Novel approach in sensory color masking: Effects of colored environments on chocolates with different cocoa content

In sensory practice, the color masking procedure is executed almost exclusively with red light, regardless of the color of the product. In this work the practical application of a novel approach for sensory masking of color differences, hence the applicability of a spectrally tunable light booth was investigated. The aim of the research was to find the most appropriate masking environment, which i) can be considered as the most effective mask and ii) provides the best conditions regarding chromatic adaptation. The main idea of the novel approach in sensory color masking is that the instrumental characterization of all parameters influencing the perception of the sensory assessor. The comparison of the potential masking illu-minations based on instrumental results (considering masking effect and chromatic adaptation) and the definition of the most appropriate mask should be accomplished based on sensory tests. The sensory effects of colored environments were tested on chocolate samples containing different amount of cocoa. CIELAB dE color differences between the samples were calculated based on their spectral reflectance. Four monochromatic (red, green, blue, amber) and six combined masking illuminations were developed in a spectrally tunable light booth and were instrumentally characterized by their spectral power distribution and luminance values. The chocolates were ranked according to their darkness by the assessors. The masking environments were compared based on the analysis of sequence correctness, decision time, and their combination (ANOVA, Survival-analysis). In Experiment 1 the monochromatic Blue light was not only more effective, than the commonly used Red light, but it was the most effective mask regarding color differences: the sequence error increased to 61% compared to the standard white (D65) illumination, and the decision time increased to more than double (15.74 s → 35.7 s). In Experiment 2 green and amber lights were combined with the blue mask, increasing the degree of chromatic adaptation: decision time decreased while the effectiveness of the mask did not decrease based on the sequence errors. Following the novel approach product specific masking environment considering effectiveness and chromatic adaptation was realized verified by statistical analysis of sensory test results.


