Non-Invasive Use of Imaging and Portable Spectrometers for On-Site Pigment Identification in Contemporary Watercolors from the Arxiu Valencià del Disseny

Abstract

Imaging techniques have revolutionized cultural heritage analysis, particularly for objects that cannot be sampled. This study investigated the utilization of spectral imaging for the identification of pigments in artifacts from the Arxiu Valencià del Disseny, in conjunction with other portable spectroscopy techniques such as XRF, Raman, FT-NIR, and FT-MIR. Four early 1930s watercolors were examined using point-wise elemental and molecular spectroscopic data for pigment classification. Initially, the data cubes obtained with the spectral camera were processed using various methods. The spectral behavior was analyzed pixel-point, and the reflectance curves were qualitatively compared with a set of standards. Subsequently, a computational approach was applied to the data cube to produce RGB, false-color infrared (IRFC), and principal component (PC) images. Algorithms, such as the Vector Angle (VA) mapper, were also employed to map the pigment spectra. Consequently, 19th-century pigments such as Prussian blue, chrome yellow, and alizarin red were distinguished according to their composition, combining the spatial and spectral dimensions of the data. Elemental analysis and infrared spectroscopy supported these findings. In this context, the use of reflectance imaging spectroscopy (RIS), despite its technical limitations, emerged as an essential tool for the documentation and conservation of design heritage.

1. Introduction

Non-invasive analytical techniques have made significant contributions to the study of cultural heritage in the 21st century. During this period, the development of digital imaging technologies has facilitated the characterization of cultural heritage, enhancing our understanding of artistic objects at the structural, morphological, and compositional levels. Reflectance imaging spectroscopy (RIS), referred to as multispectral (MSI) and hyperspectral imaging (HSI) in the literature [1,2,3], has gained widespread acceptance in the field of conservation. This is because current technology makes it possible to analyze cultural objects using lightweight and portable devices that combine spatial information obtained from a two-dimensional image with spectral information [4]. The number of bands has often been associated with the terminology: multispectral imaging is considered an acquisition of tens of contiguous records of bands characterized by the same spectral width: hyper- with hundreds, and ultra- with thousands of bands [5]. Nevertheless, despite the discrepancies in the terminology [6], RIS has emerged as an evolution of technical multiband imaging (MBI), a set of scientific images taken within a few ranges of the electromagnetic spectrum (UV-Vis-NIR), which is extensively used in museums and collections for artistic diagnosis [7,8].

An advantage of RIS is that the information is recorded in the form of a three-dimensional array or data cube. The data cube is a numerical matrix that collects the values of each pixel in terms of spatial and spectral dimensions. Therefore, a data cube can be analyzed as a set of images, one for each wavelength according to the range of the instrument, or as a collection of spectra, one spectrum for each pixel according to the resolution, as explained by Cucci et al. [9]. Using these data, different types of processing can be performed [10]. These technical features are useful for non-invasive identification of pigments or stone deterioration [11] according to their reflectance behavior, that is, without contact with the artifact surface and without the need for sampling. The interaction between electromagnetic radiation and matter can lead to different fundamental processes including refraction, reflection, scattering, emission, absorption, reflection, and transmission. The former three radiation–matter interactions are particularly relevant in the visible region of the spectrum, where the optical behavior of artistic materials is determined by electronic transitions, including charge transfer, conduction band transitions, and ligand field effects [12,13,14].

The signal captured by the sensor, which is primarily composed of reflected radiation, constitutes the foundation of reflectance spectroscopy employed in current imaging technologies. However, the resulting reflectance spectrum also includes contributions from scattered light [15]. For example, sample refractive index, particle size, and concentration have a great influence on the analysis, making characterization and interpretation of the results challenging. Scattering can significantly affect measurements by modifying the intensity and distribution of the reflected signal. However, some data quality issues, such as the level of captured details and the improvement of the signal-to-noise ratio (SNR), can be addressed by the correct management of the acquisition distance and the integration times for a given set of illumination, respectively [16].

Material identification using portable RIS systems remains a complex task due to the variability in spectral data, which is influenced by factors such as acquisition geometry, illumination conditions, and the trade-off between spatial and spectral resolution in certain devices [17]. Factors such as signal-to-noise ratio, wavelength range, and spectral resolution can limit the performance of portable cameras. However, their main strength lies in the integration of spatial and spectral dimensions within a single dataset. This integration enables the immediate visualization of physicochemical information—specifically, the material’s reflectance associated with electronic transitions—superimposed on the sample image and displayed directly on the device screen [18].

A few solutions have been proposed to validate the characterization of painting materials using RIS, such as the multi-analysis approach [19], which combines collected data with other non-invasive spectroscopic techniques, such as Raman spectroscopy, Fourier transform infrared (FTIR) spectroscopy, and X-Ray Fluorescence (XRF), for the interpretation of spectral results [20,21]. Other studies involve comparative methods that allow algorithms to be designed for the automatic identification of pigments [22]. The processing of false-color maps and images [23] is also commonly used to help visualize spectral results and algorithms, such as Spectral Angle Mapping (SAM) and the Spectral Correlation Mapper (SCM). SCM and SAM both compare two spectra over the full wavelength range by measuring the similarity between the spectrum of a selected point and a reference according to its spectral angle [24,25].

