Size Perception

Abstract

Size perception is a form of space perception: the visual recovery of an object's real physical extent from a retinal image whose size depends jointly on that extent and on distance. Because the same object casts a retinal image half as wide at twice the distance, size cannot be read off the retina directly; the brain must rescale retinal angle by perceived distance, a computation captured by the size–distance invariance hypothesis and expressed behaviourally as size constancy. This article sets out the determinants of apparent size, the invariance hypothesis and its limits, the geometric illusions explained as inappropriate constancy scaling, the moon illusion, the dissociation between size for perception and size for grasping, and the neural evidence that primary visual cortex tracks perceived rather than retinal size.

Keywords: size perception, size constancy, size–distance invariance

Hold a coin at arm's length and it blots out a distant car, yet no one supposes the coin is the larger object. The visual system routinely recovers the true size of things from a retinal image that, taken alone, is deeply ambiguous about size — because that image shrinks with distance in exactly the proportion by which real objects recede. Size perception is the set of processes that resolve this ambiguity, turning a variable retinal angle into a stable estimate of physical extent (Sperandio & Chouinard, 2015). It is a problem of inference rather than measurement, and the errors it makes under carefully arranged conditions — the classical size illusions — are among the most revealing phenomena in perception, because they expose the assumptions the system uses when it works correctly.

What Size Perception Is

Size perception is the recovery of an object's physical extent from vision. Its central difficulty is that the proximal stimulus — the visual angle the object subtends at the eye — is not determined by physical size alone but by physical size divided by distance. A one-metre object at two metres and a two-metre object at four metres cast the identical retinal image, so any given retinal angle is consistent with infinitely many object sizes, one for each possible distance. To perceive size at all, the visual system must break this ambiguity, and the way it does so is to combine the retinal angle with an independent estimate of the object's distance (Ittelson, 1951). Perceived size is therefore not a direct readout but the product of a computation, and everything distinctive about size perception — its accuracy across enormous ranges of distance, and its characteristic failures — follows from the properties of that computation. Figure 1 shows the ambiguity the system must resolve.

The empirical study of the problem began with the question of what actually determines apparent size when distance is varied. In a classic corridor experiment, observers matched the size of a distant test disc to a nearby comparison, and their matches tracked the physical size of the test almost perfectly when the corridor was rich in distance cues — binocular disparity, motion, and the texture of the floor and walls. As those cues were progressively removed, ending with monocular viewing through a reduction tube that hid everything but the disc itself, the matches drifted away from physical size and toward the value predicted by retinal angle alone (Holway & Boring, 1941). The result fixed the terms of the field: apparent size lies on a continuum between physical size and retinal size, and where it falls is governed by the information available about distance.

Size Constancy

Size constancy is the name for the perceptual achievement the corridor experiment measured: the tendency to perceive an object as having a stable physical size despite large changes in the size of its retinal image as it moves nearer or farther. Constancy is rarely perfect and rarely absent; it is typically high, so that a person walking away from us appears to keep their size rather than shrink, even though the retinal image is halving with every doubling of distance (McKee & Welch, 1992). The phenomenon shows that the visual system is not reporting the proximal stimulus but a distal property inferred from it, and it is the everyday, correct-functioning counterpart of the illusions considered below.

Constancy is not all-or-none but graded, and its degree indexes how fully distance has been taken into account. Under full-cue viewing, size matches are close to the physical size of the target and the precision of those matches is high, degrading only modestly as distance grows (McKee & Welch, 1992). Under reduced cues, matches regress toward retinal size, and the classic determinants — binocular disparity, motion parallax, the gradient of the ground texture, and familiar size — each contribute a portion of the distance signal on which the rescaling depends (Holway & Boring, 1941). These cues are not merely added: when several estimates bear on the same property the visual system weights each by its reliability and combines them much as a statistically optimal, maximum-likelihood estimator would, a principle Ernst and Banks established directly for size by pitting vision against touch and finding that each sense's influence on the combined size estimate tracked its measured precision (Ernst & Banks, 2002). Table 1 organises these determinants and the direction in which each moves apparent size.

