RESEARCH DOSSIER
Perceptual Constancy and Ambiguity
Abstract. Perceptual constancy and ambiguity concern complementary problems in visual inference: the preservation of comparatively stable object and surface properties despite changing sensory conditions, and the coexistence or alternation of perceptual interpretations under insufficient or unchanging evidence. This monograph reconstructs the experimental history of size, shape, lightness and color constancy; distinguishes appearance matching, surface identification, and discrimination across illuminants; reviews reversible figures and binocular rivalry; and evaluates cue integration, adaptation, competition, multistability, cortical dynamics and probabilistic explanations. The synthesis draws on original psychophysical and neuroscience experiments together with critical reviews and recent naturalistic investigations. The evidence supports property-specific and task-dependent stabilization, as well as multiple interacting mechanisms behind perceptual transitions. Important unresolved questions concern generalization across displays and environments, the causal location of perceptual switches, quantitative cross-task predictions, and comparisons with artificial vision.
Keywords: perceptual constancy; perceptual ambiguity; color constancy; lightness constancy; size constancy; shape constancy; Necker cube; reversible figures; binocular rivalry; multistable perception; perceptual inference; visual illusions; psychophysics; Bayesian perception; visual awareness
The Scientific Problem of Stable Properties and Ambiguous Evidence
Perceptual constancy is the empirically measurable persistence of an estimated object or surface property when viewing conditions alter the signal available to the observer. An object's physical width does not change as the observer moves, but the retinal angle it subtends does. The reflectance of paint may remain constant while illumination changes its cone excitations. The projected contour of a tilted rectangle changes although the rectangle's geometry remains fixed. Constancy is therefore always relative to a specified property, transformation, task, and dependent measure. A classification response about material identity, an asymmetric appearance match, and an action guided by apparent size each answer a distinct question. The distinction is central to the experimental and theoretical analysis of color constancy Foster (2011) and to the older quantitative treatment of apparent size Holway & Boring (1941).
Perceptual ambiguity describes a different relation between sensory evidence and its possible causes: the same or insufficiently informative input supports more than one plausible interpretation. A Necker cube supports alternative depth orders; incompatible images shown separately to the eyes can alternate in dominance; an illumination change can resemble a surface reflectance change. These cases share underdetermination, yet the resulting behavior varies. Some ambiguous stimuli yield a stable choice, others switch, and others produce mixed or uncertain reports. The experimental distinction between ambiguity as availability of interpretations and reversibility as experienced alternation is established in the synthesis of Long & Toppino (2004).
The central question is how perception maintains useful stability when irrelevant conditions vary while remaining capable of revising an interpretation when evidence or internal state warrants change. This question cuts across psychophysical measurement, scene organization, adaptation, attention, probabilistic inference, and neuroscience. The present dossier reviews original experiments and evidence syntheses, comparing claims at the level of their operational tests. It concentrates on visual property constancy and multistability; recognition, sensory thresholds, and broad theories of perception are considered where they explain these phenomena. A scientifically complete account must connect physical transformations to observed reports, model predictions, and the conditions in which stability breaks down Kersten et al. (2004).
Distal Variables, Proximal Images, and Inverse Problems
Visual stimulation is jointly determined by the world and the viewing arrangement. For color, the spectral signal reaching the eye depends on the surface reflectance function, the spectral distribution of illumination, optical transmission, and sensor sensitivity. For apparent size, retinal angle varies with physical extent and distance. For projected shape, surface orientation determines foreshortening. These are inverse problems because a single recorded image may be consistent with multiple physically different scenes. An observer must exploit spatial relationships, temporal continuity, other depth or illumination cues, and constraints on possible environments to estimate a useful distal property Kersten et al. (2004).
Defining an inverse problem does not yet identify the algorithm used by biological observers. A probabilistic model may compute a posterior distribution over possible scene causes, whereas a relational model may preserve diagnostic ratios across surfaces, and an ecological account may rely on information available during movement. These models sometimes predict similar behavior in familiar conditions. Their scientific separation requires experiments that vary one cue independently of another or construct circumstances where their quantitative predictions diverge. Depth-cue experiments demonstrate this principle through controlled manipulations of disparity, edges, and shading Bülthoff & Mallot (1988).
The observation-level distinction is equally important. A physical specification such as photometric luminance, measured reflectance, viewing distance, or object geometry defines a reference variable. A reported perceptual match identifies how observers experience or select among candidate targets. A response can show invariance because participants recognize a familiar object, remember its typical appearance, or adopt a verbal rule. Conversely, observers may identify a surface correctly while experiencing a visible appearance change. Convergent evidence comes from varying instructions, reference objects, response format, stimulus duration, and the information available to the observer. These distinctions underlie the operational account advanced by Craven & Foster (1992).
Origins of Experimental Constancy Research
Historical research on constancy emerged from attempts to explain the relation between stimulus geometry and perceived properties. The classical size problem was given a rigorous experimental form by Holway & Boring (1941), who varied distance and visual conditions to examine how apparent visual size depends on the availability of distance information. Such work forced a distinction between visual angle and perceived size and made cue availability an experimentally controllable factor. It also revealed a recurring difficulty: reducing cues to isolate one mechanism may simultaneously change how the observer understands the task.
Lightness research addressed the stability of surface appearance across changes in illumination. Wallach (1948) provided a historically significant experimental treatment of brightness constancy and achromatic colors. Later accounts examined how local luminance relationships and larger-scale scene frameworks determine perceived lightness. Gilchrist et al. (1999) integrated a large body of experiments in an anchoring theory, arguing that lightness depends on how surfaces are organized within perceptual frameworks. This tradition moved beyond the assumption that the luminance of a patch alone determines its perceived shade.
Contemporary color studies introduced calibrated illumination and more naturalistic spatial arrangements. Brainard et al. (1997) used controlled illumination and asymmetric color matches in a room; Brainard (1998) investigated achromatic settings under different illuminants and backgrounds. These programs sharpened the question of how adaptation, illuminant estimation, and spatial context contribute to stable surface-color judgments. The historical trajectory thus moved from demonstrating constancy toward explaining its mechanisms and measuring its limits. The same progression is visible in multistable perception, where the fact of perceptual switching became an experimentally tractable neural and computational problem Leopold & Logothetis (1999).