RIS has been used for the conservation and diagnosis of cultural heritage and has been successfully employed in the classification of deterioration, monitoring of cleaning processes, and authentication of artworks [26,27,28]. Its greatest potential has been proven in the identification of pigments. Li et al. [29] combined Macroscopic X-ray fluorescence (MA-XRF) and reflectance imaging spectroscopy (RIS) to create an automatic pigment scanning and mapping system. To cover the VNIR range, Cucci et al. [14] experimented with two portable RIS instruments in the 400–950 nm and 1000–2450 nm ranges. They applied this method to analyze several paintings on canvas and other artworks with the objective of identifying and mapping artistic materials. Previously, Daniel et al. [30] analyzed Goya’s painting on-site by applying two reflectance imaging spectroscopy devices, Energy Dispersive X-Ray Fluorescence (ED-XRF) and Raman. Certain authors also rely on comparative methods [31,32] to create and analyze their own reference standards, with the aim of obtaining spectral features that qualitatively define the material via RIS. Reflectance spectra characteristics of artworks are influenced by several variables, including the presence of different binders, pigment concentration, and pictorial layer thickness. Thus, studies are sometimes limited to a single chromatic range [33] or a single family of pigments [34] to reduce the sample variability. This issue is further compounded when considering the extensive range of contemporary pigments, particularly those synthesized since the 20th century [35]. On the other hand, even though these studies serve as a reference, to the best of our knowledge, there is currently no open-access contemporary pigment database meeting the standards acquired with portable RIS instruments, and the comparison of spectra is made based on UV-Vis-NIR reflectance spectroscopy [36,37].

RIS is an adaptable technique that provides valuable information on the reflectance of organic materials [38], including dyes and colorants [39], and is used in the agricultural and food industries [40,41,42]. Although several artistic typologies, such as textiles, murals, and easel paintings [43,44,45], have been studied using RIS, a significant part of the literature refers to artwork on paper, such as illuminated books and watercolor paintings [31,46,47]. These artifacts present suitable attributes for analysis, as they use a wide range of organic and inorganic pigments, mostly applied in pure color shades or with simple mixtures.

Delaney et al. [48] proposed a RIS workflow for analyzing an illuminated manuscript. The initial step involved the extraction of spectral endmembers using computational algorithms on a selected miniature. The endmembers constitute a set of individual spectra that optimally characterize the reflectance behavior of a sample. Later, these spectra were employed to generate spatial maps for the precise selection of FORS and XRF analysis points. Consequently, preliminary identification and mapping of the pigments and mixtures were conducted to establish correlations between the reflectance, elemental, and imaging data. Mounier and Daniel [49] employed a similar approach, combining RIS, FORS, Raman spectroscopy, and XRF for pigment identification in a collection of medieval illuminations from the 14th century. Furthermore, the processed RGB and IRFC images derived from the data cube were compared to formulate an initial hypothesis regarding pigment assignment. Pseudo-color or false-color composite (i.e., RGB images generated by assigning specific spectral bands to the red, green, and blue channels) are frequently used in RIS for an initial visual approach [50,51]. In cases where similar pigments yielded inconclusive results when combined with Raman and XRF, Mulholland et al. [47] employed a process of creating pseudo-color images, which were then compared with eighteenth-century watercolor palette mock-ups. These mock-ups included colors such as alizarin crimson, ultramarine violet, and yellow lake.

For this study, a selection of contemporary watercolors was analyzed to identify the pigments of the color palette, and a multi-technique approach based on a combination of RIS, and site-specific XRF, Raman, and infrared spectroscopies was employed for on-site and non-invasive classification of the materials used in the manufacture of the samples. This research aims to present a non-invasive and on-site approach for the characterization of contemporary watercolors combining RIS, XRF, Raman, and infrared spectroscopies in order to identify the materials used in their manufacture, including the employed pigments, in favor of the conservation and material integrity of unique and fragile works of art. The drawings belong to the ‘Hijos de Mariano García’ collection, which in the 1930s formed a catalog or sample showcase for the sale of furniture. These cultural assets are part of the Arxiu Valencià del Disseny (AVD) of the University of Valencia, a documentary archive that brings together a diversity of artifacts from heritage design, graphics, furniture, and fashion arts with the aim of documenting, preserving, and disseminating them.

2. Materials and Methods

This study focuses on four pieces in the collection of Mariano García’s sons (Figure 1), from the ‘Sociedad limitada de muebles y decoración Mariano García’, a furniture factory founded in 1893. The main part of the AVD collection consisted of documentation. This includes sheets, drawings, and photographs of interior design and furniture projects conducted between 1920 and 1980. The watercolors studied consisted of a primary paper support on cardboard (

Table 1. Essential technical data to identify the sample and summarize the state of conservation.

Sample (ID)	Size (mm)	Color Palette (RGB)	Condition
Numero 132 (MG_132)	266 × 440	Blue (12, 38, 74); lavender (141, 129, 139); ochre (223, 172, 69); orange (190, 74, 26); brown (128, 52, 27); green (22, 71, 37); carmine red (195, 14, 21).	‘Ref. 128’ in pencil. Paper acidification, color fading, stains, fingerprints.
N° 145 (MG_145)	281 × 455	Brown (200, 68, 35); bronze (187, 30, 21) blue shades (39, 76, 106); violet (152, 47, 50); green (46, 74, 60); carmine red (186, 1, 11).	‘Ref. 228’ in pencil. Paper acidification, foxing, fingerprints, material loss.
Numero 147 (MG_147)	270 × 465	Brown (130, 60, 24); blue shades (11, 53, 100); purple (140, 44, 78); ochres (230, 183, 107); carmine red (194, 4, 10).	‘Ref. 230’ in pencil. Paper acidification, foxing, material loss, color fading.
Número 148 (MG_148)	275 × 455	Brown (117, 60, 35); blue shades (14, 61, 109); ochres (176, 126, 63); yellow (224, 168, 25); green shades (23, 70, 64); purple (125, 82, 112); carmine red (176, 35, 45).	‘Ref: 122’ in pencil. MARIANO GARCIA MUEBLES-VALENCIA sealed. Paper acidification, adhesive stains, foxing.

