Spectral Optimization to Minimize Light Absorbed by Artwork

ABSTRACT

Light is needed to appreciate artwork, but visible radiation causes damage by photochemical action. Photons trigger photochemical action only when they are absorbed by the object. Damage to sensitive materials can be reduced by optimizing the spectrum of a light source to the reflectance factor of a museum artifact. Absorption minimization approaches can be utilized to quantify and optimize typically conflicting parameters, such as damage due to light absorption, color quality of artwork, and energy consumption. In this project, seven narrowband light emitting diodes (LEDs) are optimized using a multi-objective genetic algorithm to reduce light absorption and energy consumption, while maintaining the color appearance of five paints (blue, green, yellow, orange, red). Results indicate that optimized test spectra can reduce light absorption between 18% and 48%, without causing perceptible color or hue shifts in the paintings. Parallel to absorption, energy consumption is almost halved for all of the paints (between 42% and 48%). Optimizing the light source spectrum can aid in conservation of art collections in museums by reducing damage caused by optical radiation while preserving color fidelity.

KEYWORDS:

• Art conservation
• color perception
• color rendition
• LED sources and systems
• museum lighting

1. Introduction

Light is necessary to display museum objects. However, exposure to visible and invisible radiation damages artworks (Beek and Heertjes 1966; Cuttle 1996; Feller 1964). Visual appearance and damage are two competing criteria in the display of art. The optimal lighting for preservation would be total darkness, which would minimize irreversible damage to artwork but prevent visibility. Current conservation standards are based on the Harrison damage curve, which states that energy in shorter wavelengths has more potential to cause damage (CIE 2004a). However, the Harrison damage curve is based on low-grade paper and is not representative of all materials in museums (Cuttle 1996). The damage curve was first proposed in 1953 and was influential in the adoption of incandescent light sources in museums, due to their limited emitted energy at shorter wavelengths (Cuttle 1996). However, incandescent sources have low luminous efficacy and emit energy mostly in longer wavelengths, including infrared (IR), which also causes damage to artwork.

Light absorbed by an object turns into heat and damages pigments. Absorption of visible radiation causes deterioration of the artifacts. Therefore, the optimal lighting for a museum object requires the minimization of absorbed light while maintaining the color appearance of the artifact, which is critical for the appreciation of artwork. Because solid-state lighting devices allow better control over the spectral output of a light source and light causes damage to artwork only when it is absorbed, it is possible to envision a lighting system that detects the reflectance characteristics of an artwork and emits spectrally optimized light to each colored part to reduce damage. This article investigates the optimization of light source spectra using a nonlinear optimization tool to minimize damage to artwork while maintaining color appearance.

2. Background

Photochemical action is the process of chemical changes in a molecule when it absorbs a photon, which can only be initiated by radiation (Schaeffer 2001). Although photochemical actions do not always cause immediate changes, modified molecules may alter the appearance of an object over time, due to additional chemical reactions (Schaeffer 2001). Radiation of short wavelengths, especially ultraviolet, has greater energy per photon than longer wavelength light, and it is more likely to cause photochemical action due to the excitation state of molecules and energy level of photons. Although it has lower energy levels, IR also contributes to damage by increasing the surface temperature of an object, which may result in surface hardening, discoloration, and cracking from accelerated chemical reactions (CIE 2004a).

The magnitude of photochemical action depends on irradiance, the spectral power distribution (SPD) of radiation, duration of exposure, and action spectrum of the material (i.e., relative spectral responsivity of a receiving material; CIE 2004a). An important visible result of photochemical reaction is discoloration of artifacts, which hinders the appreciation of artwork. Changes in the colors of individual pigments can be quantified to measure the damage caused by light absorption. To address these challenges, the CIE (2004a) recommends filtering ultraviolet and IR and limiting lighting exposure to 50 lx for materials that have medium and high responsivity to light (or 15,000 lx h y⁻¹ for highly responsive materials and 150,000 lx h y⁻¹ for medium responsive materials). However, color discrimination is diminished at lower illumination levels, which results in differences in perceived color between museum conditions (indoors and low light levels) and the conditions in which the artist created the artwork (presumably under daylight). In addition to reduced color discrimination, lower illumination levels cause a decrease in colorfulness (Hunt effect; Hunt and Pointer 2011) and induce hue shifts (Bezold-Brucke hue shift) for certain wavelengths (Pridmore 1999).

