Visual crowding

Visual crowding is impaired recognition of a target caused by nearby visual elements, often called flankers. It is especially important in peripheral vision: an object that is recognizable alone can become difficult to identify among its neighbours, even while remaining visible. In XR, crowding has been studied both as a limit on perception in complex scenes and as a possible consideration for image processing in headsets.

Reviewed 21 October 2026. Original crowding research, spacing and grouping models, scene recognition and headset-study limitations About review dates.

Recognition and visibility

Crowding is distinct from the inability to resolve an isolated small target, measured by visual acuity. It is also distinct from ordinary masking that reduces a signal's visibility. Pelli, Palomares and Majaj's experiments compared identification with detection: neighbouring letters could strongly impair identifying a letter while having little effect on detecting its presence. The distinction depends on the task; their account associates crowding with combining information from multiple detected features.

For XR interface design, this distinction implies that testing an isolated symbol does not establish its recognizability among surrounding items.

Critical spacing

Eccentricity is the target's angular distance from fixation. Critical spacing describes the target-flanker separation needed to avoid, or reach a specified criterion for, interference. In classic letter experiments, spacing is measured between centres, not between edges. Critical spacing increased approximately in proportion to eccentricity, with a value near half the eccentricity in the conditions studied.

This relationship is commonly called Bouma's law. It can be expressed as s ≈ bE, where s is critical spacing, E is eccentricity and b is a proportionality factor. The factor is not a fixed constant for every scene or task. Soo, Chakravarthi and Andersen's 2018 experiments jointly varied flanker contrast, display duration, backward masking, preview and target-flanker contrast polarity. Combinations of these properties changed critical spacing, so individual effects could not simply be added on that measurement scale.

Those researchers proposed critical resolution, proportional to the inverse square of critical spacing, as an alternative measure. On this scale, the tested combinations were approximately additive. Their results came from controlled letter-orientation tasks on a monitor; they do not supply a single spacing value for all XR menus. Comparisons between studies also need to account for the performance criterion used to estimate a threshold.

Explanations and the surrounding layout

Pooling accounts explain crowding through combining target and flanker information over a region of the visual field. Grouping and segmentation accounts also consider which elements are assigned to the same perceptual group. These approaches need not be mutually exclusive: a grouping stage can determine which elements subsequently interfere.

Bornet and colleagues compared computational models against human results for dense arrays of oriented bars, using a genetic algorithm to select configurations. In that paradigm, models with a dedicated grouping stage reproduced the principal human patterns, whereas the tested feedforward pooling models did not. The paper also discusses uncrowding, in which adding elements can improve target discrimination because the overall configuration changes. Thus, the number or proximity of distractors alone is insufficient to predict every crowding result. This comparison concerns particular models and tasks; it does not identify one settled neural mechanism for all crowding.

Complex scenes

Ringer and colleagues' 2021 experiments used images of computer-generated living rooms with realistic objects. Four flankers surrounded each target, with their separation varied between image versions while retaining the scene context. One experiment briefly presented the image; another used eye tracking to control fixation and remove the scene when the participant moved their eyes. Recognition accuracy declined with closer flankers in both experiments, despite the objects fitting the scene semantically.

These were monitor-based scene-image experiments, rather than unrestricted interaction in a VR room. They show that coherent scene context does not necessarily remove crowding. The authors also caution that their tested spacing range did not establish the full recognition-performance curve or a universal critical distance for realistic scenes.

Blur perception in VR

Andriamanalina, Larabi and Le Moan's 2025 study used an HTC Vive Pro Eye to compare blur discrimination for isolated natural-texture patches and patches surrounded by four sharp neighbours. Thirty-two people participated; one was excluded after response-quality analysis, with 13-15 valid observations retained per texture. The experiment used twelve textures, six per participant, and a fixed target eccentricity of 12.5 degrees. Participants compared two successive presentations to identify the blurrier central patch.

Overall, stronger filtering was needed to detect blur with flankers, but the effect varied by texture. Associations with target-flanker similarity were examined after excluding two luminance-outlier textures from that analysis; a separate exploratory regression included all twelve. These findings suggest a possible role for crowding in foveated rendering and compression, but the study did not benchmark a deployed rendering or streaming system. It tested one blur filter and controlled peripheral patches, leaving other distortions and broader content for further research.

Headsets as research instruments

A July 2026 study by Volkova, Dowd and Grubb reproduced crowding and attentional modulation in Apple Vision Pro. Fourteen participants judged the orientation of a peripheral T flanked by H characters. Accuracy improved with greater separation; model-based critical distances were smaller when an uninformative peripheral cue appeared on the target side rather than the opposite side. The separate polar-angle experiment in the same paper addressed a different visual phenomenon.

The crowding application used visionOS 2.5 and required gaze at a central cross followed by a pinch to initiate each trial. Brief stimulus presentation limited opportunities for an eye movement. The researchers lacked continuous real-time eye-position access in that setup, and reported positions in visionOS points without a precise conversion to degrees of visual angle. They therefore bounded their conclusions to relative comparisons within the device. This method is not a substitute for continuous fixation monitoring or exact angular calibration in every experiment.