Definitions, Measures, and Competing Research Questions
Constancy can be quantified by estimating how much a perceptual judgment changes relative to a defined change in sensory stimulation, but there is no task-independent universal constancy coefficient. Color research uses asymmetric matching, achromatic settings, identification of surface colors, and discrimination between illumination and reflectance changes. Size research employs matching and adjustment across viewing distances. Shape research uses slant, surface-depth, and object-identity judgments. Every estimate depends on a baseline, a comparison scale, adaptation, and observer instructions. The measurement of color constancy is especially sensitive to whether researchers assess stable appearance or correct attribution of a change Craven & Foster (1992) Foster (2011).
Ambiguity is measured using a different family of observables: proportion of time an interpretation dominates, distribution of dominance durations, transition frequency, mixed percepts, initial choice, adaptation effects, and report confidence. In binocular rivalry, a key methodological distinction separates dominance periods from transitions and occasional return transitions. Brascamp et al. (2006) showed that these aspects can have different contrast dependencies, constraining models that explain all switching with a single process. Measuring only the number of button presses loses information about how the state evolves.
Experimental questions must therefore be specified independently of familiar terminology. Does a change in illumination alter perceived surface identity or merely its immediate hue? Does depth-cue manipulation change geometric interpretation, response criterion, or certainty? Does attention alter the probability of one percept, the duration of a dominance interval, or the time required to report a change? Does a neural signal precede a perceptual transition or reflect its subsequent detection? Once these questions are separated, apparent disagreements among studies can become differences in operational target rather than mutually exclusive claims Brascamp et al. (2018).
Evidence Strategy and Scope of the Synthesis
This monograph is a critical narrative synthesis of documented experimental, psychophysical, computational, and neuroscientific research. The evidence corpus includes landmark measurements of size and lightness, controlled studies of color and depth, investigations of ambiguous figures and binocular rivalry, neuroscientific comparisons, and reviews that situate individual findings within competing research traditions. Sources are attributed at the claims they support; reviewed papers are distinguished from experiments that produced original data. The dossier does not represent a preregistered systematic review, a new meta-analysis, or an original participant experiment.
Evidence quality is evaluated according to stimulus control, transparency of the response measure, independence of tested cues, comparison conditions, explanatory specificity, and replicability across tasks or laboratories. A measured perceptual switch is an observation; a putative neural switching mechanism is an explanatory proposal that requires a stronger causal design. An association between background chromaticity and an achromatic match does not by itself establish conscious illuminant inference. Similarly, a mathematical fit to dominance durations provides a constraint on competing models but is not evidence that an inferred internal variable is implemented by a single cortical area Brainard (1998) Brascamp et al. (2006).
Constancy and ambiguity are integrated through a shared scientific problem: mapping incomplete retinal evidence to useful descriptions of environmental properties. Their specialized experimental programs remain distinct because constancy tests invariance across altered inputs, whereas multistability tests changes in interpretation with relatively stable or competing inputs. The synthesis examines both stable identification and switching, assesses where probabilistic and competitive mechanisms can be compared directly, and closes with methodological priorities for naturalistic and artificial perceptual systems. Research belonging principally to language ambiguity, general perceptual learning, and clinical visual disorders lies outside the core boundaries of this dossier.
Size Constancy and the Recovery of Spatial Scale
Size constancy refers to the relative stability of perceived physical extent despite changes in distance that alter the retinal angular extent of an object. A simple geometric approximation relates visual angle to size divided by distance for objects subtending small angles; it identifies the variables that the visual system must jointly constrain. An object that occupies a smaller retinal image at greater distance need not be judged physically smaller. Historical experiments systematically manipulated the distance information accompanying a test object and showed that perceived size depends on the available cues Holway & Boring (1941).
Distance information can derive from binocular disparity, perspective, texture gradients, accommodation, motion, and contextual relationships. No single cue remains equally informative at all distances, so a claim about size constancy must report the viewing range, illumination, monocular or binocular conditions, and availability of surrounding structure. A constrained laboratory corridor and an ordinary environment offer different opportunities for estimating distance. Performance under conditions designed to remove depth cues should be interpreted as evidence about that manipulated stimulus arrangement, not as a universal deficit of everyday visual constancy.
Size estimates also depend on the question asked. Matching apparent angular size is different from matching a remembered physical object, and verbal reports of an object's familiar size may recruit semantic knowledge. The successful identification of a distant person does not establish perfect size appearance invariance. A stronger design compares instructed apparent-size and objective-size judgments under identical visual conditions, includes unfamiliar objects, and estimates response variability. This distinction connects classical size psychophysics with general principles of scene inference Kersten et al. (2004).
Shape Constancy, Slant, and the Geometry of Projection
Shape constancy concerns the retention of a useful representation of object geometry when surface orientation and perspective change its image. A door remains a rectangle while its retinal projection becomes progressively narrower as it opens; a tilted disk produces an elliptical outline. Inferring the shape of the surface therefore requires information about slant, depth, contour, and often contextual geometry. Whether an observer sees a stable shape depends on the cues specifying the orientation of the object and the reference frame used for judgment. Shape reports should be distinguished from the recognition of an object's category.
Experimental manipulation of depth information demonstrates why the problem is richer than reversing a known projection rule. Bülthoff & Mallot (1988) used controlled computer-generated ellipsoids to vary binocular disparity, shading and edge information. Their results show that different cues can contribute unequally to reconstructed depth and that introducing edges can override information supplied by other cues. Such evidence resists accounts in which every depth cue is assigned a permanent, context-independent weight.
The weak-fusion modeling program synthesized by Landy et al. (1995) formalized cue promotion, dynamic weighting, and robustness as problems for depth and shape estimation. This approach makes experimental predictions about how a depth judgment should shift when one cue is perturbed while the others remain fixed. The broader implication is that shape constancy is an achievement of structured multisource inference: the retinal image supplies constraints, and the system estimates the spatial arrangement that best explains those constraints under particular conditions.
Color Constancy as Illuminant–Reflectance Separation
Surface color cannot be read directly from the spectrum reaching the retina. Light reflected toward an observer depends jointly on illumination and reflectance; a change in the spectral distribution of the illuminant can alter cone excitations without changing the surface material. Color constancy denotes stability across this transformation, but experimenters distinguish matching the surface's perceived color from identifying whether a scene change was caused by illumination or reflectance. Craven & Foster (1992) demonstrated that observers can reliably attribute some sufficiently large changes to the appropriate cause even under conditions where strict appearance invariance is imperfect.