). The drawing was traced using a pencil and inked. All the samples were made between 1930 and 1936 and share the same chromatic palette.

2.1. Reflectance Imaging Spectroscopy

Spectral imaging data were acquired with a SpecimIQ^® Hyperspectral camera (Spectral Imaging Ltd., Oulu, Finland by Konica Minolta) with a spectral range of 400–1000 nm, a spectral resolution of 7 nm for a total of 204 bands in 3.5 nm steps, and 512 pix spatial sampling. The camera records three different data cubes: a raw data cube, dark frame (dark signal from the sensor), and white reference, which is the signal obtained from a Lambertian, highly reflective (~99% in the 400–1000 nm region), and spectrally flat surface. Although each data cube can be processed separately, the camera performs an automatic normalization process or Reflectance Transformation according to the following Equation (1):

$$R={\displaystyle \frac{{Raw\_data}^{t1} - {Dark}^{t1}}{{White}^{t2} - {Dark}^{t2}}}\times {\displaystyle \frac{t2}{t1}}$$

(1)

where R is the normalized reflectance, Raw_data, Dark, and White are the original three data cubes, t1 is the dataset integration time (IT), and t2 is the white reference integration time. Spectral images were acquired using the ‘Custom White Reference Mode’, which enables data recording without positioning the white reference next to the target, thereby avoiding unwanted stray light scattered by the reference itself. This acquisition mode involves recording separate integration times for the white reference and target data during two different-scene acquisitions. One acquisition was carried out with the reference card, and the other was performed without it under the same light conditions.

One of the primary challenges encountered when using the SpecimIQ^® hyperspectral camera is the presence of stray light artifacts. These unwanted effects, which originate from internal reflections and scattering within the optical system, cannot be entirely eliminated and are influenced by the quality of the anti-reflective coatings and materials employed in the instrument design [52]. This effect was more pronounced at both ends of the wavelength range, where the spectral response was weakest. In the presented data, stray light artifacts are observed as an elevated reflectance slope between 400 and 420 nm, leading to a loss of information. Additionally, increased noise and variations in reflectance were detected beyond 960 nm.

For data acquisition, the samples were positioned in front of a camera next to two halogen lamps (150 W, color temperature 2800–3200 K) at 1.5 m. The light sources were projected symmetrically onto the surface at a 40° angle to avoid specular reflectance. To calibrate the scene, a Teflon (PTFE) white reference card was used, which reflects ~99% in the 400–1000 nm wavelength range and was provided by Specim^®, Spectral Imaging Ltd. The pigments in the four watercolors were identified by analyzing the primary colors (blue, red, and yellow) and the secondary colors (green, purple, and orange). Arbitrary color points per sample were compared using Pigment Checker version 5.0 [53], which counts at least 65 artists’ pigments from the antiquity since the early 1950s and different spectral libraries for qualitative data interpretation [54,55,56]. The normalized reflectance data (R) or ‘apparent reflectance’ (i.e., reflectance values corrected with dark and white signals, with true reflectance being the one that takes into account the influence of illumination geometry, sample surface characteristics, and environmental conditions) per pixel were plotted as percentages (%) to classify the materials and compare their spectral features. Reflectance spectra were acquired and processed using proprietary software (SpecimIQ^® Studio 2019, OriginPro^® 2022) and an open-source software (Hypercube v.11.52). As a complementary method of pigment classification, IR false-color images were generated by mixing selected spectral images as RGB channels from IR bands (900 nm) to red and red (650 nm) to green and green (540 nm) to blue [57].

2.2. X-Ray Fluorescence

The watercolors were analyzed to determine the main elemental characteristics of the paint layers and paper support using portable X-Ray Fluorescence (pXRF). Each analysis point was measured twice using a 40 kV rhodium (Rh) anode X-ray tube and a Silicon Drift Detector (SDD) on a Vanta C series Handheld XRF spectrometer by Olympus (Waltham, MA, USA). The measuring area has a single click collimation with a diameter of 3 mm; however, an internal VGA CMOS camera was used to clearly identify the color analysis points. The Geochem 2-beam method was used. It is characterized by 60 s of acquisition time, with the first 30 s being to determine heavy elements (Beam 1: 30 s, 40 keV) and the following ones to determine light elements (Beam 2: 30 s, 10 keV). The detection limit is higher for light elements below atomic number 14 [Z]. The spectra were analyzed using the pXRF software, Vanta Desktop App 3.34.102.

2.3. Raman and Infrared Spectroscopies

Portable i-Raman^® Plus Spectrometer (model: BWS465-785S) by B&W Tek (Plainsboro, NJ, USA) was employed for the Raman spectroscopy analysis. The spectrometer is characterized by a laser of 785 nm with a maximum power of 495 mW at the laser port and was coupled with a 20× magnification microscope. Each measurement consists of an average of 3 replicates carried out with an integration time from 5 s to 10 s and a power from 1% to 5% of the maximum power of the spectrometer. A dark scan with the same integration time was subtracted to each spectrum.

Fourier Transform Mid-Infrared (FT-MIR) spectra in the region of 4000 to 650 cm⁻¹ were obtained employing a portable spectrometer 4300 Handheld FT-IR by Agilent Technologies (Santa Clara, CA, USA) equipped with a diffuse reflectance sampler. Each spectrum was averaged over 50 scans and was expressed as the absorbance [A = log(1/R)]. The spectral resolution is 4 cm⁻¹ and Boxcar apodization was used. A background spectrum was obtained by employing reference material between the measurements and was automatically subtracted. Each color point was measured twice, and the obtained spectra were averaged.

