Assessment of VINO filters for correcting red- green Color Vision Deficiency

In our ongoing research on the effectiveness of different passive tools for aiding Color Vision Deficiency (CVD) subjects, we have analyzed the VINO 02 Amp Oxy-Iso glasses using two strategies: 1) 52 observers were studied using four color tests (recognition, arrangement, discrimination, and color-naming); 2) the spectral transmittance of the lenses were used to model the color appearance of natural scenes for different simulated CVD subjects. We have also compared VINO and EnChroma glasses. The spectral transmission of the VINO glasses significantly changed color appearance. This change would allow some CVD subjects, above all the deutan ones, to be able to pass recognition tests but not the arrangement tests. To sum up, our results support the hypothesis that glasses with filters are unable to effectively resolve the problems related to color vision deficiency. © 2019 Optical Society of America under the terms of the OSA Open Access Publishing Agreement

these glasses the dichromats improved their red-green discrimination but for the anomalous trichromats did not. Patterson found a rotation of the confusion axis from protan to deutan for the two protan observers. Patterson concluded that this improvement is achieved by the change in contrast provided by the lenses rather than the change in chromatic sensitivity.
In this study we focus on the "Color Blind Glasses" VINO 02 Amp Oxy-Iso by the VINO company, but without focusing on the enhancement of the O 2 signal from hemoglobin under the skin application. Two strategies are used for this purpose. The first is to perform an objective study (Section 3) with a larger set of observers (52 CVD people) using three classical color tests: recognition (Ishihara), arrangement (Fansworth-Munsell) and discrimination (Oculus anomaloscope). In addition, a subjective color-naming test based on X-Rite Color Chart has been added. The second approach is to evaluate the color performance of the filters by modeling the appearance of natural scenes as seen through the filter by different simulated subjects (Section 4). We also compared our results for the VINO glasses with the results obtained for the EnChroma glasses.
Finally, we sum up the conclusions of this research, analyzing the relative efficiency of the most common passive aids that aim to correct CVD after the comparative analysis of the EnChroma and VINO glasses. Figure 1 shows the spectral transmittance of the O2 Amp Oxy-Iso VINO glasses (purchased in December 2017), from 380 to 780 nm measured by a Thorlabs CCS200 spectrometer with a UV/VIS/NIR BDS130 Edmund Optics light source. In addition, VINO provided us with the spectral transmittance, also shown in Fig. 1. The differences found between these two curves are due to the effect of the curvature and varying thickness of the lenses on the transmittance measurements. In our simulations (section 4) we have used the transmittance provided by the VINO company.

Materials
As Fig. 1 shows this transmittance is close to zero from 530 nm to 580 nm, which results in a purplish (or pinkish) color of these glasses contrary to the neutral-bluish color of the EnChroma Cx-65 glasses (see Fig. 3). The low-transmittance region in the VINO glasses, radically affects the subject's perceived luminance or object lightness, since it overlaps with the peak of the human luminous efficiency curve. At the same time, it makes predominantly green-yellowish colors appear very dark, which would not in principle benefit protanopic subjects. Fig. 1. Spectral transmittance of O2 Amp Oxy-Iso VINO glasses (measured and provided by manufacturer) and LMS cone spectral responsivity of CIE31 2° Standard Observer (relative units) [19].

CVD observers
52 volunteers, 2 women and 50 men with ages between 10 and 64 years and who were aware of their CVD condition, participated in this research. 19 of these observers also participated in the study of the EnChroma glasses [15]. All the participants signed an informed consent for all the research procedures adhering to the tenets of the Declaration of Helsinki.

