywords: object recognition; top-down; feedback; orbitofrontal cortex; low spatial frequencies; visual
context; parahippocampal cortex; retrosplenial cortex; visual associations; priming
Successful interaction with the visual world de-
pends on the ability of our brains to recognize
visual objects quickly and accurately, despite infi-
nite variations in the appearance of objects and the
settings in which they are encountered. How does
the visual system deal with all of this information
in such a fluent manner? Here we consider the
cortical mechanisms and the type of information
that they rely on to promote highly efficient visual
recognition through top-down processes. The ev-
idence we review, from studies by our lab and
others, suggests that top-down facilitation of rec-
ognition can be achieved through an object-based
mechanism that generates predictions about an
object’s identity through rapidly analyzed, coarse
information. We also review evidence that top-
down facilitation of recognition can be achieved
through the predictive information provided by
contextual associations between an object or scene
!
Corresponding author. Tel.: +1-617-726-7467; Fax: +1 617-
726-7422; E-mail: bar@nmr.mgh.harvard.edu
DOI: 10.1016/S0079-6123(06)55001-0 3
and the other objects that are likely to appear
together in a particular setting. In the following
sections, we first consider each of these forms of
top-down enhancement of object recognition sep-
arately, and then consider how these object- and
context-based mechanisms might operate together
to promote highly efficient visual recognition.
An object-based cortical mechanism for triggering
top-down facilitation
The traditional view regarding visual processing is
that an input image is processed in a bottom-up
cascade of cortical regions that analyze increas-
ingly complex information. This view stems from
the well-defined functional architecture of the vis-
ual cortex, which has a clear hierarchical structure.
However, several models propose that both bot-
tom-up and top-down analyses can occur in the
cortex simultaneously (Grossberg, 1980;Kosslyn,
1994;Ullman, 1995;Desimone, 1998;Engel et al.,
2001;Friston, 2003;Lee and Mumford, 2003), and
recent evidence suggests that top-down mecha-
nisms play a significant role in visual processing
(e.g., Kosslyn et al., 1993;Humphreys et al., 1997;
Barcelo
´et al., 2000;Hopfinger et al., 2000;
Miyashita and Hayashi, 2000;Gilbert et al.,
2001;Pascual-Leone and Walsh, 2001;Mechelli
et al., 2004;Ranganath et al., 2004a). Neverthe-
less, the way in which such top-down processing is
initiated remains an important outstanding issue.
The crux of the issue concerns the fact that top-
down facilitation of perception requires high-level
information to be activated before some low-level
information.
Recently, Bar (2003) proposed a model that spe-
cifically addresses the question of how top-down
facilitation of visual object recognition might be
triggered. The gist of this proposal concerns a cor-
tical (or subcortical) ‘‘short-cut’’ of early informa-
tion through which a partially analyzed version of
the input image, comprising the low spatial fre-
quency (LSF) components (i.e., a blurred image), is
projected rapidly from early visual areas directly to
the prefrontal cortex (PFC). This coarse represen-
tation is subsequently used to activate predictions
about the most likely interpretations of the input
image in the temporal cortex. For example, if an
image of an oval blob on top of a narrow long blob
were extracted from the image initially (Fig. 1),
then object representations sharing this global pro-
file would be activated (e.g., an umbrella, a tree, a
mushroom, a lamp). Combining these top-down
‘‘initial guesses’’ with the systematic bottom-up
analysis could thereby facilitate visual recognition
by substantially limiting the number of object rep-
resentations that need to be tested.
Orbitofrontal involvement in top-down facilitation
of recognition
The proposal that object recognition might benefit
from a cortical ‘‘short-cut’’ of early, cursory in-
formation about the input image implies that there
should be an additional cortical region that shows
recognition-related activity before other object-
processing regions. The cortical regions most often
associated with visual object recognition are situ-
ated in the occipito-temporal cortex (Logothetis
et al., 1996;Tanaka, 1996). Within this region, the
fusiform gyrus and the lateral occipital cortex are
especially crucial for visual recognition in humans
Fig. 1. Schematic illustration of a cortical mechanism for trig-
gering top-down facilitation in object recognition (Bar, 2003). A
low spatial frequency (LSF) representation of the input image is
projected rapidly, possibly via the magnocellular dorsal path-
way, from early visual cortex to the prefrontal cortex (PFC), in
addition to the systematic and relatively slower propagation of
information along the ventral visual pathway. This coarse rep-
resentation is sufficient for activating a minimal set of the most
probable interpretations of the input, which are then integrated
with the relatively slower and detailed bottom-up stream of
analysis in object-processing regions of the occipito-temporal
cortex (e.g., fusiform gyrus).
4
(Kosslyn et al., 1995;Martin et al., 1996;Bar et al.,
2001;Grill-Spector et al., 2001;Malach et al.,
2002). The PFC has also been shown to be in-
volved in visual recognition (e.g., Bachevalier and
Mishkin, 1986;Wilson et al., 1993;Parker et al.,
1998;Freedman et al., 2001), and recent evidence
suggests that the orbitofrontal cortex (OFC) might
be specifically related to top-down visual process-
ing (Bar et al., 2001, 2006;Bar, 2003). Bar et al.
(2001), for example, used functional magnetic res-
onance imaging (fMRI) to compare the cortical
activity elicited by trials in which objects were
successfully recognized with that elicited by the
same pictures under identical conditions when they
were not recognized. As expected, this contrast
showed differential activity in the occipito-tempo-
ral regions previously associated with object rec-
ognition. However, successful recognition was also
associated with increased activity in a site within
the OFC (Fig. 2). The connection between activity
in the OFC and successful object recognition
makes it a prime candidate for being involved in
top-down facilitation of visual recognition.
Rapid recognition-related activity in orbitofrontal
cortex
If the recognition-related region of the OFC iden-
tified in the Bar et al. (2001) study is critical for
early top-down facilitation, then recognition-
related activity should develop in this site before
other object-processing regions of occipito-tempo-
ral cortex. To test this prediction, Bar et al. (2006)
used the same object recognition task as Bar et al.
(2001) while obtaining magnetoencephalography
(MEG) recordings, which provide millisecond-
resolution measurements of cortical activity. As
predicted, the contrast between recognized and
not-recognized trials in this study revealed differ-
ential activation in the OFC 50 ms before it devel-
oped in the object-processing regions of the
occipito-temporal cortex (Fig. 3). Moreover, a
time-frequency, trial-by-trial covariance analysis
of the MEG data demonstrated strong synchrony
between occipital visual regions and the OFC at a
relatively early stage (beginning at approximately
80 ms after stimulus onset), and a strong sync-
hrony between the OFC and the fusiform gyrus
activity at a relatively later stage (130 ms after
stimulus onset). Taken together, these results pro-
vide strong converging support for the proposal
that this frontal region plays a critical role in top-
down facilitation of object recognition.
Early orbitofrontal activity is triggered by low
spatial frequencies
The finding that early activity in the OFC is re-
lated to successful object recognition is consistent
with the model of top-down facilitation proposed
by Bar (2003). This model posits that early activity
in the OFC is related to the direct projection from
early visual areas of an LSF representation of the
input image (Bar, 2003). The projection of this
early and rudimentary information to the OFC
would thereby allow it subsequently to sensitize
the representation of the most likely candidate
objects in the occipito-temporal cortex as ‘‘initial
guesses.’’ The possibility that early OFC activity is
driven by LSF information is supported by phys-
iological findings that the magnocellular pathway
conveys LSF information (i.e., a blurred image)
early and rapidly (Shapley, 1990;Merigan and
Maunsell, 1993;Bullier and Nowak, 1995;Grab-
owska and Nowicka, 1996;Leonova et al., 2003),
and by evidence from anatomical studies showing
bidirectional connections between early visual
Fig. 2. A group averaged statistical activation map showing
recognition-related cortical activity (adapted from Bar et al.,
2001) on the left hemisphere (ventral view) of a brain that has
been ‘‘inflated’’ to expose both sulci (dark gray) and gyri (light
gray). Successful object recognition produced greater activity
than unsuccessful recognition attempts in inferior regions of
both the occipito-temporal (e.g., fusiform gyrus) and the frontal
(e.g., orbitofrontal cortex) lobes.
