Color, Light and Materials

Talk Session: Tuesday, May 19, 2026, 5:15 – 7:15 pm, Talk Room 1
Moderator: Rhea Eskew, Northeastern University

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

Talk 1, 5:15 pm, 55.11

Blurred foveal vision makes the periphery appear blurrier: Rapid recalibration of peripheral appearance under degraded foveal input

Seonggyu Choe^1,2,3 (), Alex R. Bowers^2,3, MiYoung Kwon¹; ¹Department of Psychology, Northeastern University, Boston, MA, USA, ²Schepens Eye Research Institute of Massachusetts Eye and Ear, Boston, MA, USA, ³Department of Ophthalmology, Harvard Medical School, Boston, MA, USA

The fovea and periphery differ markedly in sampling resolution, resulting in characteristic declines in acuity and contrast sensitivity with increasing eccentricity. Yet perceptually, the visual field appears seamlessly clear and coherent. This uniformity is thought to arise from calibration processes that align peripheral appearance with high-resolution foveal vision. When foveal input is degraded, as in conditions causing central vision loss, this calibration hierarchy is disrupted. This raises the question of whether peripheral appearance is recalibrated to match a compromised fovea. Here we measured perceived peripheral blur using the point of subjective equality (PSE), with the fovea (normal or degraded) serving as the reference. Normally-sighted observers viewed naturalistic dead-leaves textures with varied peripheral blur and judged whether the periphery appeared blurrier than the fovea. Foveal degradation was simulated by applying a 1.33-cpd low-pass filter to the central 5° region in a gaze-contingent display, yielding blur level comparable to ~20/450 acuity. PSEs were obtained before-and-after 20-minute adaptation during which participants performed demanding recognition tasks under foveal blur. A separate experiment confirmed that perceived clarity of the normal fovea was comparable to 20/20 acuity. We observed pronounced shifts in both the PSE and the psychometric slope as foveal clarity decreased. Under normal-fovea viewing, a peripheral blur level of 5.23 cpd±0.34 (~20/120) was perceived as matching the intact fovea. This six-fold difference in physical resolution indicates that normal foveal-vision strongly anchors perceived peripheral clarity. In contrast, under degraded-fovea, a peripheral blur level of ~1.39 cpd±0.16 (~20/430) was perceived as equal to the foveal blur. PSEs in degraded-fovea condition were unaffected by adaptation, suggesting this calibration occurs immediately. Our findings suggest that the visual system rapidly down-regulates perceived peripheral clarity when foveal input is degraded, thereby maintaining perceptual balance between the fovea and periphery. They also offer insight into perceptual changes associated central vision loss.

Talk 2, 5:30 pm, 55.12

Natural retinal statistics and the design of an optimal trichromatic array

Deepthi Bannai^1,2, Alexander Belsten^1,2, Bruno Olshausen^1,2, Jacob Yates^1,2; ¹UC Berkeley, ²Redwood Center for Theoretical Neuroscience

Human color perception arises from the comparison of signals from long-(L), medium-(M), and short-(S) wavelength-sensitive cones, which form a mosaic that tiles the retina and encodes visual input. Interestingly, the human cone mosaic contains few, spatially distributed S-cones (Curcio et al. 1991). Previous work attributed S-cone sparsity to longitudinal chromatic aberration (LCA), but neglected to incorporate the full optical constraints of the eye, utilize a trichromatic array, or learn spatial arrangements (Garrigan et al. 2010). We address these limitations by proposing an efficient-coding model that optimizes a population of linear-nonlinear neurons to maximize information transfer between its inputs and photoreceptor outputs ((Karklin & Simoncelli 2011). Each cone in the mosaic learns a spatial location and spectral type. Model inputs are derived from hyperspectral natural images (Arad et al. 2022) processed through a full model of the human optics (Cottaris et al. 2019). Power analysis of hyperspectral, optically-processed, and cone-response images demonstrated that optical and spectral filtering alters image statistics in a wavelength-dependent manner. Hyperspectral images had a 1/f^2 distribution, which only medium wavelengths retained up to a cutoff after optical processing. Due to high LCA, shorter wavelengths had the lowest amount of power, which translated to S-cones having significantly lower power than L- and M-cones. Optimal mosaics learned by the efficient-coding model had few, distributed S-cones, similar to observed human mosaics. Finally, we observed a negative relationship between degree of LCA at shorter wavelengths and proportion of S-cones in learned mosaics. Taken together, this work provides a framework for describing the formation of an optimal trichromatic cone mosaic. By incorporating natural images processed by realistic model of the human, the model predicts optimal mosaics which match those of the human retina. Future work will apply this framework to the optical systems of other species.

