Perceptual Organization: Neural mechanisms

Talk Session: Friday, May 15, 2026, 4:15 – 5:45 pm, Talk Room 2
Moderator: Daniel Kaiser, Justus Liebig University Giessen

Schedule of Events | Search Abstracts | Symposia | Talk Sessions | Poster Sessions

Talk 1, 4:15 pm, 15.21

Characterizing figure-ground response modulation in human visual cortex

Emil Olsson¹ (), Michael Epstein^2,3, Juneau Wang³, Karen Tian³, Brian Maniscalco¹, Jennifer Motzer³, Angela Shen¹, Minh Nguyen¹, Meera Sriram¹, Sydney Liu¹, Olenka Graham Castaneda¹, Maggie Zhang¹, Yuzheng Wu¹, Diana Gamboa³, Richard Brown⁴, Victor A. F. Lamme⁵, Hakwan Lau^6,7, Biyu J. He⁸, Jan W. Brascamp⁹, Ned Block^10,11, David Chalmers^10,11, Rachel Denison³, Megan Peters^{1,12,13,14,15}; ¹Department of Cognitive Sciences, University of California Irvine, ²Center for Brain Imaging, New York University, ³Department of Psychological & Brain Sciences, Boston University, ⁴Department of Humanities, LaGuardia Community College; The Graduate Center, City University of New York, ⁵Department of Psychology, University of Amsterdam, Amsterdam, ⁶Department of Biomedical Engineering and Department of Intelligent Precision Healthcare Convergence, Sungkyunkwan University, ⁷RIKEN Center for Brain Science, ⁸New York University Grossman School of Medicine, ⁹Department of Psychology, Michigan State University, ¹⁰Department of Philosophy & Center for Neural Science, ¹¹Department of Psychology, New York University, ¹²Department of Logic & Philosophy of Science, University of California, ¹³Center for the Neurobiology of Learning & Memory, University of California, ¹⁴Center for Theoretical Behavioral Sciences, University of California, ¹⁵Program in Brain, Mind, & Consciousness, Canadian Institute for Advanced Research

Recurrent processing in early visual cortex underlies figure-ground segregation and has been proposed as a neural correlate of visual awareness. Human fMRI studies have demonstrated figure-ground modulation in early visual cortex, but how this modulation varies with stimulus strength and across visual areas V1-V3 is not well characterized. Here, we measure figure-ground modulation across human V1-V3 using carefully controlled stimuli. We collected fMRI data from 27 participants across two independent sites (BU n=15, UCI n=12). Using an event-related design, participants viewed briefly-presented (250 ms) texture-defined figures at five stimulus strength levels, behaviorally calibrated per participant. Critically, the stimuli control for low-level confounds, with matched luminance across strength levels and elimination of T-junctions at figure edges. Using preregistered behavioral and analytic pipelines with pRF mapping and a spatial localizer, we identified voxels in V1-V3 responsive to figure locations and computed figure-ground modulation (response to figure-present minus figure-absent). Voxels were grouped as “figure-positive” (enhanced response to figure presence) or “figure-negative” (suppressed response) based on this modulation. Across V1-V3, roughly half of the figure-responsive voxels showed figure-positive modulation and half showed figure-negative modulation, a ratio that was consistent across visual areas and robust to different classification criteria. We examined how figure-ground modulation varied as a function of stimulus strength and across the visual hierarchy. In both figure-positive and figure-negative voxels, figure-ground modulation responses remained stably present across areas V1-V3 (with numerically largest responses in V1), and increased modestly with stimulus strength, suggesting that figure-ground segregation involves coordinated and stable excitatory and inhibitory processes. This characterization of distinct figure-positive and figure-negative response populations, and their consistency across visual areas and stimulus strengths, sets the stage for future studies testing the functional role of figure-ground modulation in visual processing and experience.

Acknowledgements: Templeton World Charity Foundation 0567 to BH, JB, NB, DC, MP, and RD, startup funding from the University of California Irvine and support from the Canadian Institute of Advanced Research to MP, startup funding from Boston University to RD, and BU UROP funding to DG.