5
areas and the prefrontal cortex (e.g., humans:
Oenguer and Price, 2000; macaque: Rempel-Clower
and Barbas, 2000) (although information about di-
rect connections between occipital visual areas and
the OFC is still lacking). These findings suggest that
the neural infrastructure is indeed in place for LSF
information to trigger prefrontal activity subserving
top-down facilitation.
To test the prediction that early recogni-
tion-related activity in the OFC should depend
on the availability of LSF information in the im-
age, Bar et al. (in press) used both fMRI and MEG
to compare activity in the OFC elicited by filtered
images of objects (Fig. 4) containing pre-
dominantly LSFs or predominantly high spatial
frequencies (HSFs). They found that the same re-
gion of the OFC that shows early recognition-re-
lated activity (Fig. 3) also shows early differential
activity for LSF vs. HSF images. This differential
activity associated with spatial frequency content
was found to peak around 115 ms from stim-
ulus onset. Moreover, a trial-by-trial covariance
analysis of the MEG data indicated that there
was a clear interaction between the occipital visual
regions and the OFC, and between the OFC
and the fusiform gyrus for LSF images, as sug-
gested by phase synchrony, but no significant
synchrony existed between these regions for the
HSF stimuli.
Taken together, the results reviewed in this sec-
tion provide strong support for the proposal that
visual object recognition benefits from the rapid
analysis and projection of coarse LSF information
of an input image to orbitofrontal regions of pre-
frontal cortex. Importantly, this object-based
mechanism allows early information about an ob-
ject to act as the catalyst for top-down enhance-
ment of its own bottom-up cortical processing.
A context-based cortical mechanism for triggering
top-down facilitation
In addition to the top-down benefit provided
by prefrontal analysis of cursory object informa-
tion, recognition efficiency can be increased
through processes that take advantage of natu-
rally occurring regularities in the environment
(Gibson, 1969). A lifetime of visual experience can
thereby guide expectations about which objects are
likely to appear in a given setting to aid subsequent
recognition of those objects through their contex-
tual associations (Biederman, 1972, 1981;Palmer,
1975;Biederman et al., 1982;Bar and Ullman,
1996;Davenport and Potter, 2004). A beach
setting, for example, typically contains objects
such as beach chairs and beach umbrellas. Recog-
nizing a scene with ocean surf breaking on the
Fig. 3. Cortical dynamics of object recognition. Bar et al. (2006) found that recognition-related activity in the orbitofrontal cortex
(OFC) precedes that in the temporal cortex. Anatomically (MRI) constrained group averaged statistical activation maps were cal-
culated from MEG at three different latencies from stimulus onset. Recognition-related activation (recognized vs. not-recognized)
peaked in the left OFC 130 ms from stimulus onset, 50 ms before it peaked in recognition-related regions in the temporal cortex.
6
sand, combined with this prior knowledge about
the scene’s typical content, may act to rapidly
sensitize the representations of these contextually
related objects and facilitate their subsequent rec-
ognition. Indeed, several studies have shown that
scenes can be recognized in a ‘‘glance’’ and
that their content is extracted very rapidly (Bieder-
man et al., 1974;Schyns and Oliva, 1994;Oliva
and Torralba, 2001), lending support to the
notion that early analysis of the surrounding
contextual information may facilitate object rec-
ognition. Other studies have directly examined
the interaction of scene and object processing,
showing that objects are recognized faster when
located in a typical environment than in an un-
usual setting. Palmer (1975), for instance, showed
that recognition of briefly presented objects fol-
lowing the presentation of a contextual scene was
significantly more accurate when the target object
was contextually consistent with the preced-
ing scene (e.g., kitchen — bread) than when it was
inconsistent (e.g., kitchen — mailbox). In an-
other series of experiments, Biederman et al. (1982)
demonstrated that violating various types of
contextual relations (e.g., relative position or size)
between an object and its environment hinders
subjects’ ability to detect the object. Similar
effects were reported by Davenport and Potter
(2004). Taken together, these findings suggest that
objects and their settings are processed interac-
tively, supporting the view that contextual infor-
mation directly facilitates the perceptual processes
involved in visual object recognition (but see Hol-
lingworth and Henderson, 1998, for a different
perspective).
Cortical analysis of contextual information
To understand how cortical analysis of contextual
associations might generally facilitate perception,
and object recognition in particular, it is useful to
first consider how the brain analyzes these associ-
ations. Bar and Aminoff (2003) recently addressed
this issue in a series of fMRI studies. They iden-
tified those regions of the brain that are sensitive to
contextual information by comparing cortical ac-
tivity during recognition of individual objects with
strong contextual associations with that for objects
with weak contextual associations. For example,
the association of a toaster with a kitchen setting is
relatively stronger than the association of a camera
with the various contexts in which it tends to oc-
cur. Bar and Aminoff found that such objects with
strong contextual associations elicited greater ac-
tivity than those with weak contextual associa-
tions, and did so primarily in the parahippocampal
cortex (PHC) and the retrosplenial complex
(RSC).
1
They found similar context-related acti-
vation in these regions for each subject and across
several related experiments, showing clearly that
the PHC and the RSC comprise a cortical ‘‘context
Fig. 4. Examples of the visual stimuli used in the Bar et al. (2006) MEG and fMRI studies. Unfiltered stimuli (middle) were used as
intact controls for comparison with activity for low filtered (left) and high filtered (right) stimuli.
1
The cortical region we refer to as the retrosplenial complex
extends beyond the callosal sulcus, which is where recent cyto-
architectonic investigations have shown the retrosplenial cortex
to be almost entirely located (Vogt et al., 2001). The re-
trosplenial region that is involved in contextual analysis (Bar
and Aminoff, 2003) has a broader extent that is typically cen-
tered near the junction of the callosal sulcus, the precuneus, the
caudal-ventral portion of the posterior cingulate gyrus, and the
ventral-medial portion of the subparietal sulcus.
7
network’’ that processes contextual associations
during object recognition.
In addition to this newly established role in
processing contextual associations (Bar and Amin-
off, 2003), the PHC and RSC have both previously
been shown to mediate the processing of spatial,
place-related information (e.g., Aguirre et al.,
1996, 1998;Epstein and Kanwisher, 1998;Levy
et al., 2001). Indeed, a region within the PHC
typically shows a preferential response to pictures
of scenes and topographical landmarks, and has
therefore been termed the parahippocampal place
area (PPA, Epstein and Kanwisher, 1998). Inter-
estingly, in Bar and Aminoff’s (2003) studies, the
same region of PHC showed robust context-
related activity for individual objects presented in
isolation (see Fig. 5). How could individual objects
activate a region of the cortex that is typically
sensitive to scenes and landmarks? One possibility
is that the perception of a strongly contextual ob-
ject (e.g., a bowling pin) indirectly activates the
representation of the place with which it is asso-
ciated (e.g., a bowling alley). However, it is also
possible that the PHC and RSC process contextual
associations more generally. Thus, while places
and landmarks may often correspond with specific
sets of associations, context-related activity in the
PHC and RSC might not be restricted to these
highly spatial, place-specific contexts. This issue
directly concerns the nature of the information
that is processed by the context network.