National Science Foundation, grant # 2313149

Talk 3, 5:45 pm, 55.13

Analysis of natural images reveals response polarity differences in the cones and in the chromatic channels

Yangyi Shi¹, Rhea T. Eskew, Jr.¹; ¹Northeastern University

Human visual perception exhibits distinct asymmetries in the processing of increments and decrements. Psychophysical studies show that perceptual scales of achromatic and S-cone increments follow a saturating curve, while their decrements both follow a reversed, inverted S-shaped curve; discrimination thresholds on increment pedestals are higher than on decrement pedestals (Shi & Eskew, 2024, 2025, Gabree, Shepard & Eskew, 2018), and field spectral sensitivities are different for S increment and decrement tests (McLellan & Eskew, 2000). The present study investigates whether these psychophysical polarity differences could be related to natural image statistics. Hyperspectral images of natural and urban scenes (Cauwerts & Jost, 2019; Nascimento et al., 2016) were analyzed to characterize the distribution of cone response contrasts. Spectral data were converted to L, M, and S cone excitations, and then filtered with Difference of Gaussians (DoG) using different spatial scales for each type of cone to simulate cone response contrast. Post-receptoral channel responses were then calculated using weighted sums of cone contrasts. The analysis revealed a consistent statistical dominance of decrements over increments across L, M, and S cones, at multiple spatial scales. In post-receptoral channels, the Red-Green channel showed a prevalence of "greenish" over "reddish" responses, while the Yellow-Blue channel showed a prevalence of "yellowish" over "bluish" responses, with few exceptions. Both Luminance (L+M) and Achromatic (L+M+S) channels exhibited a strong bias toward decremental responses. These results align qualitatively with previous psychophysical asymmetries. The observed decrement dominance may be attributed to the physics of material reflectance: objects typically exhibit low reflectances due to light-absorbing pigments (Richards, 1982). Furthermore, consistent with the efficient coding hypothesis, the high sensitivity observed at low contrasts for both polarities in psychophysics can be explained by the prevalence of low-contrast content in the natural environment.

NSF BCS-2239356

Talk 4, 6:00 pm, 55.14

Diverse and functionally distinct contrast response functions of macaque V1 neurons

Cai-Xia Chen¹, Shu-Chen Guan¹, Nian-Sheng Ju¹, Sheng-Hui Zhang¹, Xin Wang¹, Shi-Ming Tang¹, Cong Yu²; ¹Peking University, ²Zhejiang University

Contrast response functions (CRFs) of V1 neurons are typically described as fast rising at low contrasts and saturating at high contrasts. These CRFs are well captured by the Naka–Rushton model and are effective in explaining low-contrast discrimination, but how well they support high-contrast discrimination remains unclear. Additionally, intermediate- and high-threshold neurons have been reported, yet their prevalence and functional roles are not well established. We used two-photon calcium imaging to measure CRFs of V1 neurons in five fixating macaques. Among ~10,000 identified neurons, 54.2% were low-threshold (C10<0.25) with steep, shallower, or linear slopes, 21.7% were intermediate-threshold (C10=0.25-0.60) with steep or shallower slopes, and 9.7% were high-threshold (C10>0.60) with steep slopes. Because the Naka–Rushton model’s C50 is less meaningful for intermediate- and high-threshold classes, we adopted a revised version (Heeger, 1992) replacing C50 with a threshold term, such that CRF arises when stimulus contrast exceeds threshold. This revised model fits the data well (R2=0.74). To characterize functional roles of each CRF class, we trained binary logistic classifiers to discriminate adjacent contrast levels with high accuracy. Furthermore, ablation experiments revealed that removing intermediate- and high-threshold neurons selectively impaired contrast discrimination at their corresponding contrast ranges. However, removing low-threshold neurons impaired discrimination at all contrasts, though less so at intermediate and high contrasts, likely because low-threshold neurons’ sheer number and many show shallower or even linear slopes to convey higher contrast information. CRF slope was also critical: removing steep-slope neurons markedly reduced discrimination across the full contrast range, while removing shallower-slope neurons had minimal impact. We conclude that (1) low-threshold neurons may not fully decode intermediate and high contrasts without contributions from higher‑threshold neurons; (2) The revised Naka–Rushton model is more interpretable describing CRFs of all neurons; (3) Neurons with steep CRF slopes play a dominant role in contrast decoding.

