Measurements of chromatic adaptation and luminous efficiency while wearing colored filters

Abstract

The visual system adapts dynamically to stabilize perception over widely varying illuminations. Such adaptation allows the colors of objects to appear constant despite changes in spectral illumination. Similarly, the wearing of colored filters also alters spectral content, but this alteration can be more extreme than typically encountered in nature, presenting a unique challenge to color constancy mechanisms. While it is known that chromatic adaptation is affected by surrounding spatial context, a recent study reported a gradual temporal adaptation effect to colored filters such that colors initially appear strongly shifted but over hours of wear are perceived as closer to an unfiltered appearance. Presently, it is not clear whether the luminance system adapts spatially and temporally like the chromatic system. To address this, spatial and temporal adaptation effects to a colored filter were measured using tasks that assess chromatic and luminance adaptation separately. Prior to and for 1 hour after putting on a pair of colored filters, participants made achromatic and heterochromatic flicker photometry (HFP) settings to measure chromatic and luminance adaptation, respectively. Results showed significant chromatic adaptation with achromatic settings moving closer to baseline settings over 1 hour of wearing the filters and greater adaptation with spatial context. Conversely, there was no significant luminance adaptation and HFP matches fell close to what was predicted photometrically. The results are discussed in the context of prior studies of chromatic and luminance adaptation.

Introduction

The visual system dynamically adapts to changes in illumination to maintain largely consistent color percepts over a wide range of lighting conditions. For example, when transitioning from indoor to outdoor environments, the color of an object remains quite stable, though the illumination spectrum and intensity have changed. This phenomenon of being able to “discount the illuminant” in color perception (Helmholtz von, 2013) is known as color constancy and is driven by a process known as chromatic adaptation (for recent reviews on color constancy, see Foster, 2011; Werner, 2014). While many factors affect the degree of chromatic adaptation, the illuminant is never totally discounted and adaptation is never fully complete. 

The complexity of surrounding contextual cues is an important factor in color constancy (Blackwell & Buchsbaum, 1988; Hansen, Walter, & Gegenfurtner, 2007; Kraft & Brainard, 1999; Pearce, Crichton, Mackiewicz, Finlayson, & Hurlbert, 2014), and the degree of color constancy is greater within contextually richer real-world environments versus the more impoverished visual displays often found in laboratory studies. For example, when viewing a uniform chromatic disc on a black background under a strongly colored illuminant, adaptation may be weak because of sparse surrounding contextual cues and the color appearance of the disc shifts more toward the color of the illuminant (Kraft & Brainard, 1999). In contrast, viewing the same stimulus in the natural world or in rich virtual reality scenes produces strong constancy effects (Gil Rodríguez et al., 2022). A common way to report the magnitude of a color-matching experiment under different illuminants is by using a “constancy index,” which is a proportion of the color constancy observed in a given study. The constancy index assumes an “ideal match,” which is equal to 1, is reflective of total adaptation, while an index of zero indicates no adaptation/discounting of the illuminant (Arend & Reeves, 1986; Foster, 2011). An analysis of color constancy experimental results over many studies reported color constancy indices as low as 0.11 and as high as 0.91, depending on the stimuli and methodology of different labs (Foster, 2011). 

Another factor that can reduce color constancy is an extremely unnatural change in the illuminant, such as being in a room lit with a strongly colored light bulb or wearing a colored filter. Studies on memory color matching with chromatic adaptation under different-colored illuminants showed that the level of adaptation drops as the illuminant becomes more chromatic (Smet, Zhai, Luo, & Hanselaer, 2017a, Smet, Zhai, Luo, & Hanselaer, 2017b). This reduction in the degree of adaptation as the color of the illuminant changes is observed as variations in the size of regions in Commission Internationale d’Eclairage (CIE) color space associated with specific color names (Hansen et al., 2007; Morimoto, Yamauchi, & Uchikawa, 2023) as well as reduced color discrimination assessed with color arrangement tests (Thomas & Kuyk, 1988; Coffey, Abel, Karas, Gavrilescu, & Douglass, 2023). 

The reduced color constancy while wearing colored filters can be partially reversed by allowing participants time to adapt to the illuminant change. Daily exposures of several hours to red or green environments result in hue shifts of unique yellow (a pure yellow that has neither red nor green in it) to compensate for the altered chromatic environments (Neitz, Carroll, Yamauchi, Neitz, & Williams, 2002). Unique yellow can be measured by having observers adjust the color along a red/green (R/G) dimension of color space until the stimulus appears as a yellow, which has no red or green in it. Li, Tregillus, Luo, and Engel (2020) and Li, Tregillus, and Engel (2022) found a sizable degree of adaptation to red filters occurred over the first hour of wear with little additional adaptation occurring between 2 and 5 hours of wear. This adaptation was observed as changes in the hue angle of unique yellow toward the unfiltered yellow settings. They also found that the degree of adaptation when the filters were first put on was higher on the second day and that this initial adaptation continued to increase over the course of 5 days of wearing the red filters for 5 hours per day. They concluded that participants had learned a new adaptation mode and developed the capacity to switch to it rapidly. 

In addition to shifting chromaticity (and hence color perception), colored filters also reduce the luminance of stimuli. The luminance of a stimulus is a measure of the visual effectiveness of its light and is determined by weighting the radiance of the light by the sensitivity of the human eye to different wavelengths (Lennie, Pokorny, & Smith, 1993). A standard spectral sensitivity function for the human eye, V(λ), was adopted by the CIE in 1924. V(λ) was derived primarily using heterochromatic flicker photometry (HFP), in which the relative intensity of a patch of light, quickly flickering between two colors, is adjusted until the flickering is minimized (Abney & Festing, 1886; Ives, 1912). The minimized flicker indicates the two colored patches are equal in luminance, which has been associated with minimized neural activity in the magnocellular visual pathway (Lee, Martin, & Valberg, 1988). While parvocellular neurons also respond to a flickering stimulus, they do not show the same null response at equiluminance as do magnocellular neurons (Lennie et al., 1993). 

