Professional Documents
Culture Documents
www.elsevier.com/locate/visres
a
Department of Psychology, University of California, Santa Barbara, CA 93106, USA
b
Psychology Institute, Federal University of Rio De Janeiro, Ribeir~ao Preto 14040-901, Brazil
c
Department of Psychology, University of Sao Paulo at Ribeir~ao Preto, RJ 22290-240, Brazil
Received 26 March 2003; received in revised form 8 July 2003
Abstract
The question of how perceived extents are related to the corresponding physical extents is a very old question that has not been
satisfactorily answered. The common model is that perceived extent is proportional to the product of image size and perceived
distance. We describe an experiment that shows that perceived extents are substantially larger than this model predicts. We propose
a model that accounts for our results and a large set of other results. The principal assumption of the model is that, in the com-
putation of perceived extent, the visual angle signal undergoes a magnifying transform. Extent is often perceived more accurately
than the common model predicts, so the computation is adaptive. The model implies that, although the perception of location and
the perception of extent are related, they not related by Euclidean geometry, nor by any metric geometry. Nevertheless, it is possible
to describe the perception of location and extent using a simple model.
Ó 2003 Elsevier Ltd. All rights reserved.
It is a common observation that objects appear least to Malebranch in 1674 (Pastore, 1971). This is
smaller when they are far away from us than when they sometimes expressed in the following equation:
are near. This is the most obvious example of the fact
that the perceived size of things does not generally equal S 0 ¼ D0 ðS=DÞ; ð1Þ
their physical size. It is natural to wonder how the
perceived size depends on physical size and the stimuli where S 0 is perceived size, D0 is perceived distance, and S
that carry information about space. and D are the corresponding physical magnitudes. In
The retinal image, which was first correctly described words, if distance is misperceived, size will be misper-
by Kepler in 1604, decreases in size in proportion to ceived proportionally and their ratio will always equal
object distance. That might explain the perceptual the ratio of physical size to physical distance. For an
shrinking as distance increases, except that perceived object that is straight ahead and subtends an angle of H
size changes much less than image size. As long ago as at the eye (Fig. 1(A)), S=D ¼ 2 tanðH=2Þ, which for
300 BC, it was noted in Euclid’s Optics that perceived small angles is approximately equal to H in radians. In
size does not decrease in proportion to distance (Boring, practice one or the other of these expressions is often
1942). Indeed in many well lighted situations perceived substituted for S=D in Eq. (1), so that S 0 is expressed as a
size changes little with distance when distances are short function of D0 and H. The visual angle, H is propor-
(Harvey & Leibowitz, 1967). tional to retinal image size and both are inversely pro-
Distance is also often under-perceived and the mag- portional to distance. Eq. (1) implies that perceived size
nitude of the error depends on lighting conditions and will fill the visual angle at the perceived distance. This
available cues. This led to the idea that perceived size model is described in most textbooks on perception. The
depends on perceived distance, an idea that goes back at purpose of this paper is to show that this common
model is substantially wrong, to describe an alternative
model, to show that the alternative model gives a much
*
Corresponding author. Fax: +1-805-893-4303. better account of judgments of extent in both dark and
E-mail address: foley@psych.ucsb.edu (J.M. Foley). well lighted viewing conditions, and to describe some of
0042-6989/$ - see front matter Ó 2003 Elsevier Ltd. All rights reserved.
doi:10.1016/j.visres.2003.09.004
148 J.M. Foley et al. / Vision Research 44 (2004) 147–156
1.3. Design sual distance than binocular viewing here. This is not a
common result and it may be due to individual differ-
There were four experimental conditions: ences between the observers in the different conditions.
