amit et al

38 Pages·2012·0.27 MB·English

Journal section: Behavioral/Systems/Cognitive

Do object-category selective regions in the ventral visual stream represent perceived distance
information?

Elinor Amit1, Eyal Mehoudar 2, Yaacov Trope3, and Galit Yovel2
1Harvard University
2Tel-Aviv University
3New York University
Address for correspondence:
Elinor Amit
Department of Psychology
Harvard University
William James Hall 1484
33 Kirkland Street
Cambridge, MA 02138
Tel. [phone]
Email: [email protected]

Acknowledgments: We would like to thank Yonantan Douek and Vadim Axelrod for valuable help
with fMRI data analysis, Peter Huy Wong for assisting with data collection and stimulus preparation of
the psychophysical studies, Jonathan Oron for help with collection of eye tracking data, and
Tejaswinhi Srinivas for proofreading. We also like to thank Russell Epstein for valuable comments.
This work was funded by a grant from the Adams Super Center for Brain Research to G.Y.
Abstract
It is well established that scenes and objects elicit a highly selective response in specific brain
regions in the ventral visual cortex. An inherent difference between these categories that has not been
explored yet is their perceived distance from the observer (i.e. scenes are distal whereas objects are
proximal). The current study aimed to test the extent to which scene and object selective areas are
sensitive to perceived distance information independently from their category-selectivity and
retinotopic location. We conducted two studies that used a distance illusion (i.e., the Ponzo lines) and
showed that scene regions (the parahippocampal place area, PPA, and transverse occipital sulcus,
TOS) are biased toward perceived distal stimuli, whereas the lateral occipital (LO) object region is
biased toward perceived proximal stimuli. These results suggest that the ventral visual cortex plays a
role in representing distance information, extending recent findings on the sensitivity of these regions
to location information. More broadly, our findings imply that distance information is inherent to
object recognition.

Key words: ventral visual stream, distance, objects, scenes
Running title: distance information in the ventral visual cortex
1. Introduction
It is well established that scenes and objects elicit a highly selective response in specific brain
regions in the ventral visual cortex (e.g., Kanwisher et al, 1996; Epstein and Kanwisher, 1998).
Notably, those categories differ is their perceived distance from the observer (i.e. scenes are distal
whereas objects are proximal). The goal of the current study is to test to what extent category selective
areas in the ventral visual stream play a role in representing distance information, such that scene
regions are biased towards distal information whereas object regions are biased towards proximal
information, independently from their category-selectivity and retinotopic location.

1.1. Past Research on the Representation of Object Categories in the Ventral Visual Stream
Neuroimaging studies in the past 15 years have established that certain object categories elicit
a highly selective response in specific brain regions in the ventral visual stream (e.g., Kanwisher et al,
1996; Epstein and Kanwisher, 1998; Aguirre et al., 1998; for a recent review, see Op de Beeck et al,
2008). Such selective neural response has been reported for faces (Kanwisher & Yovel, 2006), places
(Epstein and Kanwisher, 1998), body parts (Downing et al, 2001), and words or letter strings (Cohen et
al, 2000). The underlying assumption of this line of research is that responses of category-selective
regions in the ventral visual stream can be explained by higher-level object representations. For
example, a “scene” contains information about the spatial structure of the local environment, and this
triggers the response in scene regions (the parahippocampal place area, PPA, and transverse occipital
sulcus, TOS, Epstein, 2005; Epstein et al, 2007). With respect to objects, Epstein (2005) suggested that
objects are spatially compact entities that one acts upon, in contrast to scenes, which are spatially
distributed entities that one acts within. Likewise, Mazer and Gallant (2000) suggested that in order to
recognize a single object, constituent patches must be segmented from the background and grouped
together into a coherent whole. In other words, in both Epstein and Mazer et al’s definitions, objects
2
are defined by their inverse relationships with background.
Consistent with the research on the ventral stream, the ventral/dorsal stream theory (Ungeleider
and Mishkin, 1982) suggests that as visual information exits the occipital lobe, it follows two main
paths, or "streams." The dorsal stream terminates in the parietal lobe and represents the location of an
object (the "where pathway") or an “action” towards the object (Goodale et al, 1991). The ventral
stream in the temporal lobe is involved with object identification (the "what pathway"). Lesions to this
region of the brain often produce difficulties in recognizing, identifying and naming different
categories of objects (Habib & Sirigu, 1987; McNeil & Warrington, 1993; Moscovitch et al, 1997).