It is assumed that spectral sensitivity as measured by HFP is additive, and when two lights are mixed, their luminances will add together, which is not the case for other measures of spectral sensitivity such as brightness matching (Abney & Festing (1886; Abney, 1913; Lennie et al., 1993; Wagner & Boynton, 1972). From this standpoint, one might hypothesize that a colored filter would simply subtract a given amount of luminance from the stimulus and the resulting HFP minimum flicker match would adjust for that. On the other hand, since chromatic adaptation occurs while wearing colored filters (Li et al., 2020; Li et al., 2022; Tregillus, Werner, & Webster, 2016), one might expect similar adaptation effects to occur for luminance and the HFP match would remain relatively unchanged by the filter. 

While HFP luminance additivity may hold if two flicker matches are done under the same illumination, several studies have shown that spectral sensitivity is dependent on the spectral composition of the background on which the flickering stimulus is superimposed, implying that chromatic adaptation has an effect on luminous efficiency (De Vries, 1948; Eisner & MacLeod, 1981; Marks & Bornstein, 1973; Stockman, Jägle, Pirzer, & Sharpe, 2008; Swanson, 1993). If a flickering stimulus is superimposed on a test field that strongly stimulates one cone type, the response of that cone type will be suppressed in the resulting spectral sensitivity curve for the flickering stimulus (Eisner, 1982; Eisner & MacLeod, 1981; Ikeda & Urakubo, 1968). Consider an HFP experiment where red and green lights are being exchanged. Under a neutral state of adaptation (with no filter), it is expected that V(λ) would accurately predict the red/green luminance ratio, which should be close to 1. A cyan filter (like the one in the current study) that blocks long wavelength (red) light will cause the visual system to need more intense long-wavelength (red) light to equate with the middle-wavelength (green) light. If something akin to chromatic adaptation were to affect the luminance mechanism, this should result in a shift over time in red/green luminance ratios from an elevated ratio back toward 1. 

In contrast, other studies have found little evidence of luminance adaptation in the presence of chromatic adaptation. Kuriki (2006) looked at changes in achromatic settings and HFP matches under various chromatic lights and found achromatic matches were dependent on chromaticity and intensity of illumination, but HFP matches were not. This is supported by studies that found different adaptation mechanisms for chromaticity and luminance (Ahn & Macleod, 1993; Kinnear, 1979; Laxar, Kass, & Wooten, 1984; Ware, 1983). In addition to short-term adaptation effects, several studies looked at long-term adaptation after wearing colored filters. For example, Neitz et al. (2002) found large differences in color perception after wearing colored filters for several days; however, they also measured a flicker electroretinogram on one participant after the extended filter wear and found no difference between baseline and adapted spectral sensitivity. Another study that assessed contrast adaptation effects on the steady-state visual evoked potential found adaptation effects for isoluminant chromatic stimuli but not achromatic luminance stimuli (Zhang, Valsecchi, Gegenfurtner, & Chen, 2023). Although these effects were on a much shorter time scale (150 seconds) than in the Neitz et al. (2002) study, both studies support the concept that luminance and chromatic adaptation mechanisms behave differently. 

One aim of the current study was to measure the time course of chromatic and luminance adaptation over a 1-hour period. We quantified chromatic adaptation and determined its time course by having participants make achromatic settings by adjusting the color of a patch until it appears devoid of color, or achromatic (Helson & Michels, 1948; Werner & Schefrin, 1993). We measured luminance adaptation by having participants make HFP matches. Both achromatic settings and HFP matches were conducted before putting on a colored filter and at different times over a 1-hour period after putting on the filter. The general hypothesis for chromatic adaptation is that achromatic settings while wearing the filter should be shifted over time toward the unfiltered achromatic setting and away from the chromaticity measured through the filter. Similarly, the general hypothesis for luminance adaptation is that the HFP flicker match should be shifted over time toward the unfiltered luminance match and away from the photometric filtered luminance. 

Another aim was to measure the extent that surrounding spatial context contributes to chromatic and luminance adaptation effects. We tested how spatial context modified adaptation by presenting test stimuli on either a black background or a background consisting of a colored airplane cockpit scene. We expected that if chromatic and luminance adaptation were observed, it would likewise be greater in the condition with richer surround context. Finally, we quantified chromatic and luminance aftereffects subsequent to participants removing the filter. Several studies looking at long-term adaptation to colored illumination reported color shifts from baseline after wearing a colored filter (Belmore & Shevell, 2008; Eisner & Enoch, 1982; Li et al., 2020; Neitz et al., 2002). 

Methods

Participants

Nineteen volunteers participated in the study (5 female, 14 male, average age 41, SD = 14.8). Participants were recruited through word of mouth among the Tri Service Research Laboratory community at Fort Sam Houston in San Antonio, Texas. All participants gave informed consent to participate in the study. The study was approved by the United States Air Force Research Lab Institutional Review Board protocol number FWR20210082E and is in adherence with the Declaration of Helsinki. All participants had normal color vision as assessed with the Ishihara 2010 concise edition (Ishihara, 1917) and first edition Standard Pseudoisochromatic (Ichikawa, Hukami, Tanabe, & Kawakami, 1979) plate tests. All participants either had 20/20 visual acuity or corrected 20/20 acuity. 

