Color, Light and Materials: Cones to cognition

Talk Session: Sunday, May 18, 2025, 5:15 – 7:30 pm, Talk Room 2

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

Talk 1, 5:15 pm

UNIFORM DISTRIBUTION OF SPECTRAL TUNING IN THE MOUSE EARLY VISUAL SYSTEM

Juan Santiago Moreno^1,2,3,4, Daniel Denman^1,2,3,4; ¹University of Colorado Anschutz Medical Campus, ²Medical Scientist Training Program, ³Neuroscience Graduate Program, ⁴Department of Physiology and Biophysics

Like other dichromats, mice expresses two wavelength-sensitive cone opsin proteins: a medium (M) opsin centered in the green region of the light spectrum and a short (S) opsin shifted in sensitivity toward the ultraviolet (UV) region. However, these opsins are not uniquely expressed by dedicated cone cells, nor are they expressed uniformly across the retina, yet they still support color discrimination. A variety of potential mechanisms for color opponency have been shown in the mouse retina, but few studies have captured distribution of spectral tuning downstream in lateral geniculate nucleus (LGN) or the primary visual cortex (V1) populations, nor the relationship between the two. Here, we used high-density electrophysiology to record from LGN (n = 888 units) and all layers of V1 (n = 2729 units) during the presentation of chromatic, achromatic, spatially structured, and uniform stimuli (n = 8 mice, 8 recordings). We defined spectral tuning based on the response-weighted average of M and S-opsin contrasts presented, providing a distribution in M-S contrast space that represents both color opponency (M-S) and luminance preference (M+S). Additionally, we developed a selectivity index that labels neurons as either color preferring (negative) or luminance preferring (positive). As opposed to a model where color and luminance are processed by distinct circuits, single neuron spectral tuning in LGN and V1 formed uniform distributions in contrast space, with their selectivity indices skewing towards luminance. We find an expansion in spectral tuning in V1 compared to LGN (LGN: range = [-0.184, 0.259], skew = 0.687; V1: range = [-0.328,0.621], skew = 1.488), suggesting a de novo origin of color information in cortical processing. Our findings provide a framework for color vision, where spectral tuning in single cells multiplexes color and luminance, and populations collectively encode the full breadth of the mouse visual spectrum without specialized parallel circuits.

NEI R00EY028612, NEI R00EY028612-S1, 1F30EY034775

Talk 2, 5:30 pm

Variation of small spot color appearance with local L/M ratio

Maxwell Greene¹, Vimal Pandiyan², Ramkumar Sabesan², William Tuten¹; ¹University of California, Berkeley, ²University of Washington

Suprathreshold color appearance is surprisingly robust to considerable variations in L/M cone ratio between trichromatic observers. Previous studies suggest chromatic sensitivity may be reduced in individuals with biased cone proportions (Gunther & Dobkins, 2002; Hood et al, 2006). Due to technical limitations, these studies relied on indirect L/M ratio measurements. Additionally, variation of color perception with L/M ratio was assessed between subjects, making it hard to disentangle the effect of cone demographics from other individual differences. Building on this earlier work, we examined how local variations in cone spectral topography influenced responses to small, brief (67 ms) increments of red (680 nm) or green (543 nm) light in two male trichromats whose cone mosaics had been spectrally classified by optoretinography. Flashes subtending 2.25 arcmin (covering ~5 cones) were presented through an adaptive optics scanning laser ophthalmoscope to targeted retinal loci ~2° from fixation. On each trial, the subject indicated whether the stimulus appeared red, green, or achromatic, or whether it went unseen. Stimulus intensities were concentrated near threshold, with occasional high intensities to gauge suprathreshold color appearance. Overall, achromatic percepts predominated (73.9% of seen trials). We fit a generalized linear mixed-effects model to examine how the the probability of categorizing the 543 nm stimulus as “green” or the 680 nm stimulus as “red” depended on i) stimulus intensity (normalized by sensitivity) and ii) the spectral heterogeneity (i.e., proximity to a 1:1 cone ratio) of the stimulated retinal locus. As expected, the likelihood of responding “red” or “green” increased with intensity. Interestingly, we found that the ability to categorize stimuli under the expected hue name improved significantly as the L/M ratio approached unity. Hence, at a fine spatiotemporal scale, the arrangement of cones imposes a limit on the postreceptoral mechanisms subserving color appearance.