Talk 2, 4:30 pm, 15.22

Investigating Neural Basis of Serial Dependence Using EEG and Ocular Tracking Task

Bao Hong^1,2 (), Jing Chen², Li Li^1,2; ¹East China Normal University, Shanghai, China, ²New York University Shanghai

Serial dependence is a phenomenon where perception is biased toward recent experience. While this effect is well documented behaviorally, its neural processes remain unclear. To investigate this, we combined eye-tracking and EEG recordings while participants tracked the step-ramp motion of a moving target with its motion direction randomized on each trial. Behavioral analysis revealed robust serial dependence: pursuit directions at initiation were systematically biased toward previous-trial target motion direction, followed by a later repulsive adaptation effect. This dynamic pattern suggests that serial dependence in ocular tracking arises at early stages of visual processing. To uncover its neural basis, we conducted time-resolved multivariate pattern analysis (MVPA) on the EEG signals. While current-trial motion direction could be decoded from EEG signals, previous-trial motion direction also showed a higher-than-chance decoding accuracy, emerging around ~100 ms after stimulus onset and strongest for occipital-parietal electrodes. This suggests that perceptual history might be reactivated within visual cortex early in processing. We next analyzed EEG activity during both stimulus presentation and the inter-trial interval (ITI) using a time-resolved regression with motion direction as the predictor. During stimulus presentation, the regression weights increased sharply after stimulus onset and decreased at the end of trials. During the ITI, the weights showed a gradual rise and exhibited rhythmic fluctuations. Spectral analysis confirmed a significant alpha-band component peaked ~10.5Hz, strongest for occipital-parietal electrodes. Sliding-window analysis revealed increasing alpha amplitude during the ITI, suggesting that past motion information might be retained through rhythmic activity. Together, these findings provide converging behavioral and neural evidence that serial dependence in ocular tracking arises early in visual processing. The retaining of information from previous trials might be linked to rhythmic alpha-band activity in the visual cortex. These results offer insight into how the brain dynamically preserves and reactivates recent motion history to guide ongoing perceptual responses.

Talk 3, 4:45 pm, 15.23

Slow change blindness from serial dependence

Haley Frey^1,2, David Whitney^1,2,3; ¹Herbert Wertheim School of Optometry and Vision Science Program, University of California, Berkeley, ²Department of Psychology, University of California, Berkeley, ³Helen Wills Neuroscience Institute, University of California, Berkeley

Slow change blindness, when attentive observers fail to notice large changes that happen gradually, raises questions about how dynamic visual information is combined across time. One plausible strategy for integration is serial dependence: using information from the recent past to inform current perception. Here, we investigate continuous serial dependencies in perception of a cartoon object that slowly changes hue. Across three experiments, we measure perception at each moment in a morph by accumulating one-shot hue reports from a large number of observers who each viewed a different extent of the total slow hue change. Across participants, the entire morph was probed. We find that overall reported hue perception is biased towards the past (Experiments 1 and 2). Additionally, and more importantly, we find that the bias increases as more of the morph is experienced, consistent with a dependence on the preceding information at each moment (Experiments 1 and 2). By explicitly asking whether observers noticed the change, we verify that the resulting illusion of stability produces slow change blindness (Experiment 2). We show that reducing serial dependence via a change in object identity reduces the bias in hue reports, confirming that the measured bias is indeed serial dependence, rather than hysteresis or averaging (Experiment 3). Lastly, in a multi-trial replication, we show that observers who experience repeated trials report a smaller yet still significant bias, suggesting that continuous serial dependence persists despite knowledge of the task (Experiment 4). Overall, we provide evidence that serial dependence actively biases perception during gradual changes, producing slow change blindness.

National Institutes of Health grant T32EY007043; National Institutes of Health grant R01CA236793

Talk 4, 5:00 pm, 15.24

Feedback-Mediated Prior Integration Unifies Gestalt Perceptual Organization

Tahereh Toosi¹ (), Kenneth D. Miller²; ¹Zuckerman Mind Brain Behavior Institute, Columbia University, New York, NY USA, ²Kavli Institute for Brain Science, Columbia University Irving Medical Center, New York, NY USA

Gestalt principles such as closure, similarity, continuation, and figure-ground segregation have traditionally been characterized as distinct perceptual rules. We present evidence that these phenomena emerge from a single computational mechanism: iterative prior integration using cortical feedback pathways. We developed Prior-Guided Drift Diffusion (PGDD), an algorithm that repurposes the same feedback connections used during learning to refine neural activations during inference. PGDD iteratively updates representations by gradient ascend updates moving away from a noisy representation of the input, constrained to remain near the sensory evidence. Applied to pre-trained convolutional neural network trained for robust object recognition on ImageNet (e.g. ResNet50), PGDD reproduced classic Gestalt demonstrations without additional training or architectural modifications. For closure, networks presented with Kanizsa square inducers generated clear illusory contours completing missing edges. We tested large scale recurrent models (cornet-based or predictive processing) as controls, and they showed no square completion. Similarity grouping and good continuation emerged through the same mechanism, linking elements sharing visual features or completing interrupted line segments across gaps. Figure-ground segregation experiments revealed prior-dependent interpretation: when presented with the Rubin face-vase illusion, object-trained networks generated vase-like patterns while face-trained networks produced face-like features through identical PGDD processing. Control experiments confirmed this specificity: networks trained on faces or scenes failed to generate contour completion for Kanizsa square input, while object-trained networks succeeded. These effects matched biological signatures of Gestalt processing, including delayed response profiles and feedback dependence. The results suggest that diverse Gestalt phenomena reflect a unified process of integrating learned statistical regularities with sensory input through feedback pathways. Rather than requiring specialized mechanisms for each principle, perceptual organization emerges from how neural networks dynamically access implicit priors acquired during training to interpret ambiguous or incomplete sensory evidence.