The next question is whether Eq. (2) or Eq. (4) de-
(1) Binocular, 18 m to nearest stake. scribes how the visual extents among the stakes depend
(2) Binocular, 5 m to nearest stake. on their perceived distances and the visual angles that
(3) Monocular, 18 m to nearest stake. they subtend. In making this and other fits we used the
(4) Monocular, 5 m to nearest stake. perceived distances determined by fitting Eq. (5) to the
binocular and monocular data separately. Fig. 3 shows
A different group of 10 university students performed all the reported extents plotted against the extents pre-
in each condition. We used a between groups design dicted by Eq. (2). Most of the reports are greater than
with each observer making one judgment of the 105 the predictions. Many of the exceptions are egocentric
extents in one of the four conditions to avoid the carry extents (shown as open symbols) for which both models
over effects that would be expected if the same observers make the same prediction, since, Hij ¼ 0 for these ex-
performed in the different conditions or made repeated tents. So Eq. (2) fails when Hij > 0. The root mean
judgments of the same extents. square error (RMSE) of the fit is 2.19 m.
Fig. 4 shows reported extent as a function of predicted
extent determined using the tangle model (Eq. (4)). In-
2. Results dividual fits to the binocular and monocular data
showed that the best predictions are very similar, so a
We computed both the mean and the median report single fit was made to all the data. The figure shows the
for each of the 420 conditions. The mean standard de- fit of each of the four conditions separately. In the 18 m
viation was 4.16 m and the mean standard error was 1.32 conditions, there is almost no systematic error. With the
m. Our analysis is based on the median reports of the 10 same parameters the tangle model slightly under-pre-
observers in each condition. We used medians because a dicts reported extent in the 5 m binocular condition and
few observers in some conditions gave reports that over predicts in the 5 m monocular condition. The fit is
seemed to be outside the normal distribution of the other much better than that of Eq. (2) in all four conditions.
reports. The medians are much less influenced by these The RMSE of the fit is 1.09 m. The parameters of the
outliers. The same model fits the mean reports well. The best fit are P ¼ 0:574 and Q ¼ 0:314.
median reported extents are given in Table A.1. We first Since the common model is a special case of the tangle
fitted the egocentric distance function (Eq. (5)) to the model with Q ¼ 0, we were able to test the hypothesis
reports. It was fitted jointly to the data from the two that the improvement in fit with the tangle model can be
binocular conditions and likewise for the two monocular attributed to chance, using a test that takes into account
conditions. So we have two egocentric distance functions the difference in the number of free parameters (Khuri &
(Fig. 2). In spite of the abundance of distance cues and Cornell, 1987). F ð2; 414Þ ¼ 628:59, p ¼ 0, so the im-
the presence of a known standard, all the median re- provement in fit is highly statistically significant. Other
ported distances are less than the physical distances. models are possible, but this one fits the data well with
Monocular viewing produced less foreshortening of vi- very little systematic error.
35
Reported Egocentric Distance (m)
35
30 30
25 25
20 20
15 15
10 10
5
5
0
0
0 5 10 15 20 25 30 35 40
0 5 10 15 20 25 30 35 40
Physical Egocentric Distance (m) Physical Egocentric Distance (m)
Fig. 2. Median reported egocentric distance to the stakes: (A) binocular conditions, (B) monocular conditions. The smooth curves correspond to the
best least squares fit of Eq. (5) to the two data sets. Only the curve for Hij ¼ 0 is shown so the model is better than it appears here. Binocular:
F ¼ 1:2517, G ¼ 0:00725, RMSE ¼ 0.93 m. Monocular: F ¼ 1:2621, G ¼ 0:000314, RMSE ¼ 1.29 m.
J.M. Foley et al. / Vision Research 44 (2004) 147–156 151
A Binocular 18 B Monocular 18
30 30
25 25
15 15
10 10
5 5
0
0
0 5 10 15 20 25 30
0 5 10 15 20 25 30
Predicted Extent (m) Predicted Extent (m)
C Binocular 5 D Monocular 5
30 30
25 25
Reported Extent (m)
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Predicted Extent (m) Predicted Extent (m)
Fig. 3. Median reported extents between pairs of points plotted against the extent predicted by the generalized common model (Eq. (2)). The open
symbols correspond to egocentric extents. (A) Binocular with nearest stake at 18 m. (B) Monocular with nearest stake at 18 m. (C) Binocular with the
nearest stake at 5 m. (D) Monocular with the nearest stake at 5 m. Reported extents consistently exceed predicted extents except when extents lie on
or near radial lines from the observer. The RMSE of the fit is 2.19 m.