The nature of the information processed by the
context network
Does context-related activity in the PHC and RSC
reflect the processing of contextual associations in
general, or the processing of particular associa-
tions with spatial, place-specific contexts? To dis-
tinguish between these alternatives, Bar and
Aminoff (2003) used objects with weak contextual
associations as a baseline and then compared the
relative increase in fMRI activation for objects
with strong contextual associations with spatial,
place-related contexts (e.g., farm) or more abstract
nonspatial contexts (e.g., romance). As before, the
PHC and RSC produced significantly greater ac-
tivity for strongly contextual objects than for
weakly contextual objects. Importantly, robust
context-related activity was found for both types
of contexts (Fig. 6). These results provide strong
support for the hypothesis that the context
network mediates the processing of contextual
Fig. 5. Group averaged statistical activation maps (medial view) showing greater fMRI response in the parahippocampal cortex (PHC)
and the retrosplenial complex (RSC) to individual objects with strong contextual associations (strong context) than to objects with
weak contextual associations (weak context). The black outlines in the PHC show the boundaries of the parahippocampal place area
(PPA), which was defined in a separate set of subjects by comparing cortical activation for indoor and outdoor scenes with that for
objects, faces, and scrambled items (i.e., as in Epstein et al., 2003). Robust context-related activity occurs within the PPA for strong
context vs. weak context individual objects.
8
associations in general, and that the representation
of contextual associations in these regions is not
confined to spatial, place-related contexts. Finding
robust activation for objects from nonspatial con-
texts clearly shows that context-related activity in
the PHC in response to individual objects is not
merely a reflection of the indirect activation of the
specific places with which they are associated.
Bar and Aminoff’s (2003) comparison of corti-
cal activity elicited by objects associated with spa-
tial vs. nonspatial contexts revealed an interesting
characteristic of the functional organization of the
PHC. That is, the activity elicited by objects as-
sociated with nonspatial contexts was found to be
confined to the anterior half of the PHC, while
activity elicited by the objects associated with spa-
tial contexts extended into the posterior half of the
PHC (Fig. 6). This pattern of activation suggests
that the representation of contextual associations
in the PHC is organized along a hierarchy of in-
creasing spatial specificity for more posterior rep-
resentations. Additional evidence for this type of
hierarchical organization has since been found in
subsequent studies showing similar differences in
the sensitivity of posterior and anterior regions of
the medial temporal lobe to spatial information
(Du
¨zel et al., 2003;Jackson and Schacter, 2004;
Pihlajamaki et al., 2004). This characteristic pat-
tern of cortical activity during contextual analysis
indicates that the nature of the information proc-
essed by the context network involves both spatial
and nonspatial associations.
The involvement of spatial information in con-
textual representations might go beyond mere as-
sociations, and extend to representing the typical
spatial relations between objects sharing the same
context. A traffic light, for example, is usually lo-
cated relatively high above a street and is typically
fixed to a standard that rises from the corner of a
sidewalk, with crosswalk markings and pedestrians
below it, and cars driving underneath or stopped in
front of it. Are these spatial relations represented in
the associations for spatial contexts? The results of
several behavioral studies indicate that this type of
spatial information can indeed be associated with
an object’s representation and thereby impact its
recognition (Biederman, 1972, 1981;Palmer, 1975;
Biederman et al., 1982;Bar and Ullman, 1996). Bar
and Ullman (1996), for example, demonstrated
that spatial contextual information is important
when identifying an otherwise ambiguous object.
Thus, a squiggly line that cannot be identified in
isolation can readily be identified as a pair of eye-
glasses when presented in the appropriate spatial
configuration with a hat (Fig. 7). These results
support the hypothesis that spatial relations be-
tween objects are part of contextual representa-
tions that bind information about typical members
of a context as well as the typical spatial relations
between them, termed context frames.
Disentangling the processing and representation
of spatial and nonspatial contextual associations is
complicated by the fact that real-world objects are
encountered in specific places and spatial loca-
tions, even those that are otherwise most strongly
associated with nonspatial contexts. For example,
while a heart-shaped box of chocolates might be
Fig. 6. Group averaged statistical activation maps (left hem-
isphere, medial view) showing greater fMRI response for ob-
jects strongly associated with spatial and nonspatial contexts
than for objects with weak contextual associations. Bar graphs
show the results of a region of interest analysis for the PHC.
The magnitude of fMRI signal change was largest for spatial
context objects in posterior PHC and for nonspatial context
objects in anterior PHC.
9
most strongly related to the context of romance,
people may also associate it with the store where it
was sold or the room where the chocolates were
eaten. Nevertheless, if the PHC is sensitive to both
spatial and nonspatial associations through visual
experience, such sensitivity should also develop in
highly controlled situations using novel, unfami-
liar contextual associations.
This type of experimental approach can maximize
control over spatial and nonspatial contextual as-
sociations, and will be important for further exam-
ining the nature of the contextual information that
is processed and represented in the PHC (Aminoff
et al., submitted). For example, contextual associ-
ations could be formed through an extensive train-
ing phase in which subjects are repeatedly exposed
to stimuli from each of three different conditions:
spatial, nonspatial, and nonassociative (Fig. 8A).
Under the spatial condition, individual groups
of three shapes would be shown together in a con-
sistent spatial configuration and with each shape
always appearing in the same display location. In-
dividual groups of three shapes would also be
shown together under the nonspatial condition, but
in a random spatial configuration on each trial.
Finally, under a nonassociative condition, single
shapes would be presented individually in a ran-
domly determined location on each trial. These in-
dividual shapes would therefore not be associated
with any other shape, nor any particular location.
Consequently, these individual shapes could serve
as control stimuli, analogous to the weak contextual
objects used in the previous experiments.
After such a training session, fMRI could be
used to detect differences in cortical activity for the
contextual shapes relative to that for the indivi-
dual, nonassociative shapes. To do this, a single
shape from the previous training session could be
presented during each trial in the scanner. Thus,
the only difference for these individual shapes
from the different conditions during scanning
would be the prior experience the subject had with
each shape (see Fig. 8B). In line with our previous
results, we would expect the results of such a study
to reveal a robust fMRI response in both the PHC
and the RSC when subjects viewed shapes with
newly formed contextual associations (spatial and
nonspatial) relative to that for the nonassociative
control stimuli. Importantly, the same anterior-
to-posterior progression in spatial specificity in
context-related activity of the PHC would be evi-
dent if the magnitude of the fMRI response were
greatest in the posterior portion of the PHC for
shapes from the spatial condition, and greatest in
the anterior portion of the PHC for shapes from the
nonspatial condition. Finding a similar anterior-to-
posterior progression of activity in the PHC for
nonspatial and spatial associations using otherwise
Fig. 8. Isolating spatial and nonspatial components in the for-
mation of contextual associations. (A) During training, groups
of three colored abstract shapes (examples shown in grayscale)
are repeatedly presented together in the same specific spatial
locations (spatial) or in random locations (nonspatial). A non-
associative control condition (individual) is formed by present-
ing single objects in random locations. (B) fMRI trials involve
individual presentations of a single shape from each training-
session display, regardless of condition.
Fig. 7. Example of the importance of spatial context in object
recognition. An ambiguous line (A) cannot be interpreted un-
equivocally in isolation, but (B) is readily identified as a pair of
eyeglasses when shown in the appropriate location below a
drawing of a hat.
10
novel abstract stimuli would provide converging
evidence that the representation of contextual in-
formation in the PHC is organized along a hierar-
chy of spatial specificity.
The general conclusion drawn from the results
of the experiments reviewed in this section is that
the PHC and RSC together mediate the processing
of contextual associations, and that the nature of
these associations involve both spatial and non-
spatial information. Characterizing the role of
these regions in associative processing provides a
critical foundation for exploring the cognitive and
neural mechanisms underlying contextual facilita-
tion of object recognition. The evidence that the
PHC and RSC mediate associative processing also
provides a framework for bridging previous find-
ings about the function of the PHC and RSC.
