RESEARCH DOSSIER
Depth and Three-Dimensional Perception
Abstract. Depth and three-dimensional perception reconstructs spatial layout, relative depth, surface slant, and volumetric structure from projective sensory signals whose physical causes are underdetermined by any individual retinal image. This fundamental research monograph synthesizes classical stereoscopic experimentation, random-dot stereograms, motion-parallax studies, occlusion and pictorial depth research, binocular cortical physiology, psychophysical measurement, quantitative models of reliability-sensitive cue integration, development, and stereoscopic display science. The evidence supports the independent contributions of binocular, pictorial, and dynamic signals, together with condition-dependent combination and learned geometric assumptions. It also reveals dissociations between early neural disparity responses and perceived depth, and between sensitivity, scale calibration, and behavioral action. Critical analysis incorporates 2021–2025 experiments on disparity detection versus depth-direction judgments, shading-induced deformation of stereoscopic shape, contextual neural processing, natural-image variability, spatial transformations in virtual reality, and the relation between stereopsis and short-term memory. These findings test the limits of ideal-observer models and establish new measurement problems. The concluding synthesis places the subject within Perception and Psychophysics, Cognitive Psychology, and Human Factors and Engineering Psychology, with a carefully bounded comparative program for artificial three-dimensional perception.
Keywords: depth perception; three-dimensional perception; stereopsis; binocular disparity; stereo correspondence; random-dot stereogram; occlusion; visual surface completion; linear perspective; texture gradient; motion parallax; shape from shading; cue integration; Bayesian perception; stereoacuity; vergence; accommodation; psychophysics; visual neuroscience; artificial perception
Depth as a Psychological Research Object
Depth and three-dimensional perception concerns the estimation of spatial layout, surface relief, object separation, and the relative or absolute location of environmental structures from sensory evidence. An image on either retina is a perspective projection whose dimensions are compatible with multiple arrangements in space. The problem is therefore one of inverse inference: an observer must recover enough structure to guide grasping, locomotion, orientation, and recognition while the distance, viewing angle, illumination, and motion of the observer vary. The scientific object includes perceived depth ordering, metric distance, surface slant, curvature, and volume, which must be distinguished because success on one judgment does not imply accuracy on another. The comprehensive analyses of visual depth in Howard (2012a) and Howard & Rogers (2012b) establish the breadth of mechanisms under investigation.
A depth cue is a measurable relation in sensory input that constrains some property of three-dimensional geometry. Binocular disparity provides differences between corresponding projections; occlusion constrains ordering; perspective, texture, and shading constrain surface orientation under scene assumptions; self-motion generates systematic retinal transformations. These sources vary in scale, ambiguity, and reliability, and may provide ordinal rather than quantitative information. A robust psychological description therefore specifies the task and the physical mapping between the stimulus and the target property rather than treating all cases as a single faculty of seeing distance. Research on cue combination explicitly separates the information made available by a cue from the weight the observer assigns to it Landy et al. (1995).
The monograph synthesizes historical experiments, geometry-calibrated psychophysics, neural physiology, probabilistic modeling, developmental work, and display research. The interpretation of each finding follows the actual evidence: controlled demonstrations can establish sufficiency under artificial conditions, threshold experiments quantify sensitivity, and cortical recordings constrain possible representations. A report of compelling relief cannot by itself identify a unique neural computation. Likewise, a statistically efficient behavioral estimate within a task is evidence for a predictive model of performance, not a demonstration that every perceptual operation implements that model. These distinctions guide evaluation of the original stereoscope, random-dot displays, motion-parallax manipulations, cue-conflict paradigms, and contemporary stereoscopic technologies Julesz (1964) Banks et al. (2012).
The Projective Geometry of Seeing in Three Dimensions
The retinal projection of a scene depends on object position, eye position, eye rotation, and the optical properties of the eye. Pictorial projection makes physically large distant objects capable of producing the same retinal extent as smaller nearby objects. For this reason retinal size alone does not determine physical size or distance. Perspective convergence, changes of projected texture density, horizon relations, and the angular orientation of contours constrain candidate scenes only with assumptions about the distribution and regularity of objects. The geometrical approach remains crucial because it predicts which changes in an experimental stimulus should change the observer's depth estimate and which should be irrelevant under a specified model Howard (2012a).
An occlusion contour is especially informative about depth order: when one continuous surface interrupts the visible boundary of another, the interrupted surface is ordinarily interpreted as farther away. At an occluding edge, information about the concealed region is absent, yet the observer may perceive continuation of the surface behind the foreground object. This surface-completion problem belongs to three-dimensional organization because the perceived scene contains regions not specified in either image. The difference between intrinsic object boundaries and extrinsic boundaries created by occluders was analyzed through studies of stereoscopic grouping and recognition, which showed how depth relations can organize figure structure Nakayama et al. (1989).
Distance information should also be described relative to the geometry of the observer. Head movements change viewpoint, gaze shifts change retinal coordinates, and fixation affects the relative location of binocular image features. Consequently experiments that compare judgments across targets must control eye position or estimate it separately. Geometrical calibration is not a cosmetic property of a stimulus: inaccurate viewing distance changes disparity-to-depth conversion and can confound putative changes of perceptual sensitivity. The design of stereo media makes this problem visible when screen position, viewing distance, and camera geometry generate conflicting depth information or visual discomfort Banks et al. (2012).
Wheatstone and the Experimental Birth of Stereopsis
Charles Wheatstone's 1838 investigations transformed the relation between binocular geometry and visual experience into a controlled laboratory question. His mirror stereoscope supplied each eye with a different drawing corresponding to a separate viewpoint, allowing an observer to experience a solid arrangement although each physical drawing remained on a plane. The decisive result was that differences between the two monocular perspectives could generate relief, placing binocular comparison at the center of experimental depth research. Wheatstone's original article appeared in volume 128 of Philosophical Transactions of the Royal Society of London, pages 371–394, and remains available through historical archival collections Wheatstone (1838).
The historical experiment mattered because it isolated a source of spatial information from the many cues normally correlated in the world. Everyday objects simultaneously produce accommodation, shading, disparity, motion, and contextual familiarity; the stereoscope manipulated binocular projection while holding the material plane of the images fixed. This separation established a methodological principle still central to psychophysics: a visual cue can be probed through an apparatus that changes its value independently of other sources. The method also exposed a recurring interpretive difficulty. A stereoscopic image may depict an apparent three-dimensional object without specifying correct accommodation, and the strength of relief does not imply that the depicted object's metric scale has been recovered Wheatstone (1838) Banks et al. (2012).
Wheatstone's findings are often described as the discovery of stereopsis, but the substantive contribution concerns experimentally demonstrated binocular relief from paired perspective images. Later scientific work refined the geometry, correspondence mechanisms, and neural processing that his apparatus left unresolved. Historical attribution should not compress the entire subsequent program into the first device: random-dot stereograms clarified the role of interocular correlation, visual-cortex recordings tested disparity sensitivity, and computational theories analyzed false matches. The lineage from nineteenth-century demonstrations to contemporary research is best understood as a sequence of progressively sharper experimental separations of spatial information and its mechanisms Julesz (1964) Poggio & Poggio (1984).
Image Disparity, Fixation, and Depth Sign
Binocular disparity describes the difference in retinal location of corresponding scene features under simultaneous viewing with the two eyes. A point lying at the fixation distance can have zero disparity relative to the particular reference geometry, whereas objects at other depths usually yield crossed or uncrossed relative disparity according to their position and configuration. The magnitude of disparity does not translate to a fixed physical depth independently of eye separation, viewing distance, and gaze. Therefore a measured angular offset and a judged metric distance are analytically different quantities. Experimental stereoscopy manipulates offset with precise paired images while controlling the conditions under which an observer should interpret the offset Howard & Rogers (2012b) Banks et al. (2012).