The nearly natural image studies introduced experimentally controlled illumination in settings more complex than isolated uniform patches. Brainard et al. (1997) measured asymmetric color matches across an illumination gradient and compared descriptions based on receptor-channel gain changes with accounts that estimate an illuminant. Brainard (1998) used achromatic settings with varied illuminants and spatial backgrounds and found constancy that simple local-surround comparisons could not fully explain. These results constrain mechanistic accounts without settling which calculations each neural population performs.
Adaptation, relational spectral information, spatial context, and assumptions about illuminants may support similar performance. The comparative review by Foster (2011) shows why the degree of reported constancy varies with both task and environment. Researchers should therefore describe the stimulus spectra, display calibration, surface set, adaptation history, and whether observers judged appearance, identity or causal change. The general principle is to separate the physical problem of illuminant–reflectance ambiguity from the operational problem of assessing how observers resolve it.
Lightness and Brightness: Context as an Explanatory Variable
Lightness normally concerns perceived surface reflectance on an achromatic scale, while brightness concerns apparent intensity or luminance-related experience; terminology and instructions differ across experimental traditions. The luminance received from a region is affected by illumination, material and surrounding context, so a small retinal patch provides incomplete information about surface shade. Wallach (1948) provided a landmark investigation of brightness constancy. The later anchoring theory Gilchrist et al. (1999) proposed that local and global perceptual frameworks jointly determine lightness assignments, accounting for both achievements and systematic failures of constancy.
Contextual organization can alter appearance even when the local target luminance is fixed. Adelson (1993) showed that perceptual grouping and scene interpretation strongly modulate brightness judgments in displays whose local image changes alone cannot explain the perceived effect. These experiments are particularly informative because they turn organization into an experimental variable: change the relationships that make patches appear to belong to a surface or lighting framework, and the judged shade changes even when the relevant photometric quantity is held constant.
Chromatic context can influence judgments about achromatic targets as well. Lotto & Purves (1999) documented effects of equi-luminant colored surrounds on apparent brightness, supporting the view that the system exploits relationships extending beyond local luminance contrast. These results do not imply one universally accepted theory of lightness. They identify causal contributions of context and organization whose quantitative weighting and generalization remain matters of model comparison. A useful explanation must predict the direction and size of each effect across novel surfaces and lighting arrangements.
Illusions as Tests of Constancy and Scene Organization
An illusion is scientifically informative when the experimenter specifies both the controlled stimulus and the physical property against which a response is compared. Size illusions can involve inferred depth, perspective, comparison context and response conventions; lightness illusions may expose how surfaces are grouped into different perceived illuminations. The same display can yield different judgments depending on whether the participant is asked to match apparent brightness, material shade or physical luminance. Such tasks probe the operations by which a visual system ordinarily reduces environmental ambiguity Adelson (1993) Gilchrist et al. (1999).
Calling an illusion a failure does not explain its production. A systematic difference between perceptual and physical measurements can be predicted by a processing rule well suited to some ordinary environments. At the same time, adaptive plausibility is a hypothesis requiring separate ecological evidence. Researchers must avoid inferring evolutionary origin from a compelling demonstration or assuming that observer expectations explain an effect without an experimental manipulation of those expectations. Quantitative tests can distinguish low-level adaptation, contextual grouping, and scene-model predictions through counterbalanced stimulus arrangements.
Illusions also show why a stable experience can be incorrect according to a chosen external measure. Some displays create remarkably durable judgments despite explicit knowledge of the true geometry; others yield rapid perceptual reversals. Their mechanisms must be investigated separately. Cross-condition comparisons of contextual lightness effects and reversible figures illuminate the distinction between an inference that is persistently biased and a competition between plausible interpretations that alternates over time Long & Toppino (2004).
Naturalistic Constancy and Contemporary Empirical Tests
Contemporary research revisits the ecological validity of classic constancy estimates. Gegenfurtner & Weiss (2024) studied color constancy with familiar real objects in ordinary environmental conditions. The task involved identifying the color of an absent known item under different illuminations. Performance was highly stable within the limits of visual memory, illustrating how a natural identification task may yield greater constancy than some constrained appearance-matching paradigms. The finding is important because it exposes task dependence; it does not establish that appearance itself is invariant in every scene.
Further experiments isolate difficult natural arrangements. Morimoto et al. (2025) examined cast shadows, where illumination inside and outside a shadow differs in both intensity and chromaticity, using controlled projector-based scenes and subjective white-point judgments. This design addresses an ecological variation that flat uniformly illuminated patches omit. Its inferential value lies in specifying the illumination transition and the measurement of appearance under that transition. A naturalistic task and a carefully structured laboratory manipulation should therefore be treated as complementary tests with distinct standards of control.
Computational comparisons can support these investigations while retaining a clear distinction between biological and algorithmic performance. Toscani et al. (2025) evaluated traditional color constancy algorithms using spectrally rendered scenes with ground-truth illumination information, illustrating how benchmark choice and physical image realism affect comparative conclusions. An algorithm's illuminant-estimation error is not the same observable as a human observer's color match. Comparative conclusions require compatible stimuli, task objectives, error functions, and evidence about the observer or system actually performing the task.
Reversible Figures and the Architecture of Multistability
Reversible figures provide a distinctive observation: one substantially unchanged drawing can support different mutually exclusive organizations. The Necker cube affords two dominant depth orientations, while other figures permit alternative object or figure-ground interpretations. The observer may initially experience either solution, maintain it for a period, and then report an unsolicited reversal. Long & Toppino (2004) established a useful conceptual distinction between ambiguity, the availability of multiple interpretations, and reversibility, the observed alternation among them. This distinction prevents the mistaken assumption that every uncertain image must produce spontaneous switching.
Interpretations depend on the structural relationships in the display and on the observer's perceptual state. Adaptation, attention, eye movements, prior exposure, and the time allotted for observation can influence the distribution of interpretations. A single static snapshot of an ambiguous drawing is therefore insufficient to characterize its behavior. Experimental series can measure first-percept probability, latency to first reversal, subsequent dominance durations, and the proportion of mixed or uncertain reports while maintaining the retinal pattern as constant as practicable Leopold & Logothetis (1999).
Reversible figures are scientifically useful because stimulus-driven changes can be contrasted with spontaneous ones. An experiment can introduce disambiguating edges or shading, or physically replay a sequence of clear alternatives. These controls allow investigators to ask whether a signal reflects sensory interpretation, perceptual reorganization, report preparation, or an unexpected external change. The interpretive challenge is to avoid treating every difference between ambiguous and unambiguous conditions as evidence for a unique neural source of the conscious percept Kornmeier & Bach (2005).