1.2. Biases in the Ventral Visual Stream Beyond Shape
Recent studies have shown that at least some of the object category areas in the ventral “what”
stream are sensitive not only to shape, but also to retinotopic location – that is, the location of a target
relative to fixation (center vs. periphery). Specifically, it was found that scene-selective regions are
biased toward stimuli presented in the upper visual field, whereas object-selective regions are biased
toward stimuli presented in the lower visual field (Schwarzlose et al, 2008). Furthermore, object areas
were found to be biased towards foveal information, whereas scene areas are biased towards peripheral
information (Hasson, Levy, Behrmann, Hendler, & Malach, 2002).
In addition to these retinotopic biases, recent studies also suggest that these category-selective
regions are sensitive to the size of the stimuli. Specifically, it was shown that scene regions (PPA and
TOS) are biased toward stimuli that are generally classified as large (henceforth, “large stimuli”),
whereas object region (lateral occipital, LO) is biased toward small stimuli (Cate, Goodale, and
Kohler, 2011; Konkle & Oliva, 2010). In sum, the findings above suggest that the ventral stream
includes information other than just the category of the target (e.g., face, object, or place) but also its
size and location. However, a unified account of these location and size biases and the selectivity for
objects and scenes is still lacking.
3
1.3. Perceived Distance as an Integrating Variable
One factor that may be common to all four types of biases (i.e., object category, location,
retinotopy, and size) is the perceived distance between the observer and the visual stimuli – that is,
egocentric distance. In particular, scenes are typically farther away whereas objects are more proximal.
Furthermore, distal stimuli usually occupy the upper part of the visual field (e.g., Gibson, 1950, p.180;
Yonas, Elieff, & Arterberry, 2002; Bruno & Cutting, 1988; Cutting & Vishton, 1995; Epstein, 1966)
and do not require foveal vision for recognition, whereas proximal stimuli usually occupy the lower
part of the visual field and do require foveal vision for recognition (Levy, Hasson, Avidan, Hendler, &
Malach, 2001). Thus, the location and eccentricity biases in scene and object regions may reflect an
egocentric distance bias. Finally, size and distance are inherently related to each other (in the realm of
visual perception). Specifically, when presented with two items identical in size, the distal item is
perceived to be larger than the proximal item (e.g., Murray et al, 2005, Smith, 1958). Thus, it is
possible that effects that were attributed to size could actually be explained by distance. While size and
distance are typically confounded, given that object and scene selective areas also showed biases to
location (down /up) and eccentricity – with both being associated with distance but not size – we
believe that it is distance and not size that may underlie all reported biases in object and scene areas.
In sum, perceived distance can account for the location, eccentricity, and size biases reported
above. Consistent with the view of the ventral/dorsal stream theory (Ungeleider and Mishkin, 1982), it
is possible that this distance bias in the “what” stream reflects the ideal distance of a given stimulus
from the observer to allow for intact identification.
The current study tested whether category selective areas in the ventral visual stream represent
egocentric distance information. In order to test the hypothesis, in both Experiments 1 and 2, we took
advantage of the well-known distance illusion: the Ponzo illusion (e.g., Yi Li & Guo, 1995). This
illusion is comprised of two lines that converge at the upper end (see Figure 1 for an illustration). As a
result, the upper end is perceived as more distal than the lower end. The evidence for the existence of
4
such an illusion comes from the estimation of a “distal” (upper) item as larger than the “proximal”
(lower) item, when in reality they are equal in size. We first pretested the effectiveness of the distance
illusion using a psychophysical size estimation task and then examined the response of object and
scene category regions to the distance manipulation.
Our main hypothesis concerns the representation of perceived distance in the ventral visual
stream. Although there are several candidate regions for testing the distance hypothesis, we chose,
based on the literature reviewed above, four representative regions: scene regions (the
parahippocampal place area, PPA, Epstein & Kanwisher, 1998; and transverse occipital sulcus, TOS,
Epstein et al, 2007) and object regions (LO; and posterior fusiform gyrus: pFs; both comprises together
the LOC: the Lateral Occipital Cortex, e.g., Malach et al 1995). We hypothesized that object regions
(LO and pFs) would be biased toward perceived proximal stimuli, whereas scene regions (PPA and
TOS) would be biased toward perceived distal stimuli.
In Experiment 1 we tested this hypothesis by presenting buildings and objects within the
Ponzo lines, in a “distal” (upper) or “proximal” (lower) locations. Objects and buildings were used as
they were found to elicit a strong response in the LOC object area and the scene areas respectively
(e.g., Malach et al, 1995 Epstein et al, 1998). From a theoretical point of view, the PPA response to
buildings may be due to the fact that buildings are stable objects that often play a role in defining the
spatial structure of the environment (Epstein, 2005). In Experiment 2 we tested the effect of distance
independently of location of the target item, by presenting the stimuli in a fixed location at the center
of the screen, while the Ponzo lines are presented in either the upper half or the lower half of the
screen. Importantly, in both experiments, our perceived distance manipulation entailed egocentric and
not retinotopic distance. In other words, stimuli were always presented foveally and perceived distance
was defined as a function of the location of the stimulus in the Ponzo lines rather than retinotopic
location. Finally, we collected eye position data outside of the scanner with the exact same design and
stimuli as the experiment in the scanner, to confirm that subjects fixate on the stimuli and to rule out
5
the possibility that retinotopic and not spatiotopic effects accounts for our results.

