Comparative Analysis of Low-Cost Portable Spectrophotometers for Colorimetric Accuracy on the RAL Design System Plus Color Calibration Target

Abstract

Novel low-cost portable spectrophotometers could be an alternative to traditional spectrophotometers and calibrated RGB cameras by offering lower prices and convenient measurements but retaining high colorimetric accuracy. This study evaluated the colorimetric accuracy of low-cost, portable spectrophotometers on the established color calibration target—RAL Design System Plus (RAL+). Four spectrophotometers with a listed price between USD 100–1200 (Nix Spectro 2, Spectro 1 Pro, ColorReader, and Pico) and a smartphone RGB camera were tested on a representative subset of 183 RAL+ colors. Key performance metrics included the devices’ ability to match and measure RAL+ colors in the CIELAB color space using the color difference CIEDE2000 ΔE. The results showed that Nix Spectro 2 had the best performance, matching 99% of RAL+ colors with an estimated ΔE of 0.5–1.05. Spectro 1 Pro and ColorReader matched approximately 85% of colors with ΔE values between 1.07 and 1.39, while Pico and the Asus 8 smartphone matched 54–77% of colors, with ΔE of around 1.85. Our findings showed that low-cost, portable spectrophotometers offered excellent colorimetric measurements. They mostly outperformed existing RGB camera-based colorimetric systems, making them valuable tools in science and industry.

1. Introduction

Colorimetry quantifies and studies human perception of colors. If two objects are perceived as having the same color, they shall have the same colorimetric value [1]. Conversely, measured color differences should correlate with the visual color difference. Colorimetry is, thus, crucial in industrial quality control, where multiple batches of the same product are manufactured, for example, in textiles, paints, printing, and cosmetics [1].

Color perception depends on (1) a light source, (2) object properties affecting light reflectance or transmittance, and (3) an observer [2,3,4]. Predefined light sources, called standard illuminants, exhibit specified spectral power distribution. Some of the most popular standard illuminants [5] are the International Commission on Illumination’s (CIE) illuminants: A (related to a tungsten filament lamp), D50 (horizon light), and D65 (daylight). The CIE-defined observer (with a 2° or 10° visual field) is based on light-mixing and color-matching experiments [1]. Mathematically, perceived color is a sum of three tristimulus values: X, Y, and Z. Each tristimulus value is a product between a spectral color-matching function of the chosen standard observer ($\overline{x}\left(\lambda \right)$, $\overline{y}\left(\lambda \right)$, or $\overline{z}\left(\lambda \right)$), power distribution of the illuminant S(λ), and the object’s reflectance or transmittance spectrum R(λ).

The popular CIELAB color space transforms tristimulus values into coordinates L* (lightness), a* (opponent colors: red-green), and b* (yellow-blue), providing a more visually uniform color space [1]. L* varies between 100 (perfect white) and 0 (perfect black). The value of a* is positive for red and negative for green shades. Similarly, b* is positive for yellow and negative for blue colors. The first attempts to quantify the difference between two colors (ΔE) in the CIELAB color space were based on a simple Euclidian distance: ${\mathsf{\Delta }E}_{CIE76}=\sqrt{\mathsf{\Delta }{L}^{*2}+\mathsf{\Delta }{a}^{*2}+\mathsf{\Delta }{b}^{*2}}$. However, the CIELAB color space was less perceptually uniform than intended, particularly close to saturated values. Thus, the newest color differences formula of CIEDE2000 (ΔE₀₀) addressed perceptual non-uniformities by replacing the a* and b* coordinates with chroma (C*) and hue (h*), and adding several corrections to better quantify human eye perception [6]. The human eye can spot color differences of 2.3–5.0, called the just-noticeable difference (JND) [7].

To determine a specific color, spectrophotometers acquire missing information, i.e., light reflectance or the transmittance spectrum. The spectrophotometer’s applicability, functionality, and portability are defined by its geometry, specified by the light source and observer’s angles. Most commonly, 45°/0° or 0°/45° geometry is used for matt surfaces due to excluded gloss [8]. An alternative for shiny surfaces is an integrating sphere spectrometer, which provides diffuse illumination and collects the light at an 8° angle (i.e., diffuse/8° or d/8° in short). Finally, multi-angle spectrophotometers are used for more complex surfaces or industrial requirements.