To explore the potential of near-infrared analysis for the detection of compounds associated with pigments and raw materials, Fourier Transform Near-Infrared (FT-NIR) spectroscopy measurements were performed using a Bruker (Ettlingen, Germany) MPA—Multipurpose Analyser Fourier Transform Near-Infrared spectrometer, equipped with an integrating sphere and a fiber-optic probe for the acquisition of diffuse reflectance spectra. OPUS software 6.5 (Bruker) was used for instrument control and data acquisition. Color samples and paper support were directly measured twice using a fiber-optic probe to obtain NIR spectra by diffuse reflectance. Spectra were recorded in Kubelka–Munk units, in the 14,000–4000 cm⁻¹ spectral region, using a resolution of 4 cm⁻¹, 50 scans per spectrum. The background spectrum was acquired from a closed integrating sphere under the same instrumental conditions as the samples. This instrument proved to be effective in characterizing paper support.

3. Results and Discussion

The points of analysis are shown in the Supplementary Online Materials.

Qualitative analysis was performed using RIS on arbitrary points of primary and secondary colors. For this purpose, the absorption bands, inflection points (i.e., wavelengths where spectral curvature changes from concave to convex or vice versa), and reflectance behavior in the VNIR (400–1000 nm) were described and expressed in terms of apparent absorbance with log(1/R) conversion to highlight absorbing species [58]. Spectral features can change significantly in a color mixture or when another color is added to obtain brightness. In this sense, absorption features are more constant in terms of band position and are easier to identify in mixtures than reflectance features. Similarly, the first derivative of the spectra was plotted to accurately determine the positions of the absorption bands and inflection points.

Table 2. Assignment of primary and secondary pigments. All spectroscopic analyses were significantly influenced by the presence of chemical elements and absorption bands associated with the paper substrate.

Color (Primary)	Elements (pXRF)	Raman	RIS	Color Shift (IRFC)	Pigment Classification
Red (R)	Fe (Kα = 6.40 keV, Kβ = 7.06 keV), Ba (Lα = 4.45 keV, Lβ1 = 4.83 keV, Lβ2 = 5.16 keV)	2130 cm−1, 2070 cm−1	Ref. 435 nm, Abs. 470~550 nm, Ip. 595 nm	Yellow, orange	Alizarin red
Yellow (Y)	S (Kα = 2.30 keV, Kβ = 2.46 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper)		Abs. 420~450 nm, Ip. 470~490 nm	White	Yellow (A): organic yellow
	Cr (Kα = 5.41 keV), Pb (Lα = 10.55 keV, Lβ = 12.61 keV)		Abs. 430~480 nm, Ip. 530 nm	White, yellow	Yellow (B): chrome yellow
Blue (B)	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper)	2130 cm−1, 2070 cm−1	Ref. 480 nm	Violet, dark	Prussian blue
	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper)	2130 cm−1, 2070 cm−1	Abs. 590 nm, Ip. 680 nm	Red, dark	Ultramarine (artificial)
Color (secondary)	Elements (pXRF)	Raman	RIS	Color shift (IRFC)	Possible classification
Violet (V)	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper)		Ref. 480 nm, Ip. 600~620 nm	Orange	Red and blue mixtures
Orange (O)	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper)	2130 cm−1, 2070 cm−1	Ip. 560 nm	Green, light	Organic orange
Green (G)	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper) and Cr (Kα = 5.41 keV)	2130 cm−1, 2070 cm−1	Ref. 450~550 nm	Violet, light	Green (A): yellow and Prussian blue
	S (Kα = 2.30 keV, Kβ = 7.06 keV), K (Kα = 3.31 keV), Ca (Kα = 3.69 keV, Kβ = 4.01 keV), Ti (Kα = 4.50 keV, Kβ = 4.93 keV), Fe (Kα = 6.40 keV, Kβ = 7.06 keV) (paper) and Cr (Kα = 5.41 keV)	2130 cm−1, 2070 cm−1	Ref. 450~550 nm, <640 nm	Violet, red	Green (B): yellow and Ultramarine

Ref. (maximum reflectance); Abs. (absorption band); Ip. (inflection point).

summarizes the pigment classifications using imaging and portable spectroscopy.

The main elements detected on the paper support were S, K, Ca, Ti, Fe, and small amounts of Br, Mn, and Pb (Figure 2, P1). Calcium sulphate dihydrate [CaSO₄·2H₂O] and calcium carbonate [CaCO₃] are standard products used in paper manufacturing as fillers that provide opacity and whitening. On the other hand, Fe was found in historical and contemporary paper by entering through the water during wet processes, from paper machinery, or as contaminants [59]. Other elements such as K, Ti, Mn, and Pb were detected on the paper support related to their function as components, contaminants, or degradation products [60]. Thus, potassium hydroxide [KOH] is used in the paper industry for pulping, breaking down the lignin in the wood chips, and leaving behind cellulose fibers. KOH helps to dissolve the lignin and separate the fibers that can be used to make paper and as a fire retardant [61].

Figure 3 shows the FT-NIR spectra of the support and the painted areas. These are very similar to those in paper (S1), possibly due to the absence of features in the near-infrared region or the material concentration. The samples were painted with thin layers, in accordance with the watercolor technique. Hence, the bands around 6750 cm⁻¹ were caused by OH vibrational modes due to cellulose hygroscopicity [62]. Otherwise, the sharp peaks at ~7185 and ~6750 cm⁻¹ suggest the presence of clay minerals, such as kaolin [63]. The presence of clay minerals is also corroborated by a doublet at approximately 4530 cm⁻¹ [64]. Aluminosilicates can be part of the material used as paper fillers or extenders [65]. Wide bands peaking at approximately 5180 cm⁻¹ and 4750 cm⁻¹ are attributed to OH and CO vibrational modes in cellulose [66], and bands at ~4400 cm⁻¹ and 4280 cm⁻¹ are related to CH vibrations in cellulose glucose molecules [67].