2. Materials and Methods
Our study included two fMRI experiments. Each fMRI experiment was preceded by a
psychophysical experiment that took place outside the scanner, in which we assessed whether the
particular Ponzo display that we used in the fMRI experiment indeed generates a distance illusion.
Finally, we collected eye position data outside of the scanner.

2.1. Methods for Psychophysical Experiments
To assess whether the Ponzo lines generated a distance illusion we took advantage of the well-
established effect of distance on size perception (e.g., Murray, Boyaci, & Kersten, 2005; Smith, 1958)
(see Figure 1 for an illustration). Participants were asked to compare the size of two same-identity
stimuli, which appeared simultaneously in the upper and lower parts of two lines that converged
toward the upper end (i.e., the Ponzo lines). The participants were asked to judge which stimulus was
larger. A distance illusion is evident if the stimulus at the upper end of the Ponzo lines is judged to be
larger than the stimulus at the lower end of the Ponzo lines, despite the fact that they are identical in
size.

2.1.1. Participants
Ten New York University undergraduates participated in the two psychophysical manipulation-
check experiments, for partial fulfillment of course credit. The study was approved by the University
Committee on Activities Involving Human Subjects (UCAIHS).
2.1.2. Stimuli
The stimuli consisted of black and white photographs of 20 objects and 20 buildings. Images of
buildings were taken from the site http://www.zoomap.co.il/default.asp, and were selected randomly
6
(See Figure 2 for an illustration). Objects were adapted from Epstein & Kanwisher (1998). A pair of
vertically oriented converging straight lines, the Ponzo illusion, was used to create the illusion of
depth. In Experiment 1, the Ponzo lines transversed the entire computer screen. In Experiment 2, the
Ponzo lines appeared either at the upper half or the lower half of the screen. In both experiments, two
same-identity stimuli appeared simultaneously at the upper end and at the lower end of the Ponzo lines.
The stimuli appeared in five different sizes. In Experiment 1 (full screen Ponzo), the standard image
that was used in the fMRI experiment and labeled here “100%,” was 4.5 cm in width, and 4.5 cm in
length. In Experiment 2 (half screen Ponzo) the width of the standard image was 2.6 cm and the length
was 2.6 cm. We created four additional stimuli that were 98%, 99%, 101% and 102% of the size of the
standard stimuli.
2.1.3. Procedure
Stimuli were presented using Superlab software. Trials of objects and buildings were presented
in separate blocks, with a rest period in the middle of each block. Each trial consisted of two stimuli
that appeared simultaneously at the upper end and at the lower end of the Ponzo lines. The two stimuli
included the standard stimulus (size 100%) and same-identity stimulus in one of the five different sizes
(98%, 99%, 100%, 101% or 102%). The location of the standard stimuli was at the upper part of the
screen on half of the trials, and the lower part on the other half.
In Experiment 1, the design was as follows: 2 Categories (Buildings, Objects) x 5 Sizes (98%,
99%, 100%, 101%, 102%) x 2 Locations of the standard stimulus (Upper, Lower). Each of the
combinations was presented twice for each of the 20 Building and the 20 Object stimuli, making a total
of 800 presentations. On each trial, the Ponzo lines with the two stimuli embedded in them appeared
for 800 ms.
In Experiment 2, the design was as follows: 2 Categories (Buildings, Objects) x 5 Sizes (98%,
99%, 100%, 101%, 102%) x 2 Locations of the standard stimulus (Upper, Lower) x 2 Locations of the
Ponzo lines (Upper screen, Lower screen). Because this experiment included an additional condition
7
(the location of the Ponzo), to keep the length of this experiment similar to Experiment 1, each of the
unique combinations was presented once for each of the 20 Building and 20 Object stimuli, making a
total of 800 presentations. On each trial, the Ponzo lines with the two stimuli embedded in them
appeared for 800 ms.
In both experiments, after viewing the stimulus, the following question was presented: “Please
indicate which of the two stimuli seemed larger: Press (A) if it was the upper, (H) if the lower or (L) if
they seemed equal.”