Binocular Rivalry and Competition Between Visual Representations
Binocular rivalry occurs when different incompatible patterns are presented to corresponding regions of the two eyes, yielding alternating periods in which one image dominates experience and the other is partially or largely suppressed. It is a controlled example of competition between alternative perceptual organizations. Blake & Logothetis (2002) synthesized evidence that rivalry engages multiple processes and levels of visual organization. Consequently the term does not refer to a single neural switch or a universal duration distribution. Stimulus contrast, spatial configuration, timing, adaptation, and reporting method all constrain the observed alternations.
Rivalry differs from ambiguous-figure viewing in the origin of conflict: discrepant images are delivered to the eyes instead of multiple interpretations being supported by one drawing. Both produce multistability, but similarity of their temporal signatures does not prove identical implementing circuitry. A useful comparison manipulates strength, attention, and exposure with comparable response measurements and includes controls for binocular interaction and monocular rivalry. The relevant common problem is selection among perceptual interpretations under conditions that resist a continuously stable unambiguous solution Leopold & Logothetis (1999).
Dominance duration, transition duration, and return transitions are distinct empirical observables. Brascamp et al. (2006) systematically tested their contrast dependence and found patterns implicating stochastic dynamics in addition to adaptation and mutual inhibition. This work places constraints on models that describe competition solely as two reciprocally inhibiting units. A successful model must account not only for mean switching rate but also for temporal distributions, dependence on stimulus parameters, and the character of intervening perceptual states. Observers' button reports remain an imperfect temporal record of internal transitions.
Adaptation, Inhibition, and Stochastic Transition Dynamics
Adaptation offers one possible reason that a currently dominant interpretation loses stability. Neural responses to sustained or repeated stimulation change over time, reducing the effective advantage of one pattern and allowing another to become more competitive. Reciprocal inhibition offers a mechanism through which competing representations suppress one another. Noise can initiate or modulate transitions when the system approaches a decision boundary. Theoretical models frequently combine these ingredients, and their adequacy depends on the distinctive predictions they make for duration distributions and contrast manipulations Brascamp et al. (2006).
Different mechanisms can generate qualitatively similar alternations. For example, slowly accumulating adaptation followed by deterministic switching and fluctuating evidence combined with a probabilistic decision rule may both produce variable reversal times. Distinguishing them requires interventions such as changing stimulus strength, brief interruption of stimulation, or evaluating serial dependence and return transitions. Model recovery tests are needed because parameter estimates can become misleading if a candidate model approximates data that actually arose from another process. The experimental program therefore involves quantitative identification, not merely fitting a plausible curve.
The presence of stochasticity in a successful behavioral model should not be interpreted as proof of one neural noise source. Variability may arise across sensory transduction, recurrent activity, attention, eye movement, memory, or response timing. A strong mechanistic account specifies which latent processes produce measurable transition signatures and seeks independent physiological or experimental evidence. The multistability literature consequently supports the interaction of processes across temporal scales while leaving contested the exact mapping from behavior to cellular mechanisms Brascamp et al. (2018).
Attention, Expectation, and Voluntary Influence
An observer can sometimes prolong a particular interpretation of a reversible figure by directing attention or attempting to hold it, although spontaneous transitions commonly remain. Such effects indicate that perceptual competition is sensitive to cognitive state, yet the manipulation of attention must be separated from instructions to report a favored percept. Long & Toppino (2004) reviewed findings consistent with the interaction of sensory and cognitive influences rather than a single explanatory mechanism. The key experimental issue is whether changes occur in the perceptual state itself or in the threshold and latency of a response.
Attention may alter processing gain, selection of diagnostic image regions, stability of the current interpretation, or the speed with which a transition enters awareness. Expectation can bias the initial interpretation without necessarily changing the dynamics of later rivalry. Eye position and fixation shifts introduce additional sensory changes that may imitate top-down control. Researchers accordingly measure eye movements, counterbalance cues, include physically disambiguated controls, and compare voluntary and incidental attention. When a result survives these controls, it supports a more specific claim about modulation of inference.
Contextual influence need not mean conscious reasoning. Learned scene statistics, short-term adaptation, and expectation can operate without explicit verbal deliberation. Conversely, an instruction to reinterpret a picture may recruit response strategies that alter reports. The mature explanatory goal is to determine which perceptual component is altered, at what timescale, and by which measured intervention. Accounts that assign all reversal control to either early visual processing or deliberate cognition are too coarse for the range of reported behaviors Brascamp et al. (2018).
EEG Evidence for Perceptual Reorganization
Electrophysiology contributes temporal constraints by measuring event-related activity associated with reported changes in ambiguous percepts. Kornmeier & Bach (2005) developed a paradigm that related an early occipital response at approximately 120 milliseconds to perceptual reversal in Necker-cube viewing. The result is a measured signal under a particular experimental design. It does not establish that this component constitutes the subjective transition itself or that all ambiguous stimuli switch at the same processing stage. Event timing depends on how the moment of perceptual reversal is operationally defined.
Ambiguous and disambiguated images can also be compared while minimizing differences in task requirements. Kornmeier & Bach (2009) reported a later ERP component around 400 milliseconds that varied strongly between ambiguous and unambiguous variants of more than one figure class. The authors interpreted this response in relation to the evaluation of competing interpretations. These temporal signatures are valuable precisely because different components may reflect distinct operations: formation of a percept, assessment of its reliability, reorganization, or later access and report.
EEG studies must consider stimulus-locked versus response-locked averaging, temporal uncertainty in subjective reports, changes in eye position, and the contribution of motor preparation. A difference between scalp waveforms cannot by itself identify a unique anatomical source or establish consciousness. Stronger tests compare spontaneous reversals, controlled stimulus changes, no-report proxies, and conditions matched for button use. These designs make temporal evidence informative within a model of multiple sequential and recurrent operations rather than treating a peak latency as a complete mechanism.
Neuroimaging, Report Confounds, and Causal Interpretation
Functional imaging has associated spontaneous perceptual reversals with distributed changes in brain activity. Kleinschmidt et al. (1998) found transient changes across visual and parietal areas during reversals of ambiguous figures, providing evidence that perception-related transitions involve multiple interacting regions. Sterzer & Kleinschmidt (2007) compared spontaneous switches with physically replayed disambiguated motion and found relative inferior frontal activation with timing evidence interpreted as possible involvement in initiating spontaneous changes. Such results raised important hypotheses about feedback and high-level perceptual inference.