Familiar size stands apart from the other determinants because it draws on memory rather than on current stimulation. Knowledge of an object's typical extent — a playing card, an adult human figure — can supply the distance term even when geometric cues are sparse, and under reduced viewing this prior can dominate the estimate. Gogel and Da Silva's theory of off-sized perceptions holds that a familiar object tends to be seen at the distance its assumed size implies, so that an object secretly built larger or smaller than normal is misperceived in both its size and its distance (Gogel & Da Silva, 1987). The Ames room is the vivid demonstration: by holding the observer's assumptions about a room's shape fixed, it makes two people of equal height, standing in a distorted room contrived to project a normal one, appear grossly unequal in size.

The first demonstration lets the reader move an object in depth and watch perceived size hold constant while the retinal image shrinks, then remove the distance cues and watch constancy fail.

The Size–Distance Invariance Hypothesis

The computation underlying constancy was given a formal statement in the size–distance invariance hypothesis, which holds that a given retinal angle specifies a fixed ratio between perceived size and perceived distance (Epstein, Park, & Casey, 1961). For a small visual angle, perceived size is approximately perceived distance multiplied by the retinal angle in radians; the retinal angle is held constant by the stimulus, so perceived size and perceived distance are locked in proportion. The hypothesis explains constancy directly: as an object recedes, its retinal angle shrinks in proportion to its growing distance, and if perceived distance grows correspondingly the two changes cancel and perceived size stays put. It also makes a strong, testable prediction in the other direction — that whenever perceived distance is wrong, perceived size must be wrong by the matching factor, because the invariant ratio leaves no other outcome.

That prediction is the engine of the size illusions, and it is also where the hypothesis meets its limits. Reviewers noted early that observers can sometimes report an object's angular size and its physical size independently, and that judgements of size and of distance do not always move in the lockstep the strict hypothesis requires, so the invariance holds for registered distance under good conditions rather than as an exceptionless law (Epstein et al., 1961). Reaction-time work bears out that size and distance are integrated rather than merely read off together: responses to a target are fastest when its retinal size and its depth are consistent with a single physical size and slow when the two conflict, indicating an active combination of the two signals rather than a fixed geometric identity (Plewan & Rinkenauer, 2017). The invariance hypothesis is thus best read as a description of the rescaling the system attempts, not a guarantee that it always succeeds.

Geometric Illusions and Constancy Scaling

If perceived size is set by rescaling retinal angle with distance, then a stimulus that supplies misleading distance information should produce a predictable size error, and this is the basis of the most influential account of the geometric illusions. Richard Gregory argued that figures such as the Ponzo and Müller-Lyer illusions contain perspective cues — converging lines, corners implying near and far edges — that trigger the constancy mechanism inappropriately: the visual system treats a part of the flat figure as more distant and, applying the invariance rule, scales up its perceived size even though the figure lies at a single fixed distance (Gregory, 1963). On this view the illusion is not a malfunction but the normal size-scaling process misapplied to a stimulus engineered to fool it, so the very mechanism that produces veridical constancy in the world produces error in the picture.

The account makes the illusions continuous with constancy rather than a separate curiosity: the two lines of the Ponzo figure that flank the same physical length are read as lying at different depths along the implied corridor, and the “farther” one is enlarged exactly as a truly more distant object of equal retinal size would be. Inappropriate constancy scaling remains contested — some size illusions survive when obvious depth cues are removed, and low-level contour interactions contribute alongside the depth account — but the core insight, that depth cues drive size rescaling whether or not real depth is present, is well supported and unifies a wide range of effects (Gregory, 1963). The second demonstration builds a Ponzo figure whose converging cues the reader can strengthen or remove, changing the size illusion in step.