Binocular 18 Monocular 18
30 30
25 25
Reported Extent (m)
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
A Predicted Extent (m) B Predicted Extent (m)
Binocular 5 Monocular 5
30 30
25 25
Reported Extent (m)
Reported Extent (m)
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
C Predicted Extent (m) D Predicted Extent (m)
Fig. 4. Reported extent vs extent predicted by the tangle model (Eq. (4)) for all point pairs: (A) binocular 18 m, (B) monocular 18 m, (C) binocular
5 m, (D) monocular 5 m. A single fit was made to all four data sets. The parameters of the best fit are P ¼ 0:556 and Q ¼ 0:310. The RMSE of the fit
is 1.09 m.
152 J.M. Foley et al. / Vision Research 44 (2004) 147–156
Effective Visual Angle for Size Computation that depth extents are under perceived relative to frontal
2 extents (Foley, 1968; Loomis et al., 1992; Norman,
Todd, Perotti, & Tittle, 1996). The tangle model ac-
1.8
counts qualitatively for this phenomenon as well. How
Effective Visual Angle (radians)
the triangle inequality (Eq. (6)) was satisfied. However, cal and perceived location and the relation between any
Kelly, Beall and Loomis (personal communication) have pair of locations and the perceived extent between them.
done a similar experiment using angles up to 2.35 rad The common assumption is that perceived direction re
(135 deg) and at this visual angle they found violations straight ahead equals physical direction, but there is
of this inequality in 7 of 10 cases. The model also pre- evidence that perceived direction may be slightly greater
dicts that the perceived lengths of adjacent line segments than this and that consequently perceived visual angles
along a non-radial line will not generally sum to equal may be slightly greater than the corresponding physical
the perceived length of the entire line. angles (Bock, 1993; Foley, 1975; Foley & Held, 1972).
As a consequence of these and other implications, the Either assumption can be incorporated into the model
tangle model is inconsistent with all geometries of the without affecting the nature of the space.
class referred to as metric geometries, including the Ri- The model implies that, given an array of points in the
emannian geometries of constant curvature that have visual field, there are two modes in which they can be
frequently been proposed for visual space (Aczel, Boros, processed (Foley, 1972). We can localize the points and
Heller, & Ng, 1999; Blank, 1958a, 1958b; Heller, 1997; we can perceive the extent between them. In applying
Higashiyama, 1976, 1982; Indow, 1974, 1991, 1997; the model to any task, one must first determine whether
Luneburg, 1947, 1950) and Riemannian geometries of it is a location task or an extent task. To describe and
varying curvature (Koenderink, van Doorn, & Lappin, predict perceived locations, one needs only the trans-
2000). The model is also inconsistent with the affine form between physical and perceived coordinates. For
transform and vector contraction models of Wagner perceived extent, one must also apply the perceived
(1985). extent function.
Nevertheless, if we add to our model a plausible as- What, if any, advantage does this strange geometry
sumption about the relation between perceived direction give us? When egocentric distance is under-perceived,
and physical direction, and substitute perceived visual the tangle visual extent function plays an adaptive role.
angle for physical visual angle in Eq. (3), the model is It usually makes visual extents more similar to physical
then a complete model of the relation between physi- extents than they would be if computed with the
A Binocular 18 B Monocular 18
30 30
25 25
Reported Extent (m)
Reported Extent (m)
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40
Physical Extent (m) Physical Extent (m)
C Binocular 5 D Monocular 5
30 30
25 25
Reported Extent (m)
20 20
15 15
10 10
5 5
0 0
0 5 10 15 20 25 30 0 5 10 15 20 25 30
Physical Extent (m) Physical Extent (m)
Fig. 6. Reported extent as a function of physical extent for all point pairs: (A) binocular, 18 m; (B) monocular, 18 m; (C) Binocular, 5 m; (D)
Monocular, 5 m.