Relative disparity between two features may support judgments of their ordering or separation even when the observer has imperfect knowledge of absolute fixation. Absolute disparity relative to gaze incorporates the eye's current reference position, and its use becomes especially relevant in calibration of a screen or virtual scene. The distribution of disparities over a surface also conveys orientation and relief through gradients, rather than simply a collection of isolated object depths. Such complexity explains why stereopsis research includes contour correspondence, disparity gradients, surface slant, and disparity-defined shapes. A single stereoacuity threshold, measured at one central location and viewing distance, cannot summarize every aspect of this geometric ability Poggio & Poggio (1984) Howard & Rogers (2012b).
Vergence alters the angular relations between the two retinal images as the eyes rotate toward a common target. Accommodation adjusts the optical focus of the crystalline lens, and the two responses are ordinarily linked during natural near viewing. The correlation between vergence, accommodation, and disparity supports multiple kinds of distance information, but each has different reliability and dynamics. In fixed-focus stereoscopic displays the relationship may be altered, creating a vergence–accommodation conflict. Psychophysical studies of focus cues and engineering analyses show why apparent depth, binocular comfort, and focus behavior must be considered together rather than treated as interchangeable outcomes Watt et al. (2005) Banks et al. (2012).
Random-Dot Stereograms and the Isolation of Binocular Structure
Béla Julesz's random-dot procedures made it possible to examine stereopsis while greatly reducing familiar monocular object information. Corresponding random patterns shown separately to the two eyes produce a coherent depth-defined region when the interocular displacement is appropriately organized. The methodological advantage is substantial: a shape that cannot be recognized from either monocular half-image can emerge as a binocular percept. This demonstrates that visual depth can depend on comparisons made between inputs without first requiring the observer to recognize a familiar object in either input. Julesz's 1964 Science article documented controlled random-dot and tachistoscopic methods to separate binocular organization from familiarity cues Julesz (1964).
The random-dot result is an existence demonstration for binocular extraction of depth from correlation structure; it is not proof that all natural stereopsis proceeds independently of monocular segmentation. Natural images contain texture gradients, boundaries, shadows, and occlusion, each of which changes which elements are plausible binocular correspondences. In the laboratory, the density, size, contrast, and correlation of dots affect discrimination and fusion. A random-dot threshold therefore measures sensitivity under a specific statistical distribution of image features. Studies of local and global stereopsis using sparse and vernier-like random structure reinforce this dependence on the exact stimulus and task Julesz & Spivack (1967) Howard & Rogers (2012b).
One major research question is how the brain distinguishes a true image correspondence from visually similar but geometrically impossible alternatives. When many local elements are alike, matching a point in one eye to the wrong point in the other can generate inconsistent surfaces. Psychophysical fusion limits and neural recordings help discriminate candidate computational solutions, and later models explicitly include disparity selectivity that may respond to impossible pairings as a step in rejecting them. These results illustrate how a laboratory stimulus can expose both the computational problem and the limits of a simple interpretation of the neural response Read & Cumming (2007) Cumming & Parker (1997).
Correspondence, Fusion, and Binocular Competition
Stereo correspondence is the assignment of features in one eye to features originating from the same environmental points in the other eye. Image similarity supplies candidate matches, while continuity, orientation, spatial frequency, disparity gradients, and occlusion restrict the set of possible surfaces. Matching is difficult when periodic textures create repeated alternatives, and when some surfaces are visible to only one eye. A successful theoretical account must explain fused depth, double vision at large disparity, and the treatment of half-occluded regions. The proposal that occlusion relations are detected during early binocular processing addresses phenomena that ordinary feature-pairing accounts leave unexplained Anderson & Nakayama (1994).
Fusion denotes coordinated binocular representation compatible with a unified percept, but images can also compete when features are too inconsistent to combine. Binocular rivalry, suppression, and diplopia reveal that the relation between inputs is conditional rather than a uniform averaging operation. Gain-control models additionally consider how interocular contrast changes the influence of each eye on a common representation. In experiments in which eye-specific contrast and content are varied, a binocular response can depend strongly on normalization rather than on disparity alone. These effects matter for interpreting clinical stereo tests, display artifacts, and neural estimates obtained under unequal monocular visibility Ding & Sperling (2006) Anderson & Nakayama (1994).
Neural sensitivity to the disparity of two features is a necessary candidate substrate for stereopsis, yet a signal tuned to image correlation need not directly represent consciously perceived distance. The distinction was made concrete by recordings showing disparity-related activity in primary visual cortex under anticorrelated random-dot stimulation that did not produce the expected depth percept. The physiological signal therefore requires subsequent transformations and selection mechanisms before being equated with perception. This dissociation counsels against interpreting a single cortical tuning function as an exhaustive explanation of three-dimensional experience, and it motivates convergent measurements of neural responses and psychophysical behavior Cumming & Parker (1997) Read & Cumming (2007).
Occlusion, Surface Completion, and Depth Ordering
Occlusion supplies one of the most stable ordinal constraints in natural viewing. When the contour of one surface terminates against a second continuous boundary, the first may be represented as continuing behind the second. The resulting percept is an organized scene with partially invisible surfaces, not a list of two-dimensional patches. Stereoscopic experiments show that depth relations help distinguish intrinsic contours belonging to an object's actual outline from extrinsic contours created by foreground occluders. This distinction is important for recognition of partially hidden objects because a visible fragment must be assigned to a plausible complete surface Nakayama et al. (1989).
Binocular viewing complicates the classical rule of occlusion because an object may obscure a feature for one eye while leaving it visible to the other. Such monocularly visible regions, sometimes called half-occlusions, are geometrically structured and can themselves specify depth. Nakayama and Shimojo examined unpaired binocular image points in the context of da Vinci stereopsis, providing evidence that observers exploit information that cannot be expressed as a disparity between two matched points. Their results challenge any theory treating unmatched regions solely as discarded noise and demonstrate the need for a joint account of matching and occluding surfaces Nakayama & Shimojo (1990) Anderson & Nakayama (1994).
Surface completion is also a test of how perception uses assumptions about possible worlds. The same local contour can arise from different shapes, so continuity, convexity, and contextual layout restrict interpretation. Naturalistic assumptions are sometimes useful, but they can generate illusions when an experimental arrangement violates them. The inference should be made from controlled changes in grouping and occlusion rather than from intuition about how an image ought to look. The theoretical treatment of generic viewpoints links experienced surface structure with the statistical regularities of optical sampling across observer positions Nakayama & Shimojo (1992) Kersten et al. (2004).
Linear Perspective, Texture, and Projective Shape
Linear perspective constrains layout because parallel lines in a physical scene generally approach common vanishing directions under projection. The relation carries strong information when the observer assumes that pictured lines represent parallel environmental structures. A perspective image can nevertheless be constructed from an artificial geometry in which those assumptions fail. Texture gradients have a similar conditional structure: projected size, spacing, and density vary with surface slant and distance under assumptions about the distribution of repeated elements on the surface. A laboratory study of these cues should specify the statistical properties of texture and the surface coordinate system, allowing physical and perceived slant to be compared Howard (2012c) Knill & Saunders (2003).
Texture is particularly valuable in cue-integration experiments because its reliability changes as a function of stimulus geometry. Knill and Saunders measured slant discrimination from texture and stereo information as well as their combined contribution. The published results were consistent with reliability-sensitive weighting, though the authors reported an apparent modest underweighting of texture relative to the specified optimal predictor. The importance of the study lies in comparing individual cue thresholds with the corresponding weights rather than merely observing that a scene looks more three-dimensional when two signals are present Knill & Saunders (2003).