Subjective experiment
The subjective experiment consisted of two sessions. The first session was the evaluation of the color deficiency of the volunteers with three different tests: the Ishihara (recognition test) [20], Fansworth-Munsel 100 Hue (FM-100, arrangement test, X-Rite USA) [21], and anomaloscope (discrimination test, OCULUS HMC-Anomaloskop USA). The Ishihara and FM-100 tests were performed in a Verivide CAC 60 lighting booth equipped with a D65 simulator light source located in a dark room.
In addition to these routine tests, we designed a color-naming test with 21 colors from an X-Rite GretagMacbeth Chart. These colors were displayed on a 55x55 mm square patch isolated with a black background on a HP 2510i 25" monitor, calibrated with X-Rite Eye One device. The participants were asked to name the color by using the 11 color names proposed by Berlin and Kay (black, white, grey, red, green, blue, yellow, orange, brown, purple and pink) [22]. The answers were compared with the responses from a control group of normal color vision subjects. With a reduced number of subjects (7 CVD observers and 2 with normal color vision) we repeated the color naming test both in the RGB monitor and in the lighting booth with the real chart under a D65 simulator. The average number of changes is kept in both monitor and light booth, with a difference below 1 change. The same trend is found when counting the number of errors with and without glasses. This shows that, even if spectrally the light signals shown to the observers are different comparing the displayed samples in the monitor and the real samples in the lighting booth, the number of changes and errors remain the same doing the experiment in any of the two conditions.
During the second session, performed at least 2 weeks after the first one to minimize the effect of memory, the observers repeated the Ishihara, FM-100 and color naming tests, after wearing the VINO glasses for 30 minutes (adaptation time). We have to mention that the anomaloscope was not used in this second session because the combination of the filters of the glasses and the monochromatic lights of this device resulted in incongruous results. Table 1 shows the results of the first (VINO glasses off) and second (VINO glasses on) sessions. The classification of the observers has been made based on the result of the anomaloscope, which is the best test, amongst all the others, to classify CVD [23,24]. This method classifies the observers into four categories (first column of Table 1): 23 protanopes, 19 deuteranopes, 4 protanomalous, and 6 deuteranomalous in our study.

Subjective evaluation results
The third column shows the number of fails (the number of plates that the observer failed to identify correctly) in the Ishihara test, which is a widely used test for screening but is not very reliable for diagnosis [25].
Columns four to seven show the four most relevant parameters for quantifying the severity of a CVD provided by the FM100, although this test does not give a reliable diagnosis of the type of CVD [26]. The first parameter is the SQR which is the square root of the total number of errors in the test. The other three parameters are obtained after analyzing the color coordinates of the chips in CIELUV color space: the confusion index (CI) describes the severity of a color loss, the scatter index (SI) quantifies the randomness of cap arrangement and Angle identifies the orientation of the cap arrangement. As a reference, a perfect arrangement has a CI = 1 and SI = 1.28.
The eighth column shows the results for the color-naming test as the number of fails, which are the number of patches for which the observer gave a different name than that given by the normal-observers panel. Comparing the results wearing or not the VINO glasses, Table 1 shows that the number of fails in the Ishihara test has decreased (p = 1,3974x10 −35 ); while the results of the FM100 and color naming tests have worsened. This is explained by the SQR, CI and SI parameters. They indicate a more severe degree of CVD in the second session (p sqr = 1,5799x10 −9 , p CI = 1,1183x10 −13 , p SI = 7,5x10 −5 ). This result can be interpreted considering that the VINO glasses are able to produce a darkening of the greenish colors and an increase in chroma for the reddish colors at the same time. This transformation breaks the working principle of the pseudoisochromatic plates, revealing the hidden figures and the traces that are meant to be confused by the red-green CVD subjects. Regarding the FM100 results, the CI and SI values indicate that the observers find it more difficult to carry out the ordering of the samples with the VINO glasses on. Regarding the Angle parameter, there is a counterclockwise rotation for deutan observers and a clock-wise rotation (decreasing angle values) for the protan observers. This basically means that both groups are more clearly separated with the VINO glasses (p<0,001 in all cases).
These results suggest that VINO glasses are able to improve only the results of the Ishihara test. This is mainly due to their transmittance spectra decreasing the lightness of the green colors and not because the observers' color vision is more similar to a normal subject when he or she wears the VINO glasses. This fact is supported by the results of the color naming and color arrangement tests.