Stimuli/equipment/apparatus

Stimuli for the achromatic setting task and the HFP task were 2° discs presented on a calibrated display (EIZO ColorEdge CG319X, refresh rate 60 Hz) and generated by a computer running MATLAB with the Psychtoolbox-3 extension (Kleiner, Brainard, & Pelli, 2007). The discs were presented on a black background or within an image of an airplane cockpit (Figure 1A). The average chromaticity of the cockpit was (u′, v′) = [0.20,0.48] with an average luminance of 25.78 cd/m², while the luminance of the black background was less than 0.01 cd/m². The visual angle of the cockpit background was 28.85° wide by 23.2° tall. In the cockpit scene, there was a local black ring surrounding the stimulus disc, which was 2.2° wide. The display was calibrated to the Adobe RGB wide gamut color space with the maximum white having D65 chromaticity and set at 100 cd/m² luminance. The chromatic disc was located toward the bottom of the display, and all calibration measurements were performed on this region. 

Figure 1.

 

(A) Stimuli used in the current experiment. In one condition, the disc was presented inside an image of a cockpit (top). In another condition, the disc was presented on a black background (bottom). (B) Picture of cyan filter used in the study. (C) Spectral plot of the relative irradiance of the red, green, and blue display primaries (red, green, and blue lines) along with the transmittance spectrum of the colored filter (cyan line).

The colored filter that participants wore was a commercial eye protection goggle meant for blocking long-wavelength lasers (see Figure 1B). The transmittance spectrum for the filters were measured using a spectrophotometer (Cary 6000i). The spectrum of the R, G, and B outputs of the display were measured with a spectrometer (Ocean Insight Flame) and luminance with a colorimeter (Konica Minolta CS200). Figure 1C shows spectral plots of the relative irradiances of the red, green, and blue display primaries at the maximum level. The transmittance spectrum of the filter used in the study is also shown (cyan line, Figure 1C). This filter had a photopic luminance transmittance of 45.7% for illuminant D65. 

The cyan filter blocks most of the light present at wavelengths between 600 and 700 nm, largely attenuating the red primary while transmitting most of the energy in the green and blue primaries. The CIE u′v′ values for the red, green and blue primaries as well as the monitor white are plotted in Figure 2A. The filter shifts the chromaticity of the red primary more than the blue and green primaries. This can be seen by comparing the unfiltered gamut (black dashed triangle) to the filtered gamut (red dashed triangle). The unfiltered white point (black “W,” (u′, v′) = [0.20, 0.47]) falls just outside of the filtered gamut. The red “P” represents the closest filtered chromaticity to the unfiltered white point and represents a “no-adapt prediction” for the achromatic setting. The black “P” (u′, v′) = [0.44, 0.51] represents the unfiltered chromaticity needed to achieve this “no-adapt prediction” through the filter. In other words, if participants are not chromatically adapting, their achromatic setting with the filters on should be close to the black “P” in Figure 2A because the chromaticity of this stimulus, when measured through the filter, is close to the unfiltered D65 chromaticity. 

Figure 2.

 

(A) u′v′ diagram showing unfiltered and filtered monitor gamut (black and red dashed line triangles, respectively) as well as unfiltered and filtered monitor white points (black and red “Ws,” respectively). The black and red “P” letters represent a null hypothesis physical point that the filtered chromaticity (red “P”) is closest to the unfiltered white. (B) Gamma functions for the red, green, and blue primaries with no filter (lighter colors) and with filter (darker colors).

For the achromatic settings experiment, the luminance of the disc was set at 30.0 cd/m² without the filter and started at a random chromaticity in u′v′ space. The random unfiltered u′ value could be between 0.1 and 0.3 and the random unfiltered v′ value between 0.35 and 0.45. These unfiltered chromaticity starting point settings were used in both filtered and unfiltered conditions. 

Figure 2B plots gamma functions representing the luminance (y-axis in cd/m²) as a function of R, G, or B value (0–255, x-axis) for the red, green, and blue primaries. The darker-colored lines are the gamma functions with the filter and the lighter-colored lines without the filter. Similar to the chromaticity shifts, the luminance of the red is reduced the most by the filter. 

The HFP experiment used the red and green primaries as stimuli. For Condition 1, the red was fixed at its maximum 30.75 cd/m² (unfiltered) and the green was adjusted by the participant. For Condition 2, the green was fixed at 3.1 cd/m² (unfiltered) and the red was adjusted by the participant. These different luminance values for the fixed color in each condition were chosen to allow for individual differences and because the filter reduced red luminance more than the green. Pilot studies indicated that matches could be made with and without the filters for these luminance values. 

Procedure

After giving informed consent and passing the color and acuity screening, participants performed three practice sessions for each task (achromatic setting and HFP). Once they were comfortable with the tasks, the experiment began. The HFP and achromatic settings tasks were presented in a random order throughout the experiment. 

For the achromatic settings task, participants were instructed to use the up, down, left, and right arrow keys to adjust the chromaticity of a disc until it appeared achromatic, or gray. These keys made the disc more yellow, blue, green, and red, respectively. These key presses changed the chromaticity of the patch in u′v′ color space, with luminance held constant (with the filters on, the stimuli were not equiluminant). Participants were free to adjust the u′v′ coordinates anywhere within the monitor gamut, and the program beeped if they reached the edge of the gamut, indicating they could move no further in that direction. Each key press moved 0.005 units in u′v′ space. Once they were satisfied with their setting, participants pressed a button to move on to the next trial. 

For HFP conditions, participants viewed the disc, which alternated between red and green at a rate of 15 Hz. They were instructed to adjust the relative brightness of either the red or green phase of the stimulus by using the left and right arrow keys until the flickering appeared to be minimized. The left arrow increased luminance while the right arrow decreased luminance. In half of the HFP trials, the red phase had fixed luminance and the participant adjusted green luminance (Condition 1). The other half of HFP trials had the green phase at a fixed luminance and the participants adjusted red luminance (Condition 2). Without a filter, these two conditions had different mean luminances, which fell on different slopes of the gamma functions. This caused the luminance step for each key press to be approximately 0.5 cd/m² around the mean luminance in Condition 1 and 0.1 cd/m² around the mean luminance in Condition 2. Once they were satisfied with their setting, participants pressed a button to move on to the next trial. 