Perspective and texture have a dual theoretical significance. They can serve as individually informative image structures and can constrain interpretation of binocular disparity by specifying how a coherent surface should extend in the scene. Inconsistent cues may yield weighted compromise, apparent distortions, switching, or suppression depending on separation and task demands. Consequently adding realistic textures to a stereogram is not a neutral enhancement: it can modify the depth estimate itself. Quantitative modeling should evaluate uncertainty in both sources and ask whether apparent improvements reflect more precise estimates, changed response criteria, or a transformed representation of surface geometry Landy et al. (1995) Hillis et al. (2004).
Shading, Lighting Assumptions, and Shape from Image Intensity
Shading provides gradients of image intensity that vary with surface orientation, illumination, and material reflectance. To infer shape from these gradients, vision must make assumptions about which changes belong to geometry and which arise from lighting or surface pigmentation. A convex bump illuminated from above and a concave indentation under a different light direction can sometimes produce similar luminance patterns. Shape-from-shading therefore illustrates the fundamental ambiguity of inverse optics: an image does not reveal its generating physical causes by inspection alone. Analyses of probabilistic object perception identify lighting, reflectance, and scene priors as critical constraints for the interpretation of such image information Kersten et al. (2004).
Combining shading with occlusion and perspective often stabilizes perceived relief because the cues constrain different dimensions of the possible scene. Yet the agreement among several cues must be evaluated at the level of experimental control, not assumed from a naturalistic impression. For example, an image can contain a realistic cast shadow whose position suggests depth while the underlying retinal disparity remains fixed. The shadow may change judgments by modifying expected light direction, object contact, or the segmentation of adjacent surfaces. The methodological lesson is to measure the response to each candidate source separately and to document the geometry used to render the stimulus Howard (2012c) Landy et al. (1995).
Surface material further complicates quantitative depth measurement. Specular highlights, transparency, and patterned reflectance produce local image signals that cannot always be attributed to a single solid surface. The observer may perceive multiple layers or reinterpret contours in light of apparent transparency, thereby changing depth organization. These findings extend the research problem beyond elementary lists of monocular cues: the critical computation involves attributing image evidence to physical sources. Models of object inference describe the challenge in terms of selecting likely scenes under uncertain image formation, while psychophysical experiments determine which assumptions observers actually use Kersten et al. (2004) Nakayama & Shimojo (1992).
Motion Parallax and Depth from Changing Viewpoint
Motion parallax arises when observer movement changes the relative image positions of environmental objects at different distances. Under common translations of the eye, nearby and distant points undergo different retinal transformations, permitting estimates of relative depth even when static perspective and binocular disparity are unavailable. The information depends on knowing or estimating observer motion, because retinal motion could also arise from independently moving objects. Experiments must therefore distinguish retinal velocity fields from observer-controlled movement and specify the geometrical reference frame in which depth is inferred. Ecological accounts emphasize that perception unfolds during activity, and motion-parallax experiments permit that principle to be isolated and measured Rogers & Graham (1979) Gibson (1979).
Rogers and Graham used computer-generated random-dot transformations to manipulate the visual motion associated with three-dimensional surfaces. Their 1979 experiments showed compelling depth from motion parallax in displays designed to minimize alternative cues, and a stereoscopic matching procedure related the perceived depth to the magnitude of simulated image displacement. The article is an important counterexample to a theory that assigns all precise visual relief exclusively to stereopsis. Yet the specific dynamics of the display and the reported perceptual matches do not establish complete metric veridicality across natural environments or arbitrary observer trajectories Rogers & Graham (1979).
Visual systems must separate translation, rotation, object movement, and depth-induced flow components when the observer moves. Rotation can change retinal direction without the same dependence on scene depth as translation; independently moving objects add additional transformations. Perceived three-dimensional structure is therefore linked to motion estimation, stabilization, and the observer's ability to predict changes across successive views. Theories emphasizing active optical sampling treat this interaction as a source of evidence rather than as contamination of a static visual representation. The relevant empirical tests compare known simulated self-motion with judgments of depth and examine the effects of head movement and task demands Nakayama & Shimojo (1992) Howard (2012c).
Depth Constancy, Scale, and Distance Calibration
A depth estimate may preserve ordinal relations while exhibiting substantial errors in metric scale. Stereo matching can indicate which of two surfaces is nearer, but conversion from angular disparity to distance depends on viewing geometry and the calibration of vergence. Similarly, texture and familiar size can indicate a probable distance only through assumptions about actual object dimensions. Absolute distance, perceived size, and perceived surface curvature are related outputs of scene interpretation rather than interchangeable names for one variable. This distinction is vital when comparing experiments: a task asking observers to match visible separation may yield different results from one requiring a grasping movement or a numerical distance estimate Howard (2012a) Banks et al. (2012).
Depth constancy refers to stability of perceived structure when changes in viewing distance or gaze change the retinal information. It requires compensating transformations whose effectiveness can vary with available fixation, perspective, and environmental context. Stereo-camera geometry offers a concrete illustration: images photographed with a particular separation and displayed at a different viewing distance may yield altered apparent depth and scale. Banks and colleagues analyzed these relationships in the context of stereoscopic media, showing how errors in the mapping from capture to display geometry can distort apparent spatial structure and influence visual comfort Banks et al. (2012).
Quantitative psychophysical tasks should therefore distinguish threshold from bias. A person can detect very small depth differences and still systematically misestimate the size of a larger interval, or estimate mean depth accurately while showing poor precision on repeated trials. Forced-choice discriminations can estimate sensitivity while comparison or adjustment tasks estimate point of subjective equality. If a cue-conflict study measures only the mean response, changes in variability may be mistaken for better calibration. Methodological frameworks for depth cue combination explicitly recommend perturbation analysis and separate assessment of the constituent cues to identify how weights and biases change Landy et al. (1995) Knill & Saunders (2003).
Depth Illusions as Tests of Geometric Assumptions
Depth illusions reveal dependencies in visual inference when controlled displays produce a systematic difference between physical and perceived structure. A conflict among perspective, shading, occlusion, or disparity can shift the estimate of orientation or spatial order, often in predictable directions. Rather than classifying illusions as failures of vision, psychophysics uses them to identify the assumptions under which a signal is informative. A cue that is usually reliable may mislead in a specially constructed image. The critical scientific step is to relate a manipulated stimulus dimension to a measured response while considering alternate cues that covary with it Kersten et al. (2004) Howard (2012c).
Binocular and pictorial effects are particularly informative when they disagree. In a stereoscopic display, disparity can indicate one surface slant while the projected texture pattern indicates another. The resulting percept can shift toward either source depending on its precision, the base slant, and the observer's prior exposure. Hillis and colleagues evaluated such combined slant signals through a quantitative cue-integration design, identifying cases where behavior followed maximum-likelihood predictions and discussing discrepancies with potentially unmodeled cues. The findings support a conditional model of cue weighting rather than a universal ranking of visual signals Hillis et al. (2004).
Illusory depth also occurs because the observer represents layers and occluded regions, not solely visible features. When cues assign a foreground object and a continuing background, a contour can become perceptually extrinsic and lose influence over perceived object shape. This illustrates how surface organization changes the meaning of an image boundary. Studies of grouping and stereoscopic recognition connect this organization with the reconstruction of partially hidden objects. Such results remain task-dependent: an apparent three-dimensional surface in a display does not automatically imply that the same inference would guide accurate behavior in unrestricted natural scenes Nakayama et al. (1989) Anderson & Nakayama (1994).
Reliability-Weighted Depth Cue Integration
Scenes normally provide several correlated estimates of their spatial structure. A common quantitative strategy models each cue as an uncertain observation of a latent geometrical variable and assigns greater weight to the more precise estimate. For independent Gaussian errors, the minimum-variance combined estimate uses weights proportional to inverse variance. This formulation makes concrete predictions: changing the reliability of texture while holding disparity reliability fixed should change their relative contribution to judged slant. It also predicts a reduction of response variance relative to either isolated source under its independence and unbiasedness assumptions. The influential formulation of measurement, cue promotion, dynamic weighting, and robust fusion was presented by Landy and colleagues Landy et al. (1995).