Specifically, in addition to contextual analysis, the
PHC and RSC have both been implicated in the
processing of spatial or navigational information
(Aguirre et al., 1998;Epstein and Kanwisher,
1998;Maguire, 2001), as well as in episodic mem-
ory (Valenstein et al., 1987;Brewer et al., 1998;
Wagner et al., 1998;Davachi et al., 2003;Morcom
et al., 2003;Wheeler and Buckner, 2003;Kirwan
and Stark, 2004;Ranganath et al., 2004b). Thus,
at first glance it would seem that the same cortical
regions have been assigned several different and
seemingly unrelated functions. This potential con-
flict is resolved, however, when one considers that
all of these processes rely on associations as their
building blocks. Place-related and navigational
processing requires analysis of the associations
between the objects and landmarks appearing in a
specific place, as well as the spatial relations
among them. Episodic memories also rely on fa-
miliar associations. For example, an episodic
memory of last night’s dinner is a conjunction of
associated constituents (e.g., the location, the com-
pany, the food, the dishes, etc.). Thus, describing
the key role of the PHC and RSC in terms of as-
sociative processing, rather than limiting this role
to spatial or episodic information, provides a com-
mon foundation from which various functions can
be derived. As we discuss below, the ability of
these regions to rapidly process such associations
is critical for triggering top-down contextual facil-
itation of visual object recognition.
Contextual facilitation of object recognition
In the preceding sections we have considered evi-
dence that object recognition can be facilitated in
the presence of contextually related information.
We have also described the cortical network that is
specifically involved in analyzing contextual asso-
ciations. But how does the processing of contextual
associations lead to facilitated visual recog-
nition? We propose that the contextual associa-
tions analyzed during the recognition of an object
or scene are represented and activated in a corre-
sponding ‘‘context frame’’ in the PHC. Context
frames can be considered to be contextual repre-
sentations (or schema) that include information
about the objects that typically appear within that
context (see also Bar and Ullman, 1996;Bar, 2004).
The precise nature of the information represented
in context frames is still being explored, but the
results reviewed above suggest that it includes both
spatial and nonspatial aspects of contextual rela-
tions. In this way, the PHC might serve as a
switchboard of associations between items that are
represented in detail elsewhere (Bar, 2004). The
specific associations mediated by an activated con-
text frame might act to sensitize, or prime, repre-
sentations of other contextually related objects
(Fig. 9). Spreading activation of contextually re-
lated object representations might thereby help to
facilitate subsequent recognition of such objects as
a function of their contextual associations. Inter-
estingly, while the functional role of such efficient
representation of contextual associations is hy-
pothesized to benefit perception, evidence from
visual false memory studies suggests that certain
memory distortions can arise as a byproduct of
activating these associations. For example, when
tested for their memory of a previously shown pic-
ture, subjects in these studies often ‘‘remember’’
having seen objects that are contextually related to
the picture but that were not actually in the picture
(Miller and Gazzaniga, 1998).
Bar and Aminoff’s (2003) studies showed that
robust context-related activity can be elicited by a
single key object from a particular context. This
finding suggests that recognizing a single object
might be sufficient to activate a corresponding
context frame, prime the associated representations
11
and thereby facilitate recognition of the other con-
textually related objects. The hypothesis that initial
recognition of even a single strongly contextual
object can facilitate subsequent recognition of a
different, but contextually related, object was tested
directly by Fenske et al. (submitted). While their
study focused primarily on the neural substrates of
contextual facilitation, it is interesting to note that
no prior study had addressed even the behavioral
characteristics of this type of object-to-object con-
textual priming using foveal presentations (cf.,
Henderson et al., 1987;Henderson, 1992).
Fenske et al. (submitted) presented individual
objects with strong contextual associations within
a visual priming paradigm, and asked subjects to
make a simple size judgment about each object.
Recognition improvement due to contextual prim-
ing effects was assessed by comparing response
times (RTs) and the corresponding change in
fMRI signal for objects that were immediately
preceded by a contextually related object relative
to that for objects from a new unprimed context
(Fig. 10). The important result of this study was
robust contextual facilitation of recognition-
related RTs and corresponding fMRI response re-
ductions in both context- and object-processing
regions. Cortical response reductions were found
in bilateral PHC and in the left anterior fusiform,
lateral occipito-temporal, and inferior frontal cor-
tices. At the behavioral level, these findings repli-
cate previous studies showing more efficient and
rapid recognition of contextually related, than un-
related, items. At a neural level, the novel and im-
portant finding of the Fenske et al. study was the
robust cortical response reduction obtained in the
context network and object-processing regions.
Such cortical response reductions are important,
as traditional priming studies have shown them to
be a hallmark of enhanced recognition following
prior exposure to the same, or a related, stimulus
(Schacter and Buckner, 1998;Koutstaal et al.,
2001;Vuilleumier et al., 2002;Henson, 2003;
Simons et al., 2003;Lustig and Buckner, 2004;
Maccotta and Buckner, 2004;Zago et al., 2005).
Fig. 9. A model of how contextual associations in the PHC might activate visual representations of contextually related objects in
inferior temporal cortex to facilitate object recognition (adapted from Bar, 2004). Individual objects (e.g., a straw hat) can be
associated with multiple context frames (e.g., a beach, a store, a closet). The experience-based set of associations represented in a
specific context frame is activated as soon as the context has been established through recognition of a strongly contextual object (e.g.,
a palm tree) or through other contextual cues from the scene.
12
Finding a response reduction in a primary com-
ponent of the context network (i.e., PHC) supports
the hypothesis that context-related facilitation of
object recognition results from an enhanced
processing of an object’s contextual associations.
Likewise, the response reduction seen in the left
anterior fusiform gyrus implies that encountering
a strongly contextual object activates representa-
tions of other contextually related objects (such as
the contextually related targets), while the re-
sponse reduction in the lateral occipital cortex, an
area implicated in the processing of perceptual
features, reflects activation of specific perceptual
representations associated with these target ob-
jects. Finally, the response reduction in left inferior
frontal cortex presumably reflects sensitization of
semantic associations between context-related ob-
jects. The selective contextual priming effects
found in these specific regions therefore provide
important insight into the types of representations
involved in the contextual facilitation of visual
recognition. Importantly, these results suggest that
there are many different types of information that
are connected by the proposed context frames
stored in the PHC, but that the various types of
representations (perceptual, conceptual, etc.) are
stored elsewhere.
The results of the Fenske et al. (submitted) study
provide support for our proposal that contextual
analysis during initial recognition of a highly con-
textual object serves to activate a corresponding
context frame and, through the inherent associa-
tive connections, sensitizes the representations of
other contextually related objects. Recognition is
therefore enhanced when these objects whose rep-
resentations have been primed in this manner are
subsequently encountered. However, objects rarely
appear in isolation, so it makes sense that context
frames might also be activated by the information
contained in a scene image, per se. Indeed, just as a
coarse, LSF representation of an input image is
often sufficient for rapid object recognition, the
LSF content in a scene image might also be suffi-
cient for deriving a reliable guess about the context
frame that needs to be activated. The work of
Schyns, Oliva, and Torralba (Schyns and Oliva,
1994;Oliva and Torralba, 2001) supports the pro-
posal that LSF information from a typical scene is
Fig. 10. Contextual facilitation of object recognition. Objects preceding repeated-context target items were always contextually related;
objects preceding new-context items were never contextually related. After each picture presentation, participants were required to
respond ‘‘bigger than a shoebox,’’ or ‘‘smaller than a shoebox,’’ by a key-press. Contextual priming was reflected by faster response
times (RT) and reduced cortical fMRI response for repeated-context objects than for new-context objects. Contextual priming related
fMRI response reductions occurred in the parahippocampal component of the context network, and in object-related regions of
occipito-temporal and inferior frontal cortices.
13
sufficient for a successful categorization of its con-
text (e.g., a street, a beach), and the statistics of
these images can help produce expectations about
what and where is likely to appear in the scene.
This work converges with psychophysical and
computational evidence that LSFs are extracted
from scene images earlier than HSFs (Palmer,
1975;Metzger and Antes, 1983;Bar, 2003). The
proposal that emerges from this is that contextual
facilitation of object recognition, just like object-
based top-down facilitation, can be triggered rap-
idly on the basis of early cursory analysis of an
input scene or object image.