There are several portable spectrophotometers on the market. The scientific community often considers CM-700d (Konica Minolta KK, Tokyo, Japan) a reference when calibrating colorimetric systems or performing colorimetric measurements [9,10]. CM-700d includes the d/8° geometry, pulsed xenon lamp, and 31 light-collecting channels (10 nm steps between 400–700 nm). X-rite (Grand Rapids, MI, USA) is another popular manufacturer of portable spectrophotometers such as Ci64 (d/8°) or eXact™ (45°/0°). The listed inter-instrument agreement of these high-end state-of-the-art spectrophotometers is around 0.2 ΔE₀₀. However, standard spectrophotometers are relatively expensive, with listed prices between USD 5000 and 10,000. Furthermore, they provide an average measurement at a single point, limiting their feasibility in the colorimetric study of an area.

We can also colorimetrically calibrate commercial RGB cameras to achieve lower costs and spatial measurements. Color calibration relies on a model that maps measured (RGB) colors with the standard color space, such as CIELAB. The model is trained on color calibration targets (CCTs) or color reference charts, which can be in a single-page or fan format. The most popular CCTs are ColorChecker Classic (X-rite, Grand Rapids, MI, USA) [11,12], IT8 Targets (LaserSoft Imaging AG, Kiel, Germany) [13], RAL Design System Plus (RAL+, RAL gGmbH, Bonn, Germany, Figure 1) and Pantone (Pantone LLC, Carlstadt, NJ, USA). In biomedicine, various colorimetric systems with calibrated (smartphone) RGB cameras achieved colorimetric accuracy (ΔE) between 2.2 and 8.4 [11,12,14,15,16,17]. The color constancy of the existing systems is not optimal because most of these color differences could be spotted by the human eye [7].

The recently available low-cost (USD < 1200) and portable spectrophotometers could fill the gap between standard spectrophotometers (=high colorimetric accuracy) and RGB camera-based colorimetric systems (=affordability). Popular low-cost spectrophotometers, such as Color Muse (Variable Inc., Chattanooga, TN, USA), Nix Mini 2 or Pro (Nix Sensor Ltd., Hamilton, ON, Canada), and Cube (Palette Pty Ltd., Melbourne, Victoria, Australia), have been applied to the monitoring of food quality, environment, and radiation [18,19,20,21]. Compared to the standard spectrophotometers on food and soil, their colorimetric accuracy (ΔE₀₀) was, on average, between 4.3 and 11.4 [20,21]. The color differences decreased to 2.1–6 and ~1.0–1.7 * (* our estimates) when low-cost spectrophotometers were tested on custom-made and RAL+ CCTs, respectively [22,23].

Most existing studies tested the colorimetric accuracy of low-cost spectrophotometers on non-standardized objects (food, soil, custom-made CCT). Furthermore, new versions of affordable devices like Nix Spectro 2 (Nix Sensor Ltd., Hamilton, ON, Canada) and Spectro 1 Pro (Variable Inc., Chattanooga, TN, USA) feature resolution with 31 channels and low inter-instrument agreements (ΔE₀₀ < 0.35), comparable to the standard spectrophotometers. Therefore, in this study, we tested the colorimetric accuracy of four new low-cost, portable spectrometers in matching and evaluating RAL Design System Plus (RAL+) colors. The results provided important insights into the performance of low-cost, portable spectrophotometers, especially compared to the custom-made colorimetric systems based on the calibrated RGB cameras. This study helps professionals and researchers in fields such as biomedicine, food hygiene, and environmental science find accessible but accurate tools for colorimetry.

2. Materials and Methods

2.1. Spectrophotometers

This study enrolled four spectrophotometers (

Table 1. Low-cost portable spectrophotometers evaluated in this study.

Spectrophotometer	Nix Spectro 2 a	Spectro 1 Pro b	ColorReader c	Pico d
Price (USD)	~1200	~300	~130	~120
Resolution	31 channels (10 nm steps between 400–700 nm)	31 channels (10 nm steps between 400–700 nm)	3 (RGB)	3 (RGB)
Geometry	45° (ring)/0°	Diffuse/0°	~35° (ring)/0° e	~35°/~35° e
Illumination	8 CRI LEDs	Full-spectrum LEDs	6 CRI LEDs	3 (LED) floodlights
Illuminants	A, C, D50, D55, D65, D75, F2, F7, F11	A, F2, D50, D65	/	/
Observer	2°, 10°	2°, 10°	/	/
Calibration steps	1	3	1	1
Claimed average inter-instrument agreement	0.35 ΔE00	0.35 ΔE00	/ f	/ f
Libraries	Several (incl. RAL, Pantone, NCS)	Several (incl. RAL, Pantone, NCS)	Several (incl. RAL, NCS)	Several (incl. RAL, NCS)