154 J.M. Foley et al. / Vision Research 44 (2004) 147–156
Euclidean metric. However, since the correction is visual system and spatial knowledge that we have (Gogel,
angle dependent, the visual extent function does not 1998). At some level the system compensates for the fact
provide a generally accurate compensation for the that perceived egocentric distances are often shorter
foreshortened egocentric distance. Fig. 6 shows reported than physical distances. We continue to misperceive
extent as a function of physical extent. Most of the re- distance, but this misperception is taken into account in
ported extents are less than the corresponding physical the computation of perceived extent. The compensation
extents, and there is a wide spread in the reports for is far from exact, but it is better than no compensation.
comparable physical extents showing that physical ex- Thus, our space perception system is not very elegantly
tent is less closely related to reported extent than is designed, nor is it very accurate, but it is good enough to
predicted extent from the tangle model. Nevertheless, keep most of us alive most of the time.
reported extents are closer to the physical extents than
are the predictions of the common model. In cases where
perceived egocentric distance is over-perceived or accu-
rately perceived, the perceived extent function is Acknowledgements
counter-adaptive.
What processes underlie the counterintuitive relation Supported by FAPESP-Brazil. We thank Albert
between perceived location and perceived extent? We Blank, John Cotton, Brian Foley, William Jacob, Jon
think that the explanation of this relationship lies in the Kelly, Jack Loomis, and Srinivasa Varadharajan for
conflicts between the signals that arise in our visual comments on the manuscript or the analysis.
Table A.1
Median reported extent in meters
B C D E F G H I J L M N O P
Binocular viewing, far condition
A 9.00 10.00 12.25 13.50 11.50 3.00 8.00 16.50 5.50 6.50 6.00 15.00 12.00 19.00
B 8.00 3.75 6.00 3.38 7.50 6.00 10.00 5.00 4.00 8.75 9.00 5.50 21.50
C 13.50 8.50 6.00 6.00 3.00 10.75 8.00 10.00 4.75 7.50 10.50 14.00
D 9.00 6.75 14.00 9.00 13.00 8.00 6.75 14.00 12.50 6.50 25.00
E 3.75 13.00 7.50 3.50 11.00 10.00 10.00 3.50 3.13 18.00
F 8.50 4.50 7.50 6.70 7.50 7.50 6.75 2.65 18.00
G 5.25 14.00 3.00 7.00 2.00 10.50 12.00 15.75
H 10.00 4.00 7.50 2.75 7.25 8.00 15.50
I 13.75 13.75 12.00 3.25 6.00 18.50
J 2.63 4.25 12.00 9.00 17.75
L 9.25 13.25 9.00 21.00
M 9.50 12.00 15.00
N 7.50 15.50
O 22.00
Monocular viewing, far condition
A 10.00 12.00 14.00 17.50 12.50 4.50 9.00 18.75 5.50 7.00 7.50 21.50 17.50 20.50
B 8.50 3.75 10.50 4.50 10.00 5.50 13.50 7.00 5.50 10.00 11.00 7.50 25.50
C 12.50 8.50 7.25 9.00 5.00 11.50 8.25 11.00 6.50 6.75 11.00 15.00
D 11.00 6.00 13.50 10.00 13.50 11.00 8.00 15.00 13.00 6.00 30.00
E 5.00 14.50 10.00 4.00 12.50 13.50 13.00 4.50 4.50 20.50
F 10.50 5.00 10.00 9.50 10.50 9.00 7.50 4.00 22.50
G 6.50 18.00 4.00 7.50 3.50 11.00 17.00 18.50
H 10.00 5.00 7.25 4.50 9.50 11.00 18.50
I 15.50 18.50 15.00 4.25 8.00 20.50
J 4.00 5.50 14.00 10.00 20.50
L 10.00 17.00 11.50 24.00
M 12.50 13.50 19.00
N 7.00 20.00
O 24.00
Binocular viewing, near condition
A 8.75 10.50 12.00 15.75 11.50 3.25 8.50 18.25 4.25 7.00 5.00 15.50 15.00 13.25
B 8.50 3.00 8.75 3.75 8.50 5.25 11.50 5.00 4.00 9.50 11.00 5.50 12.00
C 12.00 8.00 7.50 6.00 2.75 9.25 7.50 11.00 4.25 6.75 9.25 4.50
D 9.00 7.00 13.50 10.00 14.00 9.00 7.00 12.75 15.25 6.00 19.00
E 3.50 11.25 7.25 3.25 9.50 13.50 12.50 3.00 3.50 11.50
J.M. Foley et al. / Vision Research 44 (2004) 147–156 155
Higashiyama, A. (1982). The effects of familiar-size cue on children’s Loomis, J. M., Da Silva, J. A., Fujita, N., & Fukusima, S. S. (1992).