Similarly, the FT-MIR spectrum (Figure 4) did not show any differential features concerning paper support. A wide band at ~3500 cm⁻¹ was related to OH stretching, and small shoulders at ~3690 and ~3620 cm⁻¹ are more precisely attributable to Al-OH stretching, which could indicate the possible presence of clay minerals used as fillers [65]. The bands at ~2905 cm⁻¹ are linked to CH stretching in cellulose, hemicellulose, and lignin [68]. The combination band peaking at ~1645 cm⁻¹ was caused by the presence of water, although the contribution of CO stretching vibrations caused by cellulose oxidation could be suggested by the shoulder at approximately 1590 cm⁻¹ [68]. The features of the combination bands at ~1465, ~1450, and ~1430 cm⁻¹ can be attributed to CH bending in cellulose and lignin [69]. The band at ~995 cm⁻¹ was attributed to starch [70] which was used as a binder in paper manufacturing [71]. The combination band at ~2135 cm⁻¹ is due to CC stretching [72], but the most relevant features of this region can be observed in the blue (B1) and green (G1) points, which show the characteristic band of cyanide (CN) stretching vibrations close to 2095 cm⁻¹, according to Rosi et al. [73]. This suggests the use of Prussian blue both alone and in a mixture to obtain green shades. This feature is also present as a weaker band in the red spot (R2) and as a shoulder in the orange one (N1), possibly due to contamination.

Raman spectra are shown in Figure 5. Support (P1) does not show any clear characteristics. However, a band at about 2130 cm⁻¹ can be seen in R1, R2, G1, O1, and more clearly in B1, which also shows a feature close to 2070 cm⁻¹. Both these bands can be attributable to CN stretching vibrations and would confirm the presence of Prussian blue [74]. The use of this pigment in blue and green areas confirms what is suggested by pXRF and FT-MIR. The possible presence of a weak feature in the orange (O1) and red (R1) areas would also be consistent with the FT-MIR results.

3.1. Yellow Pigments

The identification of yellow pigments by reflectance spectroscopy presents challenges owing to the alteration of their spectral features when combined with white pigments for enhanced color opacity. Yellow pigments absorb electromagnetic radiation in the region between violet and green and exhibit an increase in reflectance across the remainder of the spectrum (Figure 6a), resulting in a sigmoidal spectral curve. As described by Aceto et al. [58], yellow pigments can be categorized into three types based on their reflectance features: (1) lead-containing pigments (lead-tin yellow I and II, Naples yellow) and orpiment, all semiconductors, exhibit a flat sigmoid curve spectrum; (2) iron oxide yellows (ochre, natural umber, and sienna) are similar to the red ones, demonstrating a sigmoid curve with two absorption bands in the red and infrared regions; and (3) in addition to the sigmoidal curve, yellow dyes (weld, saffron, stil-de-grain) present an absorption band between 400 and 450 nm, and significant diagnostic bands in the UV region, between 200 and 400 nm.

The paper support and sample MG_147 exhibited an absorption band between 420 and 450 nm, followed by an increase in reflectance from approximately 450 nm onwards. Although this behavior was associated with the yellowing of the support due to the acidification process, the presence of stray light effects at the beginning of the spectra (400–420 nm) prevented the reliable determination of the absorption band or the presence of other diagnostic features. The inflection point (Ip.) was determined for all spectra in bands between 470 and 490 nm, and the MG_132 spectra exhibited a maximum absorption in log(1/R) at approximately 450 nm (Figure 6b), occupying the violet and blue regions of the spectrum and corresponding to the weld lake (Reseda luteola) standard. While these spectral behaviors are similar to those of inorganic pigments such as Naples and lead-tin (II) yellows, and some colorants such as saffron and turmeric, they all exhibit subtle differences in maximum absorption bands and inflection points. Weld lake was reported to exhibit an absorption band centered at 410 nm and an inflection point at 460 nm [58], while absorption maxima at 455 nm and an inflection point at approximately 490 nm were stated by de La Codre et al. [39] in textile samples. Weld dyes exhibited alterations in absorption bands depending on the type of mordant, and accelerated aging also had an impact on the reflectance features, with observations indicating a loss of the main absorption bands at short wavelengths.

The modern yellow pigments that were widely used in the 19th and 20th centuries were cadmium, cobalt, and chrome yellow. Two analysis points are shown in the MG_148 sample (Figure 7a) and are classified by their spectral behavior in (A) and (B). Yellow (A) shows spectral features close to the paper support and yellow organic pigments, and (B) to the yellow chrome standard [75], with an absorption area of 430–480 nm, a sigmoid curve near 520 nm, in the green region (500–570 nm), and reflectance from the yellow region (570–590 nm) to the infrared region. As the yellow chrome standard shows an inflection point at ~515 nm in the first derivative (Figure 7b), yellow (B), and yellow (m), a mean curve plotted from 5 × 5 adjacent pixels for comparison [32] was observed at ~530 nm. Feller [76] stated that the wavelength at which inflection of reflectance begins to occur is a characteristic that can aid in the classification of yellow pigments in the visible spectrum. That is, at shorter wavelengths the hue is green, while at longer wavelengths it is orange. This is also true for both the orange and red pigments. Nevertheless, the inflection point observed for the yellow (B) samples and the absorption band centered at 450–470 nm were associated with the chromate (VI) ion chromophore [77], a transition metal oxyanion characteristic of lead chromate.