^a Nix Sensor Ltd., Hamilton, ON, Canada, www.nixsensor.com/nix-spectro-2/, ^b Variable Inc., Chattanooga, TN, USA, www.variableinc.com/spectro-1-pro.html, ^c Datacolor GmbH, Marl, Germany, www.datacolor.com/colorreader/products/colorreader, ^d Palette Pty Ltd., Melbourne, Victoria, Australia, palette.com.au/pico, ^e our estimation according to the illuminator/sensor geometry, ^f not found.

, Figure 2), covering a price range between USD 100 and 1200. The devices typically require Bluetooth connection with smartphones to display data and alter spectrophotometer settings. Smartphone applications provide details on colorimetric measurements like color values (e.g., CIELAB, sRGB, etc.) and the closest matches with the corresponding ΔE to standard CCT colors (e.g., RAL, Pantone). This study also included a smartphone Asus Zenfone 8 (ASUSTeK Computer Inc., Taipei, Taiwan) combined with a DermLite DL1 dermatoscope (DermLite LLC, Aliso Viejo, CA, USA) to guarantee stable illumination conditions [11].

2.2. Measurements

Colorimetrical accuracy was tested on RAL+ colors, which are labeled with seven-digit codes (HHH LL CC) representing hue (H), lightness (L), and chroma (C). RAL+ includes 1825 colors, covering most of the CIELAB color space (i.e., H: 0–360°, L: ~20–90, C: ~0–70). However, the RAL+ manufacturer provides only color light reflectance values(LRV) [24], indicating that the “real” RAL+ colors differ from their labeled CIELAB values. Therefore, some portable spectrophotometer manufacturers evaluated actual RAL+ color values independently, and their colorimetric assessments are sometimes directly available in smartphone applications that control the portable spectrometers.

For the sake of clarity, we employ the following terminology for the CIELAB values (L*, a*, b*) of RAL+ colors:

• Labeled values were calculated from the RAL+ color label (e.g., 010 60 15 translates to 60, 14.8, 2.6),
• Assessed values are the actual RAL+ color values, as evaluated by the portable spectrophotometer manufacturers (e.g., Spectro 1 Pro manufacturer foresaw 62.1, 14.3, 2.0 for the RAL+ color 010 60 15),
• Measured values were estimated in this study using portable spectrophotometers or a smartphone (e.g., 59.4 15.2 2.4 for the RAL+ color 010 60 15 by Nix Spectro 2 spectrophotometer),
• Normalized are measured values, normalized to labeled values (as defined in Section 2.4).

To rationalize measurements, we downsampled the number of enrolled colors to 183. By using random permutation and a 10-fold decrease (randperm function, Matlab R2017b, MathWorks, Natick, MA, USA), we retained the original RAL+ color distribution. Downsampling kept the differences in the L*, a*, and b* value histograms (10-unit wide bins) between entire and downsampled RAL+ colors below 1.1%. On the other hand, we acquired all 1825 RAL+ colors by the smartphone to prevent model overfitting when transforming measured colors into labeled ones (Section 2.4). We also excluded 27 smartphone measurements due to image saturation.

First, we calibrated the spectrophotometers according to the manuals. The color measurements were conducted in the ascending order of H, L, and C values in one continuous session. For each RAL+ color, we recorded (1) the measured CIELAB value, (2) the matched RAL+ color, and (3) ΔE₀₀ to the assessed RAL+ color (the Pico spectrophotometer did not provide ΔE).

2.3. RAL+ Color-Matching Accuracy

The strict RAL+ color-matching accuracy was calculated as a ratio between correctly matched and all enrolled colors (n = 183). For ColorReader, we excluded two faulty measurements from further analysis (n = 181). The Pico spectrophotometer did not offer color matching; thus, we selected the color with a minimal ΔE₀₀ between the measured and labeled RAL+ color values. The same procedure was applied to the smartphone. Since the Spectro 1 Pro did not display information on 19 (assessed) RAL+ colors, we additionally evaluated color-matching accuracies excluding these missing colors. On the other hand, Nix Spectro 2 and ColorReader presumably included all colors.