Introduction
The appearance of food products plays a particularly important role in sensory evaluation.The color of a product is one of the most salient visual cues a consumer faces before it is smelled or tasted.Numerous articles support that consumers "buy with their eyes" and color is a key element in visual appearance that influences and underpins expectations about the expected smell and taste (e.g., Brice, 1954;Foster, 1956;Hutchings, 1977;Tom et al., 1987;Wei et al., 2012).De Melo et al. (2009) examined the sensory acceptance and liking of light brown, calorie-reduced chocolates.Their result shows, that the brown color of the calorie-reduced chocolates is not less popular than the color of the traditional reference chocolate.Piqueras-Fiszman and Spence (2012) examined the effect of hot chocolates' cup color on consumer preferences and sensory properties.It was revealed that samples tasted from dark cups were perceived as sweeter and more intense in aroma as samples tasted from red, orange, or white cups.
Of the many visual cues, such as transparency or homogeneity, color plays a highlighted role, since it is an indicator of edibility (Lavin & Lawless, 1998;Zampini et al., 2007).Various researches report that visual characteristics of the product under study affect other sensory properties i.e., one modality is cross-modulated with another (Spence et al., 2010;Spence, 2011).If the color of the food sample causes expectation error, the objective assessment of sensory characteristics is distorted (ISO 6658:2017;ISO 8589:2007;Meilgard et al., 2006;Lawless & Heymann, 2010).Several studies report that increased color level is associated with higher aroma values (e.g., Zellner & Kautz, 1990;Spence 2019).Moreover, taste identification is aided by the color characteristic of the product, making it more difficult to identify the taste of colorless food (Dubose et al., 1980;Stillman, 1993).As a result, many researchers emphasize that color plays a particularly important role in shaping the taste experience (Zampini et al., 2007, Wang & Spence 2019).
Standards describing sensory tests state that if the evaluation should not be based on visual assessment, it will be necessary to create conditions in which the color stimulation of the products does not affect other sensory characteristics (ISO 6658:2017, ISO 8589:2007).Currently, the problem of color masking is not given sufficient weight in the sensory literature and practice.So far, there have been several attempts to cover the visual differences of food products (Taylor & Linforth, 2009;Carpenter et al., 2012).Even though, most of the commonly applied methods carry systematic errors and their scope of application is strongly limited (Sipos et al. 2021).Colored illuminations are the most widely applied methods, of which single-wavelength, usually red-light systems are used in the practice of accredited sensory laboratories (ISO 17025:2018, ISO 8589:2007).The spectral content of the fluorescent lamps mentioned in the standard is not tunable; therefore the spectral power distribution of the light source limits the number of the products for which the mask can be effective, hence fail product-specific efficiency testing.To date, there have been quite a few scientific publications in the international literature dealing with the masking of the color of chocolates or cocoa powders.These studies use almost exclusively red illumination as the examples of Benedetti, Pompei, & Mannino, 2004;Chapman, Rosenberry, Bandler, & Boor, 1998;Hough, Sánchez, Barbieri, & Martínez, 1997;Lenfant et al., 2013, andTolve et al. (2018) show.In Kilcast and Clegg's (2002) study green light was applied to mask the color discrepancy between various foods, including chocolate mousses.Liu et al. (2015) chose an unusual method: the evaluations of chocolates were performed under a dim light.Besides the importance of masking color differences between the observed products, the quality of the illuminaton regarding the human visual system in the testing environment is also very important.One of the processes permanently operated by the human visual system is chromatic adaptation, which allows us to adapt to the color and intensity of the surrounding illumination.The perceptible result of this process is color constancy, i.e., that the colors of the objects remain (approximately) constant under illuminations with different spectral content (Foster, 2011).Of course, even adaptation has its limits and can not be completed under any conditions.Performing visual tasks under illumination providing incomplete chromatic adaptation might lead to sensory fatigue.Most metrics of the degree of adaptation consider only the luminance of the illumination.Even though, studies of chromatic adaptation can be found considering both neutral and colored illuminants (Smet et al., 2017a;Smet et al., 2017b) showing that the adaptation can be strongly affected by the purity and the chromaticity of the illumination.As described above, in masking environments mostly lights with narrow spectral content are applied, therefore the effect of the chromaticity of the illumination on the degree of adaptation should be taken into consideration.In summary, in previous researches, the general practice to mask color discrepancy was to use monochromatic red illumination.Although, the masking effect of the applied illuminations was not reported.Unfortunately, the research results so far are characterized by the fact that the luminaire, the color environment of the testing and the geometry of the sensory examination are not sufficiently defined.During literature review, we did not find any publications investigating the masking efficiency of color environments produced by combining multiple lights.
The aim of our research was to find the most appropriate masking environment, which on the one part can be considered as the most effective mask (covering the color differences between the examined products), and on the other part provides the best conditions regarding chromatic adaptation (causing minimal sensory fatigue of the visual system).

Spectrally tunable light booth
To eliminate the inaccuracy mentioned above, a self-developed spectrally tunable light booth was applied to create the appropriate illuminations.The light booth was built at the Department of Mechatronics, Optics, and Mechanical Engineering Informatics, University of Technology and Economics, Budapest.The installation has enclosing dimensions of 1.5 × 1 × 1 m, in which there are sets of 5 types of LEDs (red, green, blue, white and amber) mounted on 4 fixed panels.The material of the installation is white wood (furniture board) which is diffuse and provides uniform reflectance.The panels can be easily rotated by hand or accessed through two doors.In addition, there are two deflectors in the booth that prevent the light sources from direct viewing and also improve the homogeneity on the bottom plate.The LEDs are located in the corners, which emits direct light onto the white walls, thus the light enters the workspace after multiple reflectances.Access to the booth is solved by an opening in the front surface.The structure of the light booth has been designed to provide a spacious and comfortable workspace for the sensory assessors.Particular care was taken to position the panels so that the luminance distribution was even on both the horizontal bottom surface and the white walls.The schematic structure of the system is shown (Fig. 1).As illustration the light booth with blue masking environment is also shown (Fig. 2).
Control is via two Arduino Uno microcontrollers.The luminous flux of the LED channels is adjustable with relative control values ranging from 0 to 255, which implies the maximum level of luminance of a channel.LEDs are generally characterized by a range of lanes, where most of their power is emitted.Thanks to the semiconductor technology, the bandwidths are quite thin, thus a typical wavelength can be specified for each LED.The parameters of the LEDs used in the booth are shown (Table 1).
The degree of adaptation considering the chromaticity of the illumination was calculated for each masking environment with the LuxPy Python Toolbox (Smet, 2018;Smet, 2019).