Databases, CVD model, and CIECAM02 computations
As we intend to simulate as much as possible the color vision of normal and CVD observers, under two conditions: without and with VINO glasses, the analysis of the simulations was carried out using CIECAM02 color attribute values computed for two data sets (D1 and D2). Data set D1 is composed of 124 reflectances from three different sets of samples: X-Rite Color Checker chart, FM-100 test, and the CRI set for computing Color Rendering Indices. Data set D2 is extracted from 5 natural scenes from a hyperspectral database of urban scenes [27], and comprises 565500 reflectances. A complete description of these two data sets can be found in [15].
The lightness (J), chroma (C) and hue angle (h, with values between 0 and 360) attributes were computed using the CIECAM02 model, the latest color appearance model recommended by the International Commission on Illumination (CIE), for the seven different vision conditions (normal, protanopy, deuteranopy, mild and medium protanomaly and deuteranomaly), with and without the VINO glasses.
A color appearance model simulates the color vision, considering chromatic adaptation and many other visual effects, better than any basic colorimetry system. The set of parameters considered in CIECAM02, depending of the viewing conditions, have been optimized for each data set, simulating different situations. On the one hand, samples in data set D1 have been simulated to be observed in a lighting booth VeriVide CAC under a D65 simulator. On the other hand, the images in data set D2, have been simulated to be observed in the real viewing conditions of the capture moment. The white point and luminance of the background have been measured in the lighting booth for samples in data set D1, and estimated in each of the images of the data set D2. Table 2 shows the parameters for each case: the absolute luminance of the adapting field (L a ), the relative luminance of the background (Y b ), the relative tristimulus value of the white point, and the degree of adaptation to the white point (D). As we will compare later VINO and EnChroma glasses, the optimized parameters for EnChroma glasses have also been estimated.
As it is shown, the parameters change when any of the glasses are worn. In general, the absolute luminance of the adapting field decreases, and of course, the white point changes, more in the case of the VINO glasses, which are more chromatic than EnChroma's.  [28]. To ensure a better simulation we have estimated the factor D using the "chromatic adaptation factors" (F ρ , F γ , and F β ) of the Hunt Color Appearance Model [28], the most extensive one, capable of considering any kind of illuminant. These factors take into consideration not only the level of luminance of the adapting field but also the chromaticity of the illuminant. The chromatic adaptation factors are designed such that chromatic adaptation is always complete for the equal-energy illuminant. Therefore, after computation of the chromatic adaptation factors, based on their values, we have estimated the final value of the factor D (lower D values corresponding to F ρ , F γ , and F β values further apart from 1). The CVD conditions were simulated using the Lucassen et al. model with parameter d = 0.3 for mild conditions, d = 0.6 for medium conditions and d = 1.0 for dichromats [29]. As we are using CIECAM02 framework, the CVD is applied inside the CIECAM02 model, specifically after computation of R',G',B' values in the fundamental cone space. The severity parameter d used is not directly linked to the loss of chromatic discrimination of a CVD observer, as described in [30]. The values used for d = 0.3 and d = 0.6 would correspond to spectral separations between R' and G' in the R',G',B' space of 20 and 10 nm for protanomalous, and 30 and 20 nm for deuteranomalous observers. As a reference, the spectral separation in the R',G',B' space for normal observers is 50 nm. One of the limitations of the approach presented is the fact that the simulation model used did not allow for a direct determination of the severity of the CVD simulated observer based on spectral separation of responsivities in the cone response space, which can be related to color discrimination loss, as shown in [30].
The data distributions with and without the VINO glasses were then compared to determine the effect of adding the filter on the color attributes in the different data sets. The data was analyzed using, according to the data distribution type, either parametric or nonparametric statistical tests, and circular statistics for the h parameter, as described in [15].