Loose color-matching accuracy additionally accepted the adjacent RAL+ color as the correct one. We defined an adjacent color as gradually changing in only one coordinate, i.e., hue, lightness, or chroma (for example, 020 30 30 vs. 030 30 30). These looser metrics are practical because most spectrophotometers match RAL+ colors according to their assessed and not labeled values. According to the Spectro 1 Pro database, the median difference between labeled and assessed RAL+ values is larger (ΔE₀₀ = 2.6) than between adjacent RAL+ colors (ΔE₀₀ = 2.1).

2.4. Color Difference

Color differences (ΔE₀₀) were calculated between the measured and (1) labeled or (2) assessed RAL+ color values. However, we should note that the labeled RAL+ CIELAB color values are based on a particular measurement geometry. Therefore, the color difference between the measured and labeled colors cannot be a simple measure of spectrophotometer performance but rather an indicator of similarities between the device and RAL+ measurement geometry. Thus, we additionally normalized the measured against the labeled values and then calculated ΔE₀₀ between them.

Normalization was based on a regression model [11] estimating the relationship between the measured (C_m) and labeled RAL+ CIELAB color values (C_RAL):

C_RAL = f(C_m), C_RAL ∈ {L_RAL*, a_RAL*, b_RAL*}

(1)

which served to normalize C_m:

C_{m_norm} = f(C_m).

(2)

The spectrophotometers measured color values (C_m) in the L*, a*, and b* coordinates, while the smartphone retrieved the RGB values. Potential overfitting on only 183 samples was avoided by (1) training and testing the model with a leave-one-out-cross-validation (LOOCV) approach and (2) choosing a simple linear function for f:

C_RAL = p₀ + p₁ L_m* + p₂ a_m* + p₃ b_m*, C_RAL ∈ {L_RAL*, a_RAL*, b_RAL*}.

(3)

The LOOCV approach enabled the same 183 RAL+ colors to be used for the regression model’s training and the color difference calculation between the normalized and labeled values.

On the other hand, the smartphone normalization model was trained on 1798 RAL+ colors (27 saturated images were excluded) but tested only on 183 selected colors. Since overfitting was not likely, a more complex polynomial was selected for f:

$$C_{RAL}={\sum }_{k=0}^3{p}_k{C}^k+p_{11}RG+p_{12}RB+p_{13}GB, C\in \left\{R, G, B\right\}$$

(4)

3. Results and Discussion

3.1. Color-Matching Accuracy

First, we evaluated the accuracy of four spectrophotometers and a smartphone in matching 183 RAL+ colors (

Table 2. Accuracies (in %) of four spectrophotometers and a smartphone in matching 183 RAL Design System Plus (RAL+) colors when (a) only studied colors (“Strict”), or (b) also adjacent colors (i.e., being gradually different in only one coordinate, i.e., hue, lightness, or chroma) were considered true (“Loose”).

Color-Matching Accuracy (%)	Nix Spectro 2	Spectro 1 Pro	ColorReader a	Pico b	Smartphone b
(a) Strict	98.4	76.0 (86.3 c)	78.5	24.0	47.5
(b) Loose	99.5	86.3	86.2	53.6	76.5

^a Two faulty measurements were excluded from the calculations (n = 181). ^b Matching was based on the minimal ΔE₀₀ to the labeled RAL+ colors. ^c Matching accuracy when 19 missing RAL+ assessed colors were dismissed.

). The strict color-matching accuracies ranged from 47.5 to 98.4%. Nix Spectro 2 performed the best with only three mismatched RAL+ colors. However, strict color-matching criteria could be too rigorous because the median difference between two random adjacent RAL+ colors (i.e., gradually different in only one coordinate) is smaller than that between the labeled and assessed RAL+ colors (see Section 2.3). Therefore, loose color-matching accuracy (i.e., adjacent RAL+ colors are also a correct match) can be considered a more practical measure.

With the loose approach, the color-matching accuracies significantly improved, ranging between 72.1% and 99.5%. Again, Nix Spectro 2 performed the best with almost perfect execution (99.5%). The only mismatched RAL+ color was opulent green, mistaken for intense green, which gradually differs in two coordinates (i.e., 160 20 20 vs. 170 20 15). Since the color difference (ΔE₀₀ = 3.74) between both labeled colors falls in the JND range [7], the color shades would probably be distinct to the human eye. Extrapolating its accuracy to the whole RAL+, Nix Spectro 2 would mismatch around nine RAL+ colors (out of 1825).