2.1. Damage and color quality

Much research on lighting and damage to artwork has focused on irradiance and color quality. Cuttle (2000) matched the illuminance of a three-band light source with incandescent sources of 2856 K and 4200 K, resulting in equal visual satisfaction and reduced incident irradiance on three paintings, with varying levels of detail. Although three-band light sources were created by using filters, which reduced luminous efficacy, the potential to use noncontinuous spectra for reducing damage was highlighted.

Narrowband light sources can also be used to enhance the color appearance of degraded museum objects. Five narrowband light emitting diodes (LEDs) were mixed to rejuvenate the appearance of aged color prints by increasing chroma (Viénot et al. 2011). CIE 2004b (L*, a*, b*) was used to predict color shifts successfully, and visual examinations supported the proposition that color degradation can be counteracted with chromatic lighting (Viénot et al. 2011). However, illuminating a faded museum specimen by optimizing the light source only for a few colors may cause shifts in the color appearance of the other parts of the artwork. Hue and chroma shifts for the whole object should be assessed to prevent reduction in the color quality of an item.

2.2. Optimization for color quality and damage

Computational tools can be used to optimize light sources for minimal damage and maximal color quality. In one study, combinations of four LEDs (red, green, blue, warm white) were optimized to reduce the difference in the color appearance of 26 pigments from the frescoes in the Sistine Chapel in the Vatican under a reference daylight illuminant and test light sources of 3400 K (Schanda et al. 2016). The optimal solution had an average color difference of ∆E_UCS,ave = 1.6 and a maximum color difference of ∆E_UCS,max = 3.2, although pigments had varying lightness levels (i.e., red to yellow pigments were higher in lightness and green to blue pigments were lower in lightness under the optimized spectrum; Schanda et al. 2016). In addition to colorimetric computations, visual assessments were carried out by museum curators, who reportedly accepted the optimized spectrum as the ideal solution.

Other researchers considered luminous efficacy and color rendering in the optimization process, where the spectrum of a test light source was optimized using filters instead of mixing narrowband LEDs. Filtered test light sources were found to be significantly beneficial in reducing discoloration after exposing museum artifacts (i.e., old masters’ drawings) to illumination between 1.4 × 10⁶ lx h and 9.8 × 10⁶ lx h (Delgado et al. 2011). In visual assessments, the unfiltered reference illuminant scored highest in overall satisfaction, observers found the reference and test source to be identical in terms of brightness, and the filtered light source performed better in “perception of details” (Delgado et al. 2011, p. 250).

In another study, three-band LEDs were optimized for color rendering, luminous efficacy of radiation (LER), and color fidelity (i.e., decreased shifts in the color appearance of acrylic paints under the optimized LEDs relative to a D65 reference illuminant; Berns 2011). Irradiance, which can cause damage through photochemical action, was successfully reduced and the color appearance of paints was maintained (average ∆E*₉₄ < 1.0) with high LER (between 279 lm W⁻¹ and 340 lm W⁻¹; Berns 2011).

Similarly, other researchers optimized four LED channels (red, amber, green, blue) for damage factor, chromaticity, and color quality, while maintaining a constant irradiance (Tuzikas et al. 2014). The reference point for the relative damage factor (RDF) is considered as RDF = 1.0 for an incandescent illuminant, and lower values indicate reduced damage. Optimized test spectra resulted in decreased damage for low-grade paper ranging from RDF = 0.23 to RDF = 0.43 and from RDF = 0.77 to RDF = 0.89 for oil paint on canvas (Tuzikas et al. 2014). Color quality results were reported using color rendition metrics, and the results were varied (color rendering index R_a = 16 to R_a = 96, color quality scale (CQS) Q_f = 35 to Q_f = 92; Tuzikas et al. 2014). Although color rendition metrics express the general color quality of an object’s appearance under a light source, color difference analysis for individual pigments is needed to communicate color shifts with precision.

2.3. Optimization for absorption

Previous studies utilized damage factor, color quality metrics, and luminous efficiency to optimize spectra instead of reducing the spectral absorption of a museum artifact. A photon causes a photochemical reaction only if it is absorbed by a pigment. A study investigating various museum materials has shown that a pigment’s responsivity is largely dependent on its spectral reflectance function (Saunders and Kirby 1994). Miller (1993) proposed using a reflected energy matching factor, which indicates the extent of the match between the illumination from a light source and reflected light from an artifact. Reflected energy matching is designed to be proportional to the lifetime of a museum object under a light source relative to unfiltered daylight, by matching the “illumination color to the reflected color of the artifact” (Miller 1993, p. 14). However, matching the chromaticity of a light source to the light reflected from the object causes visible color shifts. Shifts in the color appearance caused by changes in the illuminant and light absorption should be quantified by using colorimetric tools.