The Moon Illusion

The oldest recorded size illusion is the moon illusion: the moon looks markedly larger near the horizon than high in the sky, although its distance from the observer, and therefore its retinal angle, is essentially unchanged between the two positions. The dominant explanation is again a size–distance effect. Kaufman and Rock proposed the apparent-distance theory: the horizon sky, seen across a filled expanse of terrain, is perceived as farther away than the empty sky overhead, so a moon of fixed retinal angle at the horizon is registered as more distant and, by the invariance rule, is scaled up in perceived size (Kaufman & Rock, 1962). Their experiments, using artificial moons whose apparent distance could be manipulated, showed that increasing perceived distance increased perceived size in the way the theory requires.

The illusion has a paradoxical feature that has kept it under debate for centuries: asked directly, many observers say the larger horizon moon also looks closer, the opposite of what the apparent-distance theory posits as its cause. The standard resolution distinguishes the registered distance that drives the size computation from the distance a person reports after the enlarged moon has already been perceived — a large object is judged near, reversing the felt distance without touching the mechanism that produced the size (Ross & Plug, 2002). A rival family of explanations dispenses with apparent distance altogether and appeals to relative size: on Restle's account the moon's perceived size is fixed by the ratio of its angular extent to that of the objects framing it, so the horizon moon, measured against nearby terrain, looms larger than the zenith moon adrift in an empty sky (Restle, 1970). No single account commands universal assent — the apparent-distance and relative-size mechanisms may both contribute, alongside oculomotor factors — but the moon illusion remains the most vivid natural demonstration that perceived size is constructed rather than read from the retinal image alone (Ross & Plug, 2002).

Perception Versus Action

Perceiving an object's size for report is not the only thing the brain does with size; it also scales the hand to the object when reaching to grasp it, opening the grip to an aperture matched to the target's width well before contact. A striking body of work shows that these two uses of size can come apart. When observers grasp a disc embedded in a size-contrast illusion such as the Ebbinghaus display, the maximum grip aperture their hand adopts is governed largely by the disc's real size, even as they consciously perceive — and report — the illusory size that the surrounding context imposes (Haffenden & Goodale, 1998). The dissociation was taken as evidence for two visual streams, a ventral pathway computing the size available to perception and memory and a dorsal pathway computing the metrics that guide action in real time.

The dissociation is not absolute, and the conditions under which grasp resists the illusion have been debated, with obstacle-avoidance and calibration accounts qualifying the original claim. But the deeper lesson has held: the size that reaches awareness and the size that calibrates the hand can be computed differently and can disagree. More recent work has begun to specify what each computation uses. When vision is restricted, providing distance through proprioception — the felt position of the arm — restores near-perfect size constancy in the scaling of the grasp while leaving perceptual size constancy incomplete, showing that the action system can draw on non-visual distance signals that perception does not fully exploit (Chen, Sperandio, & Goodale, 2018). The third demonstration places an object inside an adjustable size-contrast context and reads out a distorted perceptual estimate alongside a comparatively veridical grip aperture.

The Neural Basis of Perceived Size

If perceived size is rescaled retinal size, a natural question is where in the brain the rescaling shows up, and modern imaging locates its signature strikingly early. Primary visual cortex, area V1, is retinotopically organised: nearby points on the retina map to nearby points on the cortical sheet, and a larger retinal image activates a larger cortical area. The surprise is that V1 activity tracks perceived size beyond retinal size. Using a scene in which two spheres cast identical retinal images but are made to look different in size by their depicted distance, the extent of V1 activation was larger for the sphere that looked larger, even though its retinal image was the same, so the cortical representation followed the percept rather than the stimulus (Murray, Boyaci, & Kersten, 2006).

Converging evidence from afterimages makes the point without any change in the retinal stimulus at all. An afterimage has a fixed retinal size, yet by Emmert's law it looks larger when projected onto a far surface than a near one, the relation Boring set out in formalising size constancy (Boring, 1940); retinotopic V1 activity for such an afterimage expands and contracts with its perceived size, tracking the felt size rather than the constant retinal image (Sperandio, Chouinard, & Goodale, 2012). Individual differences point the same way: the physical surface area of a person's V1 predicts how strongly they experience size illusions, with a smaller V1 associated with larger illusory size differences (Schwarzkopf, Song, & Rees, 2011). The rescaling is not instantaneous or purely local, however. Manipulating real viewing distance shows that constancy-consistent activity in early visual cortex evolves over time as the computation resolves (Chen, Sperandio, Henry, & Goodale, 2019), and the temporal order of processing places higher-tier lateral occipital cortex ahead of the size-specific modulation seen in early visual areas, consistent with feedback carrying a distance-corrected size signal back to V1 (Zeng, Fink, & Weidner, 2020).