Air Force Office of Scientific Research (FA9550-20-1-0195, FA9550-21-1-0230), the National Institutes of Health (R01EY02359, T32EY007043, U01EY032055, P30EY001730), Research to Prevent Blindness Unrestricted grant, the Hellman Fellows Fund, and the Alcon Research Institute.

Talk 3, 5:45 pm

Neural Mechanisms of color saturation

Robert Shapley¹, Valerie Nunez², James Gordon³; ¹Center for Neural Science, New York University, ²Center for Neural Science, New York University (now at Albert Einstein College of Medicine), ³Hunter College, City University of New York

Color saturation is a quantitative estimate of how colorful something looks. In an attempt to understand the neural mechanisms of color appearance, we measured perceived color saturation in human observers with hue and saturation scaling, as used by Gordon et al. (1994). The color targets were equiluminant color-gray checkerboards (spatial frequency 2-3 c/deg) presented on a calibrated OLED monitor. The colors lay along the two cardinal axes of DKL color space: “Red/Green” (L-M/M-L) and “Blue/Yellow” (+S/-S). Five or six cone contrasts were used for each stimulus ranging from 0-10% for the L-M/M-L stimuli and from 0-50% for the +S/-S stimuli. The observers estimated saturation as the percentage of the entire sensation, chromatic and achromatic, that was chromatic. To compare color perception with activity in early visual cortex, we also measured the chromatic visual evoked potential (cVEP) over the same range of cone contrast and for stimulus patterns similar to those used in the behavioral experiments. The main results are: 1) saturation varies with the magnitude of cone contrast, and therefore equiluminant complementary colors appear equally saturated; 2) the slope of the saturation vs cone contrast line is 6-8 X shallower for the +S/-S stimuli than for L-M/M-L; 3) cVEP amplitude's dependence on cone contrast resembled the scaling data. It was interesting that, for many observers, cVEPs to +S and -S were approximately equal in amplitude when stimuli were equated for cone contrast magnitude. This result is somewhat surprising because the neurons that carry -S (“Yellow”) signals in the LGN might be expected to respond weakly to the checkerboard stimuli. It raises the possibility that there is a cortical contribution to -S signals. The results also suggest that saturation is likely to be a result of integration of color-evoked responses over the entire population of neurons in early visual cortex.

Our research was supported by a grant from the US National Science Foundation

Talk 4, 6:00 pm

Neural gain compensation for color deficiencies: evidence from color contrast adaptation

Fatemeh Charkhtab Basim¹, Arsiak Ishaq², Erin Goddard², Michael A. Webster¹; ¹Graduate Program in Integrative Neuroscience, University of Nevada, Reno, NV, USA, ²School of Psychology, University of New South Wales, Sydney, Australia

Anomalous trichromacy (AT) is characterized by a reduced separation between the spectral peaks of the normal medium (M) and long (L) wavelength cone photopigments, resulting in a reduced LvsM comparison signal. Several studies suggest that the color vision of ATs is compensated for this sensitivity loss, so that their color perception is more similar to color-normal (CN) observers. However, it is often difficult to distinguish whether the compensation observed is sensory (e.g., neural gain) or post-perceptual (e.g., learning). To directly probe sensory gains, we compared contrast adaptation in 11 color-normal (CN) and AT observers (10 deutan and 5 protan), by measuring LvsM contrast thresholds before or after adapting to 1 Hz modulations of LvsM contrast for 120 sec. Adaptation increases the thresholds, and the losses increase with the adapting contrast. In ATs, LvsM sensitivity is lower, but adaptation should also be weaker, because the adapt contrast is a lower multiple of threshold. This predicts that the change in thresholds (post/pre adapt) should be weaker for ATs. The predicted relation was quantified by measuring threshold changes in CNs after rescaling stimulus contrasts by varying amounts to simulate different levels of sensitivity loss. All deutan observers exhibited stronger adaptation than their threshold sensitivity losses predicted, consistent with partial (but incomplete) compensation for their reduced LvsM separation. The results for 4 protan observers were instead consistent with no compensation (with a 5th outlier in the direction of over-compensation). Chromatic contrast adaptation is thought to have an early cortical locus. The results thus provide direct psychophysical evidence for compensatory neural gain adjustments at or before early visual cortex that may partially offset the altered cone signals in some anomalous trichromats, and reinforce neuroimaging (fMRI) measures which have also implicated a cortical site of compensation for color deficiencies (Tregillus et al 2021).