This study was supported by a Ministry of Science and Technology, China, STI2030-Major Projects grant (2022ZD0204600).

Talk 5, 6:15 pm, 55.15

The Geometry of Color Space: Suprathreshold Differences from Discrimination Thresholds

Fangfang Hong¹ (), Jason Chow², Phillip Guan², David H. Brainard¹, Alex H. Williams³; ¹Department of Psychology, University of Pennsylvania, ²Reality Lab Research, Meta, ³Center of Neural Science, New York University

A foundational question in psychophysics is whether a geometric perceptual metric can predict both discrimination thresholds and suprathreshold difference perception. Investigation of this question is greatly facilitated by comprehensive measurement of both, but this has been intractable because of a large number of psychophysical trials required. Here, we first fully characterized one participant’s color discrimination thresholds throughout a 3D color gamut (~30,000 trials) using a 3AFC oddity task and fitting an innovative Wishart Process Psychophysical Model. This model derives color discrimination thresholds in all directions around any reference stimulus, and characterizes the local geometry of color space. We then evaluated how well this local geometry can be used to predict suprathreshold color differences. Suprathreshold differences were measured in the same participant using a 2AFC task with stimuli closely matched to the threshold task (~7,200 trials). On each trial, a reference stimulus was cued, and the participant judged which of two comparisons appeared more different. We interleaved 12 conditions, each defined by a fixed reference (R_fixed) and a comparison (C_fixed). For each condition, the second comparison (C_variable) was adaptively sampled to measure the contour at the same perceptual distance from R_fixed as C_fixed. The difference between R_fixed and C_fixed was varied across conditions from ~2 to ~6 just-noticeable differences (jnds). The measured contours were compared to equal-geodesic-distance predictions derived from the threshold data. Specifically, for each R_fixed, we traced Riemannian geodesic paths in many directions, identified where the geodesic distance from R_fixed matched that of C_fixed, and connected those points to form the equal-geodesic-distance prediction. This prediction holds for short distances between R_fixed and C_fixed (< ~2 jnds), but breaks down as this distance increases. The deviations appear systematic, with the measurements falling inside the predicted equal-geodesic-distance contours. These measurements provide important constraints for developing global geometric metrics of color comparison.

Meta

Talk 6, 6:30 pm, 55.16

Task-Dependent Compensation in Anomalous Trichromacy

Fatemeh Charkhtab Basim¹, Erin Goddard², Kenneth Knoblauch³, Kimberly Jameson⁴, Michael Webster¹; ¹Department of Psychology, University of Nevada, Reno, NV, US, ²School of Psychology, University of New South Wales, Sydney, Australia, ³Univ Lyon, Université Claude Bernard Lyon 1, Inserm U1208, Stem Cell and Brain Research Institute, Bron, France, ⁴Institute for Mathematical Behavioral Sciences, University of California, Irvine, CA, USA

A number of studies have shown that anomalous trichromats may have stronger color percepts than predicted by their threshold sensitivity losses. However, the form and nature of this compensation remains poorly understood. Prior work has typically examined isolated tasks, leaving open how compensation varies across different levels of perceptual and cognitive processing. In the present study, we used common stimuli and compensation metrics to quantify anomalous observers’ performance across a range of different tasks. 17 anomalous trichromats (11 deuteranomalous, 6 protanomalous) and 15 normal trichromats completed a battery assessment for stimuli that varied in nominal LvsM chromatic contrast. These included color classification (achromatic boundary placement), focal-hue selection, reaction-times, contrast adaptation, and magnitude estimation (MLDS). For each, we computed the contrast-scaling factor required to align an observer’s responses with the average values for the color normal observers, normalized by scaling relative to the threshold elevation (resulting in a compensation index ranging from 0=no compensation to 1=full compensation). The average index varied strongly with task, Appearance-based measures (e.g. focal hues) showed the strongest effects, along with contrast scaling (MLDS). Alternatively, contrast adaptation exhibited significant but smaller improvements, with weakest compensation found for reaction times. The degree of compensation also showed large observer variability, which was unrelated to the magnitude of the individual’s threshold loss. Across tasks, deuteranomalous observers generally showed stronger compensation than protanomalous observers, particularly for naming and reaction times, whereas group differences were weaker for adaptation, hue selection, and MLDS. Overall, the results show that compensation in anomalous trichromacy may involve multiple distinct processes that are differentially engaged by perceptual tasks.