Worked Example

The size–distance invariance hypothesis can be worked by hand, and the first demonstration reproduces the calculation. Consider a person 1.7 metres tall standing on a ground plane. At a distance of 10 metres the person subtends a visual angle of 2 × arctan(0.85 / 10) = 0.1697 radians, about 9.7 degrees. Move the person to 20 metres and the angle falls to 2 × arctan(0.85 / 20) = 0.0849 radians, about 4.9 degrees: the retinal image has halved, exactly as the doubling of distance requires.

Now apply the invariance rule, perceived size ≈ perceived distance × retinal angle. If the ground texture and disparity let the observer register the true distances, perceived size at 10 metres is 10 × 0.1697 = 1.697 metres and at 20 metres is 20 × 0.0849 = 1.698 metres: the two changes cancel and perceived size is constant at the person's real height, which is size constancy. Suppose instead the far distance is underestimated as only 14 metres while the retinal angle stays 0.0849 radians. Perceived size becomes 14 × 0.0849 = 1.188 metres, so the person now looks about 30 per cent shorter — the regression toward retinal size that reduced-cue viewing produces. The moon illusion is the same arithmetic run the other way: two moons of equal angular size 0.0087 radians (half a degree), one registered at a relative distance of 1.0 and the other, at the horizon, at 1.5, are scaled to perceived sizes in the ratio 1.5 to 1, so the horizon moon looks half again as large from nothing but a difference in registered distance.

Discussion

Size perception is a case study in perception as inference. The retinal image fixes only the ratio of physical size to distance, so a determinate percept of size requires the visual system to supply the missing term, distance, and to combine it with retinal angle by something close to the size–distance invariance rule (Ittelson, 1951; Epstein et al., 1961). When the distance estimate is good, the combination yields size constancy, the quiet competence that lets us see a receding friend as constant in size rather than dwindling (Holway & Boring, 1941; McKee & Welch, 1992). When the distance estimate is manipulated — by a picture's perspective, by the terrain beneath the horizon moon, by a surface onto which an afterimage falls — the same rule produces a proportional error, which is why the illusions are not exceptions to the account but its clearest confirmation (Gregory, 1963; Kaufman & Rock, 1962).

Two developments complicate and enrich the classical picture. The first is that there is no single size: the size that reaches awareness and the size that scales the reaching hand can be computed by different pathways from different information and can disagree, so “perceived size” must be indexed to the use to which it is put (Haffenden & Goodale, 1998; Chen et al., 2018). The second is that the rescaling is now visible in the brain, and remarkably early: activity in primary visual cortex reflects perceived rather than retinal size, shaped by feedback from higher areas that carry the distance-corrected estimate (Murray et al., 2006; Sperandio et al., 2012; Zeng et al., 2020). Size perception thus turns out to be neither a low-level readout nor a purely cognitive judgement but a computation distributed across the visual hierarchy, in which the earliest cortical map already speaks the language of the perceived world rather than the retina.

Common Misconceptions

We perceive size directly from the retinal image.

The retinal image fixes only the visual angle, which equals physical size divided by distance, so it is consistent with endlessly many sizes. A determinate size percept requires the system to estimate distance and rescale the angle by it, which is why apparent size collapses toward retinal size only when distance cues are stripped away (Holway & Boring, 1941; Ittelson, 1951).

The moon illusion is caused by the atmosphere magnifying the moon near the horizon.

The moon's retinal angle is essentially the same at the horizon and overhead; no optical magnification occurs. The illusion is perceptual, arising because the horizon sky is registered as more distant, which the size-scaling mechanism converts into a larger perceived size (Kaufman & Rock, 1962; Ross & Plug, 2002).