2.2. Methods of the fMRI Experiments
The fMRI experiments tested the prediction that object regions (the two parts of the LOC: LO,
and pFs) would be biased toward perceived proximal information whereas scene regions (PPA and
TOS) would be biased toward perceived distal information. We therefore ran a localizer task to
identify object and scene regions in each subject (See Figure 3 for representative participants).
Subsequently, we examined the response of these areas to the different conditions of the main task
(i.e., perceived distance).

2.2.1. Participants
Twenty-six healthy participants participated in Experiment 1 and 14 participants participated
in Experiment 2. All participants had normal or corrected to normal vision and signed a consent form
that was approved by the Helsinki committee of the Sourasky Medical Center, Israel.
2.2.2. Stimuli
Localizer. In order to define functional regions of interest (fROI) we ran a localizer experiment.
The stimuli consisted of three categories of items: Scenes, Objects, and Scrambled objects. Scrambled
objects were designed as Fourier power-spectrum controls for the object pictures. They were
constructed by performing a two dimensional (2D) Fourier transformation on high-pass filtered
8
pictures of the same objects (Malach et al, 1995). Each category included 80 different images. The
images were digitized grayscale photographs (300x300 pixels) that were adapted from previous studies
(Epstein and Kanwisher, 1998; Epstein et al, 1999).
Main Task. The stimuli – buildings and objects - were the “standard stimuli” used in the
psychophysical experiments. Thus, the stimuli in the fMRI experiments came in one size. In
Experiment 1, the stimuli were 4.5 cm in width, and 4.5 cm in length. In Experiment 2, the stimuli
were 2.6 cm and the length was 2.6 cm. We used the same Ponzo illusion that was used in the
psychophysical experiments. In Experiment 1 (full screen Ponzo), the stimuli (buildings or objects)
were presented either at the upper (“distal”) or the lower (“proximal”) part of the Ponzo lines. In
Experiment 2 (half screen Ponzo), the stimuli, buildings and objects, were placed in a fixed location at
the center of the screen, thus appearing either at the upper end of the lower Ponzo lines (“distal”) or at
the lower end (“proximal”) of the upper Ponzo lines.

2.2.3. fMRI Data Acquisition
Scanning was performed with a GE 3.0-T scanner at the Tel Aviv Sourasky Medical Center,
Israel. For the functional scans, we collected 30 slices aligned parallel to the temporal lobe, TR = 2000
ms TE =35 ms, flip angle = 90. We also collected high-resolution anatomical images (SPGR) including
170 sagittal slices (TR = 6.5 ms; TE =2 ms, FOV = 256).
2.2.4. Procedure
Stimuli were presented using Psychtoolbox implemented in MATLAB (Brainard, 1997). Each
fMRI experiment consisted of two runs of the localizer and two runs of the experimental task, which
were presented in an interleaved manner, starting with the localizer.
Localizer Scan. The localizer aimed to localize scene and object regions, and included 2 scans,
each consisting of 17 consecutive 16 s epochs, four for each category (Scenes, Objects and Scrambled
objects) and five fixation blocks (total time 272 s). Each block included 20 stimuli. The serial position
9