T.T. is supported by NIH 1K99EY035357-01 and NSF DBI-2229929. This work was also supported by the Simon Foundation and Gatsby Charitable Foundation GAT3708.

Talk 5, 5:15 pm, 15.25

Constant Curvature Operators in Human Vision

Austin Phillips¹, Philip Kellman¹; ¹University of California, Los Angeles

A central challenge in vision science remains understanding how early, rapidly changing signals—from orientation-tuned contrast detectors, for example—are transformed into stable, symbolic descriptions of object properties. Recent research has investigated this question in the perception of contours and 2D shape. Evidence suggests that an initial symbolic representation of contour shape consists of segments of constant curvature (Baker, Garrigan, & Kellman, 2021). These representations may be produced by responses from banks of constant-curvature filters built from oriented units, linked by constant turning angles, in early visual processing (Kellman, Garrigan, & Erlikhman, 2013; Baker & Kellman, 2021). This initial symbolic encoding of arbitrary smooth contours into sequences of constant-curvature segments may provide a concise and versatile description of shape that provides a foothold for further operations. Together with neurophysiological findings regarding curvature sensitivity in V4 (e.g., Dumoulin & Hess, 2007; El-Shamayla & Pasupathy, 2016), this framework suggests that there may exist hard-wired detectors for constant curvature that operate in parallel across broad regions of the visual field. To investigate this hypothesis and explore properties of hypothesized curvature detectors in the visual pathway, we developed a variant of the path-detection task (Field, Hayes, & Hess, 1993). Observers judged small arrays containing three-element Gabor targets. Triplets forming constant-curvature arcs were contrasted with (a) relatable but non-constant-curvature arrangements with uniform curvature polarity, (b) relatable triplets with alternating curvature polarity, and (c) “null” configurations in which elements were non-relatable. Detection performance was substantially higher for constant-curvature triplets relative to all other conditions. Both relatable types exceeded performance on null triplets, and this pattern held at 600, 1000, and 1400ms exposures. These findings support the presence of dedicated curvature mechanisms that efficiently encode smooth contour structure, and may provide the earliest symbolic descriptions of contour shape.

Talk 6, 5:30 pm, 15.26

A rapid and automatic neural correlate of liking for visual artworks

Daniel Kaiser¹, Sanjeev Nara², Gustavo Menegon³, Vaishali Goyal³, Philipp Flieger¹; ¹Justus Liebig University Giessen, ²IIT Mandi, ³Lusofona University, University of Barcelona, and University of Paris Cité

When we appreciate visual art, we often take our time: We visit museums or galleries and deliberately engage with artworks and their meaning. This deliberate engagement in turn shapes whether we like an artwork or not. Yet, our liking might at least partly be shaped by incidental encoding of the artworks’ visual features, a process initially independent from sustained engagement. Here, we show that liking for visual art can be predicted from rapid and automatic responses in the visual brain. We recorded EEG responses while participants viewed 999 artworks from the Vienna Art Picture System (VAPS), which features complex artworks that span different categories (from still lives to portraits and landscapes) and art styles (from renaissance to impressionism and expressionism). Artworks were shown with a 300ms stimulus-onset-asynchrony (150ms on, 150 off), and participants responded to occasional target images. Using representational similarity analysis, we predicted the geometry of neural representations from the artworks’ similarity in liking (based on ratings from a different group of observers contained in the VAPS), while simultaneously controlling for similarities in other properties (category, art style, valence, arousal, and familiarity). Crucially, neural responses emerging within the first 150ms were robustly connected to liking, even though the task did not afford any liking judgments. Rapidly emerging visual representations thus predict liking for visual artworks, highlighting the contribution of basic visual feature processing. Strikingly, this early neural correlate of liking was also observed in a second experiment, in which we presented the same 999 artworks with a 50ms stimulus-onset-asynchrony. This shows that even under ultra-fast presentation conditions, automatic neural responses rapidly indicate liking. While fully appreciating an artwork may still need deliberate and prolonged engagement, the perceptual basis of whether we like or dislike an artwork seems to be laid in a split second.

DK is supported by the DFG (KA4683/6-1, project number 536053998) and an ERC Starting Grant (PEP, ERC-2022-STG 101076057). This work is further supported by the DFG under Germany’s Excellence Strategy (EXC 3066/1 “The Adaptive Mind”, project number 533717223).