The evidential force of an optimality test comes from independently estimating the variability associated with the component cues and subsequently predicting combined performance. Knill and Saunders applied this strategy to binocular disparity and texture information for surface slant. Their results broadly supported reliability-dependent integration while also identifying an apparent shortfall in the weight assigned to texture, which they discussed in relation to cue-isolation procedures. Hillis and colleagues extended experimental tests across viewing configurations and slants and compared behavior against maximum-likelihood predictions. These studies constitute controlled evidence for efficient combination within particular tasks; they do not justify an unqualified assertion that human depth perception is globally optimal Knill & Saunders (2003) Hillis et al. (2004).
The optimal-weight rule requires examination of correlated noise, systematic biases, and observer assumptions. If two cues share an image-processing bottleneck or respond to the same uncontrolled distortion, treating their uncertainty as independent can overstate the precision gain. A cue may also be downweighted because it is judged to originate from a separate object, not because its intrinsic sensory precision is low. In practice, fitted weights and estimated cue thresholds may depend on task instructions, stimulus range, and whether participants can detect the conflict. The modified weak-fusion account emphasized these issues, making cue combination a measured process rather than an axiom imposed on all depth judgments Landy et al. (1995) Howard (2012c).
Bayesian and Causal Models of Three-Dimensional Interpretation
A Bayesian model represents uncertainty about candidate scenes and combines likelihood information from retinal measurements with prior expectations about surfaces, illumination, and the arrangement of objects. The central object is a distribution over possible scene structures, rather than an unconditional decision that one interpretation must be true. This approach is particularly suited to depth perception because many physical scenes can yield similar images. Prior expectations about coherent surfaces and ordinary viewpoints can stabilize interpretation; changes in those assumptions can produce systematic perceptual biases. Kersten and colleagues surveyed how probabilistic object inference addresses ambiguity across shape, depth, material, and scene organization Kersten et al. (2004).
The Bayesian vocabulary becomes scientifically informative when alternative models yield separable predictions. A prior over plausible surface orientations, for example, changes the predicted response to noisy texture even if the physical stimulus remains unchanged. A model that integrates disparity and texture must also specify whether the cues are assumed to refer to the same surface. When they originate from different causes, a weighted mean may be a poor representation of the perceptual task. Experimental cue conflicts can therefore test both the numerical weights and the organization of cues into shared or separate sources. Models without independently constrained priors risk fitting almost any observation retrospectively Landy et al. (1995) Kersten et al. (2004).
Evidence for efficient estimation should be interpreted at the level at which efficiency was demonstrated. Ernst and Banks' visual–haptic experiment concerned a different physical judgment, perceived object height, but provides an important crossmodal comparison: measured sensory variances predicted how information from sight and touch was combined. It supports a general research program of reliability-sensitive integration without implying that its parameters automatically transfer to the relation between stereopsis and texture. Theoretical generality requires new tests under new cue statistics, surfaces, and actions, together with evaluation of how the observer's prior learning affects judgments Ernst & Banks (2002) Knill & Saunders (2003).
Cortical Mechanisms and the Processing of Disparity
Binocular inputs reach visual cortex through pathways that combine eye-specific signals into neurons sensitive to particular image structures. Early visual neurons may respond to orientation, spatial frequency, and binocular relations, while later populations support more elaborate surface and object representations. The classical program of analyzing stereo mechanisms related psychophysical matching problems to disparity-selective responses. However, neural disparity tuning does not directly specify which features are matched correctly in a complex scene. To interpret a cell's response as representing perceived depth, the experiment must also address possible correspondences, fusion, and the behavioral consequences of the neural activity Poggio & Poggio (1984) Read & Cumming (2007).
The distinction between disparity encoding and perceptual readout becomes especially clear in anticorrelated random-dot experiments. Cumming and Parker reported neurons in macaque primary visual cortex that preserved disparity modulation for stimuli that did not support the corresponding stereoscopic depth percept. Their result challenges a direct equivalence between early cortical tuning and conscious depth. It does not show that early signals are irrelevant; instead it identifies additional processing that must resolve candidate matches and suppress geometrically inconsistent representations. The experiment is a model of rigorous dissociation between a physiological response and a psychophysical outcome Cumming & Parker (1997).
Read and Cumming proposed that responses to phase disparities, which are problematic as literal natural correspondences, could help reject false image matches. The analysis highlights a computational possibility: a neuron may encode evidence useful for the process of finding correct correspondences rather than carrying the final distance estimate. A mature neural account therefore needs linked tasks, measurements, and predictions at several levels, from retinal geometry to population code and behavior. It must also explain the integration of occlusion and monocular information, which is inadequately captured by a network that considers only local disparity correlations Read & Cumming (2007) Anderson & Nakayama (1994).
Vergence, Accommodation, and Display-Induced Conflict
Natural viewing couples vergence, accommodation, and the binocular disparities produced by the scene, although each source operates through its own mechanism. In a conventional fixed-focus stereo screen, a viewer can converge toward a point depicted in front of or behind the display while optical focus remains at the screen plane. The resulting vergence–accommodation conflict is a concrete case in which technologically generated cues depart from their ordinary relation. It matters to a scientific account because the same display can yield strong stereoscopic relief while making sustained viewing difficult or changing perceived scale. Systematic display research provides a bridge between laboratory spatial perception and engineering applications Banks et al. (2012).
Watt and colleagues investigated focus cues and perceived depth with controlled displays, demonstrating that focus-related information can influence estimates of spatial separation. This result is methodologically important because accommodation cannot be treated as an irrelevant background variable whenever stereo images are used. Changes in aperture, optical blur, simulated distance, and the observer's refractive state may affect an apparent depth response or the ease of stereo fusion. Researchers should distinguish the cue value available in the optical stimulus from the eye's measured focusing behavior, and separate perceptual bias from discomfort or failure to fuse Watt et al. (2005).
Modern three-dimensional interfaces, medical visualization, and remote manipulation generate a practical need for faithful geometry and comfortable viewing. Banks and colleagues reviewed screen geometry, depth cue interactions, fixation and focus, and the temporal properties of stereoscopic imagery. Their analysis shows why a stereo image is an engineered sensory environment whose perceptual properties cannot be inferred from binocular disparity alone. For scientific reproducibility, a report should identify eye separation assumptions, viewing distance, stereoscopic rendering geometry, display size, frame timing, and optical correction when these variables materially affect the outcome Banks et al. (2012).
Psychophysical Tasks, Thresholds, and Measurement Validity
Stereoacuity is commonly assessed by asking how small a binocular disparity difference can be discriminated under specified viewing conditions. Such a threshold depends on stimulus duration, size, contrast, spatial frequency, eccentricity, reference disparity, binocular image correlation, and the response procedure. It should be reported with a clearly defined criterion and uncertainty interval rather than treated as a fixed trait that characterizes all three-dimensional judgments. A forced-choice task can limit some forms of response bias, while an adjustment task may more directly estimate a perceived depth match. These procedures answer related but distinct questions and should be selected according to the research object Howard & Rogers (2012b) Julesz & Spivack (1967).
Threshold estimation also requires attention to lapse rates, training, fatigue, and the number of trials. A psychometric function may relate stimulus disparity to the probability of choosing one interval as nearer, with the slope characterizing discrimination precision and the midpoint reflecting response bias. If a method ignores lapses or ceiling effects, it may misinterpret changes in task engagement as changes in the underlying sensitivity. The strongest designs measure individual baselines, counterbalance stimulus signs and positions, monitor fixation where necessary, and test whether threshold improvements persist across stimulus families Howard & Rogers (2012b) Banks et al. (2012).