There is one perceived size that governs everything we do with an object.

The size available to conscious report and the size that calibrates the grasping hand can be dissociated: an object embedded in a size illusion can be misperceived while the grip aperture stays close to its real size. Size is computed more than once, for different purposes, and the computations can disagree (Haffenden & Goodale, 1998; Chen et al., 2018).

Glossary

Afterimage.

A persisting visual impression of fixed retinal size that remains after a bright stimulus is removed; by Emmert's law its apparent size grows with the distance of the surface onto which it is projected.

Apparent-distance theory.

The account of the moon illusion on which the horizon moon looks larger because the horizon sky is registered as more distant, so a fixed retinal angle is scaled to a greater perceived size.

Bayesian cue integration.

The reliability-weighted combination of several estimates of a property, in which each cue's influence is proportional to its precision, so the pooled estimate is more reliable than any single cue; visual and haptic size are combined in just this near-optimal way.

Emmert's law.

The generalisation that the perceived size of an afterimage is proportional to the perceived distance of the surface against which it is viewed; a direct expression of size–distance scaling.

Familiar size.

Prior knowledge of an object's typical physical extent, which can serve as a cue biasing the joint estimate of its size and distance toward the remembered value.

Grip aperture.

The separation between finger and thumb as the hand pre-shapes to grasp an object; its maximum during a reach is scaled to the target's size and can be more veridical than conscious size judgements.

Inappropriate constancy scaling.

Gregory's explanation of geometric size illusions, on which depth cues in a flat figure trigger the size-scaling mechanism as if part of the figure were more distant, enlarging its perceived size.

Moon illusion.

The apparent enlargement of the moon near the horizon relative to the zenith, despite an unchanged retinal angle; the most familiar natural size illusion.

Off-sized perception.

Gogel and Da Silva's account on which a familiar object is perceived at the distance its assumed size implies, so that an object built larger or smaller than its typical size is misjudged in both its size and its distance.

Ponzo illusion.

A geometric illusion in which two equal horizontal lines placed across converging lines appear unequal, the one nearer the convergence looking larger, commonly attributed to misapplied depth scaling.

Reduced viewing.

A condition in which distance cues are removed — for example by monocular viewing through a tube — so that apparent size regresses toward the value specified by retinal angle alone.

Relative-size theory.

An account of the moon illusion on which perceived size depends on the ratio of the moon's angular extent to that of the surrounding objects, rather than on its apparent distance.

Retinal (visual) angle.

The angle an object subtends at the eye, equal to its physical size divided by its distance; the proximal stimulus from which perceived size must be recovered.

Size constancy.

The tendency to perceive an object's physical size as stable across changes in distance, despite the corresponding changes in its retinal image; the everyday achievement of the size-scaling system.

Size–distance invariance hypothesis.

The proposal that a given retinal angle specifies a fixed ratio of perceived size to perceived distance, so that perceived size is perceived distance times the retinal angle and any error in distance forces a proportional error in size.

Two visual streams.

The distinction between a ventral pathway supporting perception and recognition and a dorsal pathway supporting the visual control of action, invoked to explain dissociations between perceived size and grasp.

V1 (primary visual cortex).

The first cortical stage of visual processing, retinotopically mapped, in which the spatial extent of activation tracks an object's perceived size and not merely the size of its retinal image.

Key Researchers

Edwin G. Boring. Harvard University (1886–1968); he formalised size constancy and its link to Emmert's law, and with Holway ran the classic corridor experiment showing that apparent size follows physical size when distance cues are rich and collapses toward the retinal image as those cues are removed. Faculty Page - Wikipedia - Wikidata

Melvyn A. Goodale. University of Western Ontario, Brain and Mind Institute; through the two-visual-streams framework he showed that pictorial size illusions distort conscious perception far more than the grip aperture used to grasp an object, and later work with Chen and Sperandio traced size constancy for grasping and for afterimages to computations that draw on distance signals perception does not fully exploit. Faculty Page - Google Scholar - Wikipedia - Wikidata - ORCID