Subsequent work challenged a simple assignment of frontal activity to perceptual change. Frässle et al. (2014) used binocular rivalry and methods allowing perceptual transitions to be tracked with reduced dependence on explicit reports. The study linked much frontal activation to introspection and action rather than to perception itself. The resulting dispute illustrates why neural correlation is sensitive to task design. Frontal engagement during a switch could reflect causal influence, monitoring, attention, uncertainty, decision, or response production, depending on the experimental comparison.
Reviewing the evidence across fMRI, EEG, stimulation, and modeling, Brascamp et al. (2018) emphasized the unresolved role of frontoparietal cortex in multistable perceptual inference. Causal claims require interventions that are selective enough to separate perceptual stabilization from response and attention demands. Timing differences alone are suggestive but subject to hemodynamic, decoding, and measurement limitations. The durable conclusion is that perceptual reversals provide a powerful design for examining neural interactions, while the necessity and sufficiency of individual regions remain separate empirical questions.
Perceptual Inference and Probabilistic Explanations
Bayesian approaches to perception formalize the relation between sensory evidence, prior expectations, and hypotheses about world structure. A prior distribution describes plausibility before the current evidence, a likelihood specifies the consequences of a candidate world state for the incoming signal, and posterior inference combines them. Kersten et al. (2004) used this architecture to analyze the interpretation of complex scenes with occlusion, variable illumination, and ambiguous objects. The approach offers a common mathematical language for both constancy and ambiguity: stable properties may be inferred across nuisance transformations, while competing scene hypotheses may retain substantial probability under the same image.
An explanatory Bayesian model must state which variables are represented and what task determines its response. For color constancy, possible latent variables include reflectance and illumination; for size, physical extent and distance; for rivalry, competing perceptual states and temporal dynamics. Posterior inference does not by itself specify whether the system reports a maximum estimate, samples hypotheses over time, or integrates uncertainty into an action. These alternatives can produce different behavioral predictions even when they assign similar probabilities to the same scene.
Probabilistic models are constrained by their empirical implementation. A model fitted after seeing the experimental results may redescribe the outcome without predicting independent data. Strong tests identify priors and likelihoods through separate measurements or manipulations, predefine response rules, and assess parameter recovery. Researchers should compare probabilistic accounts with relational, normalization, dynamic, and cue-combination alternatives. Bayesian terminology earns explanatory value only when the resulting mechanism predicts specific changes in measured behavior Landy et al. (1995) Brascamp et al. (2018).
Normalization, Spatial Ratios, and Scene-Based Grouping
Relational accounts emphasize information that remains relatively stable under some transformations of illumination. For example, spatial ratios among surface signals may provide useful constraints even when their absolute magnitudes change. Craven & Foster (1992) separated stable appearance from the attribution of scene change, encouraging researchers to consider relational information across the environment. Brainard (1998) demonstrated that achromatic settings in nearly natural arrangements cannot be predicted fully by a simple relation between a test patch and its immediate surround. These results direct attention to wider spatial organization and estimation of the illumination affecting multiple objects.
Lightness anchoring theories likewise assign a significant role to contextual grouping. In the framework developed by Gilchrist et al. (1999), surfaces belong to perceptual frameworks that contribute reference values for lightness computation. In displays studied by Adelson (1993), changes in apparent organization alter brightness judgments with little corresponding change in local target stimulation. The explanatory issue is how the visual system selects relevant comparison regions and whether grouping itself is inferred from edges, junctions, depth cues, or learned scene regularities.
Normalization and grouping can be tested experimentally. Hold the luminance of a target fixed while changing the perceptual grouping of surrounding areas; manipulate whether a boundary appears to be a reflectance edge or illumination edge; preserve local ratios while altering more distant surfaces. A model that predicts only simultaneous contrast has narrower scope than one that explains constancy across multiple illuminants and frameworks. The observed success of a relational rule must also be evaluated across chromatic channels, spatial scale, and viewing conditions Foster (2011).
Multisource Cue Integration and Robustness
Real-world surfaces and objects usually provide multiple partly redundant cues. Binocular disparity, contour, shading, motion, texture, and perspective can jointly constrain depth and shape. Experimental work by Bülthoff & Mallot (1988) revealed interactions rather than a uniform sum of independent cues. The framework synthesized by Landy et al. (1995) makes explicit the possibility of dynamically weighting cues and rejecting inconsistent information. These features matter for constancy because stability requires deciding which image changes reflect actual object transformation and which reflect changes in viewpoint or observation conditions.
A convenient statistical ideal weights independent estimates according to their reliabilities, but biological cue integration often involves correlated noise, context-dependent priors, categorical interpretation, and failure modes. The empirical test is whether predictions hold when cue reliability, disagreement, and spatial arrangement are manipulated independently. Measuring only an overall depth estimate can conceal which cue dominated a decision. Investigators can use conflict trials and perturbation analysis to identify relative influences, while keeping separate a descriptive combination rule and a claim about its neural implementation.
Constancy and multistability can intersect in conflict situations. Competing depth or illumination explanations may yield a stable compromise, a categorical choice, or alternating outcomes. Which behavior occurs depends on cue strength, adaptation, and temporal structure. Models that generate both reliable invariant estimates and occasional revisions offer useful conceptual integration, but cross-paradigm predictions remain a demanding test. Evidence from one cue-integration task cannot simply be transferred to binocular rivalry without verifying that the latent variables and response operations are commensurable Brascamp et al. (2006).
Experimental Control, Calibration, and Response Validity
Visual experiments require calibration of the quantities that define their hypotheses. Color constancy studies benefit from measured spectral power distributions, controlled surfaces and illumination, and explicit records of adaptation. Size and shape studies require known physical dimensions, viewing distance, binocular conditions, and geometrically specified displays. Multistability studies need precise presentation timing, checks for mixed percepts, and measurements of eye position or response latency when these factors could alter the interpretation. The nearly natural experiments of Brainard et al. (1997) exemplify the value of controlled illumination combined with a structured scene.
Response validity is distinct from stimulus validity. A forced-choice judgment can be highly reliable while answering a different question from open-ended appearance matching. In color research, a participant may successfully identify a surface under shifted illumination while setting an achromatic reference differently. In reversible-figure experiments, an instruction to press a key at a change introduces decision and motor time, and the participant may suppress reports of intermediate states. Experimental design should specify the mapping from experience to observable, test that observers understood the task, and compare methods when possible Foster (2011) Long & Toppino (2004).