While the results of the Fenske et al. (submitted)
study clearly show that recognition of a strongly
contextual object can facilitate the subsequent rec-
ognition of other contextually related objects, and
that this contextual priming effect is associated
with changes in activity in the PHC component of
the context network, this study does not address
the relative top-down influence of different types
of contextual associations (e.g., spatial vs. non-
spatial) on subsequent recognition. Nevertheless,
the finding that contextual facilitation of object
recognition is also associated with cortical re-
sponse reductions in cortical regions associated
with different aspects of object processing (e.g.,
perceptual, semantic, etc.) suggests that contextual
facilitation may indeed be mediated through var-
ious types of object-related representations. To
maximize contextual facilitation, the context
frames that maintain the corresponding associa-
tions must therefore accommodate various types
of representations. How are context frames organ-
ized for this to be achieved? We consider this
question in the following section.
Contextual facilitation and the representation of
spatial and nonspatial associations
The context network is proposed to mediate the
processing and representation of various types of
contextual associations through context frames.
The nature of these context frames and how they
represent different types of associations remains
an open and exciting area for future research. Bar
and Ullman (1996) suggested that contextual
frames contain information about not only the
identity of objects that tend to co-occur in scenes
but also their typical spatial relations. Indeed, our
experiments have repeatedly shown that the PHC
component of the context network is involved in
the analysis of both spatial and nonspatial asso-
ciations during recognition of individual objects
(e.g., Bar and Aminoff, 2003). In addition, there is
clear behavioral evidence that spatial contextual
information can facilitate object recognition.
Several studies have shown that objects are recog-
nized more efficiently and accurately when located
in contextually consistent (expected) locations
than in contextually inconsistent (unexpected) lo-
cations (Biederman, 1972;Mandler and Johnson,
1976;Hock et al., 1978;Biederman, 1981;Bieder-
man et al., 1982;Cave and Kosslyn, 1993;Bar and
Ullman, 1996;Chun and Jiang, 2003). For in-
stance, when viewing a scene of an office, a table
lamp would typically be recognized faster when
appearing on top of a desk than when appearing
beneath it, suggesting that prior knowledge about
the spatial constraints of objects in scenes contrib-
utes to object recognition.
However, it remains unclear whether nonspatial
(e.g., semantic) information about the specific
identities of objects in scenes is indeed integrated,
as Bar and Ullman (1996) have suggested, with
knowledge about the spatial relations between
these objects. In other words, does context-based
top-down facilitation of object recognition rely on
spatial and nonspatial associations that are linked
within unified context frames or that are repre-
sented independently? This question is critical for
understanding the nature of contextual represen-
tation and the mechanisms through which associ-
ative information influences visual perception.
Consider that while spatial analysis is an inherent
part of visual processing and the interpretation of
real-life scenes, it may nevertheless rely on differ-
ent forms of cortical representation than those in-
volved in the analysis of the nonspatial content of
these scenes. That is, nonspatial and spatial con-
text effects may in fact be mediated by separate,
independent mechanisms. Thus, when viewing a
desk in an office, a nonspatial context frame may
generate predictions regarding the semantic infor-
mation in the scene (e.g., the likelihood of certain
14
objects to appear with the desk), sensitizing the
representation of a ‘‘lamp’’ regardless of its spatial
position relative to the desk. At the same time, an
independent spatial context frame might constrain
the spatial relations between these objects, en-
hancing the representation of the lamp’s location,
regardless of any particular object identity (i.e.,
based on the knowledge that objects are typically
positioned on top of desks and not beneath them).
An alternate view is that context frames contain
unified representations for both nonspatial and
spatial contextual information. Thus, when view-
ing a desk in an office, the representation of a desk
lamp would necessarily include information about
its expected location on top of the desk because of
the strong connections between this type of spatial
and nonspatial knowledge within the contextual
schema of an ‘‘office.’’ According to this view, the
contextual representation of a particular type of
scene is maintained as a single unit (i.e., a single
context frame). Any violation of the spatial rela-
tions between objects in the scene should also im-
pact analysis of the nonspatial associations
between these objects (e.g., their expected identi-
ties) and the interpretation of the scene as a whole.
Thus, seeing a desk lamp under a desk would not
only violate expectations regarding the most prob-
able location of objects in general (as objects typ-
ically appear on top of desks), but also violate
expectations regarding the specific identity or role
of the lamp (as desk lamps appear, by definition,
on desks). In other words, in a unified context
frame, both spatial and nonspatial knowledge
jointly contribute to visual comprehension, and
thus any spatial inconsistencies between objects
may affect the interpretation of these objects’
identity, or meaning, altogether.
The hypothesis that nonspatial and spatial con-
textual associations are represented independently
is supported by evidence that an object’s identity is
typically analyzed through a ventral anatomical
route including inferior occipito-temporal cortex,
while its spatial location is typically analyzed
through more dorsal fronto-parietal brain regions
(Goodale and Milner, 1992). The existence of dis-
tinct neural systems for analysis of object’s identity
and spatial location raises the possibility that
different cognitive mechanisms underlie facilitation
of object recognition through nonspatial and spa-
tial contextual associations. Indeed, one interpre-
tation of the results of Fenske et al. (submitted)
study is that the recognition of contextually related
objects is facilitated by a priming mechanism that
sensitizes the nonspatial representations of objects
that are strongly associated with a contextual cue.
In contrast, the results of studies investigating spa-
tial contextual facilitation suggest that recognition
of ‘‘properly’’ positioned objects is facilitated by a
spatial attention mechanism that enhances the
processing of information at a specific location
(Hollingworth and Henderson, 1998;Chun and
Jiang, 2003;Gordon, 2004). Prior experience with a
specific scene or fixed spatial configuration, for ex-
ample, can guide spatial attention to the location
that is most likely to contain a target of interest,
thereby facilitating visual search and target recog-
nition within that scene or configuration (Chun
and Jiang, 2003).
The two views presented here concerning the
underlying structure of context frames and the
representation of spatial and nonspatial contextual
associations (i.e., unified vs. independent represen-
tations) should not necessarily be considered to be
mutually exclusive. Indeed, contextual facilitation
of object recognition might ultimately be found to
involve the top-down influence of both independ-
ent and unified representations of spatial and non-
spatial contextual associations. Investigating this
possibility will require a study that examines the
effects of both nonspatial and spatial contextual
factors on subsequent object recognition. To de-
termine whether the two types of associative
knowledge operate within separate contextual rep-
resentations, one needs to test whether they can
influence perception simultaneously without inter-
acting. Thus, most importantly, an orthogonal de-
sign is required in which the two factors are
manipulated independently, such that the unique
effect of each factor, as well as their joint (inter-
active) effects on object recognition, can be directly
assessed (cf., Sternberg, 2001). To the extent that
the two types of associative knowledge are linked
within a combined context frame, we would antic-
ipate an interaction in reaction times to target ob-
jects. Specifically, one might expect the benefit of
identifying an object in a contextually consistent
15
location relative to a contextually inconsistent lo-
cation to be significantly greater for objects whose
identities have been contextually primed than for
contextually unprimed objects, suggesting a unified
representation for both nonspatial and spatial fac-
tors. In this situation, a similar interaction might
also be expected for corresponding cortical activity
in brain regions associated with contextual process-
ing (i.e., the PHC and/or RSC) and in object-re-
lated processing areas (e.g., fusiform gyrus), with
the largest differential cortical activation when the
target is consistent with both spatial and nonspatial
contextual information. If, however, nonspatial
and spatial contextual representations operate in-
dependently, their contribution to response late-
ncies, as well as to brain activation, will be
additive. In addition, a possible anatomical dis-
tinction within the contextual network may be
found between the two types of representations (as
found previously with spatial and nonspatial asso-
ciations, Bar and Aminoff, 2003). With such an
outcome, one could conclude that the two types of
information reside in separate representational
‘‘stores,’’ suggesting dissociable influences in con-
text-based top-down facilitation of object recogni-
tion. Finally, because spatial and nonspatial
contextual associations may involve the top-down
influence of both independent and unified repre-
sentations of spatial and nonspatial contextual as-
sociations, it is also possible that the effects of these
representations will be additive in context- and ob-
ject-related regions, but will interact at higher-level
brain regions associated with postrecognition proc-
esses. Future research is needed to address these
issues, and further increase our understanding of
the nature of contextual representations and their
effect on visual object recognition.