In a previous study, single-peak theoretical spectra optimized to reduce absorption resulted in energy savings between 38% and 44% without causing perceptible color shifts (Durmus and Davis 2015b). In a follow-up study, double-peak theoretical spectra resulted in higher energy savings (up to 71% (Durmus and Davis 2015a)). In these studies, the color appearance of 15 test samples taken from the CQS (Davis and Ohno 2010) under optimized theoretical spectra and reference illuminants (incandescent and equal-energy illuminator) was compared. In both of these studies, a linear optimization method was used to customize the test SPDs. Then, the perceived naturalness and attractiveness of real objects under commercially available narrowband LEDs and reference phosphor-coated white LEDs were investigated (Durmus and Davis 2017). In these experiments, observers found real objects to be equally natural and attractive under optimized light sources and reference phosphor-coated white LEDs. From this work, a group of researchers has proposed a point-by-point light projection system to reduce damage to artwork caused by optical radiation (Benítez et al. 2016). Another group optimized LED spectra to reduce absorption for multiple reflectance factors, instead of monochromatic samples, using nonlinear optimization tools and reduced energy consumption by 38% without causing color shifts (Zhang et al. 2017). However, absorption minimization studies have not been previously applied to quantify damage to artwork. This study builds on the previous work by focusing on spectral optimization to reduce light absorption, instead of energy efficiency, while maintaining the color appearance of artifacts.

3. Methods

Oil-based paintings have been popular among artists since the 15th century due to the flexibility, longevity, and versatility of oil paints (Jones 2002). Most digital art collections (i.e., Web Gallery of Art, Wikiart Visual Art Encyclopedia, Google Art Project, the Metropolitan Museum of Art New York, the Paul J. Getty Museum, Art UK, etc), which display hundreds of thousands of artworks, own far more oil paintings than other painting types, such as acrylic, gouache, watercolor, or tempera. Oil paintings are also more expensive and sold in larger numbers compared to other paintings thanks to their superior visual properties (i.e., vivid colors; Renneboog and Spaenjers 2013; Worthington and Higgs 2008). For example, some of the most iconic artworks, such as the Mona Lisa by Leonardo da Vinci, The Persistence of Memory by Salvador Dalí, The Birth of Venus by Sandro Botticelli, and Starry Night by Vincent Van Gogh, are oil on canvas. Here five individual oil paints spanning the hue circle (blue, green, orange, yellow, red) have been studied. Each paint was uniformly applied to separate canvases, measuring 61 cm by 76 cm. Because high-chroma objects are most likely to undergo color shifts from lighting (Davis and Ohno 2010), high-chroma oil paints were used. The spectral reflectance factors of the five paintings were measured with a Konica Minolta CM-2600d spectrophotometer. Figure 1 shows the spectral reflectance factors of these paintings.

Fig. 1. Reflectance as a function of wavelength for the five paintings.

The SPDs of seven different channels of Source Four LED Profile ×7 Color System theatrical lights were measured with a Photo Research Spectrascan PR-730 Spectraradiometer. Normalized spectral power distribution curves of the seven channels are shown in Fig. 2.

Fig. 2. Relative power as a function of wavelength for the seven narrowband LEDs that were used in the nonlinear optimization process.

A multi-objective genetic algorithm (MOGA) was used to optimize the combination of seven narrowband LEDs for damage, color quality, and energy efficiency. Genetic algorithms (GAs) are computational optimization methods inspired by the theory of evolution, explaining the origin of species, where unfit species in an environment face extinction and fit ones have a higher probability of passing their genes to future generations by reproduction (Konak et al. 2006). GAs operate by generating an initial random population with a wide range of chromosomes (possible solutions). These chromosomes produce a new generation by crossover and mutation. Fitter solutions finally converge to a single solution. Research has shown that GAs successfully reach near-optimal solutions and they have been used to solve science and engineering problems (Elbeltagi et al. 2005). GAs also perform better than other optimization algorithms, such as particle swarm optimization (Roberge et al. 2013), linear optimization (Azamathulla et al. 2008), and other nonlinear optimization methods (Fowler et al. 2008).

A MOGA is different from a generic single-objective GA in that it promotes solution diversity by including a fitness function (defined by the user) that evaluates solutions in each step (Konak et al. 2006). High-ranking offspring are selected to breed a new generation, and the process continues until a group of solutions satisfying the minimum criteria is found, a fixed number of generations is reached, or the fitness of the highest-ranking solution reaches a plateau.