Analytical practices also determine the evidentiary strength of a result. Data inclusion rules, segmentation of perceptual epochs, treatment of missing responses, model selection, uncertainty intervals, and observer-level variation should be documented. A group mean can obscure observers with distinct dominance statistics or matching strategies. Statistical comparisons should address plausible alternative models and not rely exclusively on significance thresholds. Cross-validation, independent replication, and preregistered contrasts are valuable because a flexible perceptual model can fit many tasks without revealing the operations that produced them Brascamp et al. (2018).
Behavior, Neural Measures, and the No-Report Question
Perceptual experience is usually accessed through an observer's action, choice, description, or confidence report. The same response includes sensory, decisional, attentional, and motor contributions, making the isolation of perceptual mechanisms a fundamental measurement problem. Neural recordings can complement report, but signals associated with a choice may reflect monitoring or response preparation. Research on rivalrous and reversible stimuli takes advantage of transitions occurring with relatively stable stimulation to seek differences linked to perceptual change rather than to new external input Kleinschmidt et al. (1998).
One important strategy uses involuntary proxies for what is seen, alongside trials that minimize explicit perceptual reporting. Frässle et al. (2014) challenged interpretations of frontal activation by comparing binocular-rivalry conditions that differed in introspection and action requirements. Such designs improve inference about report-related signals, although a proxy itself must be validated against independent perceptual judgments and may be influenced by factors other than conscious experience. No-report findings narrow candidate explanations; they do not automatically identify an observer-independent measurement of subjective experience.
Intervention is stronger evidence of causal involvement than correlation when the manipulation affects a sufficiently specific process. Temporally resolved perturbations, selective stimulation, pharmacological methods where appropriate, or controlled sensory changes can test whether a candidate circuit is necessary for a transition. Yet interventions can also change vigilance, eye movements, or response criteria. A complete account should combine task-matched behavior, physiological time courses, and manipulations with distinct predicted consequences Sterzer & Kleinschmidt (2007) Brascamp et al. (2018).
Perceptual Suppression, Awareness, and Competing Interpretations
Visual suppression paradigms show that a physically present stimulus need not dominate reportable experience. Wilke et al. (2003) developed generalized flash suppression, in which an abruptly introduced surrounding texture can make a salient target disappear from report under defined conditions. The effects depended on ocular and spatial arrangements and prior adaptation, indicating interactions between competition and relatively early sensory representations. The phenomenon offers a way to test the dynamic availability of a target despite continuing retinal stimulation.
Continuous flash suppression is a different controlled arrangement in which rapidly changing patterns delivered to one eye suppress awareness of an image in the other. Tsuchiya & Koch (2005) reported reduced negative afterimages during suppression, showing that perceptual suppression is associated with measurable changes in later visual response. The study supports investigation of interactions between awareness and visual processing; it does not imply complete elimination of sensory activity whenever an observer fails to report an image. Detection thresholds and task-specific indirect measures remain essential.
These paradigms illuminate ambiguity because competition and suppression can prevent one interpretation or object from becoming perceptually dominant. Their outcomes cannot be merged without attention to stimulus structure. A rotating ambiguous figure, monocular presentation of a Necker cube, conventional binocular rivalry, and continuous flash suppression provide different manipulations of available evidence. Comparing them helps identify shared principles of selection and competition, while preserving the boundaries of each experimental mechanism Blake & Logothetis (2002) Brascamp et al. (2018).
Where Constancy and Ambiguity Converge
Constancy and ambiguity arise from a shared information problem: the light reaching the eyes underdetermines the physical scene, yet behavior requires sufficiently reliable estimates of surfaces and objects. Their observable outcomes differ because their tasks differ. A constancy experiment asks whether a property estimate persists when illumination, viewing distance, or perspective changes. A multistability experiment asks why competing interpretations alternate while some stimulus variables remain stable. Probabilistic scene inference offers a common conceptual vocabulary, but it does not automatically imply a common switching mechanism Kersten et al. (2004) Brascamp et al. (2018).
The connection becomes experimentally fruitful when variables can be crossed. A surface can be shown under competing illuminant interpretations while observers indicate both perceived color and inferred illumination; an ambiguous three-dimensional object can be rotated while its shape and viewpoint cues change independently. Repeated judgments reveal whether an initial scene hypothesis stabilizes estimates across transformations and whether conflict induces uncertainty, switching, or averaging. Such approaches can connect work on relational color attribution Craven & Foster (1992) with experimentally measured perceptual competition Brascamp et al. (2006).
A synthesis should distinguish three explanatory scales: constraints of the physical stimulus, computational operations that infer or select scene properties, and biological interactions that implement those operations. Stability may reflect robust cues, adaptation, learned regularities, or persistent competition outcomes. Reversals may reflect changing evidence, adaptation, fluctuations, attentional modulation, or combinations of them. The empirical requirement is to predict how the same observer responds when a variable is perturbed, including when a representation remains stable and when it changes.
Individual Variation, Learning, and Environmental Context
Perceptual performance is an interaction between a task and an observer with a particular sensory history. Variability in color judgments can follow differences in adaptation, illumination experience, sensor physiology, or interpretation of response instructions. Size estimates depend on the availability and reliability of spatial cues; dominance durations in rivalry vary from episode to episode and across observers. A single group-average constancy index cannot capture this diversity. Hierarchical analyses and repeated individual testing can reveal whether variation lies mainly in perceptual sensitivity, criterion, prior information, or temporal competition Foster (2011) Brascamp et al. (2006).
Learning and environmental familiarity can influence which causes of sensory input are treated as plausible. Gegenfurtner & Weiss (2024) illustrates how familiar real objects and natural conditions alter the operational demands of color identification. Such findings make observer experience an essential candidate variable. However, recognition memory, labeling, and task strategy can also account for improved identification, so a learning effect should be decomposed through complementary judgments. Cross-cultural and developmental generalizations require appropriate populations and methods rather than inference from a single adult laboratory sample.
The environment is itself a component of the experiment. Illumination spectra, cast shadows, clutter, movement and viewing geometry differ between simplified displays and ordinary scenes. Morimoto et al. (2025) supplies a controlled investigation of shadow-related color constancy, while Brainard et al. (1997) and Brainard (1998) demonstrate how nearly natural arrangements can remain experimentally precise. A theory that performs well in laboratory patches has a genuine achievement, but generalization to complex settings must be shown with validated environmental sampling and matched dependent measures.