Funding: EY010834

Talk 5, 6:15 pm

Prolonged fixation durations in color-deficient observers reflect a generalized visuo-motor adaptation

Doris I. Braun¹ (), Karl R. Gegenfurtner¹; ¹Giessen University

About 300 million men worldwide have inherited color vision deficiencies (CVD). While their difficulties in distinguishing shades of red and green are well documented, much less is known about the broader aspects of their visuo-motor behavior. We examined eye movements of 27 color-normal and 24 color-deficient observers of different types and severities. They viewed high-quality digital reproductions of still-life paintings, presented both in their original colors and in grayscale. We measured saccade counts and amplitudes, fixation positions and durations, explored image areas (spread), scan paths, fixation heatmaps, and the chromatic properties of fixated regions. Color-deficient participants exhibited an average increase of about 15 ms in fixation duration compared to color-normal observers. Notably, this increase also appeared when viewing grayscale images, suggesting it is not solely related to color discrimination. The effect was especially pronounced for the very first fixation. Despite this increase in fixation duration, we observed remarkably little difference in the overall spatial distribution of fixations or the color distributions of fixated regions. This may reflect the distinct boundaries of objects within these artworks, allowing both groups to identify salient features similarly. Preliminary data from natural image viewing also show prolonged fixation durations, especially for initial fixations, while no such differences emerge during reading tasks. These findings suggest that color-deficient observers may have adapted a slightly different fixation strategy to cope with their challenges in processing visual information. Unlike previous accounts that focused on specific perceptual deficits, our results indicate a more general alteration in visuo-motor behavior.

ERC AdG Color 3.0 and DFG SFB/TRR 135 Cardinal mechanisms of perception.

Talk 6, 6:30 pm

Sparse coding of chromatic natural images recovers universals in color naming and unique hues

Alexander Belsten^1,2 (), Paxon Frady³, Bruno Olshausen^1,2,4; ¹Redwood Center for Theoretical Neuroscience, University of California, Berkeley, ²Herbert Wertheim School of Optometry and Vision Science, University of California, Berkeley, ³Neuromorphic Computing Lab, Intel, ⁴Helen Wills Neuroscience Institute, University of California, Berkeley

Understanding the transformation from cone activations to color appearance is a central question in vision research. Here, we study the color statistics of natural images and propose a model of this transformation based on principles of efficient coding. We show that this model replicates universals in color naming from the World Color Survey (WCS) (Kay and Cook, 2023) and aligns with unique hues (Hering, 1878). Whether or not the unique hues have a privileged status in perception has been subject to debate (Conway et al., 2023), and our model provides additional data and computational analyses that help address this question. Using simulated long-, medium-, and short-wavelength (LMS) cone activations in response to natural scenes, we compute a decorrelating transform that spheres the activations (i.e., achieves unit variance in all directions). Utilizing overcomplete, non-negative sparse coding models, we derive optimal bases for representing the data distribution and analyze their tuning properties. The decorrelating transform aligns with the DKL color space (Derrington, Krauskopf, Lennie, 1984). The data distribution in this space exhibits asymmetrical, heavy-tailed structure within the chromatic plane, but not along the luminance axis. Adapting sparse coding models to this structure recovers bases that align with universal color categories identified by the WCS. Notably, the six basis-vector model aligns with the unique hues and replicates features of psychophysical characterizations of these hues, such as mutual exclusivity between opponent colors (e.g., red and green). This work demonstrates that a model based on principles of efficient coding can provide an account for the physiology of color processing and the psychology of color appearance. Specifically, sparse coding models adapted to the sparse structure of the LMS distribution recover bases aligned with universal color categories and the unique hues, supporting previous reasoning that the unique hues are a well-suited basis for describing color appearance.