In multi-objective optimization problems, parameters are usually in a trade-off relationship. When there are several conflicting parameters, an infinite number of optimal solutions exist. The Pareto efficiency (Pareto-optimal solution) is calculated when it is not possible to improve one objective without worsening others (Deb 2001). The Pareto front is the set of all resulting individuals that meet the initial optimization criteria. This method is applied to identify and select alternatives when no single optimal solution exists. Pareto front set and MOGAs have previously been used to optimize light source spectra for color rendition metrics (He et al. 2011; Smet et al. 2012), lighting solutions for uniformity and energy efficiency (Madias et al. 2016), and lighting conditions for visual preference and energy efficiency (Villa and Labayrade 2013).

In this project, Matlab’s controlled elitist multi-objective genetic algorithm function (Deb and Goel 2001), which is a variant of the nondominated sorting genetic algorithm (Deb 2001), is used to optimize light sources and find optimal solutions (test SPDs). Elitism is the method of carrying elite individuals (previously found good solutions) of a population to the next generation to ensure that the fitness of the best-solution population (i.e., Pareto front) does not deteriorate (Deb 2001). The controlled elitist approach restricts the number of elite individuals to increase the successful convergence to a better Pareto front or to prevent a premature convergence to a suboptimal solution set (Deb and Goel 2001). Matlab’s default Pareto fraction (35%), which is the percentage of elite individuals in the Pareto front, was adopted in the optimization process. A MOGA using elitist strategies has been found to outperform nonelitist MOGAs (Van Veldhuizen and Lamont 2000; Zitzler and Thiele 1999). Because visibility and damage to artwork caused by visible radiation are inversely related, a controlled elitist MOGA stands out as an ideal tool to optimize light source spectra to achieve optimal results where optimization parameters are in a conflicting relationship. The optimization problem can be stated as

(1) $\left\{\begin{array}{c}optimize\\ subjectto\\ \end{array}\begin{array}{c}f{\left(x_i\right)}_{}=y\\ g_j\left(y\right)\le z\\ x\in \left[0,100\right],i=7\end{array}\right.$(1)

where f(x_i) is the optimization function that generates test SPDs, x is the power of each LED channel as a percentage, subscript i denotes the number of narrowband LEDs, y is the test SPD, g_j(y) is the fitness function(s), subscript j denotes the number of fitness functions, and z is the fitness function constraint. The optimization function is

(2) $f\left(x_i\right)=\sum _i{S}_i\left(\lambda \right)x_i$(2)

where S_i(λ) is the SPD of the ith narrowband LED channel. The optimization function, f(x_i), is equal to the test SPD, y = S_test(λ), and y is the input of the fitness function. Five fitness functions (four inequalities and one equality) were used to ensure that light absorption, energy consumed by lighting, and the shift in the color appearance of artwork were minimal. The nonlinear inequality functions were color difference, hue difference, light absorption, and energy consumption. Color difference is widely used in conservation studies to quantify the magnitude of change (damage) caused by optical radiation (CIE 2004a; Schaeffer 2001) and to evaluate the impact of a light source on the color appearance of objects (Houser et al. 2016; Jost-Boissard et al. 2015; Royer et al. 2017). In addition to the color difference, hue shifts were reported to provide more specific information about the color quality of the artwork. Because hue discrimination thresholds are lower than saturation discrimination thresholds (Danilova and Mollon 2016), hue shift calculations were added to capture the most essential dimension of the color quality of the artwork under optimized lighting conditions. Color (∆E₀₀) and hue (∆H′) differences were calculated using CIE 1976 (L*, a*, b*) and CIEDE2000 color difference formulae (CIE 2004b). Although ∆E₀₀ = 1.0 approximates a just-noticeable difference, CIE’s color difference magnitude recommendation for using CIEDE2000 (∆E₀₀ < 5; CIE 2004b) was used as a fitness function constraint in the optimization process.