Integrated object- and context-based top-down
facilitation of recognition
In this overview, we have described how top-down
facilitation of recognition can be achieved either
(1) through an object-based mechanism that gen-
erates ‘‘initial guesses’’ about an object’s identity
using rapidly analyzed coarse information about
the input image or (2) through the predictive
information provided by contextual associations
between an object or a scene and the other objects
that are likely to appear together in a particular
setting. However, it is clear that objects do not
appear in isolation, and that both object-based
and context-related information is typically avail-
able to the visual system during recognition. It
therefore makes sense that an optimized system
might take advantage of both forms of top-down
facilitation to maximize the efficiency of recogni-
tion processes. How is this achieved? In this sec-
tion, we consider how both object-based and
context-based influences might operate together
to facilitate object recognition.
Central to this discussion is the observation that
recognition of an object can be enhanced through
neural mechanisms that sensitize the cortical repre-
sentations of the most likely candidate interpreta-
tions of that particular object before information
processed through the bottom-up stream has suffi-
ciently accumulated. We propose two key mecha-
nisms through which this ‘‘sensitization’’ is achieved.
As reviewed in the first part of this paper, an object-
based mechanism is proposed to capitalize on a
rapidly derived LSF representation of the object it-
self to generate ‘‘initial guesses’’ about its identity.
The back-projection of these candidate interpreta-
tions to object-processing regions thereby sensitizes
the corresponding cortical representations in ad-
vance of the bottom-up information that continues
to activate the correct representation. The context-
based mechanism, reviewed in the second part of
this chapter, is proposed to sensitize the representa-
tions of contextually related objects whose associa-
tions are activated through context frames stored in
the PHC (see Fig. 9). Importantly, we propose that
context frames can be activated following prior rec-
ognition of strongly contextual object (as in Fenske
et al., submitted), or through early, coarse informa-
tion about a scene or an object. This includes the
possibility that contextual analysis begin promptly,
even before a scene or an object is fully recognized.
Considered together, the mechanisms for top-down
facilitation of object recognition that we describe
include one that is based on information from the
‘‘to-be-recognized’’ object itself and the other based
on information about the context in which the
object appears. Given the intricate links between
16
objects and the settings in which they appear, these
mechanisms typically should not be expected to op-
erate in isolation. Indeed, provided that sufficient
information is available for activating the appropri-
ate context frame, and that an LSF image of a single
target object is sufficient for limiting its possible in-
terpretations, the intersection of these two sources of
information would result in a unique, accurate iden-
tification (Bar, 2004). An exciting direction for fu-
ture research will be to assess how information
about additional objects or the scene in which a
target object appears may be processed in parallel to
increase the opportunities for such valuable inter-
sections of object- and context-based information.
The interactive nature of the two sources of top-
down facilitation of object recognition that we
have described emerges when the input to either
the object- or context-based mechanism is ambig-
uous. For example, when the coarse object-related
input to the prefrontal cortex is ambiguous (e.g., a
blurred image of an umbrella also resembles a
mushroom, lamp, and parasol), the benefit of hav-
ing activated the appropriate context frame will be
relatively greater than if the LSF profile of an ob-
ject is completely diagnostic of the object’s identity
(e.g., a blurred image of a giraffe only resembles a
giraffe). In addition, if ambiguous information is
projected to the PHC, then this can result in the
activation of multiple context frames. From this
possibility emerges a rather counterintuitive pre-
diction of our model. Consider, for instance, that a
picture of a gun, when projected rapidly in a
blurred (i.e., LSF) form to the PHC, may be in-
terpreted also as a drill and a hairdryer (Fig. 11).
Fig. 11. In the proposed interactive model of object- and context-based top-down down facilitation, ‘‘initial guesses’’ about an object’s
identity are rapidly generated from cursory analysis of its LSF information. These ‘‘initial guesses’’ are processed by the context
network to help determine the possible context, and thereby facilitate the recognition of other objects that may also be present.
Contextual information helps to further constrain the ‘‘initial guesses’’ about an object’s identity, just as early object-based information
helps to determine which context frames should be activated. This interactive model accounts for the otherwise counterintuitive finding
that a brief presentation of a gun can facilitate the subsequent recognition of a hairbrush (Fenske and Bar, submitted), despite the lack
of any other perceptual, semantic, or contextual relation between these items.
17
These three objects are associated with three
different context frames, and will subsequently
trigger the activation of three sets of objects. Con-
sequently, a gun will not only prime the recogni-
tion of a police car (i.e., contextual priming), but
also the recognition of a hairbrush (i.e., a member
of the context frame activated by the hairdryer).
Importantly, because there is a complete lack of
any other perceptual, semantic, or contextual re-
lation between a gun and a hairbrush, finding sig-
nificant priming for this condition is best explained
through the interactive model of top-down facil-
itation that we propose. Indeed, we found signifi-
cant priming for this condition (Fenske and Bar,
submitted), and found further that this form of
indirect priming existed for relatively short prime
exposures (120 ms) but not for longer prime expo-
sures (2400 ms). This finding underscores the in-
teractive nature of the object- and context-based
mechanisms we have described, and supports our
notion that the arrival of additional information
leaves active only the most relevant ‘‘initial guess’’
about an object’s identity and only the most rel-
evant context frame.
In conclusion, the models and research find-
ings reviewed here emphasize the importance of
top-down facilitation in visual object recognition.
Building on the previous work (Bar, 2003), we
have examined evidence for an object-based
mechanism that rapidly triggers top-down facili-
tation of visual recognition using a partially ana-
lyzed version of the input image (i.e., a blurred
image) that generates reliable ‘‘initial guesses’’
about the object’s identity in the OFC. Impor-
tantly, we have also shown that this form of top-
down facilitation does not operate in isolation.
Work from our lab indicates that contextual as-
sociations between objects and scenes are analyzed
by a network including the PHC and the RSC, and
that the predictive information provided by these
associations can also constrain the ‘‘initial
guesses’’ about an objects’ identity. We propose
that these mechanisms operate together to pro-
mote efficient recognition by framing early infor-
mation about an object within the constraints
provided by a lifetime of experience with contex-
tual associations.
Abbreviations
fMRI functional magnetic resonance
imaging
MEG magnetoencephalography
OFC orbitofrontal cortex
PFC prefrontal cortex
PHC parahippocampal cortex
PPA parahippocampal place area
RSC retrosplenial complex
RT response time
Acknowledgments
This work was supported by NINDS R01-
NS44319 and RO1-NS050615, NCRR P41-
RR14075, and the MIND Institute.
References
Aguirre, G.K., Detre, J.A., Alsop, D.C. and D’Esposito, M.
(1996) The parahippocampus subserves topographical learn-
ing in man. Cereb. Cortex, 6: 823–829.
Aguirre, G.K., Zarahn, E. and D’Esposito, M. (1998) An area
within human ventral cortex sensitive to ‘‘building’’ stimuli:
evidence and implications. Neuron, 21: 373–383.
Aminoff, E. Gronau, N. and Bar, M. (submitted for publica-
tion). The parahippocampal cortex mediates spatial and non-
spatial associations.
Bachevalier, J. and Mishkin, M. (1986) Visual recognition im-
pairment follows ventromedial but not dorsolateral prefron-
tal lesions in monkeys. Behav. Brain Res., 20: 249–261.
Bar, M. (2003) A cortical mechanism for triggering top-down
facilitation in visual object recognition. J. Cogn. Neurosci.,
15: 600–609.