Depth matching and cue-conflict experiments can estimate perceptual weights only when the geometry and noise of the isolated cues have been assessed. Landy and colleagues proposed perturbation-style analyses in which one cue is shifted while other signals are controlled. Knill and Saunders compared isolated cue precision and combined slant judgments, providing a stronger test of model predictions than a single comparison between cue-rich and cue-poor displays. Data analysis should distinguish within-observer variability, between-observer heterogeneity, and systematic miscalibration. A group mean that fits an ideal-observer equation can conceal observers using qualitatively different strategies Landy et al. (1995) Knill & Saunders (2003).
Experimental Evidence and Replication Boundaries
The strongest empirical conclusions in depth research concern the existence and conditional effectiveness of specific cues. Stereoscopic relief can be generated by interocularly displaced images; a depth surface can emerge from random-dot correspondence; motion parallax can convey relative three-dimensional structure in otherwise impoverished displays; and texture information can alter perceived surface slant. These are supported by distinct experimental traditions using different apparatus, stimuli, and dependent measures. Their convergence supports a multifactorial understanding of depth perception, while each result retains the limits of its own task and sample Wheatstone (1838) Julesz (1964) Rogers & Graham (1979) Knill & Saunders (2003).
Replication assessment must examine more than whether a later article reproduces a compelling visual demonstration. Small changes in apparatus calibration, head movement, disparity sign, luminance, image correlation, or the definition of threshold can modify estimated performance. Conversely, different displays can lead to similar perceptual effects through different mechanisms. A transparent evidence synthesis separates demonstrations of cue sufficiency from quantitative model comparisons and from neural evidence, records the viewing geometry and observer population, and reports contradictions rather than replacing them with a narrative of inevitable theoretical progress Cumming & Parker (1997) Banks et al. (2012).
Claims of perfect Bayesian behavior require especially specific tests. The decisive evidence concerns predicted changes of weights and response variance across different stimulus reliabilities, estimated from independent measurements. A fitted probability model may achieve high descriptive accuracy without demonstrating the biological implementation of its inference algorithm. Also, one observer's performance may be near-optimal in a controlled slant discrimination while remaining biased in a natural distance judgment. For these reasons the present synthesis treats efficient integration as a supported conditional pattern and retains open questions about correlated cues, learned priors, scene segmentation, and generalization Hillis et al. (2004) Landy et al. (1995) Kersten et al. (2004).
Development of Spatial Perception and Cue Coordination
Developmental research asks when observers acquire usable depth information and when they learn to integrate independently available signals. A child may discriminate a visually specified drop, detect a moving surface, and employ stereo information in ways that follow distinct developmental trajectories. These abilities should be examined separately rather than assigned a single age at which depth perception is said to begin. The visual-cliff paradigm associated with Gibson and Walk made the relation between visual structure and exploratory behavior a central empirical question. Yet approach or avoidance also depends on locomotor experience, motivation, and the experimental setting, so behavior on the apparatus is not a pure measurement of stereoacuity Gibson & Walk (1960).
The development of cue integration provides a further distinction between sensitivity to individual signals and their joint use. In navigation tasks, Nardini and colleagues compared visual landmark information with self-motion information and studied how their combination changes with age. Although navigation is broader than visual stereopsis, the findings bear directly on the question of when multiple spatial estimates are integrated in an adult-like way. They support investigating experience-dependent calibration and task constraints rather than presuming that a child who detects each source also automatically combines them with adult reliability weighting Nardini et al. (2008).
Age-related conclusions require control of the opportunities observers have had to move, view surfaces, and calibrate eye movements. A developmental shift in measured depth bias may reflect improvement in cue interpretation, changes in motor response, familiarity with the task, or the precision of fixation. Longitudinal and adaptive experimental designs can help distinguish these possibilities, while large cross-sectional differences require careful comparison of stimulus visibility. Formal models make this question testable by estimating whether precision, inferred cue weights, or an assumed common cause changes with experience Landy et al. (1995) Kersten et al. (2004).
Individual Differences, Binocular Vision, and Generalization
Performance on binocular depth tasks varies with the observer, display, optical correction, fatigue, and properties of the visual input. The practical significance of such variation is that a weak score on one stereo test does not specify which part of the perceptual chain limits performance. It may involve image quality, interocular suppression, correspondence, fixation stability, response strategy, or the capacity to use monocular alternatives. Experiments should report recruitment criteria and binocular viewing status in relation to the target task, distinguishing a psychophysical sensitivity estimate from a broader clinical characterization. The reviews of stereoscopic mechanisms and display research explain why a universal threshold detached from testing conditions is scientifically inadequate Howard & Rogers (2012b) Banks et al. (2012).
Variability may also arise from cue-specific precision and from the weight an observer gives to a source. The combined slant experiments of Knill and Saunders were informative partly because they measured each participant's isolated stereo and texture thresholds. Predicting combined behavior from those individual estimates avoids assuming that a common group-level weight applies to observers with different cue reliability. A high average correspondence between model and judgment can otherwise hide diverse individual solutions. Cross-task validation should ask whether an observer's estimated stereo weight predicts judgments of surfaces differing in spatial frequency, distance, and shape Knill & Saunders (2003) Hillis et al. (2004).
The generality of three-dimensional perception must be tested across viewing conditions. A laboratory random-dot stereogram removes many natural signals by design and is excellent for identifying binocular correspondence, yet it has limited resemblance to scenes containing transparent surfaces, variable illumination, or moving objects. An observer who succeeds on it has demonstrated a particular computational ability, not necessarily exhaustive spatial competence. Conversely, natural scene tasks permit multiple strategies, making it harder to attribute success to a particular cue. A complete research program combines controlled psychophysics with calibrated naturalistic challenges, using matched outcome measures wherever feasible Julesz (1964) Howard (2012c).
Action, Locomotion, and Ecological Spatial Information
Three-dimensional perception supports action through estimates of reachable location, free space, object boundaries, and relations between the body and surrounding surfaces. An ecological approach examines information generated during movement as a structured feature of the observer–environment relation. Gibson's account placed emphasis on optic information available to an active observer, and later parallax experiments demonstrated that transforming retinal images can independently specify relative depth. The strength of this tradition is its attention to lawful transformations of an environment during behavior; its scientific assessment still requires controlled measurement of stimulus transformations and judgments Gibson (1979) Rogers & Graham (1979).
Acting in space creates feedback between motion and vision. A head translation can produce the parallax required to judge a previously ambiguous surface, whereas fixation changes can alter the disparity distribution that will be available next. Active sampling therefore changes the sensory evidence used for subsequent decisions. Theories of spatial perception must describe this loop if they aim to explain navigation or manipulation beyond a fixed seated viewing condition. Research designs can vary permitted head movement, track gaze, and compare estimation accuracy when the same optical information is presented passively or generated by the observer Nakayama & Shimojo (1992) Howard (2012c).
Action-oriented experiments also reveal that distinct judgments may use different information. An explicit verbal report of distance, a hand-placement task, and a locomotor avoidance response involve different transformations from sensory input to behavior. Divergence among them is theoretically informative when the tasks are comparably calibrated and response constraints are measured. At the same time, an apparent difference between a conscious estimate and a movement does not alone establish completely independent visual systems: changes in task demands and feedback may suffice. Research on spatial development and visual-cliff behavior illustrates how behavioral interpretation benefits from including motor and contextual variables Gibson & Walk (1960) Nardini et al. (2008).
Naturalistic Complexity and Multiscale Spatial Structure
Real environments contain nearby surfaces, distant layouts, specular objects, texture discontinuities, moving bodies, and partially transparent layers. Each operates at a different spatial scale, and the useful evidence for depth may change rapidly with fixation and movement. Laboratory studies often restrict the field to one slanted plane or a small disparity-defined target so that a proposed mechanism can be measured. This reduction is scientifically valuable, but integration across scales remains a separate empirical challenge. A theoretical model should explain how local features contribute to surface segmentation, how those surfaces are assigned relative position, and how a coherent navigable scene persists across viewpoint changes Kersten et al. (2004) Howard (2012a).