Results of the normality tests
For data sets D1 and D2, the Jarque-Bera test was performed on the J and C distributions to test for normality, and the Watson's U square test was chosen for the hue distributions, since circular statistics were required in this case (see [20,31]).
For data set D1, the J attribute (lightness) distribution is not compatible with normality for all conditions. For C, in the "glasses off" condition, distributions for all conditions were also not normal, whilst for the "glasses on" condition the normal, protanope and deuteranope conditions (d = 1.0) were not compatible with the normality assumption. For the hue attribute, only two conditions without glasses (normal, and mild deuteranomaly) were not compatible with a distribution of the von Mises type. For data set D2, all the J and C distributions for all the conditions were not normal, and the h distributions for all conditions were von Mises type.
Given that for each condition at least one of the distributions of J and C to be compared is not normal, we have used non-parametric tests to compare the attributes obtained with and without the VINO glasses. For the h parameter, we have used either the Mardia-Watson test or Watson's U square test, depending on the distribution type, as described in [15]. Table 3 shows the statistical parameters found for each data set and attribute J. The effect of adding the VINO filters on the lightness of the samples in both data sets is a consistent trend of slightly increasing median lightness for all conditions in data set D1, and for all conditions save medium and severe protans in data set D2. The differences are not statistically significant for data set D1 except for the medium and severe protan-type observers (maximum p = 0.0280). For data set D2, all the differences are statistically significant (with p = 0). The median lightness in data set D2 is higher than in data set D1.   Table 4 shows the statistical parameters found for each data set and attribute C. Data set D2 has markedly lower C values than data set D1. The C value decreases for the CVD observers when compared with the normal ones, and progressively decreases with the severity of the condition as well, less noticeably for data set D2. The effect of adding the VINO glasses in both data sets is a noticeable increase in the chroma of the samples. This effect is more marked for normal and protan subjects, and the increase is a bit less for the deutan subjects. For data set D2, this decrease in the effect for deutan subjects is higher than for data set D1. The differences found are in all cases statistically significant (maximum p value of 2x10 −6 for data set D1 and p = 0 for data set D2).  It is remarkable that the strongest chroma increase happens for the normal subject. We could then say that comparatively the VINO glasses are less efficient in increasing the chroma for the CVD subjects, and that this attribute will contribute the most to the change in contrast induced by wearing the glasses, since the changes in lightness are much lower due to the effect of chromatic adaptation. Table 5 shows the statistical parameters found for each data set and attribute h. The effect of adding the VINO glasses on the hue of the D1 data set samples is a large increase in the h mean angular direction values, while for the D2 data set the increase appears very slightly only for the normal condition, and there is a slight decrease for anomalous trichromats. Basically, for D1 all mean angular directions are moved to the fourth quadrant, which is consistent with the reddish-magenta appearance of the simulated scenes (see Fig. 2). The mean angular direction of D2 samples is already in the fourth quadrant, which is why the variation in hue is much less noticeable for this data set. Due

Filter tran
The spectral t shown in Fig  transmittance  spectral respo EnChroma Cx leaves the gre radically diffe protanopic su spectrum, and blue region w

Subjectiv
In comparison EnChroma Cx remarkably si a slightly high that the numb higher than in If we che parameter wa trend is found the Angle par counter-clock greater produc In order t using a hyper images are fr processed wit As an examp protan subjec either EnChro  Fig. 3). In con h or reddish co oth filters will will be enhance avelength porti spectrum wou ma filter.   Table 1, taking into account that the effect of the EnChroma Cx-65 filter on the color appearance of the samples is much less intense than the effect of the VINO filter, as will be shown in the next section. So, the number of errors in the color naming task increases with the VINO glasses on because of the extreme change in appearance induced by the glasses.