The presence of Cr in the pXRF elemental analysis (Figure 2; Y1) indicates the use of chrome yellow [PbCrO4]. This point also showed the most intense peaks for Pb. However, Pb was detected at all the analysis points, possibly due to the use of lead white [2PbCO3·Pb(OH)2] mixed with other pigments or possibly present in the paper. The higher concentrations of Fe in the yellow measurement could be caused by the difficulty in isolating it from the brown areas or by intentional mixing. Indeed, the brown point (Figure 2; Br1) presented high levels of Fe compared to the other areas, which is compatible with the use of colored earth. Higher levels of K could be linked to this, while the presence of Cr could be caused by the pigment mixture, the presence of yellow in the underlying paint, or by the analytical limitations of the measurement spot area. Cr is an element present in green (Figure 2, G1). However, given the low intensity of the Cr Kα1 line, it is possible that chrome yellow was used with the blue pigment to achieve the desired green hue.

3.2. Red Pigments

Figure 8 shows the spectra obtained from the red color. All the spectra showed comparable features, with a sigmoid curve at 560–620 nm. These curves are defined by inflection points around 595 nm and 585 nm in the MG_148 sample. This behavior is depicted by many red dyes, both natural and synthetic, with the modern azo families (toluidine and naphthol red) showing a flat absorption band before 550 nm as a differentiating element in their spectrum [76].

Inorganic pigments such as vermilion (semiconductor) also present a reflectance of this type, with an inflection point between the bands 590~605 nm. Cadmium red has the same spectral features with a flatter curve, as reported by Thoury et al. [78]. In the spectra (Figure 8a), a point of reflectance increase at 435 nm was observed for samples MG_132 and MG_147, followed by an extended absorption band from this point to 550 nm, centered at approximately 505 nm.

Figure 8b shows the apparent absorbance of the red samples, revealing certain bands for their characterization. In the case of MG_145, no relevant features that could be subjected to further interpretation were observed. However, for both MG_132 and MG_147, two primary absorption bands were identified at 510 nm and 550 nm, respectively. Additionally, a shoulder was observed at approximately 475 nm in both samples. These three absorption features represent madder-based pigments [79], and are associated with the presence of artificial alizarin red, a modern pigment synthesized in 1869. Similarly, the MG_148 sample displayed a slight displacement of the curve to shorter wavelengths, exhibiting only a broad absorption band at approximately 480 nm, followed by a shoulder at 545 nm. This is related to the excessive use of calcium carbonate (CaCO₃) as a dye extender [79].

Fonseca et al. [79] used the first derivative of cochineal and madder standards to determine the nature of the red lake pigments. Relevant spectral features for identifying organic red dyes were described as absorption peaks between 500 and 600 nm, associated with the n → π* electronic transitions. For example, madder-based pigments exhibit two sub-bands (1st and 2nd inflection points) in the 510–515 nm and 540–545 nm ranges, whereas cochineal-based pigments exhibit sub-bands in the 520–525 nm and 550–565 nm ranges, which can be used to classify them. The first derivatives of the MG_132 and MG_148 spectra (Figure 8c) show weak sub-bands at ~495, ~532 nm, and ~478 and ~512 nm, respectively. The intensity of these diagnostic sub-bands is subject to absorption features in relation to the pigment concentration in the sense that a thin layer of paint can provide sharper bands than a thick layer in the first derivative transformation. This is not the case for these red colors, where these inflection points were not clearly assessed in terms of intensity and wavelength. In addition, the position of the sub-bands may depend on several factors, such as the type of mordant used, the primary support, and the chemical environment [80]; therefore, the nature of these samples could not be conclusively determined.

However, the presence of sub-bands was related to the use of an organic pigment, which is in agreement with the elemental analysis. The lack of elemental markers for mineral pigments in the red color (Figure 2, R1) suggests the possible use of modern organic pigments such as alizarin or naphthol red. Synthetic alizarin red was patented in 1868 and gradually replaced natural madder pigments, dominating the artists’ market from the beginning of the 20th century [81]. Elemental analysis also suggested the use of more contemporary naphthol red due to the presence of barium (Ba) traces in the red and brown color samples [82]. Small amounts of Ba were detected at some points (Figure 2; R1, Br1, and G1). Cesaretto et al. [82] indicate that barium sulphate (BaSO₄) was used for naphthol-based red colorants as an extender. Although the use of these types of colorants cannot be confirmed by other techniques, the use of barite as an extender has been common in the past century [83,84] for pigment and paper manufacturing [85].

FT-MIR analysis (Figure 4) shows that the shoulder at ~1465 cm⁻¹ was more intense in red and green colors, although it is difficult to isolate the latter from the former due to the spectrometer measurement area. This band, linked to CH₃ asymmetric bending, could hypothetically indicate the presence of some organic material linked to the red areas; synthetic pigments from the azo family, such as alizarin and crimson red [86]. Instead, bands from ~1375 to ~1020 cm⁻¹, as well as that at ~900 cm⁻¹ are from vibrational modes in cellulose molecules [87,88], and the minima at ~1152 cm⁻¹ observed in most points are contemplated inverted bands caused by the ‘Reststrahlen effect’ from asymmetric SO₄ stretching vibrations [89] associated with the presence of paper extenders. In addition, the ~1080 cm⁻¹ band at the brown point could be linked to silica in the colored earth. Finally, the Raman spectra for the red color (Figure 5, R1) show weak bands at ~1633, ~1569, ~1477, ~1453, ~1351, ~1330, ~1289, ~1192, and ~470 cm⁻¹, in agreement with data obtained using surface-enhanced Raman scattering (SERS) by Cañamares et al. [90] and in solid alizarin samples by Puglieri et al. [91]. Again, the band intensity shown in red in Figure 5 was not sufficient to make accurate assignments for pigment identification.

3.3. Blue Pigments

Figure 9 displays three blue spectra (A1, A2, and A3) arranged in order of brightness. PCA was applied to facilitate the visualization of areas with similar spectral features. New image cubes were produced from the original data cubes, where the new bands are the principal components and are ordered by decreased variance (PC1, PC2, PC3, …).