On the other hand, Spectro 1 Pro and ColorReader correctly matched ~86% of RAL+ colors. The misidentifications accounted for 25 colors (presumably ~250 in the whole RAL+ range). For these 25 mismatched colors, the median difference between the assessed and measured RAL+ colors was 4.1 for Spectro 1 Pro. The calibrated smartphone camera and Pico mismatched ~25% and ~45% RAL+ colors, respectively. We assume that a matching accuracy of around 75% can also be expected from the other calibrated smartphone- or camera-based colorimetric systems [12,14,16] since all tested smartphone cameras from our previous studies performed similarly [11,25]. Furthermore, Pico does not provide assessed colors. Thus, the color-matching accuracy was calculated according to the labeled RAL+ colors, which can be a contributing factor to an inferior performance.

We shall also alert readers to the proper selection of spectrophotometer settings for color matching. Nix Spectro 2 and Spectro 1 Pro enable different illuminants and observers. For example, when a different illuminant (D65 vs. D50) was selected for Nix Spectro 2, its loose matching accuracy dropped to only 88.0% (161/183), which is significantly worse than with the correct settings (Table 2). Additionally, metamerism may hinder a high-performing spectrophotometer from accurately matching the RAL+ color that visually aligns best with the unknown painted object.

3.2. Accuracy in ΔE₀₀

The most straightforward approach to colorimetrically validate spectrophotometers would be based on the standardized color difference (ΔE₀₀) between the measured and actual RAL+ color values. However, the actual RAL+ color values in the CIELAB color space remain unknown. Although the spectrophotometer’s color-matching capability (Section 3.1, Table 2) can be considered more of a qualitative than quantitative measure, these can serve to estimate an approximate colorimetric accuracy. With the almost perfect color-matching accuracy of the Nix Spectro 2 spectrophotometer, its absolute colorimetric accuracy shall be below ΔE₀₀ = 1.05, that is, 50% of the typical color difference between two RAL+ adjacent colors. Similarly, the equivalent estimate for Spectro 1 Pro was ΔE₀₀ < 1.19, based on 86.3% color matches and 13.7% mismatches exhibiting ΔE₀₀ of 4.1.

In order to verify these estimates further, we calculated median color differences (ΔE₀₀) between the labeled, assessed, measured, and normalized RAL+ CIELAB color values as measured by four spectrophotometers and a smartphone (

Table 3. Color differences (ΔE₀₀ median and, below in the squared brackets, minimum, 25th, 75th percentile, maximum) for (a) measured–labeled, (b) measured–assessed, and (c) normalize–labeled RAL+ colors (labeled: RAL+ color label, assessed: evaluated by the device manufacturers, measured: estimated by the spectrometers, normalized: mapped to the labeled RAL+ colors).

Comparison	Nix Spectro 2	Spectro 1 Pro	ColorReader	Pico	Smartphone
(a) Measured vs. labeled *	1.87 [0.38 1.32 2.42 4.34]	1.79 [0.42 1.04 3.59 9.41]	1.27 [0.33 0.84 1.83 7.19]	3.95 [1.23 2.64 5.88 12.84]	/ b
(b) Measured vs. assessed	0.52 [0.16 0.43 0.66 1.30]	1.07 [0.21 0.74 1.53 3.63]	1.39 [0.01 0.98 1.83 4.31]	/ a	/ b
(c) Normalized vs. labeled	1.01 [0.19 0.73 1.43 3.83]	1.19 [0.14 0.86 1.60 7.14]	1.11 [0.20 0.77 1.61 6.33]	1.88 [0.26 1.23 2.85 7.37]	1.84 [0.20 1.21 3.21 8.51]

* The difference between measured and labeled colors is a limited measure for colorimetric accuracy (see Section 2.4), ^a Pico does not provide assessed RAL+ color values. ^b Not applicable.

). The differences between the measured and labeled ranged between 1.27 and 3.95. The lowest color difference (ΔE₀₀ = 1.27) was provided by the ColorReader spectrophotometer, whose measurements seem adjusted to the labeled RAL+ color values the most. This hypothesis was confirmed by the fact that its color difference increased to 1.39 when the measured and assessed colors were compared. Expectedly, the detected color differences for Nix Spectro 2 and Spectro 1 Pro improved significantly to 0.52 and 1.07, respectively.