Bar, M. (2004) Visual objects in context. Nat. Rev. Neurosci., 5:
617–629.
Bar, M. and Aminoff, E. (2003) Cortical analysis of visual
context. Neuron, 38: 347–358.
Bar, M., Kassam, K.S., Ghuman, A.S., Boshyan, J., Schmidt,
A.M., Dale, A.M., Hamalainen, M.S., Marinkovic, K.,
Schacter, D.L., Rosen, B.R. and Halgren, E. (2006) Top-
down facilitation of visual recognition. Proceedings of the
National Academy of Science, 103: 449–454.
Bar, M., Tootell, R., Schacter, D., Greve, D., Fischl, B.,
Mendola, J., Rosen, B. and Dale, A. (2001) Cortical mech-
anisms of explicit visual object recognition. Neuron, 29:
529–535.
Bar, M. and Ullman, S. (1996) Spatial context in recognition.
Perception, 25: 343–352.
18
Barcelo
´, F., Suwazono, S. and Knight, R.T. (2000) Prefrontal
modulation of visual processing in humans. Nat. Neurosci.,
3: 399–403.
Biederman, I. (1972) Perceiving real-world scenes. Science, 177:
77–80.
Biederman, I. (1981) On the semantic of a glance at a scene. In:
Kubovy, M. and Pomerantz, J.R. (Eds.), Perceptual Organ-
ization. Erlbaum, Hillsdale, NJ, pp. 213–253.
Biederman, I., Mezzanotte, R.J. and Rabinowitz, J.C. (1982)
Scene perception: detecting and judging objects undergoing
relational violations. Cogn. Psychol., 14: 143–177.
Biederman, I., Rabinowitz, J.C., Glass, A.L. and Stacy, E.W.
(1974) On the information extracted from a glance at a scene.
J. Exp. Psychol., 103: 597–600.
Brewer, J.B., Zhao, Z., Desmond, J.E., Glover, G.H. and Gab-
rieli, J.D. (1998) Making memories: brain activity that pre-
dicts how well visual experience will be remembered. Science,
281: 1185–1187.
Bullier, J. and Nowak, L.G. (1995) Parallel versus serial
processing: new vistas on the distributed organization of
the visual system. Curr. Opin. Neurobiol., 5: 497–503.
Cave, C.B. and Kosslyn, S.M. (1993) The role of parts and spa-
tial relations in object identification. Perception, 22: 229–248.
Chun, M.M. and Jiang, Y. (2003) Implicit, long-term spatial
contextual memory. J. Exp. Psychol. Learn. Mem. Cogn., 29:
224–234.
Davachi, L., Mitchell, J. and Wagner, A. (2003) Multiple routes
to memory: distinct medial temporal lobe processes build
item and source memories. Proc. Natl. Acad. Sci. USA., 100:
2157–2162.
Davenport, J.L. and Potter, M.C. (2004) Scene consistency in
object and background perception. Psychol. Sci., 15:
559–564.
Desimone, R. (1998) Visual attention mediated by biased com-
petition in extrastriate visual cortex. Philos. Trans. R. Soc.
Lond. B Biol. Sci., 353: 1245–1255.
Du
¨zel, E., Habib, R., Rotte, M., Guderian, S., Tulving, E. and
Heinze, H.J. (2003) Human hippocampal and parahippo-
campal activity during visual associative recognition memory
for spatial and nonspatial stimulus configurations. J. Neuro-
sci., 23: 9439–9444.
Engel, A.K., Fries, P. and Singer, W. (2001) Dynamic predic-
tions: oscillations and synchrony in top-down processing.
Nat. Rev. Neurosci., 2: 704–716.
Epstein, R., Graham, K.S. and Downing, P.E. (2003) View-
point-specific scene representations in human parahippocam-
pal cortex. Neuron, 37: 865–876.
Epstein, R. and Kanwisher, N. (1998) A cortical representation
of the local visual environment. Nature, 392: 598–601.
Fenske, M.J., Boshyan, J. and Bar, M. (submitted) Can a gun
prime a hairbrush? The ‘‘initial guesses’’ that drive top-down
contextual facilitation of object recognition. Paper presented
at the 5th Annual Meeting of the Vision Sciences Society,
Sarasota, FL.
Fenske, M.J. and Bar, M. (submitted for publication). Can A
Gun Prime A Haitbrush? ‘‘Initial Guesses’’ that Mediate
Contextual Facilitation of Object Recognition.
Freedman, D.J., Riesenhuber, M., Poggio, T. and Miller, E.K.
(2001) Categorical representation of visual stimuli in the pri-
mate prefrontal cortex. Science, 291: 312–316.
Friston, K. (2003) Learning and inference in the brain. Neural
Networks, 16: 1325–1352.
Gibson, E.J. (1969) Principles of Perceptual Learning and De-
velopment. Appleton-Century-Crofts, New York.
Gilbert, C.D., Sigman, M. and Crist, R.E. (2001) The neural
basis of perceptual learning. Neuron, 31: 681–697.
Goodale, M.A. and Milner, A.D. (1992) Separate visual path-
ways for perception and action. Trends Neurosci, 15: 20–25.
Gordon, R.D. (2004) Attentional allocation during the percep-
tion of scenes. J. Exp. Psychol. Hum. Percept. Perform., 30:
760–777.
Grabowska, A. and Nowicka, A. (1996) Visual-spatial-fre-
quency model of cerebral asymmetry: a critical survey of be-
havioral and electrophysiological studies. Psychol. Bull., 120:
434–449.
Grill-Spector, K., Kourtzi, Z. and Kanwisher, N. (2001) The
lateral occipital complex and its role in object recognition.
Vision Res., 41: 1409–1422.
Grossberg, S. (1980) How does a brain build a cognitive code?
Psychol. Rev., 87: 1–51.
Henderson, J.M. (1992) Identifying objects across saccades:
effects of extrafoveal preview and flanker object context. J.
Exp. Psychol. Learn. Mem. Cogn., 18: 521–530.
Henderson, J.M., Pollatsek, A. and Rayner, K. (1987) Effects
of foveal priming and extrafoveal preview on object identi-
fication. J. Exp. Psychol. Hum. Percept. Perform., 13:
449–463.
Henson, R.N. (2003) Neuroimaging studies of priming. Prog.
Neurobiol., 70: 53–81.
Hock, H.S., Romanski, L., Galie, A. and Williams, C.S. (1978)
Real-world schemata and scene recognition in adults and
children. Mem. Cogn., 6: 423–431.
Hollingworth, A. and Henderson, J.M. (1998) Does consistent
scene context facilitate object perception? J. Exp. Psychol.
Gen., 127: 398–415.
Hopfinger, J.B., Buonocore, M.H. and Mangun, G.R. (2000)
The neural mechanisms of top-down attentional control.
Nat. Neurosci., 3: 284–291.
Humphreys, G.W., Riddoch, M.J. and Price, C.J. (1997) Top-
down processes in object identification: evidence from exper-
imental psychology, neuropsychology and functional anat-
omy. Philos. Trans. R. Soc. Lond. B Biol. Sci., 352:
1275–1282.
Jackson, O. and Schacter, D.L. (2004) Encoding activity in an-
terior medial temporal lobe supports subsequent associative
recognition. Neuroimage, 21: 456–462.
Kirwan, C.B. and Stark, C.E. (2004) Medial temporal lobe ac-
tivation during encoding and retrieval of novel face-name
pairs. Hippocampus, 14: 919–930.
Kosslyn, S.M. (1994) Image and Brain. MIT Press, Cambridge,
MA.
Kosslyn, S.M., Alpert, N.M. and Thompson, W.L. (1995) Iden-
tifying objects at different levels of hierarchy: a positron emis-
sion tomography study. Hum. Brain Mapping, 3: 107–132.