Selection of assessors
The requirements of the relevant international standards (ISO 8586:2012;ISO 11037:2011) were used as the guideline for testing the assessors' vision.Accordingly, in the first step, an instrumented color vision test (based on Rayleigh equation) was performed by an OCULUS 47,700 Heidelberg MultiColor anomaloscope.The aim was to eliminate people with dyschromatopsia, by defining the most important performance indicators (Anomaly Quotient, Mixing light matching range, Reference light matching range).The anomaly quotient (AQ) determines the mixing ratio of red and green as a numeric value (Fig. 3).The anomaly quotient is calculated using the formula: where -AQ: Anomaly Quotient, -E: Upper limit (73) of the mixed light scale (x-axis) with 0% green content, Á. Nyitrai et al.
Based on the AQ results, patients can be grouped according to their vision: AQ < 0.7 (Protanomaly); AQ > 1.4 (Deuteranomaly); 0.7 < AQ < 1.4 (Normal matching zone).Mixing light range is the distance between the highest and the lowest value of the X-values, where the Xaxis is a 73 point scale from pure green (0) via normal midmatch (40) to pure red (73).Reference light range is the distance between the highest and the lowest value of the Y-values, which corresponds brightness of the yellow reference field (Oculus Manual, 2015).
The assessors were selected from students and workers of Szent István University (26 participants, age: 18-27, 7 men, 16 women).After the anomaloscope examination, an achromatic grayscale test (ISO 8586:2012) was performed due to the nature of hue differentiation task.Only assessors with proven normal vision were involved in the experiment.The sensory tests were performed under colored illumination with a completely darkened background.Evaluations were made between 9 and 12 a.m.due to mental freshness and sensory condition of the eyes.The experiments were approved by the United Ethical Review Committee for Research in Psychology (EPKEB, Hungary, reference number: 112/2017).All assessors signed a written consent, agreeing to participate in the measurements after verbal and written orientation.The assessors stated that no activities that caused eye fatigue were performed 12 h prior to the test.

Table 1
The characteristic peak wavelengths of the LED channels of the spectrally tunable measuring station.

Materials
Milk chocolate bar was chosen as material, since it met the most important criteria: homogeneous surface color (condition of sensory color masking), non-perishable (no visual changes on the product during storage and testing) and can easily made a color sequence (color-scaled samples can be produced based on recipes).The goal was to produce a series of milk chocolate bar samples with monotonously increasing cocoa content.The amount of cocoa powder that almost exclusively determines the color of the chocolate was chosen to be 3.00% for the No.1 sample.Ingredients of the No.1 sample (the lightest): sugar (46.50%), cocoa butter (33.93%), skimmed milk powder (16.00%), cocoa powder (DE7800) (3.00%), sunflower lecithin (0.50%), polyglycerol polyricinoleate (PGPR) (0.05 %), crystal vanillin (0.02%).The amount of cocoa powder in each chocolate bar was then increased by 0.75%, respectively.Indeed, the increasing amount of cocoa powder resulted a decrease in sugar in the same proportion.Only the amount of sugar and cocoa powder varied for each chocolate bar, all other ingredients remained unchanged.This method provided the required color stimulus differences (ΔE ab* ) between samples.All samples were prepared following the same technological steps.The ingredients were mixed into a mass in a 4 kg plastic container, after the cocoa butter was melted.The mass was poured into a 45 • C ball mill (Wiener, 4 kg capacity) followed by grinding for 45 min.To help dissolve the total amount of sugar, 300 g of the mass was recycled back into the ball mill stub, thereby helping to dissolve the total amount into the mass.It followed another grinding for 45 min, which resulted in a chocolate mass with a particle diameter of 21 ± 3 μm, which is provided by a micrometer screw gauge (Borletti micrometer, LT-BOR-000699).Then, during tempering, 1 kg of the mass was placed in an oven (43 • C, 4-5 h).After heating, 750 g of the heated mass were poured onto a tempering marble table, which was thickened by agitation.After cooling, the tempered mass (26 • C) was poured back into the vessel, in which the still warm 250 g of chocolate warmed the 750 g taken out.During mixing, after the desired temperature of 32 • C was reached, the chocolate was quickly poured into bars.The chocolate bars were placed in a refrigerator (9 • C, 60 min), then packaged and the prepared samples were stored at 20 • C until analysis.