Five sets of trials were run in each session, beginning with a no-filter (baseline) set, followed by sets with the filter at time 0 minutes (t0), 30 minutes (t30), and 60 minutes (t60), and a set immediately after removing the filters (post). Two sessions, each on a separate day, were run with these trials. 

Each set of trials consisted of two achromatic setting conditions and four HFP setting conditions, with each condition having three repeated measures that were averaged together. The two achromatic setting conditions differed in their background: They were on the cockpit background or the black background (see Figure 1A). The four HFP conditions differed in the conjunction of the background (cockpit or black) and the flickering color (red or green) whose luminance was adjustable by the participant. 

Completing one set of trials took approximately 10 to 15 minutes depending on the pace of the participant. The order of the six conditions (i.e., two achromatic, four HFP) within a set was randomized to control for order effects. For the first seven participants, the order of the six conditions was randomized, and a new order was presented on the second day. A slight change was made for the latter 12 participants in which the six conditions were presented in a randomized Latin square design, and the same order for each participant was used over the 2 days of testing. 

All psychophysical measurements were completed in a dark room. Participants sat approximately 83 cm away from the computer monitor. After completing the baseline conditions, participants immediately donned the filters and completed the t0 trials. After the t0 trials were completed, a break was taken until it had been 30 minutes since the participant donned the filters. During this break, participants kept the colored filters on and were free to return to their desk briefly or remain in the experiment room with the lights on. After the break, the lights were turned off and participants readjusted to the dark room for 2 to 3 minutes and then completed the t30 trials, after which another break was taken until participants had been wearing the filters for 60 minutes. The same procedure was followed and the t60 trials were completed. Immediately after the t60 condition, participants removed the filters and completed the final (post) set. Once the post test was completed, the session was ended for the day, and the second session took place at least 1 day later. Seventeen of the 19 participants completed the second session 1 day after the first, while two participants were tested approximately 1 week later. 

Results

The results are presented in four sections: achromatic settings; HFP settings; constancy indices, which compare chromatic and luminance adaptation; and cone excitations. The achromatic settings and HFP sections have separate subsections that statistically compare baseline versus post conditions to each other. All statistical tests were performed using JASP software (JASP Team, 2023). All data points for each participant were the average of the three repeated trials for each condition. All post hoc tests used the Bonferroni–Holm method for correcting multiple comparisons. 

Achromatic settings: General results

The achromatic settings data were first analyzed in the u′v′ color space coordinates. Figure 3 plots the results averaged across participants over the 2 repeated days because, as reported below, there were no significant differences between days 1 and 2. Blue symbols represent matches on the black background and orange symbols matches on the cockpit background. It can be seen in Figure 3 that filter conditions (rings (t0), triangles (t30), and diamonds (t60)) fell between the baseline settings (+) and the setting predicted for no adaptation (P). This indicates that achromatic settings while wearing the filters fell more in the yellowish/reddish direction from the baseline, but not to the extent that would be predicted if no adaptation occurred. The baseline and post settings fell close to the monitor white (W), although slightly more in the bluish/greenish directions, particularly when they were on the black background. The results illustrate the effects of background and time on chromatic adaptation (see below), with the cockpit background (orange symbols) inducing greater adaptation than the black background (blue symbols) and with a time-dependent adaptation over the first 30 minutes of filter wear that then does not seem to increase much beyond the 30-minute time point. An adaptation aftereffect is also noticeable in the comparison between the baseline and post trials, with the post trials being shifted in the opposite direction to the filtered points with respect to their baselines. 

Figure 3.

 

Achromatic settings averaged over the 2 test days. The x-axis is the u′ direction (reddish/greenish) and the y-axis is the v′ dimension (bluish/yellowish). The orange symbols represent the settings when the disc was in the cockpit surround and the blue symbols when the disc was on the black surround. The “+” and “x” symbols are the baseline and post settings, respectively. The rings, triangles, and diamonds represent the matches with the filter on at t0, t30, and t60, respectively. The large black triangle represents the monitor gamut. The black “P” represents the prediction for the achromatic setting if no adaptation occurred, and the black “W” represents the unfiltered D65 chromaticity.

In order to quantify the effects of adaptation in the filter conditions, the Euclidean distances between the baseline and the t0, t30, and t60 filter conditions were calculated for the black and cockpit background conditions. Figure 4 shows the average Euclidean distance (du′v′) from baseline for the three time points while wearing the filter, for both background conditions (blue bars—black background, orange bars—cockpit background). A three-way, repeated-measures analysis of variance (ANOVA) with time (t0, t30, and t60), background (black vs. cockpit), and day (day 1 vs. day 2) as factors yielded a significant main effect of background (F(1, 18) = 23.98, p < 0.001, η_p² = 0.57) with larger distances from baseline on the black than the cockpit background and a main effect of time (F(2, 36) = 4.17, p = 0.023, η_p² = 0.19). Post hoc paired t tests revealed a significant difference between the t0 and t60 condition (t = 2.82, p = 0.023, d = 0.27) but not between t0 and t30 or t30 and t60 (p > 0.1). There was no significant main effect of day (F(1, 18) = 0.20, p = 0.66) or any interactions among the three factors (p > 0.24 for all interactions). 

Figure 4.

 

Euclidean distance from the filter conditions to the baseline for the achromatic settings experiment averaged over the 2 repeated days. The x-axis is the different time points and the y-axis is the u′v′ distance (du′v′). The orange bars represent the settings when the disc was in the cockpit and the blue bars when the disc was on the black surround. Dark dots represent individual participant responses. (Error bars ±1 SEM.)

