Introduction
Perceptual constancy refers to our ability to correctly perceive stable properties of the outside world, despite the fact that their corresponding sensory signals are in flux. Size constancy in particular refers to our visual system’s ability to correctly perceive an object’s size despite the fact that the retinal image may shrink or grow as our distance to it changes. This constancy affords us a stable impression of the physical world around us, and allows us to perform essential actions that we may take for granted like catching a ball, estimating the width of a doorway and being able to determine whether an approaching animal is a cat or a tiger.
Much of the literature on size constancy and its neural underpinnings makes use of static displays, which leaves a gap in our understanding of what happens during ordinary, everyday episodes where the distance between an object and the observer changes. The overarching question of this thesis therefore relates to the neurophysiology of size perception when things are moving.
Motivation and research Question
The literature on moving stimuli is mixed. Under impoverished cues, motion can hinder size and distance perception and encourage reliance on retinal size. In richer scenes there is evidence that movement can help maintain size constancy, but studies often use ambiguous stimuli or illusory depth. The ambiguity in prior work motivates testing in depth cue rich environments that mimic real life viewing conditions, while measuring how neural representations in early visual areas behave when stimuli move in depth with either physical size held constant or retinal size held constant.
The experiments described in this thesis address the neurophysiology of size perception right before and after movement, when the stimulus has recently been in motion in a depth cue rich environment. To probe the activity in the early visual cortex, and more specifically the size of object representations in these areas, we will make use of steady state visual evoked potentials (SSVEP) measured via electroencephalography (EEG), a method with proven effectiveness. These experiments will take place both in Virtual Reality, (VR), and in a physical setup that preserves the same geometry and depth cues. Experiments in this thesis will examine whether the primary visual cortex continues to encode physical size (as it does under static conditions), or whether it reverts to a purely retinotopic representation when motion is involved.
Conceptual Advances
Across matched experiments in VR and using a physical setup, we find the same neuroscientific result, namely that for objects that have just moved in depth, early visual cortex does not show size constant encoding. The SSVEP responses reflect retinal size after motion in both implementations of the task. By contrast, a control experiment using static but otherwise comparable stimuli shows partial size constancy of object representations in early visual cortex, replicating previously published findings. The conceptual advance is a clear dissociation between dynamic and static conditions. Following motion, early visual cortex represents the fixated object in a retinotopic way, whereas in a completely static condition, partial size constancy emerges in these same brain regions.
This dissociation marks a key conceptual advance. It suggests that motion alters the balance between feedforward and feedback processing in early visual cortex, perhaps tipping it in favor of bottom-up input at the expense of perceptual (top-down) tuning. The result is that the same region of cortex (V1) that supports size constancy under static conditions fails to do so immediately after motion—even when behavioural responses indicate correct, size constant perception. This means that size-constant representations in V1 are not a fixed property, but dynamically modulated by viewing conditions.
The static control experiment’s results also show that our SSVEP approach is sensitive to size constant signals when they are present, and demonstrates the replicability of earlier findings of this effect. This is to say that our findings are most likely related to motion rather than due to the specific visual environment being used or a failure to replicate earlier findings.
A potential explanation for these findings is that during more dynamic viewing conditions, bottom up processing in early visual cortex dominates over feedback processing, at least in the cortical layers that contribute most to the SSVEP. Our findings highlight a need to understand the temporal dynamics and laminar specificity of feedforward and feedback influences in early vision—especially under ecologically valid, dynamic conditions.
Methodological contributions
A second major contribution is methodological. This work pairs a carefully controlled VR environment with a custom apparatus that moves a physical monitor along the sagittal axis, and equates stimulus geometry, timing and depth cues across the two. This design makes it possible to test the ecological validity of the VR environment and data quality in VR without substantially changing the paradigm.
Two insights follow:
1. Behavioral validity: despite differences between VR and real life in size and distance judgements reported in the literature, we find VR to be an appropriate medium for experimentation, provided that the virtual scene is cue rich and closely mirrors the physical setup. In the behavioural tasks, errors are essentially identical between VR and the physical setup, with an average difference under one millimetre.
2. EEG signal quality: SSVEP data quality is much reduced in VR. Headset interference with electrodes results in remarkably worse signal-to-noise ratios. Although VR offers unmatched control over complex dynamic scenes, this comes at the cost of EEG data quality and necessitates careful attention to signal integrity.
The thesis also catalogues VR specific pitfalls that matter for experimental design. Distance is often overestimated in sparsely cued scenes. To keep SSVEP contrast stable, some cues (like shading) cannot be used, so scenes must be constructed with careful anchoring to preserve depth information. Comparing VR to a physical counterpart helps validate that these design choices do not distort the effects of interest.
All things considered, this work finds support for the use of VR in neuroscientific experiments in vision sciences.
Broader implications
The broader implication of these findings is that we should not assume neuroscientific findings related to perceptual constancy which are collected from experiments using static conditions generalize to more ecologically valid, dynamic contexts. While we demonstrate this distinction for size perception in particular, it should be noted that other forms of perceptual constancy (such as shape, colour or brightness constancy) tend to be studied in static conditions, while in real world viewing, these mechanisms emerge in dynamic visual environments.
Although the precise temporal characteristics of this transition remain to be mapped, and our data do not by themselves prove that this transition is caused by a shift in feedforward versus feedback balance. They do, however, motivate future experiments that are sensitive to more specific cortical layers and temporal dynamics, which could give us a clearer picture of the neural underpinnings of size perception specifically and visual perception more broadly. Ultimately, the thesis invites a rethinking of perceptual neuroscience: one that more fully embraces the complexity and dynamism of real-world vision.
Verknüpfung zu Publikationen oder weiteren Datensätzen