Supported by EY-010834

Talk 7, 6:45 pm, 55.17

Color communication across the life span

Angela Brown¹, Delwin Lindsey²; ¹Ohio State University College of Optometry, Graduate Program in Vision Science, ²Ohio State University Department of Psychology, Graduate Program in Vision Science

Color communication is more than a naming game where the informant names each color: the color terms must also be understood by the listener. Here, we studied how color naming and color term comprehension (participants’ productive and receptive vocabularies) affect communication about color. *****Human participants, 3–72 years old, played a color communication game. As “senders”, they named 15 or 30 color samples presented individually. Then, as “receivers”, they viewed or heard “message” terms from their own sender data (self-communication) or from another participant (interpersonal communication), in random order. The receiver’s task was to select, from the full array of colors, the sample that corresponded to each message. The number of color terms in participants’ data sets as senders, their fraction correct performance for self-communication, and their fraction correct for interpersonal communication, all matured continuously until at least age 21, and were highly correlated with each other. *****Simulations correctly predicted that participants’ self-communication choice performance should be directly correlated with the number of color terms they had provided as senders (their productive color vocabulary size), both of which improved with age. By contrast, simulations predicted that participants’ interpersonal choice performance should be (1) much worse than self-communication, (2) nearly unrelated to their productive vocabulary size and (3) independent of age. Contrary to these predictions, (1) choice performance for interpersonal communication was only 6% below self-communication, (2) it was strongly associated with the receiver’s productive vocabulary size (r^2=0.76), and (3) it improved throughout the age range (r^2=0.70). Further analyses attributed these effects to an age-related increase in the size of participants’ receptive color vocabularies, because participants often understood terms they had not themselves produced as senders. *****Therefore, color naming (productive color vocabulary) and color term comprehension (receptive color vocabulary) are distinct abilities that mature independently across the life span.

NIH T35 EY007151, BCS-1152841, Center Of Science and Industry, Columbus, OH

Talk 8, 7:00 pm, 55.18

Neural correlates of cross-modal expectations in material perception

Amna Malik¹, Katja Doerschner¹; ¹Justus Liebig University

Humans routinely integrate visual and auditory information to infer what objects are made of and to estimate their material properties. Although multisensory facilitation is well documented, most neuroimaging studies have investigated responses to simultaneously presented congruent versus incongruent audiovisual stimuli, where bottom-up integration and top-down processes are tightly intertwined and difficult to disentangle. Consequently, the contribution of top-down cross-modal expectations remains less explored, particularly in dynamic, complex scenes. Here, we used fMRI to investigate how such cross-modal expectations about material properties are represented in the brain. Participants viewed computer-generated animations depicting familiar objects (a wooden chair, a porcelain teacup, and a jelly) being dropped to the ground, providing visual information that elicited expectations about the object’s material and its characteristic impact sound. Just before impact, the object was masked, leaving only the auditory information. On each trial, this auditory information either met the expectations elicited by the preceding visual information (congruent) or violated them (incongruent), and participants judged whether the object was hard or soft. Multivariate pattern analyses showed that audiovisual congruency was most robustly represented in the lateral occipital cortex (LOC). Crucially, LOC activity was modulated even though expectations were violated in the auditory domain and the corresponding impact events were never presented visually. These findings indicate that object-selective visual cortex may maintain anticipatory, visually derived representations of how an object is expected to behave across modalities, and that these representations are updated when incoming auditory information violates those expectations. More broadly, they suggest that LOC may support the flexible use of context-dependent predictions to guide behavior in complex and dynamically changing environments.

This work is supported by the Deutsche Forschungsgemeinschaft (German Research Foundation, DFG) project number 222641018 – SFB/TRR 135 and DFG under Germany’s Excellence Strategy (EXC 3066/1 “The Adaptive Mind”, Project No. 533717223).