Richard L. Gregory. University of Bristol (1923–2010); he advanced the constructivist view of perception as hypothesis testing and proposed inappropriate constancy scaling as the mechanism of geometric size illusions, tying the misperceived size of illusion figures to depth cues that misfire. Obituary - Wikipedia - Wikidata

Irvin Rock. Rutgers University and University of California, Berkeley (1922–1995); with Kaufman he provided the influential apparent-distance account of the moon illusion and, in The Logic of Perception, defended an indirect theory on which perceived size is computed from perceived distance. Wikipedia - Wikidata

D. Samuel Schwarzkopf. University of Auckland (School of Optometry and Vision Science); he showed that individual differences in the surface area of primary visual cortex predict the magnitude of a person's subjective size illusions, tying the conscious experience of object size to the physical layout of V1. Faculty Page - Google Scholar - ORCID

Irene Sperandio. University of Trento (Department of Psychology and Cognitive Science); she provided direct neural evidence for size constancy, showing that retinotopic V1 activity for an afterimage tracks its perceived rather than its retinal size, and mapped the mechanisms by which the brain rescales retinal size using distance. Faculty Page - Google Scholar - ORCID

Frequently Asked Questions

What is size perception in psychology?
It is the set of visual processes that recover an object's real physical size from its retinal image. Because the retinal image size equals physical size divided by distance, size cannot be read directly; the brain estimates distance and rescales the retinal angle by it, which is what allows an object to look the same size as it moves nearer or farther (Ittelson, 1951; Sperandio & Chouinard, 2015).

What is size constancy?
Size constancy is the tendency to perceive an object's physical size as stable despite changes in the size of its retinal image with distance. It is close to complete when distance cues are rich and breaks down toward retinal matching when those cues are removed, as the classic corridor experiment demonstrated (Holway & Boring, 1941; McKee & Welch, 1992).

What is the size–distance invariance hypothesis?
It is the proposal that a given retinal angle fixes a constant ratio between perceived size and perceived distance, so perceived size is approximately perceived distance times the retinal angle. It explains constancy and predicts that any error in perceived distance produces a proportional error in perceived size (Epstein, Park, & Casey, 1961).

Why does the moon look bigger near the horizon?
Its retinal size is essentially unchanged, so the effect is perceptual. On the apparent-distance theory, the horizon sky is registered as farther away than the sky overhead, and a moon of fixed retinal angle at a greater registered distance is scaled to a larger perceived size (Kaufman & Rock, 1962; Ross & Plug, 2002).

How do depth cues cause size illusions?
Geometric illusions such as the Ponzo figure contain perspective cues that make part of a flat figure read as more distant. The size-scaling mechanism then enlarges that part as though it were a farther object of equal retinal size, producing the illusion that Gregory called inappropriate constancy scaling (Gregory, 1963).

Can we misjudge an object's size yet still grasp it accurately?
Yes. When an object sits inside a size-contrast illusion, the grip aperture the hand adopts is scaled largely to the object's real size even while the person consciously perceives the illusory size, a dissociation attributed to separate visual pathways for perception and action (Haffenden & Goodale, 1998; Chen, Sperandio, & Goodale, 2018).

Where in the brain is perceived size represented?
As early as primary visual cortex. The extent of V1 activation tracks an object's perceived size rather than its retinal size, and this holds even for afterimages of fixed retinal size, with individual V1 surface area predicting how strongly a person experiences size illusions (Murray, Boyaci, & Kersten, 2006; Sperandio, Chouinard, & Goodale, 2012; Schwarzkopf, Song, & Rees, 2011).

Is size constancy a fixed geometric rule or an active computation?
It is an active computation. Reaction times are faster when retinal size and depth are consistent with one physical size and slower when they conflict, and constancy-consistent cortical activity develops over time and depends on feedback from higher visual areas, indicating that size and distance are integrated dynamically rather than read off a fixed identity (Plewan & Rinkenauer, 2017; Chen, Sperandio, Henry, & Goodale, 2019; Zeng, Fink, & Weidner, 2020).

References