The reflectance curves of the small plate (A1) on which the jar rested and the blue reflections of the mirror (A2 and A3) were examined. The curves illustrate a general increase in reflectance with increasing brightness. The shifts in the maximum reflectance bands from 480 nm in Blue A1 to 500 nm (A2) and 550 nm (A3) indicate a transition from the blue to green region of the visible spectra. Some publications suggest that both reflectance and its respective first derivatives are sensitive to the density of the paint layer [79]. This can cause shifts in the reference bands, such as absorption and inflection points, in addition to changes in signal intensity [92]. In this case, the shift was related to both the density of the layer and the increase in paper reflectance. In watercolor paintings, the paper plays a significant role in providing light scatter, thereby acting as a white pigment. As the blue color becomes lighter, there is a shift in the maximum reflectance towards longer wavelengths and an increase in the total reflectance.

In contrast, the Prussian blue pigment was associated with this spectral behavior, especially in A1. This pigment is characterized by a slight shoulder of reflection in the blue region, between 400 and 480 nm, and high absorption in the rest of the spectrum, centered at approximately 700 nm. The high absorption is the result of intervalence charge transfer between Fe ions [74]. In this sense, three types of spectra were compared in VA images. This was performed using a Vector Angle (VA) mapper, which showed the spectral differences between the blues. The algorithm, analogous to SAM, compares image pixel vectors to selected reference vectors by calculating the angle between them in an N-dimensional spectral space. The match was determined by the closeness of the comparison, and the theoretical tolerance range was between 0 and π rad. The color map indicates the presence of an alternative pigment to Prussian blue. To differentiate between these pigments and develop accurate statements regarding the material used, spatial dimension was considered. Therefore, the areas covered by blue pigment were evidenced by processing an IRFC [84].

The IRFC image is used to support the classification of materials, particularly pigments, which exhibit distinctive behavior when the RGB color channel is manipulated [93]. To process the IRFC image, the following bands were used: R channel = 900 nm (NIR); G channel = 650 nm; and B channel = 540 nm, based on Mulholand et al. [47]. Figure 10 compares the RGB and IRFC images obtained using spectral imaging, highlighting the blue areas. The regions of interest changed color from blue to violet, corresponding to Prussian blue [34]. The figure plot (Figure 10) displays three blue spectra, corresponding to a light blue plate, cobalt blue jar, and shaded area of the jar. In this case, the infrared false-color (IRFC) image and the principal component image (PC2) are of particular interest for classification purposes. The IRFC image shows the color change occurring in Blue A and Blue B-C, with the former becoming violet, and the latter becoming reddish and dark red. Likewise, in PC2, the blues of the plate and mirror turned black, while the blues of the jar changed from light (Blue B) to dark (Blue C). On the other hand, indicating the similarity of the pigments used to paint the jar (mouth and handle) and the wooden parts of the furniture, it should be noted that the reds and coppers turn yellow in IRFC and white in PC2.

The reflectance spectra exhibited distinct behaviors among Blue A, B, and C (Figure 10). Blue A has a relatively flat curve with a maximum around 480 nm in the blue region, extending towards the green region (495–570 nm). It does not show any peculiarities, except for a slight and progressive increase in the reflectance in the IR band. Blue B presents a spectral curve with a maximum at 465 nm (blue region), a symmetrical absorption at 600 nm, and an inflection point at 680 nm (red region), with increased reflectance in the IR band. Except for Prussian blue, which absorbs long wavelengths [76], blue pigments are typically reflective in the blue and infrared regions. Blue A exhibits a peak that is shifted towards the green and yellow regions, while Blue B reflects in the blue and red regions. Blue C exhibits a low spectral resolution curve with a maximum at 450 nm between violet and blue, and a continuous absorption band throughout the rest of the spectrum. This behavior is primarily due to the saturation of the color, which is composed of several layers of dark blue watercolor. The pigment concentration is high enough to mask certain spectral features. Blue A was associated with Prussian blue, whereas Blue B had spectral features associated with ultramarine.

Based on the pXRF analysis, the blue color (Figure 2, B3) was characterized by high K levels compared to the others. High levels of potassium can be linked to a soluble variant of Prussian Blue [KFe^III [Fe^II (CN)₆]·xH₂O] [74], but iron levels are not significantly higher than those in the bare support. However, the presence of characteristic bands of cyanide [CN] stretching vibrations in FT-MIR (Figure 4) and Raman spectra (Figure 5) provides solid evidence of the presence of Prussian Blue [58,73]. Although blue phthalocyanine pigments have cyanide functional groups in their chemical structures, Cu was not detected in the elemental analysis, suggesting that this family of colorant materials was not present in the samples analyzed.

3.4. Secondary Colors

In the context of artistic color theory, secondary colors can be obtained by combining two primary colors: blue and yellow to create green; yellow and red to yield orange; or blue and red to produce violet (purple). Additionally, secondary colors can be found in their pure form as non-mixed inorganic and organic pigments, such as chromium oxide green or cobalt violet (Figure 11). In the case of the watercolors, the selected secondary colors did not display a spectral behavior that could be associated with the standards acquired or the databases reviewed. They are generally characterized by absorption bands and inflection points that, when considered individually, are related to certain primary colors. This was apparent in the case of the green and violet analyses and was not noticeable for orange and brown colors. Unmixing methods should be used to observe the behavior of these colors through their decomposition into spectral endmembers [94].