Cross-Paradigm Evidence and Explanatory Competition
An important challenge is that many theories explain overlapping sets of observations. Color gain adjustment, relational comparisons, and illuminant inference may all predict partial stability under some lighting changes. Depth cue weighting and scene priors may each support shape constancy. Rivalry models combining adaptation, inhibition, and stochastic influences can fit similar dominance statistics with different latent architectures. The existence of multiple good fits shifts the scientific task toward discriminating experiments and out-of-sample predictions Landy et al. (1995) Brascamp et al. (2006).
Theories should be evaluated against their strongest alternatives. In lightness research, contextual organization must be compared against what can be explained by receptor-level or early spatial interaction, with stimuli designed to dissociate those accounts Adelson (1993). In multistable perception, the relationship between frontoparietal activation and a reported switch must be tested against report and monitoring explanations Sterzer & Kleinschmidt (2007) Frässle et al. (2014). Neither a single dramatic visual illusion nor a localized imaging signal can decide an architecture whose competing models predict the same outcome.
Convergent evidence is strongest when different methods test an identifiable causal question. Psychophysics can constrain function, image statistics can constrain plausible stimulus causes, physiological recordings can constrain temporal and spatial implementation, and controlled interventions can test necessity or sufficiency. Such triangulation requires carefully matched tasks. Reproducibility means more than repeated significance: effect direction, magnitude, uncertainty, and explanatory prediction should remain interpretable across independent samples, stimulus families, apparatus and analysis choices.
Artificial Vision as a Comparative Research Domain
Artificial vision systems estimate color, object identity, geometry, and scene structure from sensory data. Their errors and invariances can be measured with explicit target labels or rendering-based ground truth, permitting controlled comparisons across transformations that are difficult to produce in everyday settings. Toscani et al. (2025) tested traditional color constancy algorithms on spectrally rendered scenes, a contribution to evaluating algorithmic behavior under known physical illumination. Such benchmarks make a computational function measurable, but a color-corrected camera output and a human appearance judgment remain different scientific observables.
A productive comparison specifies what remains invariant, what evidence is available, and how a response is produced. An artificial classifier may preserve identity under lighting changes while representing surface spectra inaccurately. A generative model may describe the Necker cube's alternative interpretations without exhibiting human-like spontaneous dominance intervals. These findings are meaningful descriptions of system behavior, and additional experimental criteria are required for claims about shared mechanisms or subjective perception. The problem is to establish functional correspondence, model identifiability, and robustness under controlled perturbation.
Within the Artificial Era, cross-system perception research can ask whether stable object representations, uncertainty estimates, active sensing, and temporal revisions are organized by comparable principles across biological and artificial systems. These are experimentally testable questions. The philosophical framing of Artificial Sapiens belongs to the author's wider intellectual program; empirical claims about specific artificial systems require independent behavioral and technical evidence. Comparative research gains clarity when stimulus information, training exposure, test leakage, output criteria and causal mechanisms are measured explicitly Kersten et al. (2004) Toscani et al. (2025).
Critical Appraisal of the Evidence Base
The literature establishes the reality of systematic perceptual constancies and multistable transitions with a variety of reproducible experimental paradigms. Its strength is greatest where physical variables and observer responses are precisely specified: controlled illumination changes, geometric cue perturbations, calibrated luminance contexts, unchanging ambiguous drawings, and conflicting binocular images. Studies such as Bülthoff & Mallot (1988), Brainard (1998), Brascamp et al. (2006), and Kornmeier & Bach (2009) each test a concrete mechanism-sensitive condition. Their results do not measure the same construct, so they should be integrated through explicit explanatory distinctions rather than combined into one numerical summary.
The most consequential limitations concern ecological generalization, measurement equivalence, and causal inference. A stable color identification response can coexist with changing apparent hue. A low reversal rate can arise from stimulus contrast, adaptation, attentional engagement, or the observer's reporting threshold. Neural activation associated with a switch may accompany monitoring or action instead of producing perception. The no-report controversy Frässle et al. (2014) exemplifies how methodological decisions can revise the interpretation of previously observed brain activity. Modern real-world color experiments Gegenfurtner & Weiss (2024) similarly show the importance of matching laboratory questions to everyday functions.
Review evidence should be weighed according to original study design and accessible results. The present synthesis draws on verified bibliographic records, available original full texts for selected open-access investigations, publisher descriptions, and scholarly reviews; some articles were available primarily through abstracts or metadata. No pooled effect size, unpublished sampling detail, or unverified replication count is inferred. The strongest conclusion is that perceptual stability and instability arise from interacting context-sensitive processes, while the precise implementation and cross-task generality of proposed models remain subjects of active investigation.
Research Priorities and Falsifiable Questions
A priority for constancy research is to measure multiple outcomes in the same participants and scenes: immediate appearance matches, material identity judgments, illuminant attribution, confidence, and action-guiding estimates. If a model explains one outcome but not another, the difference becomes evidence about the operations it captures. Spectrally characterized natural environments and reproducible synthetic scenes should be paired to identify which conclusions depend on physical realism, stimulus calibration, or task familiarity Foster (2011) Toscani et al. (2025).
A priority for ambiguity research is to separate perceptual change, state awareness, decision and action through convergent measurements. Time-resolved neural signals, carefully validated no-report indicators, controlled physical replay, and targeted perturbations can test whether a region changes the percept, detects a transition, or prepares the report. Computational models should predict the joint statistics of dominant states, mixed transitions, return transitions, and individual variability across independent stimulus conditions Brascamp et al. (2006) Frässle et al. (2014) Brascamp et al. (2018).
Cross-domain theory faces a further challenge: produce a model that predicts both the persistence of object properties under nuisance transformations and a transition to a different interpretation under conflict. This requires distinguishing changes in world state from changes in observer state, specifying when the perceptual system should accumulate evidence, and defining an operational loss function that can be evaluated. Artificial vision offers new models and controlled testbeds, while human psychophysics supplies richly characterized behavioral constraints. Stronger integration will follow from novel predictions validated across both rather than from conceptual resemblance alone.
Conclusion
Perceptual constancy is the context-dependent stabilization of inferred object and surface properties across changing stimulation. Perceptual ambiguity is the availability of multiple plausible interpretations of limited sensory evidence, sometimes accompanied by spontaneous alternation. These phenomena are linked through the problem of reconstructing useful environmental structure from signals that depend jointly on the world and the viewing conditions. Their observed outcomes are specific to the property being judged, the stimulus manipulation, and the reporting method.