Method
The colors of the chocolate bars were measured by a tristimulus colorimeter (X-rite RM200QC).The instrument specifies the color as the L*, a*, b* color coordinates of the CIE Lab color system.The measurements were made in ten different places on each sample, and arithmetic mean was calculated.From the obtained color coordinate values, chroma, hue and color differences (ΔE ab* ) between samples were determined (CIE, 2004).The ΔE ab* color difference was calculated as follows: Based on literature color differences perceived by standard colorimetric observers can be grouped as follows: ΔE ab* <1 observer does not notice the difference; 1 < ΔE ab* < 2 only experienced observer can notice the difference; 2 < ΔE ab* < 3.5 unexperienced observer also notices the difference; 3.5 < ΔE ab* < 5 clear difference in color is noticed; 5 < ΔE ab* observer notices two different colors (Mokrzycki & Tatol, 2011).
Spectral measurements were also performed to design the appropriate masking illumination.The spectral reflectance was measured three times on each chocolate flat surface by an AvaSpec Spectrophotometer in the range of 380-780 nm with a resolution of 10 nm.

Sensory analysis
Assessors ranked the chocolate samples from lightest to darkest based on their hue under different colored illuminations.As a reference, chocolate samples were first evaluated under validated D65 illumination simulating artificial sunlight in a PANTONE Color Viewing Light Booth (3light Unit) (ISO 11037:2011).The aim of the reference measurements was to provide benchmark data for comparison following standardized recommendations.Then, evaluations focused on the correctness of the made sequence and duration of decision making in different lighting environments.In Experiment 1, sensory tests were performed under four different monochromatic illuminataion environments (Red, Green, Blue, and Amber).In Experiment 2 assessors examined the chocolate samples under illuminations of combined colored lights, which are determined from the results of Experiment 1.The sensory analysis was executed in the Laboratory of Visual systems, at the Department of Mechatronics, Optics, and Mechanical Engineering Informatics, University of Technology and Economics, Budapest.The elements of Good Sensory Practice were taken into account in designing, performing, and evaluating sensory test.The samples given to the assessors were a row of chocolate, i.e., 4 pieces.Samples were coded with three-digit random numbers.To prevent assessors from learning the codes, 2 sets of chocolate (with different codes) were used, that were swapped between the various illumination settings.Chocolates were presented in a transparent plastic Petri dish open from above.The tests were executed by the assessors with a 0 • /45 • test geometry, which ensured that the light fell perpendicular to the sample and the assessors' line of sight was at 45 • to the surface (ISO 11037:2011).Measurements were performed in a dark room in order to avoid the effect of environmental light.Assessors were sitting in front of the darkened booth and arranged the chocolate samples in order after the adjusted masking illuminations were turned on.Instruction was given to say the word "done", when finished with the ranking.

Statistical evaluation
To make sure that the analysis gives accurate information regarding the effect of the masking illumination itself, the systematic error carried by the variability of the assessors' individual decision times should be eliminated.Therefore, the ranking times were evaluated by comparing scaled values in addition to comparing real durations.In scaling, the longest decision times of each assessor (maximum decision time, MDT) were chosen and used for normalization.As a result, a scaled value between 0 and 1 was obtained for each ranking.
An one-way fixed effect ANOVA model was used with one fixed factor (masking lights) to distinguish ranking correctness and times, after checking the assumptions regarding normality and the homogeneity of variance.In the case of significant difference (p < 0.05), the pairwise comparison was executed with Duncan's post hoc test.Survival curves (ranking time and correctness) were compared with Kaplan-Meier survival analysis.. Comparison of significant differences was calculated applying Log-rank, Wilcoxon and Tarone-Ware methods, and consensus was used to evaluate the results.A significance level of α ≤ 0.05 was determined for statistical analyzes.All calculations were performed with Addinsoft XLSTAT (ver.2016.02.27444).