Simulated data
As explained in section 4.1, the CIECAM02 model computations have been carried out using custom parameters for each scene, and in addition, the daltonization has been performed within the CIECAM02 cone response color space. This makes necessary to re-compute anew the color attributes for the EnChroma Cx-65 glasses to allow for a fair comparison with the VINO filter results, since the values shown in [15] correspond to simulated viewing under only one condition (samples displayed on a monitor screen). The new J, C and h parameter values found for the EnChroma Cx-65 filter are shown in Tables 7, 8 and 9. Comparing the effect on the J (lightness) parameter found for the VINO and EnChroma filters, in both cases the variations in J are slight (consistent with a certain degree of chromatic adaptation). The VINO filter tends to decrease J values save for D2 and medium-severe protan observers, while the EnChroma filter produces a very slight increase in general for both data sets (the values for the mild and medium protan conditions remain almost stationary). This result shows that the effect of the added filter can be different depending on the CVD condition and the region of the spectrum that the filter is suppressing, but J is the attribute which is less affected by the filters.
Regarding the hue parameter (except for the dichromat conditions), the effect of the EnChroma Cx-65 filter is to shift the mean angular direction towards the 180° axis, both for D1 and D2 samples, while VINO produces a shift towards the fourth quadrant also in both data sets. The amount of the shift is more pronounced for the EnChroma Cx-65 filter, since for data set D2 the mean angular direction was already in the fourth quadrant. The results could also be interpreted in terms of the narrowing of the standard deviation values in the hue distributions, which is found for the VINO filter but not for the EnChroma filter. This means that the VINO filter is shrinking the width spanned by the color samples in the CIECAM02 color space, while the EnChroma filter has not such a noticeable effect in the color distributions, which is expected because the VINO filter produces clearly a shift towards the purple region, while the EnChroma does not produce such a clear effect.
Regarding the C parameter (chroma), for the EnChroma filter there was a rather small chroma increase and it was only found in a consistent way for data set D1 and not for the medium and severe protan observers. With the VINO glasses, the chroma increase is much higher and consistent amongst the data sets and conditions. Since the VINO filter has a lesser transmittance in a longer range of wavelengths than the EnChroma filter, the effects tend to be more marked, making the colors more vivid in comparison with the glasses off condition. Once again, these findings can be explained by considering that the EnChroma glasses will predictably produce a lesser impact on the chroma of the majority of colors present in natural scenes, which have a lesser amount of blue in general, which is the region that is preferably enhanced by the EnChroma filter.

Conclusions
The low transmittance in the intermediate region of the spectrum in the VINO O2 Oxy-Iso glasses produces a significant change in color appearance, especially in the chroma attribute. This change allows some CVD subjects, above all the deutan subjects, to pass recognition tests such as the Ishihara but not the arrangement tests such as FM100. Similar conclusions have also been reached by other authors [16][17][18]. This extreme change in color perception also has a negative effect in the color naming test. The simulations show that the hue angle is compressed and moves towards the fourth quadrant. The chroma increases significantly for all the conditions and the for the two groups analyzed. These effects are much more significant than those produced by the Cx-65 by EnChroma model that we also analyzed using CIECAM02 parameters customized for each scene and data set. Neither of the two models we studied improves the color vision of the CVD subjects to the same level as normal subjects, nevertheless they may facilitate some discrimination tasks. Thus, the VINO O2 Oxy-Iso filters may be useful for some specific applications (like enhancing the O 2 signal from hemoglobin under the skin) as occurs with some color filters used in certain activities such as hunting, shooting, low vision etc [32].
To sum up, our results support the hypothesis that glasses with filters are unable to effectively resolve the problems related to color vision deficiency.

Funding
The Spanish State Agency of Research (AEI); the Ministry for Economy, Industry and Competitiveness (MIMECO) (Grant numbers FIS2017-89258-P and DPI 2015-64571-R); European Union FEDER (European Regional Development Funds).