Occlusion and binocular correspondence expose a particular difficulty for such scaling. A system that matches only local image fragments may generate false correspondences around repetitive textures and disappearances at boundaries. Treating every unmatched feature as noise removes information that actually specifies which surface occludes another. The empirical work on da Vinci stereopsis and the theoretical account of stereopsis, fusion, and occlusion demonstrate why a coherent spatial representation must include both matched and unmatched evidence. Generalization from simple laboratory patterns to natural scenes must preserve these relations rather than merely increase image resolution Nakayama & Shimojo (1990) Anderson & Nakayama (1994).
Quantitative comparison across natural scenes also requires distinguishing reliability from richness. A natural image provides more signals than a random-dot display, yet those signals are often correlated and subject to uncontrolled changes in lighting and perspective. Apparent robustness may come from redundancy, learned expectations, or the selective use of one dominant cue. To separate these accounts, experiments can perturb an individual signal while retaining scene realism, estimate observer uncertainty, and compare competing models out of sample. The modified weak-fusion and Bayesian traditions offer useful prediction frameworks, provided the actual covariance of cues is measured Landy et al. (1995) Kersten et al. (2004).
Artificial Vision and the Artificial Era
Artificial perception systems offer a comparative research setting because their components can be documented, altered, and subjected to controlled failures. Two-camera stereo algorithms confront correspondence, occlusion, and disparity-to-distance geometry, while active robotic sensors use motion to reduce ambiguity. Those functional parallels do not establish identity between engineered representations and human perceptual experience. Scientific comparison should specify shared inputs, target variables, uncertainty estimates, and output behavior and then determine which differences arise from sensor geometry, training, or action constraints. The visual problems described by binocular matching and naturalistic surface inference provide a well-defined basis for comparative experiments Read & Cumming (2007) Kersten et al. (2004).
In the Artificial Era, three-dimensional perception becomes a research bridge between human psychophysics and the organization of nonbiological spatial reasoning. A camera system that produces a depth map, a human who perceives a surface in relief, and a robot that avoids an obstacle may be assessed using common external measures such as depth error, occlusion handling, and robustness to viewpoint. Their internal processing should be characterized through the specific architecture and available evidence. A calibrated joint benchmark could vary disparity, texture, and parallax independently, test cue conflict, and report uncertainty without assuming that the same behavioral score implies the same internal computational mechanism Landy et al. (1995) Banks et al. (2012).
Artificial systems also make explicit the possibility of allocating perception across sensors, memory, and action. Active stereo, monocular structure-from-motion, and learned image priors can operate as components of a distributed spatial estimation process. The corresponding psychological questions concern how such components are coordinated, whether they correct each other's biases, and how calibration persists when the environment changes. These questions belong to a testable comparative program and retain their proper scientific status as hypotheses until evaluated. The historical transition From Homo to Artificial supplies the philosophical horizon of comparison, while psychophysical evidence supplies the standards for adjudicating claims about functional performance Howard (2012a) Kersten et al. (2004).
Unresolved Problems and Research Priorities
A central unresolved question concerns the stability of cue integration across context. Experiments on slant have shown reliability-dependent weighting, but natural scenes combine correlated signals and often include multiple surfaces whose causal relations are uncertain. The ability to predict cue weights under one noise manipulation does not establish their invariance under movement, lighting changes, unfamiliar materials, or changed task demands. Research can advance through preregistered model comparisons in which cue-specific uncertainty is estimated separately and then used to predict independent combined-cue conditions. Success should be assessed through predictive performance and calibrated uncertainty rather than through the flexibility of a fitted model alone Knill & Saunders (2003) Hillis et al. (2004).
Another unresolved problem concerns the transformation from neural disparity responses to perceived and behaviorally useful depth. Early cortical neurons can respond to signal configurations that do not yield the corresponding stereoscopic percept, implying that match selection, surface organization, and readout must be studied jointly. Progress will require perturbations across neural processing stages, behavioral discrimination tasks, and stimuli that separate true correspondence from locally attractive false matches. Competing hypotheses must specify what would count as evidence for rejection, rather than simply labeling a tuned neuron a depth detector Cumming & Parker (1997) Read & Cumming (2007).
Finally, spatial perception research needs stronger generalization across populations and technologies. Developmental work invites tests of how cue weights and calibrations are learned; display engineering raises questions about comfort and metric accuracy when optical relationships are altered; human–artificial comparison provides systems whose spatial representations can be tested under matched conditions. The scientific program benefits from archived stimulus geometry, eye-tracking metadata where appropriate, access to trial-level responses, and explicit reporting of calibration and exclusion criteria. These practices preserve the interpretability of both positive findings and failures to replicate a model in new environments Nardini et al. (2008) Banks et al. (2012).
Stereo-Anomaly and the Difference Between Detection and Depth Direction
Modern stereopsis research demonstrates why apparently contradictory estimates of the frequency of deficient stereo vision may reflect different measurement constructs. Baldwin and colleagues tested 228 observers using tasks that separated detecting binocular disparity from judging its direction in depth. In the authors' cohort, 98 percent consistently succeeded at detection, whereas 69 percent of those performing detection consistently succeeded in the direction-discrimination task. The corresponding reported 31 percent difficulty with direction discrimination is a task-specific finding, not a population prevalence estimate transferable without regard to sampling or measurement conditions. The methodological implication is that an experimental label such as stereo-anomaly must specify the required computation Baldwin et al. (2024).
The study's separation of detection and direction makes a conceptual distinction that classic threshold measures sometimes conceal. A disparity-sensitive system can signal that two corresponding images differ while failing to recover a stable ordering of the represented surfaces. This possibility resonates with physiological work showing disparity modulation in primary visual cortex under configurations that lack the predicted depth percept. The two bodies of evidence measure different levels of processing, and their connection is an explanatory hypothesis: the psychophysical dissociation alone does not localize the cause, while early neural dissociation alone does not determine how an individual observer performs at sign judgment Baldwin et al. (2024) Cumming & Parker (1997).
Future assessment should record whether trials require detecting a disparity-defined target, judging crossed versus uncrossed disparity, discriminating two relative depths, or matching a metric distance. Each places distinct demands on the observer and may respond differently to contrast, duration, binocular fusion, and feedback. A model explaining individual differences must predict the joint profile of these tasks and specify how a failure of one response can coexist with success on another. The 2024 study demonstrates that distinctions in task definition can change a scientific conclusion even when the underlying physical disparity cue is nominally the same Baldwin et al. (2024) Howard & Rogers (2012b).
When Shading Reshapes a Reliable Binocular Depth Signal
Aubuchon and colleagues reported three within-subject experiments in which shading patterns modified the perceived shape of surfaces carrying disparity information. In the tested conditions, shading could overcome a disparity-defined shape signal whose conventional reliability ranking would have suggested stronger binocular influence. The authors reported that these effects were not adequately predicted by the tested accounts of reliability weighting, cue vetoing, or cue promotion. Their results challenge the sufficient generality of a simple independent-measurement averaging rule without negating the earlier evidence that such a rule describes other slant tasks Aubuchon et al. (2024) Knill & Saunders (2003).
The important theoretical question is how one source changes the representation supplied by another. A shading pattern may affect segmentation, surface interpretation, or the assignment of local disparity to a coherent shape, meaning that the cues interact before a final estimate is formed. This differs from a model in which separate unbiased shape measurements are computed independently and only then averaged. The experiment therefore motivates testing interactions within cue processing and scene interpretation rather than merely assigning a lower numerical variance to one independently computed cue. Its authors presented the effect as previously unaccounted-for cue cooperation Aubuchon et al. (2024).