19
Kosslyn, S.M., Alpert, N.M., Thompson, W.L., Chabris, C.F.,
Rauch, S.L. and Anderson, A.K. (1993) Visual mental im-
agery activates topographically organized visual cortex: PET
investigations. J. Cogn. Neurosci., 5: 263–287.
Koutstaal, W., Wagner, A.D., Rotte, M., Maril, A., Buckner,
R.L. and Schacter, D.L. (2001) Perceptual specificity in visual
object priming: functional magnetic resonance imaging evi-
dence for a laterality difference in fusiform cortex. Ne-
uropsychologia, 39: 184–199.
Lee, T.S. and Mumford, D. (2003) Hierarchical Bayesian in-
ference in the visual cortex. J. Opt. Soc. Am., 20: 1434–1448.
Leonova, A., Pokorny, J. and Smith, V.C. (2003) Spatial fre-
quency processing in inferred PC- and MC-pathways. Vision
Res., 43: 2133–2139.
Levy, I., Hasson, U., Avidan, G., Hendler, T. and Malach, R.
(2001) Center-periphery organization of human object areas.
Nat. Neurosci., 4: 533–539.
Logothetis, N.K., Leopold, D.A. and Sheinberg, D.L. (1996)
What is rivalling during binocular rivalry? Nature, 380:
621–624.
Lustig, C. and Buckner, R.L. (2004) Preserved neural correlates
of priming in old age and dementia. Neuron, 42: 865–875.
Maccotta, L. and Buckner, R.L. (2004) Evidence for neural
effects of repetition that directly correlate with behavioral
priming. J. Cogn. Neurosci., 16: 1625–1632.
Maguire, E.A. (2001) The retrosplenial contribution to human
navigation: a review of lesion and neuroimaging findings.
Scand. J. Psychol., 42: 225–238.
Malach, R., Levy, I. and Hasson, U. (2002) The topography of
high-order human object areas. Trends Cogn. Sci., 6: 176–184.
Mandler, J.M. and Johnson, N.S. (1976) Some of the thousand
words a picture is worth. J. Exp. Psychol. [Hum. Learn.], 2:
529–540.
Martin, A., Wiggs, C.L., Ungerleider, L.G. and Haxby, J.V.
(1996) Neural correlates of category-specific knowledge. Na-
ture, 379: 649–652.
Mechelli, A., Price, C.J., Friston, K.J. and Ishai, A. (2004)
Where bottom-up meets top-down: neuronal interactions
during perception and imagery. Cereb. Cortex, 14: 1256–1265.
Merigan, W.H. and Maunsell, J.H. (1993) How parallel are the
primate visual pathways? Annu. Rev. Neurosci., 16: 369–402.
Metzger, R.L. and Antes, J.R. (1983) The nature of processing
early in picture perception. Psychol. Res., 45: 267–274.
Miller, M.B. and Gazzaniga, M.S. (1998) Creating false mem-
ories for visual scenes. Neuropsychologia, 36: 513–520.
Miyashita, Y. and Hayashi, T. (2000) Neural representation of
visual objects: encoding and top-down activation. Curr.
Opin. Neurobiol., 10: 187–194.
Morcom, A.M., Good, C.D., Frackowiak, R.S. and Rugg,
M.D. (2003) Age effects on the neural correlates of successful
memory encoding. Brain, 126: 213–229.
Oenguer, D. and Price, J.L. (2000) The organization of net-
works within the orbital and medial prefrontal cortex of rats,
monkeys and humans. Cereb. Cortex, 10: 206–219.
Oliva, A. and Torralba, A. (2001) Modeling the shape of a
scene: a holistic representation of the spatial envelope. Int. J.
Comp. Vis., 42: 145–175.
Palmer, S.E. (1975) The effects of contextual scenes on the
identification of objects. Mem. Cogn., 3: 519–526.
Parker, A., Wilding, E. and Akerman, C. (1998) The Von
Restorff effect in visual object recognition memory in hu-
mans and monkeys. The role of frontal/perirhinal interac-
tion. J. Cogn. Neurosci., 10: 691–703.
Pascual-Leone, A. and Walsh, V. (2001) Fast backprojections
from the motion to the primary visual area necessary for
visual awareness. Science, 292: 510–512.
Pihlajamaki, M., Tanila, H., Kononen, M., Hanninen,
T., Hamalainen, A., Soininen, H. and Aronen, H.J. (2004)
Visual presentation of novel objects and new spatial ar-
rangements of objects differentially activates the medial tem-
poral lobe subareas in humans. Eur. J. Neurosci., 19:
1939–1949.
Ranganath, C., DeGutis, J. and D’Esposito, M. (2004a) Cat-
egory-specific modulation of inferior temporal activity during
working memory encoding and maintenance. Brain Res.
Cogn. Brain Res., 20: 37–45.
Ranganath, C., Yonelinas, A.P., Cohen, M.X., Dy, C.J., Tom,
S.M. and D’Esposito, M. (2004b) Dissociable correlates of
recollection and familiarity within the medial temporal lobes.
Neuropsychologia, 42: 2–13.
Rempel-Clower, N.L. and Barbas, H. (2000) The laminar pat-
tern of connections between prefrontal and anterior temporal
cortices in the rhesus monkey is related to cortical structure
and function. Cereb. Cortex, 10: 851–865.
Schacter, D.L. and Buckner, R.L. (1998) Priming and the brain.
Neuron, 20: 185–195.
Schyns, P.G. and Oliva, A. (1994) From blobs to boundary
edges: evidence for time- and spatial-dependent scene recog-
nition. Psychol. Sci., 5: 195–200.
Shapley, R. (1990) Visual sensitivity and parallel retinocortical
channels. Annu. Rev. Psychol., 41: 635–658.
Simons, J.S., Koutstaal, W., Prince, S., Wagner, A.D. and
Schacter, D.L. (2003) Neural mechanisms of visual object
priming: evidence for perceptual and semantic distinctions in
fusiform cortex. Neuroimage, 19: 613–626.
Sternberg, S. (2001) Separate modifiability, mental modules,
and the use of pure and composite measures to reveal them.
Acta Psychol. (Amst.), 106: 147–246.
Tanaka, K. (1996) Inferotemporal cortex and object vision.
Annu. Rev. Neurosci., 19: 109–139.
Ullman, S. (1995) Sequence seeking and counter streams: A
computational model for bidirectional information flow in
the visual cortex. Cereb. Cortex, 1: 1–11.
Valenstein, E., Bowers, D., Verfaellie, M., Heilman, K.M.,
Day, A. and Watson, R.T. (1987) Retrosplenial amnesia.
Brain, 110: 1631–1646.
Vogt, B.A., Vogt, L.J., Perl, D.P. and Hof, P.R. (2001) Cytol-
ogy of human caudomedial cingulate, retrosplenial, and cau-
dal parahippocampal cortices. J. Comp. Neurol., 438:
353–376.
Vuilleumier, P., Henson, R.N., Diver, J. and Dolan, R.J. (2002)
Multiple levels of visual object constancy revealed by event-
related fMRI of repetition priming. Nat. Neurosci., 5:
491–499.
20
Wagner, A.D., Schacter, D.L., Rotte, M., Koutstaal, W., Mar-
il, A., Dale, A.M., Rosen, B.R. and Buckner, R.L. (1998)
Building memories: remembering and forgetting of verbal
experiences as predicted by brain activity. Science, 281:
1188–1191.
Wheeler, M.E. and Buckner, R.L. (2003) Functional dissocia-
tion among components of remembering: control, perceived
oldness, and content. J. Neurosci., 23: 3869–3880.
Wilson, F.A.W., Scalaidhe, S.P.O. and Goldman-Rakic,
P.S. (1993) Dissociation of object and spatial processing
domains in primate prefrontal cortex. Science, 260:
1955–1958.
Zago, L., Fenske, M.J., Aminoff, E. and Bar, M. (2005)
The rise and fall of priming: how visual exposure shapes
cortical representations of objects. Cereb. Cortex, 15:
1655–1665.
21