Achromatic settings: Baseline and post

The post trials of the achromatic settings were shifted slightly more bluish/green relative to the baseline settings, approximately the opposite direction in which the achromatic settings were made while wearing the filter. This indicates an aftereffect from wearing the colored filter similar to the aftereffects reported in prior studies (Eisner & Enoch, 1982; Li et al., 2020; Li et al., 2022; Neitz et al., 2002) in which residual adaptation induced achromatic settings in the opposite direction than the filter did. To quantify this aftereffect, vectors between baseline and post achromatic settings were calculated and the dot product was calculated with the vector from the D65 point (W in Figure 3) and the no adaptation prediction point (P in Figure 3). This dot product between the two vectors represents the magnitude of the aftereffect as a distance traveled along the W to P vector orientation. A two-way ANOVA with background and day revealed no significant differences or interaction between these factors (p > 0.46 for all tests). Figure 5A plots these data averaged across the 2 days. One-sample t tests against zero found a significant difference on the black background (t(18) = 3.12, p < 0.01, d = 0.72) but not quite significantly different from zero for the cockpit background (t(18) = 1.71, p = 0.053). 

Figure 5.

 

(A) Aftereffect magnitude after removing the filters. The x-axis is different backgrounds' y-axis is the dot product between baseline/post and W/P vectors. (B) Difference between v′ (yellowish/bluish) component of achromatic settings. The y-axis represents v′ value and x-axis background. For both graphs, blue bars represent black and orange bars cockpit background and dark dots represent individual participant responses. (Error bars ±1 SEM.)

Another observation of the baseline achromatic settings was that the matches on the black background were more blue than the cockpit settings. To test this, a paired sample t test comparing the average v′ component for the different backgrounds revealed a significantly lower v′ component for the black compared to cockpit background (t(18) = 6.11, p < 0.001, d = 1.38). These data are plotted in Figure 5B. 

HFP: General results

Figure 6 shows the results of the HFP Condition 1 (Figure 6A), in which the red luminance was fixed and green luminance adjusted, and Condition 2 (Figure 6B), where green fixed and red were adjusted. The baseline and post conditions are the leftmost and rightmost points of each plot, respectively. The red dashed line in Figure 6A (green dashed line in Figure 6B) represents the unfiltered V(λ) prediction for the standard observer. This also serves as a prediction for the filtered t0, t30, and t60 conditions if the visual system exhibited total luminance adaptation. 

Figure 6.

 

Results for HFP Conditions 1 (A) and 2 (B) in which either red primary or green primary was fixed, respectively. The y-axis is the luminance in cd/m² of the matches and the x-axis the different time conditions in chronological order. The orange bars represent the cockpit and the blue bars the black background. Dark dots represent individual participant data. The red dashed line in A (green and cyan for B and E, respectively) represents the V(λ) prediction for baseline settings and total adaptation prediction for filtered settings. The black and gray dashed lines in A, B, and E represent the no-adaptation predictions based on V(λ) and the baseline HFP data, respectively. (C) Comparison of the R/G luminance ratios for Conditions 1 and 2. (D) Correlation between baseline R/G luminance ratios between Condition 1 (y-axis) and Condition 2 (x-axis). (E) Log R/G luminance ratios of HFP data averaged across Conditions 1 and 2. (Error bars ±1 SEM.)

The gray dashed lines in Figures 6A and 6B represent the filtered V(λ) predictions for no adaptation. Since there were some deviations in the baseline HFP measures from V(λ), a second “no adaptation” prediction was derived for each condition from the observers’ average baseline HFP settings and is represented by the black dashed lines in Figures 6A and 6B. 

Baseline R/G luminance ratios were calculated by dividing the red luminance by the green luminances. Figure 6C plots the average R/G luminance ratios for Conditions 1 and 2. A paired sample t test revealed that the R/G luminance ratios in Condition 1 were significantly lower than Condition 2 (t(18) = 2.46, p = 0.024, d = 0.56). This indicates that in the high-luminance baseline condition, more green luminance was required to match the fixed red field. However, R/G luminance ratios for Conditions 1 and 2 were not significantly different from a V(λ) predicted ratio of 1 (p > 0.08 for both one-sample t tests, two-tailed). 

Figure 6D plots the correlation between Conditions 1 and 2 R/G luminance ratios for the baseline HFP values averaged across background and day. A Pearson correlation test showed a significant correlation in R/G luminance ratios between the two conditions (r(17) = 0.84, p < 0.001). Given the strong correlation, the R/G luminance ratios for the two conditions were averaged together, and the log of these data is plotted in Figure 6E. 

Figure 7 replots the data in Figure 6E, separating the filtered data (Figure 7A, left) from the baseline and post data (Figure 7B, right). In order to test for luminance adaptation effects across the 1-hour period while wearing the filter, a three-way, repeated-measures ANOVA with time (t0, t30, and t60), background (black vs. cockpit), and day (day 1 vs. day 2) as factors yielded no significant interactions (p > 0.056 for all interactions) or any significant main effects for time (F(2, 36) = 1.22, p = 0.31), background (F(1, 18) = 0.024, p = 0.88), or day (F(1, 18) = 0.36, p = 0.56). 

Figure 7.

 

Results of the HFP log R/G luminance ratios for the filtered conditions (A) and baseline vs. post conditions (B), plotted separately and zoomed in from Figure 6E: Blue bars represent results on the black and orange bars on the cockpit background. The x-axes represent the different time points. The y-axes represent the log of the R/G luminance ratios observed for the HFP settings. (Error bars ±1 SEM.)

HFP: Baseline versus post settings

Baseline and post HFP settings were also compared to see if there were any luminance aftereffects. The log of the R/G luminance ratios for baseline and post was used in this analysis (Figure 7B). A three-way, repeated-measures ANOVA that had factors of time (baseline vs. post), background, and day yielded a significant interaction between time and background (F(1, 18) = 4.75, p = 0.04, η_p² = 0.21); however, post hoc paired comparisons were not significant (p > 0.12 for all tests). There were no significant main effects (p > 0.21 for all tests). 