Generalization requires discriminating among at least three explanations: a task-specific redefinition of shape, an interaction that transforms the disparity field, and an observer's learned expectation about how shading and disparity ought to co-occur. These hypotheses should be compared with stimuli that independently vary lighting pattern, disparity amplitude, shape family, and exposure history while preserving measured performance. The contrast between optimal weighting in calibrated planar slant judgments and shading-induced warping in complex surface judgments underscores that model adequacy depends on what is represented and how that representation is experimentally read out Aubuchon et al. (2024) Hillis et al. (2004).
Natural Scene Variability and the Limits of Stereo Performance
White and Burge studied stereoscopic discrimination using images sampled from a database of natural scenes with pixel-level physical distance information. Three human observers completed two related double-pass psychophysical experiments involving repeated judgments on ten thousand distinct trials per observer. The researchers separated the contributions of variation in luminance-contrast patterns and variation in local depth structure and examined the influence of internal noise. They found that both stimulus-driven sources contributed substantially to performance differences and produced remarkably similar patterns across the three participants. The experimental depth task therefore revealed structured difficulty tied to properties of real-world images White & Burge (2025).
The double-pass design repeats the same stimulus to distinguish stimulus-specific predictability from variability in an observer's response. A stimulus that is difficult for many observers is not thereby evidence of random response behavior: particular image features may systematically challenge the estimation process. The authors reported that natural variation sometimes increased the predictability of stimulus-by-stimulus responses across observers, a result that cautions against reducing all errors to one undifferentiated noise parameter. Nevertheless, the sample of three permits detailed within-observer analysis and replication across these observers, while broader population generalization remains a separate empirical task White & Burge (2025).
This work advances the interface between classical psychophysics and naturalistic vision because it retains a controlled depth-discrimination task while introducing more realistic image statistics. It calls for image-computable models that predict which particular scene patches yield accurate or biased depth estimates and specify why. Future models should separate error caused by surface structure, texture and contrast, sensory noise, and decision policy, then evaluate predictions on independently sampled environments. A high performance score on simplified random-dot stimuli is insufficient as evidence that the system generalizes to the covariation of texture, occlusion, and local depth in natural scenes White & Burge (2025) Julesz (1964).
Context-Dependent Cortical Processing of Disparity
Duan and colleagues investigated the interaction of monocular scene content with disparity processing in human visual cortex using natural stereo-pair libraries and synthesized stimuli in which monocular appearance could be manipulated separately from depth structure. The methodological premise is that ordinary random-dot stereograms deliberately break correlations between monocular image statistics and depth that occur in the natural environment. This gives them excellent experimental control but potentially excludes contextual information that changes the neural response to a given disparity signal. The 2021 study therefore addressed disparity within realistic image content rather than treating it as an isolated numerical offset Duan et al. (2021).
The experimental distinction matters for interpreting visual-cortex mechanisms. A local response to correlated binocular dots establishes sensitivity to a controlled pattern, but a depth representation in a scene must also tolerate changes in material, texture and illumination. If cortical responses depend on those contextual patterns, a single-parameter disparity tuning model could omit essential transformations. The study's use of image content manipulations is relevant to determining which natural statistics contribute to perceived depth and which affect neural signal strength without changing the observer's eventual response Duan et al. (2021) Cumming & Parker (1997).
Theoretical synthesis must distinguish measured neural sensitivity from a demonstrated algorithm of scene interpretation. Contextual modulation can support multiple accounts, including recurrent surface grouping, nonlinear binocular filtering, and learned probabilistic expectations. Discriminating among them requires experiments that jointly record the sensory geometry, neural measures, and observer reports and evaluate cross-stimulus predictions. Natural-image stereograms make that agenda feasible while preserving the ability to perturb disparity separately from appearance. They also provide a bridge to artificial stereo systems whose correspondence algorithms encounter the same contextual variation Duan et al. (2021) Kersten et al. (2004).
Motion Parallax, Stereopsis, and the Geometry of Perceptual Space
Wang and Troje examined how binocular disparity and motion parallax transform judged spatial geometry in virtual-reality environments. Their 2024 research developed a geometric model of relief scaling and pictorial distortions and compared it with responses in two experiments where participants rotated a figure to point toward targets. The virtual frame permitted selective manipulation of disparity and motion-parallax availability. The reported model comparisons supported depth expansion and direction-related transformations, and the fitted parameters suggested distinctive roles for the two signals in maintaining spatial invariants. The results extend cue combination beyond judging the slant of one laboratory plane Wang & Troje (2024).
The study is theoretically important because the dependent variable was pointing direction under transformations of perceived space. A cue may improve one spatial invariant while introducing distortions in another; binocular disparity and parallax therefore need not be ranked by a single overall accuracy score. Its conclusions depend on the geometry of the experimental virtual scene, the response task, and the model family compared. They should not be directly converted into a claim that a signal always supplies one kind of physical truth while another supplies a different kind. A precise account asks which geometrical quantities remain stable under the tested transformations Wang & Troje (2024) Rogers & Graham (1979).
Such work also suggests a productive design for research on head-mounted displays and robotics. Rather than measuring only subjective apparent depth, an investigator can ask whether estimated direction, relative distance, apparent object shape, and action planning remain consistent as cues are independently perturbed. The effects of screen geometry, eye position, and virtual camera baseline remain critical; measurement without them can confound altered perceptual space with incorrectly rendered stimulation. Geometry-based models offer explicit competing predictions that can be compared on tasks involving rotation, navigation, and depth matching Wang & Troje (2024) Banks et al. (2012).
Three-Dimensional Organization Beyond Immediate Sensory Judgments
Three-dimensional organization can affect tasks extending beyond immediate estimates of surface depth. Liu, Laeng, and Czajkowski reported experiments with stereoscopic and monoscopic virtual displays that tested how depth separation alters visual short-term memory for object features. Their first experiment used color, whereas later experiments required memory for spatial orientation and direction of rotation. The published results indicated a stereoscopic multi-plane advantage for the spatial features while offering little comparable evidence of benefit for the color task. This contrast highlights that a depth manipulation can influence memory through the relation between spatial representation and task-relevant content Liu et al. (2025).
The result should be interpreted as a conditional effect of spatial organization on a memory task, not proof of an invariant increase in the number of remembered objects whenever depth is visible. Separating items across depth planes may change grouping, interference, attentional allocation, or the retrieval cues available at test. To identify the responsible mechanism, subsequent work could equate object visibility and grouping while manipulating whether the remembered feature is spatial or nonspatial. The dissociation between color and orientation conditions already indicates that the role of stereopsis depends on what the observer is asked to retain Liu et al. (2025).
This cognitive extension clarifies the scope of the present dossier. The object of depth perception is the organization of spatial structure from sensory evidence, while memory and attention studies ask what downstream processes use that structure. A useful interdisciplinary theory specifies the information passed from depth estimation to attention, object selection, and memory without conflating the constructs. Future research should test whether calibrated individual differences in stereopsis predict the size of memory benefits under matched displays, and whether naturally occurring depth distributions produce comparable effects in unrestricted scenes Liu et al. (2025) Baldwin et al. (2024).
Conclusion
Depth and three-dimensional perception is a coordinated psychological achievement that reconstructs spatial relationships from multiple, individually limited sources of information. Binocular disparity, occlusion, perspective, shading, texture, parallax, and eye-position cues constrain overlapping but distinct properties of the world. Classical stereoscopy established the power of binocular differences, random-dot displays isolated correspondence, motion-parallax studies demonstrated independent dynamic depth information, and subsequent cue-combination research quantified conditions under which estimates follow reliability-sensitive models. The most defensible synthesis is therefore a structured account of multiple mechanisms with task-specific evidential support Wheatstone (1838) Julesz (1964) Rogers & Graham (1979) Landy et al. (1995).