Results of the samples' color measurement
The results of instrumental color measurement confirmed that the prescribed linear change in color coordinates was mostly achieved.The fit of the trendlines were fine (R 2 L* = 0.948; R 2 a* = 0.979; R 2 b* = 0.972) (Table 2).
Color differences between samples were calculated based on their color coordinates.The aim was to include only those samples for sensory testing, between which the color difference values are in the third category (ΔE ab* = 2.0-3.5;unexperienced observer also notices the difference).Another goal was to have almost equal color difference between each chocolate bar, in a monotonically increasing row.
Instrumental measurement of the spectral reflectance of the chosen chocolates gave the result (Fig. 4).The reflectance properties of the samples are very similar in nature: below ~ 500 nm it is almost constant, and then it grows monotonically.

Results of instrumental color measurement of the environment
Observing the spectral power distributions, peak wavelengths (627 nm, 596 nm, 523 nm and 460 nm) and quasi-monochromatic nature of the 4 color LED channels can be seen.In order to accurately photometrically characterize the masking environments, the luminance values were recorded (Table 4, Fig. 5).

Results of assessors' color vision
26 subjects participated in the test, of whom 3 individuals were excluded based on the results of the anomaloscopic examination due to their identified dyschromatopsia (subject codes: 2460, 4050 and 2780) (Table 5).The grayscale test was done correctly by the 23 individuals.

Experiment 1: Sensory masking efficiency of monochromatic illuminations
In Experiment 1, the main question was if there was any difference between the masking efficiency of the different monochromatic colored lights.Results showed that everyone ranked the chocolate samples correctly and in the shortest time under the reference (D65 White) illumination.Most samples were incorrectly ranked under Blue illumination, and this masking environment prolonged the decision time the most, which were more than doubled compared to the standard D65 (white) (15.74 s → 35.7 s) (Table 6).It can be seen that assessors made significantly less correct sequences under the Blue lightthan under any other colored light.Therefore, it was the most effective masking color between all illuminations examined.Besides Blue, Green and Amber illuminations masked effectively (Table 7).
The ranking time of the assessors was also recorded.The focus of examination was on the significant differences between the times of ranking.Evaluation of both real-time and scaled data showed that setting up the sequences under Blue light was significantly the longest compared to all other illuminations.The analysis of scaled data gave some additional information: the effects of Green and Red lights can be also separated (Table 8).The order of evaluation time of scaled data (shortest to longest): D65 (White) → Amber, Red → Green → Blue.
Using survival analysis, the courses of the curves summarizing the results of the participants were analyzed.To increase the reliability of the results, the values of three trial statistics (Log-rank, Wilcoxon, Tarone-Ware) was calculated.A comparison was considered different if at least two values were found to be significant.In summary, all used colored illuminations have some masking efficiency compared to D65 (White), but they differ in the effectiveness of masking.Compared to D65 (White), all other illuminations increased decision times and errors due to masking effect.The results of the survival analysis show the significant order of the masking illuminations considering the masking effect: Blue → Green → Red, Amber (Table 9, Fig. 6).

Experiment 2: Optimization for chromatic adaptation, maintaining sensory masking efficiency
In Experiment 2 it was investigated whether it is possible to enrich the spectral content of the masking illumination to provide better chromatic adaptation (causing minimal sensory fatigue of the visual system), while maintaining the masking efficiency (covering the color differences between the examined products).
The results of Experiment 1 determined the settings of Experiment 2. Since based on the results of Experiment 1 the monochromatic Blue light was the most effective mask, the Blue LED was always on at maximum brightness.To enrich the spectral content, three Green and three Amber added settings to varying degrees (33%, 67%, and 100%) were used.After all 6 new colored light combinations were created (Table 4).The results of descriptive statistics revealed that in Experiment 2 G33 + B100 illumination had the greatest impact on ranking.The decision time was mostly increased by the G67 + B100 light combination.Based on the correctness of the ranking and the decision times, the two least effective masking light combinations were A67 + B100 and A100 + B100 (Table 11).
Results of the analysis of correctness showed, that while the illuminations mixed with amber (A67 + B100) and (A100 + B100) had more correct squences than the pure Blue (B100), so they had a weaker masking effect, the illuminations combined with Green (G33 + B100, G67 + B100 and G100 + B100) and Amber (A33 + B100) lights did not differ from the Blue (B100) in correctness of ranking.To optimize     ascendent order (Table 15).