judgments of size and distance. Japanese Journal of Psychology, Visual space perception and visually directed action. Journal of
53, 259–265. Experimental Psychology: Human Perception and Performance, 18,
Indow, T. (1974). On geometry of frameless binocular perceptual 906–921.
space. Psychologia: An International Journal of Psychology in the Luneburg, R. K. (1947). Mathematical analysis of binocular vision.
Orient, 17, 50–63. Princeton, NJ, US: Princeton University Press.
Indow, T. (1991). A critical review of Luneburg’s model with regard to Luneburg, R. K. (1950). The metric of binocular visual space. Journal
global structure of visual space. Psychological Review, 98, 430–453. of the Optical Society of America, 40, 627–642.
Indow, T. (1997). Hyperbolic representation of global structure of Norman, J. F., Todd, J. T., Perotti, V. J., & Tittle, J. S. (1996). The
visual space. Journal of Mathematical Psychology, 41, 89–98. visual perception of three-dimensional length. Journal of Experi-
Kaufman, L. (1974). Sight and mind: an introduction to visual mental Psychology: Human Perception and Performance, 22, 173–
perception. New York: Oxford University Press. 186.
Khuri, A., & Cornell, J. (1987). Response surfaces: designs and analysis. Pastore, N. (1971). Selective history of theories of visual perception:
New York: Marcel Dekker. 1650–1950. New York: Oxford University Press.
Koenderink, J. J., Van Doorn, A. J., & Kappers, A. M. L. (2002). Todd, J. T., Oomes, A. H. J., Koenderink, J. J., & Kappers, A. l. M.
Pappus in optical space. Perception and Psychophysics, 64, 380– (2001). On the affine structure of perceptual space. Psychological
391. Science, 12, 191–196.
Koenderink, J. J., van Doorn, A. J., & Lappin, J. S. (2000). Direct Toye, R. C. (1986). The effect of viewing position on the perceived
measurement of the curvature of visual space. Perception, 29, 69–79. layout of scenes. Perception and Psychophysics, 40, 85–92.
Landy, M. S., Maloney, L. T., Johnston, E. B., & Young, M. (1995). Wagner, M. (1985). The metric of visual space. Perception and
Measurement and modeling of depth cue combination: in defense Psychophysics, 38, 483–495.
of weak fusion. Vision Research, 35, 389–412. Westheimer, G., & Levi, D. M. (1987). Depth attraction and repulsion
Levin, C. A., & Haber, R. N. (1993). Visual angle as a determinant of of disparate foveal stimuli. Vision Research, 27(8), 1361–1368.
perceived interobject distance. Perception and Psychophysics, 54, Woodworth, R. S., & Schlosberg, H. (1954). Experimental psychology.
250–259. New York: Henry Holt and Company.