The alizarin red standard was compared with some arbitrary spectra in the same color range. All curves were characterized by inflection points around 600~620 nm, and an increase in reflectance in the red and infrared regions. A slight maximum reflectance was observed in the violet and blue regions, between ~430 and ~520 nm, and was centered at ~480 nm for purple and violet. This spectral behavior suggests that violet shades were created by mixing the red pigment and unidentified blue in different proportions rather than using modern cobalt and manganese violet. It is important to note that unmixed violet colors from the 19th century, such as cobalt and manganese violet, are not adequately represented by this spectral behavior due to the presence of multiple absorption bands [95] and a common maximum reflectance centered in the violet region of the spectrum (400–440 nm). Neither Co nor Mn was detected in the elemental analyses. Reflectance imaging spectroscopy did not reveal any distinctive characteristic or diagnostic bands for orange or brown color. Only the presence of Fe levels, which were higher than those in the paper support, was confirmed using pXRF. Additionally, Raman analysis identified a weak band in the orange and red areas that could be associated with modern azo pigments, including methyl, naphthol, and hansa orange and red dyes, as supported by previous research [96].

Based on pXRF measurements (Figure 2, G1), chromium-based pigments such as viridian could have been used. Green colors were represented by two types of spectra (A and B), both defined by reflectance maxima in the blue-green region, between 450 and 550 nm, and differences in absorption in the red and infrared regions. The data indicate a constant absorption from the maximum reflectance for the green type (A) sample. The green type (B) exhibited a gradual increase in reflectance from 650 nm onwards. The spectral behavior observed in green (A) was associated with the presence of ferrocyanide absorption bands, in agreement with the Raman and FT-MIR results. Consequently, it was hypothesized that Prussian blue could have been manually mixed with yellow to achieve the desired green color. Chrome green (PG15), a prepared mixture of basic lead chromate (chrome yellow) and potassium ferric ferrocyanide (Prussian blue) [76], has been used since the early 19th century as a pigment for house paintings. However, it is not employed in artist paintings because of its toxicity. Green (B) showed spectral behavior related to chrome oxide green (Cr₂O₃) and hydrated chrome green or viridian (Cr₂O₃·2H₂O). In the absence of additional evidence obtained using other spectroscopic techniques, it was not possible to reach a definitive conclusion regarding the presence of green-chrome-based pigments in the samples. Given the spectral characteristics of these materials, it is possible that green (B) was created by mixing yellow and ultramarine pigments.

4. Conclusions

The compositional materials were characterized using a compact and lightweight spectral camera. The primary and secondary chromatic ranges used to produce the watercolors were characterized by spectral imaging, combining the spatial and spectral dimensions of the data. For the qualitative interpretation of these data, the use of pseudo-color composites, such as infrared false-color processing (IRFC), principal component images, and vector mapping algorithms, are essential. These methods allow for the observation of the spectral behavior of pigments such as ultramarine, Prussian blue, chrome yellow, and alizarin red.

Two types of blue were classified in these watercolors. The first type is characterized by continuous and homogeneous absorption in the green region, whereas the second type reflects light in the red and infrared regions. These qualities have been associated with Prussian blue and ultramarine, respectively. Furthermore, a color change to dark violet was observed in the infrared false composite of the former, and to red of the latter. Although phthalocyanine blue showed a similar behavior in both the spectrum and IRFC, no traces of Cu were observed in the elemental analysis, excluding its use in the samples. The combination of all techniques served to optimize the interpretation process, but all techniques had limitations related to the primary support and its state of conservation. For example, vibrational spectroscopy results were strongly determined by the absorption bands of the paper support, including compositional materials, extenders, and additives. This was true for elemental analysis using pXRF. The reflectance of the pigments obtained from the RIS was also influenced by several additional factors, including the yellowing of the support due to acidification, the loss of saturation of the colors due to aging, and the concentration of the pigments in relation to the painting technique used (watercolor). Scattering is a key light–matter interaction in diffuse reflectance spectroscopy. The analysis of individual spectra was complicated by light scattering effects, which were attributable to both the paper substrate and the pigments themselves.

Even though the versatility of RIS makes it an ever-evolving tool for cultural heritage studies in the digital age, its limitations relate to technical reasons. Specifically, these limitations are associated with hardware, current remote sensing technologies, data acquisition, and processing methods. However, precise observations regarding the physicochemical behavior of the pigments were made on-site using a compact spectral camera in combination with other portable spectroscopies. This allowed for the classification of two types of yellow, chromium yellow and organic yellow, as well as two types of green, made from a mixture of different blue pigments and yellow, in a non-invasive way and with minimal contact with the samples. Based on the objectives proposed in this study, it can be concluded that spectral imaging has proven to be a valuable tool in the analysis and preservation of contemporary cultural heritage. By allowing for the non-invasive identification of pigments, it provides a deeper understanding of the composition and condition of artworks. The development of more specific spectral libraries tailored to contemporary art is essential as it will enhance the precision of pigment identification. This advancement emphasizes the need for ongoing research and collaboration within this field, highlighting the dynamic nature of contemporary cultural heritage. However, it must be clarified that, to obtain reliable and robust results, especially in the case of a non-invasive approach, including spectral camera data, it is necessary to carry out cross-referencing with the analytical information obtained from other complimentary techniques, such as those used in this study.

Finally, the chromatic ranges identified were contextualized within the palette of pigments commonly used or introduced during the 19th century, including alizarin red, chrome yellow, chrome oxide green, synthetic ultramarine, and Prussian blue. However, the presence of industrially synthesized pigments from the 20th century—such as toluidine, naphthol, and phthalocyanines—could not be conclusively confirmed. The results align with the historical and artistic context in which the works were produced, suggesting a deliberately limited selection of primarily chromatic pigments, likely adapted to the professional practices of the time. Moreover, the scarcity of materials during the 1930s—following the Great Depression and preceding the Spanish Civil War (1936–1939)—may have further delayed the adoption of newer synthetic pigment families, despite their availability since the 1920s.