Constancy indices

In order to compare chromatic and luminance adaptation effects to each other, both data sets were converted into constancy indices (Arend, Reeves, Schirillo, & Goldstein, 1991; Foster, 2011). The constancy index (CI) captures constancy by seeing where on the scale a data point falls, with one extreme being no adaptation and the other extreme total adaptation. If complete adaptation occurs, the constancy index will be 1. If no, or little, adaptation occurs, the constancy index will be close to zero. The current study quantified the constancy index using the following equation:  

\begin{eqnarray*} CI = 1 - \frac{{\rm{baseline}} - {\rm{test}}}{{\rm{baseline}} - {\rm{no}}\, {\rm{adapt}}} \end{eqnarray*}

where “baseline” refers to the preliminary measure, “test” refers to a given filter measure (t0, t30, or t60), and “no adapt” refers to the filtered prediction if no adaptation occurred. For achromatic settings, the “no adapt” prediction chromaticity used was (u′, v′) = [0.44,0.51], which corresponds to the black “P” drawn on Figure 3. For HFP, “no adapt” refers to the filtered luminance predictions derived from each individual participant’s baseline setting (data used to calculate the gray lines in Figures 6A and 6B). For the HFP data, the constancy indices were calculated separately for the two conditions and averaged together. 

The CI will be close to zero if the fraction term is close to 1, which would happen if the test value is close to the “no adapt” prediction. A negative constancy index (which occurred sometimes for luminance) implies that participant matches with the filter were more extremely shifted than predictions derived from their baseline measures. The CI will be 1 if there is no difference between the baseline and test. 

The results of the constancy index analysis are plotted in Figure 8. The left graph in Figure 8 shows the black and right graph the cockpit background, and orange bars represent chromatic and blue bars luminance. A three-way, repeated-measures ANOVA was done with adaptation type (luminance or chromatic), background, and time as factors. There was a significant interaction between adaptation type and background (F(1, 18) = 18.9, p < 0.001, η_p² = 0.51), driven by the constancy index being greater for cockpit than black background for chromatic adaptation but not luminance. There was also a significant interaction between adaptation type and time (F(2, 36) = 4.02, p = 0.027, η_p² = 0.18), driven by constancy index increasing with time for chromatic adaptation but not luminance. While time and background significantly influenced chromatic adaptation, they did not significantly affect luminance adaptation, which hovered around zero (one-sample t tests for luminance were not significantly different from zero; p > 0.18 for all tests). Results also showed a significant main effect of adaptation type with chromatic having greater adaptation than luminance (F(1, 18) = 245.57, p < 0.001, η_p² = 0.91), as well as a significant effect of background (F(1, 18) = 17.58, p < 0.001, η_p² = 0.49), although this effect was driven by the chromatic adaptation data. The main effect of time did not reach significance (F(2, 36) = 2.92, p = 0.07). 

Figure 8.

 

Constancy indices comparing magnitude of adaptation effects for chromaticity and luminance. The left graph represents the black and right graph cockpit background. Orange bars represent chromatic adaptation and blue bars luminance adaptation. Dark dots represent individual participants. (Error bars ±1 SEM.)

Cone excitations

An additional analysis was performed in order to investigate adaptation effects in terms of cone excitations based on the light reaching the eye. LMS cone excitations were calculated using the Smith–Pokorny cone fundamentals (Smith & Pokorny, 1975). For the t0, t30, and t60 conditions, the filtered chromaticity was transformed, and for the baseline and post conditions, the unfiltered chromaticities were transformed. The analysis followed that of Kuriki (2006). For the achromatic settings data, the XYZ tristimulus values were normalized so that Y = 1 for all conditions in order to calculate excitations based on chromaticity. This was done to ignore differences in cone excitations caused by changes in luminance due to the filter. Figures 9A, 9B, and 9C shows cone excitations for the achromatic settings data for the the L, M, and S cones, respectively. The L cone excitations (Figure 9A) show that the filtered conditions are lowered with respect to the baseline, while the M and S excitations (Figures 9B and 9C), show the opposite effect. This is to be expected, given the filter predominantly blocks long wavelengths and the adaptation effect of the visual system compensates for the loss of long wavelength light. 

Figure 9.

 

Plot of L (A), M (B), and S (C) cone excitations for achromatic settings data. The x-axes are time and y-axes cone excitations. Blue bars represent black background and orange bars represent cockpit background. Dark dots represent individual participant data. (Error bars ±1 SEM.)

To perform statistical analysis, relative L cone weights were derived by dividing M by L cone excitations and relative S cone weights by dividing M by S cone excitations and taking the log of these ratios (Kuriki, 2006). The purpose of this analysis was to determine whether the relative cone weights differ between baseline and filter conditions, which would be indicative of adaptation effects. 

Figure 10 shows the log M/L (Figure 10A) and M/S (Figure 10B) excitation ratios. The filtered conditions for the M/L data fall below the baseline and post conditions and get lower as adaptation increases such that the t60 cockpit condition has the lowest M/L ratio. In the CI analysis, this was the most adapted condition. A repeated-measures ANOVA with time and background as factors for the M/L data showed a significant interaction between time and background (F(1.98, 36.68) = 12.96, p < 0.001, η_p² = 0.42). Post hoc tests showed the cockpit background eliciting greater M/L ratios in the post condition (t(18) = 3.6, p = 0.002), but not baseline (t(18) = 1.92, p = 0.07), while the black background caused greater M/L cone ratios in the t0, t30, and t60 conditions (t(18) > 3.1, p < 0.007 for all tests). There was also a significant main effect of time (F(2.10, 37.798) = 420.69, p < 0.001, η_p² = 0.96), but no significant main effect of background (F(1, 18) = 0.072, p > 0.79). Violations of sphericity were corrected using the Greenhouse–Geiser correction. Post hoc tests comparing the filtered conditions to the baseline showed that the M/L ratio for the baseline was significantly larger than t0 (t(18) = 27.01, p < 0.001, d = 7.12), t30 (t(18) = 28.23, p < 0.001, d = 7.42), and t60 (t(18) = 28.37, p < 0.001, d = 7.46). 