Our results showed better performance for the newly tested Nix Spectro 2, while for the other devices, they are consistent with a previous study [23], which reported a colorimetric accuracy of ~1.0–1.7* (*our estimates from ΔE_CMC) for low-cost spectrophotometers. When devices were tested on visually heterogeneous objects (soil or food), ΔE₀₀ increased to 4.3–11.4 [20,21]. However, there were significant colorimetric disagreements of up to 3.8 even between standard instruments. With deducted inter-instrument color differences, the true colorimetric accuracy was probably lower, assumingly around 0.5–7.6, which is in the range of our results (Table 3).

However, relying solely on the differences between measured, labeled, and assessed colors does not allow for a colorimetric comparison with calibrated RGB cameras. Therefore, we also compared the normalized and labeled RAL+ colors, resulting in a ΔE₀₀ slightly above 1.1 for Nix Spectro 2, Spectro 1 Pro, and ColorReader, and around 1.85 for Pico and smartphone (Table 3(c)). Against the measured–labeled comparison (Table 3(a)), the color accuracy improved for around ~0.2–0.8, which is in line with the study of Kirchner et al. [23], where the normalization model improved the spectrophotometer colorimetric accuracy by ~0.7.

We shall stress that the smartphone’s median colorimetric accuracy ΔE₀₀ of 1.84 is one of the best reported. The other studies, including commercial smartphones or cameras, achieved a colorimetric accuracy above 2.2 [12,14,16]. The improved performance probably resulted from a complex calibration model trained on almost 1800 colors. Despite being effective, this approach was time-consuming and burdensome.

3.3. Overall Performance

The devices enrolled in this study exhibited three distinct colorimetric performances (

Table 4. Overall colorimetric performances (matching and evaluating RAL+ colors in the CIELAB color space) and listed prices of four spectrophotometers and a smartphone used in this study.

Device	RAL+ Color Matching	ΔE00  a	Price
(1) Nix Spectro 2	almost perfect (~99%)	0.52–1.05	~USD 1200
(2) Spectro 1 Pro, ColorReader	very good (~85%)	1.07–1.39	USD 130–300
(3) Pico, smartphone Asus 8	good (~54–76%)	~1.85	USD 120–400 b

^a Estimated from matched, assessed, and normalized RAL+ colors (Section 3.1 and Section 3.2). ^b USD 400 is the price of dermatoscope DL1, needed for colorimetric calibration of a smartphone.

). Nix Spetro 2 was the best-performing spectrophotometer, with almost perfect color matching and high colorimetric accuracy in evaluating RAL+ colors (ΔE₀₀ of mostly <1). On the other hand, it comes with the highest price, more than four times that of other spectrophotometers. Moreover, color matching requires a subscription, further widening the device price gap.

Spectro 1 Pro and ColorReader correctly matched around 85% RAL+ colors. However, absolute colorimetric accuracy (ΔE₀₀) is probably slightly above 1. ColorReader could seem the obvious choice between these two devices due to its lower price, but the spectrophotometer stopped working shortly after our study, raising concerns about its durability and quality. On the other hand, Spectro 1 Pro included small magnets near the sensor opening to secure the protective cap. Because the magnets were installed slightly above the casing level, environmental light could penetrate onto firm surfaces like RAL+ sheets.

Pico and a smartphone camera exhibited similar performances—their median colorimetric error was just below JND, and, thus, undetectable by the human eye. Although smartphones are ubiquitous, proper colorimetrical calibration is time-consuming and burdensome, and the process demands add-ons to ensure stable illumination (dermatoscopes) and reference colors (color calibration targets). However, the most significant advantage of commercial (smartphone) cameras is spatial colorimetric imaging, which can reveal colors in smaller areas than the size of a typical spectrophotometer measurement aperture (e.g., the minimal diameter of Nix Spectro 2 opening is 2 mm).

4. Conclusions

In conclusion, our study indicated that the colorimetric performance of low-cost, portable spectrophotometers correlated with their price. Furthermore, cheaper devices (<USD 300) also performed very well, beating the colorimetric accuracy of a calibrated RGB camera. Therefore, instead of relying on custom-made optical setups, future biomedical studies on color measurement shall turn to commercial, low-cost, portable spectrophotometers to achieve optimal color constancy. The notable exception is when spatial measurements are needed. In the latter, color-mapping models shall be performed on numerous CCT colors.