AFOSR FA9550-21-1-0230 and NSF IIS231-3149

Talk 7, 6:45 pm

Edge-based Image Reconstruction Provides a Unified Account of (many) Lightness Illusions

Srijani Saha¹ (), George Alvarez^1,2, Talia Konkle^1,2; ¹Harvard University, ²Kempner Institute for the Study of Natural and Artificial Intelligence at Harvard University

Lightness illusions demonstrate that how bright an object appears depends on an elaborate constructive process, to the point that the same surface can be perceived as either black or white depending on the context. Why does the biological visual system work this way? Traditionally, distinct computational goals have been proposed to account for simple lightness illusions (e.g. the Craik-O’Brien-Cornsweet Illusion) and for more complex illusions (e.g., the moon illusion: discs in different hazy backgrounds, Anderson & Winawer 2005). The Craik-O’Brien-Cornsweet illusion seems to depend on local cues — a dark/light difference at a singular edge— whereas the moon illusion seems to also require a recovered scene structure. Our work examines whether an edge-based reconstruction goal produces a range of lightness illusions. First, we trained a reconstructive U-Net to output a filled-in image from edge-only inputs of images, an objective analogous to filling-in surfaces from edge-selective neurons in the biological visual system. This model not only reconstructed images with minimal error, but also made systematic errors consistent with lightness illusions measured in people for both the Craik-O’Brien-Cornsweet illusion and the Anderson-Winawer illusion. This effect was robust across training parameter choices (32 combined variations between training datasets and model seeds) and illusion probe choices (contrast signal of edges). When the model was applied to a suite of additional lightness illusions (e.g., Adelson Haze Illusion, Snake Illusion, Koffka Illusions, and Kanizsa Square Illusion), we found that the model consistently recapitulated illusions when there are connected edges with consistent polarity bounding the illusory surface. When the same U-Net model architecture was trained with a different reconstructive goal – denoising different levels of Gaussian noise – the models did not recapitulate any illusions, indicating that edge-based reconstruction is critical and provides a plausible mechanism underlying many perceptual lightness illusions.

This work was supported by NSF PAC COMP-COG 1946308 to GAA.

Talk 8, 7:00 pm

Generalization of color-concept associations mimics color category structure

Melissa A. Schoenlein¹, Karen B. Schloss²; ¹High Point University, ²University of Wisconsin-Madison

People learn associations between colors and concepts from experiences. Evidence suggests that color-concept associations for colors seen in the input generalize to unseen colors, and this generalization pattern depends on the distance between seen and unseen colors and color category membership—typicality and boundary location (Schoenlein & Schloss, 2022). The current study aimed to build a framework characterizing how association distributions are formed by evaluating how these factors combine. First, we used an associative learning paradigm to expose participants to co-occurrences between colors (yellows/pinks) and concepts (Filk/Slub alien species, respectively). Participants saw aliens in either prototypical (saturated) or non-prototypical (desaturated) yellows/pinks (between-subjects). Then, we assessed color-concept associations for seen colors and generalization to unseen colors. Participants rated how much they associated each species name (Filk/Slub) with colors from sequences spanning pink-to-orange and yellow-to-green, including colors that varied in typicality of color category membership. We constructed a model to predict these associations from: (1) color distance from the seen colors (CIE E), (2) color typicality judgments, and (3) category boundary judgements (all judgments made by different groups of participants). Color-concept associations were correlated with each factor in isolation, but in the model, only typicality (p<.001) and color distance (p<.001) were significant predictors (no effect of category membership; p=.898). Effects of category boundaries were superseded by typicality, implying that associations extrapolated to other colors inside the category boundary shared by the seen colors, but more so for colors that were more typical of the category. Further analysis indicated that these results could not be understood merely in terms of perceptual similarity. This work demonstrates that color-concept association generalization mimics color category structure. These findings increase our knowledge of how color-concept association distributions are populated from sparse input, furthering our understanding of human judgments in visual cognition that rely on associations.

This work was funded by NSF grant BCS-1945303 to KBS