Hue is the angular component of color in three-dimensional color spaces, ranging from 0° to 360°. Hue difference is calculated as

(3) $\mathrm{\Delta }{H}^{\mathrm{\prime }}=2\sqrt{C_1^{\mathrm{\prime }}{C}_2^{\mathrm{\prime }}}sin\left(\frac{\mathrm{\Delta }{h}^{\mathrm{\prime }}}{2}\right)$(3)

where $C_1^{\mathrm{\prime }}$ and $C_2^{\mathrm{\prime }}$ are chroma values for the reference illuminant and test SPD, respectively, and $\mathrm{\Delta }{h}^{\prime }$ is the hue angle difference between test and reference light sources (Sharma et al. 2005). Though hue difference ∆H′ has no units (similar to ∆E₀₀), in contrast to the overall color difference (∆E₀₀), hue difference (∆H′) is a signed quantity that indicates the direction of hue change. Although there is no established just-noticeable difference for hue difference, research on dental ceramics suggests that the visual acceptability threshold for hue difference can be considered ∆H′ = 1.9 with 95% confidence (Perez et al. 2011). Slightly higher values for hue shifts were allowed in the optimization process, −5.0 < ∆H′ < 5.0. This was to account for the fact that experimentally derived thresholds are more conservative than is needed in typical illumination applications and to accommodate a wider range of optimal solutions.

Light absorption (A) was calculated as the ratio of the light absorbed by the paint under a test SPD to the light absorbed by the paint under a reference incandescent illuminant,

(4) $A=100\frac{\int S_{test}\left(\lambda \right)\left(1-R\left(\lambda \right)\right)d\lambda }{\int S_{ref}\left(\lambda \right)\left(1-R\left(\lambda \right)\right)d\lambda }$(4)

where S_test($\lambda )$ is the test SPD, S_ref($\lambda )$ is the SPD of the reference incandescent illuminant, and $R\left(\lambda \right)$ is the reflectance factor of paint. A constant (100) was introduced to scale the values, so that they can be reported as a percentage. A value less than 100% denotes reduced light absorption, and it was chosen as the third nonlinear inequality function constraint because absorbed light causes damage to artwork.

The fourth inequality function was energy consumption (EC), which was reported as the ratio of the integrated test SPD to the integrated reference incandescent SPD,

(5) $EC=100\frac{\int S_{test}\left(\lambda \right)d\lambda }{\int S_{ref}\left(\lambda \right)d\lambda }$(5)

and reported as a percentage. The energy consumption was limited to 100% in the optimization algorithm. Although energy consumption was not the primary focus of this study, there is a growing need to reduce the energy consumed by lighting to decrease the negative impacts of electricity use on the environment and economy (International Energy Agency 2014; Kassakian et al. 2017). The EC was reported to consider this issue and compare results with previous optimization studies.

Luminance was the only equality fitness function, which was introduced to prevent reductions in light absorption simply from reductions in painting luminance, as well as prevent color appearance shifts from the Hunt effect and the Bezold-Brucke hue shift. The total reflected light from the painting under a test SPD (ΣL_test) and a reference illuminant (ΣL_ref) was set equal. The MOGA was run with these five constraining fitness functions until optimal solutions were reached.

4. Results and discussion

The resulting optimal Pareto front sets were graphed, as shown in Fig. 3. The x-axis shows color difference values (∆E₀₀ < 5) and the y-axis shows the percentage of light absorption (A_test < 100%). The Pareto front of optimal solutions in Fig. 3 indicates an inverse relationship between color difference and light absorption (i.e., absorption is reduced when color difference is increased). Although there are other constraining functions (hue difference, energy consumption, and luminance), a strong relationship between color difference and absorption was recorded, especially for the green paint (R² = 0.97). The data for other paintings demonstrate similar, but weaker, trends (R² = 0.71 for blue paint, R² = 0.68 for yellow paint, R² = 0.45 for orange paint, and R² = 0.63 for red paint). However, the linear relationship between the first two constraints is insufficient to formulate a universal relationship between color quality and damage, due to the existence of other color characteristics such as hue and chroma shifts and differences in the spectral reflectance factor of objects.

Fig. 3. Optimal Pareto sets for five paintings. Percentage of light absorption as a function of color difference values (∆E₀₀). Linear regression (gray dotted line) formulae and the coefficient of determination (R²) values indicate the goodness of fit of a linear relationship between color difference (∆E₀₀) and light absorption.

In the Pareto set, the lowest average light absorption percentages were recorded for green and blue paints, approximately A_blue,ave = 53% and A_green,ave = 54%, as shown in Fig. 3. Average light absorption percentages were higher for yellow, orange, and red paints, approximately 68%, 74%, and 77%, respectively. Optimized test light SPDs resulted in similar average energy consumption values for all of the paints, from EC_ave = 52% to EC_ave = 55%. However, the optimal Pareto set included data points for SPDs that would cause perceptible color and hue shifts. To address this, color and hue differences were limited to ∆E₀₀ > 1.0 and ∆H′ > 1.9 within the optimal Pareto set. Then, optimized test SPDs that resulted in minimal light absorption and energy consumption for each paint were selected.