Figure 10.

 

Plot of M/L (A) and M/S (B) log cone excitation ratios. The x-axes are time and y-axes log excitation ratios. Blue bars represent the black and orange bars represent cockpit backgrounds. Dark dots represent individual participant data. (Error bars ±1 SEM.)

Another repeated-measures ANOVA was done for the M/S excitation ratios (Figure 10B). There was a significant main effect of time (F(4, 72) = 80.31, p < 0.001, η_p² = 0.82). Post hoc tests showed that baseline had significantly larger log (M/S) ratios than t0 (t(18) = 10.47, p < 0.001, d = 1.81), t30 (t(18) = 12.08, p < 0.001, d = 2.09), and t60 (t(18) = 11.34, p < 0.001, d = 1.96) conditions. There was also a significant main effect of background with M/S ratios for the cockpit being higher than the black background (F(1, 18) = 39.68, p < 0.001, η_p² = 0.69). The interaction between time and background was not significant (p > 0.05). 

For the HFP data, our analysis focused on the L and M cone excitations since the S cones are thought to play a small, if any, role in luminance (Eisner & MacLeod, 1980). The L and M cone excitations for the red and green discs were used in the following equation:  

\begin{equation*} {\omega } = \frac{L_{red} - L_{green}}{M_{green} - M_{red}} \end{equation*}

where L_red and L_green are the L cone excitations for the red and green stimuli, respectively, and M_red and M_green are the corresponding M cone excitations. Figure 11 plots the ω ratios for the HFP experiment, averaged across the 2 days and Conditions 1 and 2. A repeated-measures ANOVA with time and background as factors showed no significant main effects or interactions between these factors (p > 0.45 for all tests). 

Figure 11.

 

Plot of LM cone excitation ratios for HFP experiment. The x-axis is time and y-axis excitation ratio ω (omega). Blue bars represent black and orange bars the cockpit backgrounds. Dark dots represent individual participant data. (Error bars ±1 SEM.)

Discussion

It is generally held that the visual system recalibrates itself in response to changes in illumination. The current study aimed to measure adaptive shifts in chromaticity and luminance while wearing colored filters. Chromatic adaptation was measured by comparing baseline achromatic settings to achromatic settings made while wearing the filters. Similarly, luminance adaptation was measured by comparing baseline HFP settings to those made while wearing the filters. The results of the current study imply that while wearing colored filters and viewing stimuli on a computer display, the chromatic system recalibrates in response to changes caused by the filter while the luminance system does not. 

Baseline and post achromatic settings

We hypothesized that baseline achromatic settings would be close to the standard D65 white point, which in general held true. However, baseline achromatic settings were shifted slightly more bluish than D65, particularly baseline settings made on the black background, an effect that has been shown in other studies (Chauhan et al., 2014; Hansen et al., 2007). This may not be surprising given that desaturated bluish chromaticities tend to be seen as gray, or achromatic. This is noted in the observation that shadows, which are often composed of bluish chromaticities of reflected sky light, often appear to be achromatic (Churma, 1994; Winkler, Spillmann, Werner, & Webster, 2015). We also found a significant aftereffect on the black background after the filter was removed. This finding is consistent with other studies showing color aftereffects after chromatic adaptation to extreme environments (Belmore & Shevell, 2008; Li et al., 2020; Li et al., 2022; Eisner & Enoch, 1982; Neitz et al., 2002). 

Baseline and post HFP settings

For the baseline HFP conditions, we hypothesized that the luminance of the varying test field would be close to that of the reference test field as predicted by V(λ). This in general held true; however for the high-luminance “Condition 1,” slightly more green luminance was needed to match the fixed red luminance compared to “Condition 2” (see Figure 6C). Previous studies have found deviations in R/G luminance ratios at higher mean luminances (De Vries, 1948; Dresler, 1953; Ingling Jr et al., 1978; Ives, 1912), and an early study found this effect to be in the same direction as the current study (Ives, 1912; as discussed in Dresler, 1953; De Vries, 1948). Later studies, however, found this effect to be in the opposite direction, with more red being required to match a green of higher luminance (De Vries, 1948; Ingling Jr et al., 1978). While the reason for this discrepancy is unclear, we did find a significant deviation at higher versus lower mean luminance, with lower mean luminance being closer to V(λ). However, neither of the R/G luminance ratios were significantly different from the V(λ) prediction. 

Contrary to the baseline achromatic settings, we found no effect of background type on the baseline HFP matches and no significant aftereffect in post conditions. 

Adaptation effects while wearing the filter

Our general hypotheses were that chromatic and luminance adaptation would occur while wearing the filter and that the adaptation would be greater with a background containing complex spatial and color context compared to a sparse, uniform black background. While strong chromatic adaptation was evident in the data, luminance adaptation was not. Furthermore, increasing the complexity of the surrounding spatial context increased chromatic adaptation but had no effect on luminance adaptation. 

Strong chromatic adaptation was evident in that achromatic settings with the filter were closer to the baseline settings than would be predicted if no adaptation occurred. This was reflected in moderately high chromatic constancy indices and is consistent with the concept of imperfect color constancy (Kraft & Brainard, 1999), in which significant but incomplete adaptation occurs. Furthermore, there was a strong spatial context effect on chromatic adaptation: The cockpit background saw greater adaptation shifts than the black background, indicating that a richer context facilitates the constancy/adaptation mechanism as reported in previous studies (Hansen et al., 2007; Kraft & Brainard, 1999; Blackwell & Buchsbaum, 1988; Cornelissen & Brenner, 1995). Additionally, there was a significant effect of time on achromatic matches over the 1-hour period. This is line with other recently published reports showing increasing levels of chromatic adaptation over a 1-hour time period (Li et al., 2020; Li et al., 2022). It is worth noting that even on the black background at t0, there was on average a significant amount of chromatic adaptation. First, chromatic adaptation can occur quite rapidly (Rinner & Gegenfurtner, 2000). Second, our results suggest that even with the sparse context of the black background, there may have been enough cues for the visual system to discount the changes in illumination caused by the filter. 