Historical size and brightness experiments established the question, while controlled color matching, scene organization, depth cue combination and naturalistic tests refined the evidence. Reversible figures and binocular rivalry provided complementary access to competition, adaptation, attention, uncertainty and neural dynamics. The documented literature supports multiple interacting contributions rather than a single universal explanation: constancy can be strong in one task and incomplete in another, and switching involves processes whose sensory, cognitive and report-related components must be experimentally distinguished Gilchrist et al. (1999) Foster (2011) Brascamp et al. (2018).
The scientific synthesis yields a precise research program: specify a distal property or set of competing scene hypotheses, independently manipulate the available cues, record multiple measures of perception and action, and compare models through predictions for new conditions. A coherent account of perceptual stability must explain systematic errors and transitions as well as successful recognition. These standards extend naturally to the study of artificial perceptual systems, where functional invariance, uncertainty management and dynamic reinterpretation can be operationally assessed. The remaining task is causal and comparative: determine which computations generalize, under which environmental constraints, and through which demonstrable mechanisms.
References
- Adelson, E. H. (1993). Perceptual organization and the judgment of brightness. Science, 262(5142), 2042–2044. https://doi.org/10.1126/science.8266102
- Blake, R., & Logothetis, N. K. (2002). Visual competition. Nature Reviews Neuroscience, 3(1), 13–21. https://doi.org/10.1038/nrn701
- Brainard, D. H. (1998). Color constancy in the nearly natural image. 2. Achromatic loci. Journal of the Optical Society of America A, 15(2), 307–325. https://doi.org/10.1364/JOSAA.15.000307
- Brainard, D. H., Brunt, W. A., & Speigle, J. M. (1997). Color constancy in the nearly natural image. I. Asymmetric matches. Journal of the Optical Society of America A, 14(9), 2091–2110. https://doi.org/10.1364/JOSAA.14.002091
- Brascamp, J. W., van Ee, R., Noest, A. J., Jacobs, R. H. A. H., & van den Berg, A. V. (2006). The time course of binocular rivalry reveals a fundamental role of noise. Journal of Vision, 6(11), Article 8, 1244–1256. https://doi.org/10.1167/6.11.8
- Brascamp, J., Sterzer, P., Blake, R., & Knapen, T. (2018). Multistable perception and the role of the frontoparietal cortex in perceptual inference. Annual Review of Psychology, 69, 77–103. https://doi.org/10.1146/annurev-psych-010417-085944
- Bülthoff, H. H., & Mallot, H. A. (1988). Integration of depth modules: Stereo and shading. Journal of the Optical Society of America A, 5(10), 1749–1758. https://doi.org/10.1364/JOSAA.5.001749
- Craven, B. J., & Foster, D. H. (1992). An operational approach to colour constancy. Vision Research, 32(7), 1359–1366. https://doi.org/10.1016/0042-6989(92)90228-B
- Foster, D. H. (2011). Color constancy. Vision Research, 51(7), 674–700. https://doi.org/10.1016/j.visres.2010.09.006
- Frässle, S., Sommer, J., Jansen, A., Naber, M., & Einhäuser, W. (2014). Binocular rivalry: Frontal activity relates to introspection and action but not to perception. Journal of Neuroscience, 34(5), 1738–1747. https://doi.org/10.1523/JNEUROSCI.4403-13.2014
- Gegenfurtner, K. R., Weiss, D., & Bloj, M. (2024). Color constancy in real-world settings. Journal of Vision, 24(2), Article 12. https://doi.org/10.1167/jov.24.2.12
- Gilchrist, A., Kossyfidis, C., Bonato, F., Agostini, T., Cataliotti, J., Li, X., Spehar, B., Annan, V., & Economou, E. (1999). An anchoring theory of lightness perception. Psychological Review, 106(4), 795–834. https://doi.org/10.1037/0033-295X.106.4.795
- Holway, A. H., & Boring, E. G. (1941). Determinants of apparent visual size with distance variant. American Journal of Psychology, 54(1), 21–37. https://www.jstor.org/stable/1417790
- 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
- Kleinschmidt, A., Büchel, C., Zeki, S., & Frackowiak, R. S. J. (1998). Human brain activity during spontaneously reversing perception of ambiguous figures. Proceedings of the Royal Society B: Biological Sciences, 265(1413), 2427–2433. https://doi.org/10.1098/rspb.1998.0594
- Kornmeier, J., & Bach, M. (2005). The Necker cube—An ambiguous figure disambiguated in early visual processing. Vision Research, 45(8), 955–960. https://doi.org/10.1016/j.visres.2004.10.006
- Kornmeier, J., & Bach, M. (2009). Object perception: When our brain is impressed but we do not notice it. Journal of Vision, 9(1), Article 7. https://doi.org/10.1167/9.1.7
- 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
- Leopold, D. A., & Logothetis, N. K. (1999). Multistable phenomena: Changing views in perception. Trends in Cognitive Sciences, 3(7), 254–264. https://doi.org/10.1016/S1364-6613(99)01332-7
- Long, G. M., & Toppino, T. C. (2004). Enduring interest in perceptual ambiguity: Alternating views of reversible figures. Psychological Bulletin, 130(5), 748–768. https://doi.org/10.1037/0033-2909.130.5.748
- Lotto, R. B., & Purves, D. (1999). The effects of color on brightness. Nature Neuroscience, 2(11), 1010–1014. https://doi.org/10.1038/14808
- Morimoto, T., Sato, M., Sunaga, S., & Uchikawa, K. (2025). Human color constancy in cast shadows. i-Perception, 16(4), 1–17. https://doi.org/10.1177/20416695251349737
- Sterzer, P., & Kleinschmidt, A. (2007). A neural basis for inference in perceptual ambiguity. Proceedings of the National Academy of Sciences, 104(1), 323–328. https://doi.org/10.1073/pnas.0609006104
- Toscani, M., Chen, T., & Guarnera, G. C. (2025). Evaluation of classic colour constancy algorithms on spectrally rendered ground-truth. Perception, 54(7), 478–502. https://doi.org/10.1177/03010066251345871
- Tsuchiya, N., & Koch, C. (2005). Continuous flash suppression reduces negative afterimages. Nature Neuroscience, 8(8), 1096–1101. https://doi.org/10.1038/nn1500
- Wallach, H. (1948). Brightness constancy and the nature of achromatic colors. Journal of Experimental Psychology, 38(3), 310–324. https://doi.org/10.1037/h0053804
- Wilke, M., Logothetis, N. K., & Leopold, D. A. (2003). Generalized flash suppression of salient visual targets. Neuron, 39(6), 1043–1052. https://doi.org/10.1016/j.neuron.2003.08.003