The central theoretical achievement of contemporary research is the capacity to compare mechanistic explanations against calibrated measurements. Probabilistic models explain why ambiguous images invite expectations and weighted evidence, neural experiments show that early disparity tuning and perceived depth can diverge, and developmental and applied studies reveal the limits of isolated laboratory results. No single experiment determines the entire architecture of spatial perception. A scientifically mature account specifies which latent geometric variable is inferred, which cues are available, which measurement procedure is used, and how alternative models perform under controlled conflict and independent testing Kersten et al. (2004) Cumming & Parker (1997) Knill & Saunders (2003).
The research frontier lies in joining precision and ecological range: understanding how local binocular matches become coherent surfaces, how cues are calibrated through active viewing, how uncertainty is handled in complex scenes, and how spatial performance may be compared across biological and artificial systems. The resulting monograph situates depth perception within Perception and Psychophysics, Cognitive Psychology, and Human Factors and Engineering Psychology while preserving its independent research object. Its lasting contribution is a traceable synthesis of demonstrated phenomena, models, methodological constraints, and open questions that can support more specialized publications and comparative research Howard (2012c) Banks et al. (2012) Nardini et al. (2008).
References
- Anderson, B. L., & Nakayama, K. (1994). Toward a general theory of stereopsis: Binocular matching, occluding contours, and fusion. Psychological Review, 101(3), 414–445. https://doi.org/10.1037/0033-295X.101.3.414
- Aubuchon, C., Kemp, J., Vishwanath, D., & Domini, F. (2024). Weak overcomes strong in sensory integration: Shading warps the disparity field. Proceedings of the Royal Society B: Biological Sciences, 291(2033), Article 20241758. https://doi.org/10.1098/rspb.2024.1758
- Baldwin, A. S., Min, S. H., Alarcon Carrillo, S., Wang, Z., Cheng, Z., Zhou, J., & Hess, R. F. (2024). Stereo-anomaly is found more frequently in tasks that require discrimination between depths. iScience, 27(6), Article 109879. https://doi.org/10.1016/j.isci.2024.109879
- Banks, M. S., Read, J. C. A., Allison, R. S., & Watt, S. J. (2012). Stereoscopy and the human visual system. SMPTE Motion Imaging Journal, 121(4), 24–43. https://doi.org/10.5594/j18173
- Cumming, B. G., & Parker, A. J. (1997). Responses of primary visual cortical neurons to binocular disparity without depth perception. Nature, 389, 280–283. https://doi.org/10.1038/38487
- Ding, J., & Sperling, G. (2006). A gain-control theory of binocular combination. Proceedings of the National Academy of Sciences, 103(4), 1141–1146. https://doi.org/10.1073/pnas.0509629103
- Duan, Y., Thatte, J., Yaklovleva, A., & Norcia, A. M. (2021). Disparity in context: Understanding how monocular image content interacts with disparity processing in human visual cortex. NeuroImage, 237, Article 118139. https://doi.org/10.1016/j.neuroimage.2021.118139
- Ernst, M. O., & Banks, M. S. (2002). Humans integrate visual and haptic information in a statistically optimal fashion. Nature, 415(6870), 429–433. https://doi.org/10.1038/415429a
- Gibson, E. J., & Walk, R. D. (1960). The visual cliff. Scientific American, 202(4), 64–71. https://doi.org/10.1038/scientificamerican0460-64
- Gibson, J. J. (1979). The ecological approach to visual perception. Houghton Mifflin. https://search.worldcat.org/title/4929611
- Hillis, J. M., Watt, S. J., Landy, M. S., & Banks, M. S. (2004). Slant from texture and disparity cues: Optimal cue combination. Journal of Vision, 4(12), 967–992. https://doi.org/10.1167/4.12.1
- Howard, I. P. (2012). Perceiving in depth: Volume 1, Basic mechanisms. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199764143.001.0001
- Howard, I. P. (2012). Perceiving in depth: Volume 3, Other mechanisms of depth perception. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199764167.001.0001
- Howard, I. P., & Rogers, B. J. (2012). Perceiving in depth: Volume 2, Stereoscopic vision. Oxford University Press. https://doi.org/10.1093/acprof:oso/9780199764150.001.0001
- Julesz, B. (1964). Binocular depth perception without familiarity cues. Science, 145(3630), 356–362. https://doi.org/10.1126/science.145.3630.356
- Julesz, B., & Spivack, G. J. (1967). Stereopsis based on vernier acuity cues alone. Science, 157(3788), 563–565. https://doi.org/10.1126/science.157.3788.563
- Kersten, D., Mamassian, P., & Yuille, A. (2004). Object perception as Bayesian inference. Annual Review of Psychology, 55, 271–304. https://doi.org/10.1146/annurev.psych.55.090902.142005
- Knill, D. C., & Saunders, J. A. (2003). Do humans optimally integrate stereo and texture information for judgments of surface slant? Vision Research, 43(24), 2539–2558. https://doi.org/10.1016/S0042-6989(03)00458-9
- Landy, M. S., Maloney, L. T., Johnston, E. B., & Young, M. (1995). Measurement and modeling of depth cue combination: In defense of weak fusion. Vision Research, 35(3), 389–412. https://doi.org/10.1016/0042-6989(94)00176-M
- Liu, H., Laeng, B., & Czajkowski, N. O. (2025). Visual short-term memory is modulated by 3D depth in stereopsis. Attention, Perception, & Psychophysics, 87, 2265–2274. https://doi.org/10.3758/s13414-025-03052-3
- Nakayama, K., Shimojo, S., & Silverman, G. H. (1989). Stereoscopic depth: Its relation to image segmentation, grouping, and the recognition of occluded objects. Perception, 18(1), 55–68. https://doi.org/10.1068/p180055
- Nakayama, K., & Shimojo, S. (1990). Da Vinci stereopsis: Depth and subjective occluding contours from unpaired image points. Vision Research, 30(11), 1811–1825. https://doi.org/10.1016/0042-6989(90)90161-D
- Nakayama, K., & Shimojo, S. (1992). Experiencing and perceiving visual surfaces. Science, 257(5075), 1357–1363. https://doi.org/10.1126/science.1529336
- Nardini, M., Jones, P., Bedford, R., & Braddick, O. (2008). Development of cue integration in human navigation. Current Biology, 18(9), 689–693. https://doi.org/10.1016/j.cub.2008.04.021
- Poggio, G. F., & Poggio, T. (1984). The analysis of stereopsis. Annual Review of Neuroscience, 7, 379–412. https://doi.org/10.1146/annurev.ne.07.030184.002115
- Read, J. C. A., & Cumming, B. G. (2007). Sensors for impossible stimuli may solve the stereo correspondence problem. Nature Neuroscience, 10, 1322–1328. https://doi.org/10.1038/nn1951
- Rogers, B., & Graham, M. (1979). Motion parallax as an independent cue for depth perception. Perception, 8(2), 125–134. https://doi.org/10.1068/p080125
- Wang, X. M., & Troje, N. F. (2024). Relating visual and pictorial space: Integration of binocular disparity and motion parallax. Journal of Vision, 24(13), Article 7. https://doi.org/10.1167/jov.24.13.7
- Watt, S. J., Akeley, K., Ernst, M. O., & Banks, M. S. (2005). Focus cues affect perceived depth. Journal of Vision, 5(10), 834–862. https://doi.org/10.1167/5.10.7
- Wheatstone, C. (1838). Contributions to the physiology of vision. Part the first: On some remarkable, and hitherto unobserved, phenomena of binocular vision. Philosophical Transactions of the Royal Society of London, 128, 371–394. https://doi.org/10.1098/rstl.1838.0019
- White, D. N., & Burge, J. (2025). How distinct sources of nuisance variability in natural images and scenes limit human stereopsis. PLOS Computational Biology, 21(4), Article e1012945. https://doi.org/10.1371/journal.pcbi.1012945