The optimized spectrum that results in the lowest light absorption and energy consumption for the blue paint is shown in Fig. 4. Light absorption (A_blue,opt = 53%) and energy consumption (EC_blue,opt = 53%) were reduced without causing perceptible color (∆E₀₀ = 0.6) or hue shifts (∆H′ = −0.9), compared to the reference incandescent illuminant, when the combination of seven narrowband LEDs was optimized for the blue paint.

Fig. 4. Test SPD (gray continuous line, left y-axis) optimized for blue paint (black dashed line, right y-axis), resulting in a 47% decrease in both light absorption and energy consumption without causing perceptible color (∆E₀₀ = 0.6) or hue shifts (∆H’ = -0.9), compared to a reference incandescent illuminant (black continuous line, left y-axis).

The values shown in Table 1 describe the optimized test SPDs for each painting for which light absorption and energy consumption are minimized without causing perceptible color shifts. Among all of the paints, blue and green paints absorbed the least light, A_test,blue = 53% and A_test,green = 55%. Other paints had relatively higher light absorption, between 77% and 81%. These three paints also have higher reflectance factors; therefore, they have higher lightness than the blue and green paints. The average absorption percentages for blue and green paints (A_blue,ave = 53% and A_green,ave = 54%) were almost identical to their optimized percentages (A_blue,opt = 53% and A_green,opt = 55%). On the other hand, the differences between the average light absorption (A_yellow,ave = 68%, A_orange,ave = 74%, and A_red,ave = 77%) and optimized light absorption (A_yellow,opt = 77%, A_orange,opt = 80%, and A_red,opt = 81%) for yellow, orange, and red paints were larger. Allowing small, but likely not disturbing, color differences (∆E₀₀ < 5) can further reduce light absorption and energy consumption for these three paints. Energy consumption percentages for the optimal test SPDs (from EC_opt = 53% to EC_opt = 57%) were analogous to the average values (from EC_ave = 52% to EC_ave = 55%). Luminous efficacy of radiation of the optimized SPDs ranged between 325 lm W⁻¹ and 379 lm W⁻¹.

Table 1. Spectral characteristics of the optimized test SPDs that resulted in imperceptible color, hue, and chroma shifts, as well as minimal light absorption and energy consumption.

						Relative peak (%)
Paint	∆E00	A (%)	EC (%)	∆H′	LER (lm W^−1)	Ch. 1	Ch. 2	Ch. 3	Ch. 4	Ch. 5	Ch. 6	Ch. 7
Blue	0.6	53	53	−0.9	379	11.9	62.5	97.1	16.7	3.3	14.4	57.0
Green	1.0	55	56	0.1	334	37.1	50.2	91.2	11.6	18.3	32.7	52.3
Yellow	0.7	77	57	−0.1	329	72.5	18.7	66.5	11.0	28.2	14.3	34.7
Orange	0.9	80	57	−0.7	325	91.0	22.0	48.7	15.7	9.1	18.4	33.5
Red	0.8	81	57	0.8	353	80.3	11.3	92.0	25.7	6.3	16.7	41.5

5. Conclusion

In this research, a multi-objective genetic algorithm was used to optimize combinations of narrowband LEDs to minimize the optical radiation absorbed by paintings. Optimized spectra resulted in reduced energy consumption and absorption and therefore reduced damage caused by radiation. Results suggest that energy consumption can be nearly halved if narrowband light sources are optimized for object reflectance while preserving color appearance. There was a large variance in absorption values due to the difference in the spectral reflectance characteristics of each paint. Although light absorption for yellow, orange, and red paints was lower than that for other paints, the light absorbed by these paintings is predominantly of short wavelengths, which is more damaging. Therefore, the absolute damage reduction is greater than that captured in the relative values presented here.

It is also important to note that this study investigates absorption-minimizing spectra for single-colored objects. Because most artworks have a varying degree of chromatic complexity, the lighting system suggested here requires precise optics for high-accuracy projection. Other researchers are already investigating the feasibility of such a light projection system (Benítez et al. 2016). Future research will investigate the spatial resolution of the proposed absorption-minimizing lighting systems and the visual appreciation of museum artifacts under spectrally optimized light sources.

Disclosure statement

The authors have no interests to declare.

Additional information

Funding

The author reports no funders for the preparation of this article.