The current study did not find any significant effect across the 2 days, where previous studies did show increasing adaptation over periods of days (Li et al., 2020; Li et al., 2022). However, these studies had participants wear colored filters over 5-hour periods for 5 days, while the current study only had them wear the filters for 1 hour on 2 consecutive days. 

The lack of luminance adaptation effects observed in the current study may be surprising because many previous studies have shown that chromatic adapting fields influence the spectral sensitivity of the luminance mechanism (Stockman et al., 2008; De Vries, 1948; Eisner, 1982; Eisner & MacLeod, 1981; Ikeda & Urakubo, 1968; Ives, 1912; Swanson, Pokorny, & Smith, 1987; Swanson, 1993; Marks & Bornstein, 1973; Stromeyer, Cole, & Kronauer, 1987), and our cockpit background viewed through the filter produced what could be considered a chromatic adapting field. Evidence in favor of luminance adaptation effects usually used an intense colored background, which desensitizes one cone type more than another. This causes the weighting of the cone types to be renormalized such that a green background will desensitize the M cones relative to the L cones. One study done at lower luminance levels found more modest effects of chromatic adaptation on spectral sensitivity (Hurvich & Jameson, 1954), indicating that the luminance adaptation may be dependent on using an intense surround. This argument favors the idea that luminance adaptation found in previous studies may be happening at the level of the receptors themselves (Eisner & MacLeod, 1981; Swanson, Pokorny, & Smith, 1988). In contrast, other studies have found strong chromatic adaptation but no strong evidence of luminance adaptation due to chromatic surrounds (Laxar et al., 1984; Kuriki, 2006; Kinnear, 1979; Ware, 1983), with one speculating that the lack of an effect of chromatic adaptation on HFP may be due to the pedestal used (Kuriki, 2006). 

Cone excitations

The analysis of cone excitations revealed that achromatic percepts with the filter could be achieved despite significantly differing cone excitations compared to the baseline conditions. Since the chromaticity of the “no adapt” prediction is equal to the unfiltered D65 chromaticity, if participants were not adapting the cone excitations while wearing the filter, this should equal those of the baseline conditions after taking the luminance differences into account. The fact that participants could achieve the same achromatic percept through the filter without having to change the chromaticity to achieve equal cone excitations as without the filter is indicative of adaptation. However, for the HFP task, approximately equal cone excitation ratios were needed to achieve minimum flicker with and without the filter, indicating that observers needed to fully compensate for the filter in the HFP conditions while not needing to fully compensate with the achromatic settings. Taking a look at the M/S cone ratio for the achromatic settings, the S cones seem to play a large role in the effect seen between the different backgrounds. This can be seen in Figure 10B, which shows the significant effect of background on the M/S ratio, while there was no significant effect of background in the M/L ratio (Figure 10A). 

L/M cone ratio influence on color and luminance

A handful of studies have looked at how L/M cone ratios affect various aspects of vision with the hypothesis that if you have higher L/M cone ratios, you will be more sensitive to long-wavelength light and lower L/M cone ratios more sensitive to middle-wavelength light. Studies using genetics and electroretinogram recordings have found that among the population of humans with normal color vision, L/M cone ratio varies over a range of 0.4 to 13 (Carroll, Neitz, & Neitz, 2002). These individual differences in L/M cone ratios have been correlated with HFP flicker matches, supporting the hypothesis that higher L/M cone ratios are associated with higher long-wavelength flicker sensitivity (Rushton & Baker, 1964; Pokorny & Smith, 2020; Neitz et al., 2002). 

Concerning the effects of L/M cone ratios on color vision, the corresponding hypothesis is that the unique yellow point (a yellow that appears neither reddish nor greenish) should be shifted in the green direction for people with a higher L/M cone ratio since these observers are more sensitive to long-wavelength (red) light. In contrast to the L/M ratio × HFP correlation, correlations between unique yellow settings and L/M cone ratio have not been found (Brainard et al., 2000; Pokorny, Smith, & Wesner, 1991; Pokorny & Smith, 2020; Neitz et al., 2002). Some speculative interpretations of these findings is that chromatic adaptation is influenced by neural mechanisms that involve integration of the L, M, and S cones and allow for adaptive shifts, while luminance adaptation may be more limited by earlier receptor responses and the light reaching the eye. 

Luminance versus brightness

One final point to mention is that although we did not find evidence of luminance adaptation, this does not address whether participants experienced lightness or brightness adaptation. Two lights that are equiluminant may still differ in brightness. Many studies have shown that spectral sensitivity curves measured with brightness matching differ from those measured with flicker photometry with stronger deviations from additivity being found with brightness matching (De Vries, 1948; Kaiser, 1984; Ives, 1912; Wagner & Boynton, 1972). It is also well known that the visual system exhibits lightness constancy (Gilchrist et al., 1999; Kingdom, 2011; Murray et al., 2014), and lightness and brightness adaptation may be derived from separate neural mechanisms than luminance. 

Conclusions

In summary, we measured chromatic adaptation to a cyan filter over time and found significant adaptation occurred over 60 minutes of wear. We also found that a background with spatial context increased chromatic adaptation compared to a homogeneous black background. In contrast, we found no evidence of luminance adaptation to the cyan filter, even though it significantly reduced the intensity and altered the spectral properties of the HFP stimuli used to assess it. 

Acknowledgments

Commercial relationships: none. 

Corresponding author: Andrew J. Coia. 

Email: [e-mail]. 

Address: Science Applications International Corporation, JBSA Fort Sam Houston, 4141 Petroleum Drive, Bld. 3260, TX 78218, USA.