Discussion
In this paper, different masking environments were created in a spectrally tunable light booth.Due to reliability, instead of the commonly used Ishihara pseudo-isochromatic color test, instrumental anomaloscopic and grayscale tests were performed to filter out assessors with color vision deficiency.Assessors with proven normal vision had to rank in order five different cocoa content milk chocolate samples based on their colors.The tests were performed under standard artificial daylight (D65) as a reference, and colored masking illuminations also.The masking effects of the created illuminations were compared to each other.Under artificial daylight, the assessors performed flawlessly and made the correct order the fastest.Compared to D65 white light, all monochromatic illuminations significantly impaired the ability to     colors).Moreover, the incident radiation is absorbed more in this range than in the higher wavelengths.It is therefore advisable to look for masking illumination in this range, as this will minimize the expected difference between the color stimulus functions (and thus the perceived colors) both spectrally and in total intensity.It is always expedient to apply this principle to determine the best masking illuminations.
After further testing, it may be appropriate to integrate color masking protocols and tools into the raw material acceptance protocol, inprocess control, and finished product control of chocolate factories.Of course, masking illumination always needs to be product specific.

Conclusion
The development of sensory studies of color products requires instrumental evaluation of key factors influencing color perception (ISO 11037:2011;ISO 8586:2012).The three pillars of human act of perception: ─ instrumental characterization of sensory assessor, ─ instrumental characterization of products, ─ instrumental characterization of the test environment.This is also necessary to make sensory tests reliable, reproducible and comparable.
According to ISO 8589 (2007), in the case of colour assessment of products special lighting devices may also be needed to mask colour or visual differences (dimmer device, coloured light sources, coloured filters, black light or monochromatic light sources).As shown, the masking effect of Red light, which is most common in the practice of sensory researches, was poor.It is also shown that in order to provide higher degree of chromatic adaptation, enriching the spectral content of the most effective masking illumination is possible without loosing masking efficiency.In the future, it would be expedient to create devices suitable for creating spectrally controllable environments in sensory laboratories, so that product-specific color masks colud be performed.
The main steps of the novel approach are the instrumental characterization of the pillars of perception, which provide background to determine the potential product-specific masking illuminations.Followed by sensory evaluation by assessors the best masking illumination can be defined considering both masking efficiency and chromatic adaptation.

Fig. 2 .
Fig. 2. The work-space of the spectrally tunable lighting booth (Blue masking environment).(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) where, -L* x cthe L* value of one of the chocolate samples, -L* y : the L* value of the other chocolate sample to be compared, -a* x : the a* value of one of the chocolate samples, -a* y : the a* value of the other chocolate sample to be compared, -b* x : the b* value of one of the chocolate samples, -b* y : the b* value of the other chocolate sample to be compared.

Table 2
The CIELAB color coordinates of the milk chocolate samples range from the lightest to the darkest (L*, a*, b*, Chroma, Hue).

Table 6
Sensory effects of monochromatic illumination of chocolate samples (descriptive statistics).

Table 7
Results of the correctness of the ranking based on significant differences according to Duncan's post hoc test.

Table 8
Results of the real and scaled time data of ranking based on significant differences according to Duncan's post hoc test.

Table 11
Sensory effects of combined illumination of chocolate samples (descriptive statistics) * Values of the best masking illumination from Experiment 1.

Table 12
Results of the correctness of the ranking based on significant differences according to Duncan's post hoc test and degree of adaptation

Table 13
Results of the real and scaled time data of ranking based on significant differences according to Duncan's post hoc test.
*masking light combinations under which the masking effect did not differ significantly (p < 0.05) from the Blue (B100) light.Á.Nyitrai et al.

Table 15 Average
ΔE ab* color differences of the adjacent chocolate samples under the masking illuminations.Nyitrai et al.reported in this paper and carried out BME has been supported by the NRDI Fund (TKP2020 IES, Grant No. BME-IE-BIO) based on the charter of bolster issued by the NRDI Office under the auspices of the Ministry for Innovation and Technology.The research reported in this paper and carried out BME has been supported by the NRDI Fund (TKP2020 NC, Grant No. BME-NCS) based on the charter of bolster issued by the NRDI Office under the auspices of the Ministry for Innovation and Technology.Ákos Nyitrai expresses his gratitude to the Doctoral School of Food Science of Szent István University.