Noninvasive In Situ Study of Pigments in

Artworks by Means of VIS, IRFC Image

Analysis, and X-Ray Fluorescence
Spectrometry

Olimpia-Hinamatsuri Barbu,1* Anamaria Zahariade2

1
National Museum of Romanian History, Centre of Research and Scientific Investigation, Calea Victoriei 12, 030026, Bucharest-3 Romania

2
University of Bucharest, Faculty of Physics, Str. Atomistilor, nr. 405, 077125, Magurele, Jud. Ilfov, Romania

Received 6 October 2015; revised 2 January 2016; accepted 2 January 2016

Abstract: This article describes a preliminary method for computer vision methods of investigation have provided
pigment identification in paintings, based on the images promising results for the nondestructive characterization
obtained using digital microscopy with visible and infra- of paintings and written documents.1–6 These methods
red illumination and X-ray fluorescence spectrometry have the advantage of being a low-cost solution for art-
(XRF). About 240 standards of pigments were character- work examination, but have some limitations when pig-
ized using image processing and XRF analysis, building ments are mixed or overlapped in layers. Therefore, there
a specific database of reference images and spectra. The is a need for a noninvasive method capable of measuring
advantage of this method is that mixture of pigments small areas with sufficient precision to detect mixture of
could be discriminated and identified. An example of the pigments. Looking for a fast, cheap and easy-to-use solu-
application of this data analysis of pigments in fresco tion in the preliminary in situ study of the painting, we
paintings is presented. VC 2016 Wiley Periodicals, Inc. Col Res chose the digital microscope.
Appl, 41, 321–324, 2016; Published Online 13 February 2016 in Wiley
Online Library (wileyonlinelibrary.com). DOI 10.1002/col.22038
MATERIALS AND METHODS
Key words: color identification; color matching; pigment The database used for pigment identification was built
database; noninvasive techniques; paintings using pigments (provided by Kremer) applied with gum
Arabic on paper (partially covered by black stripes for a
better observation of the hiding power of the pigments).
INTRODUCTION For digital image acquisition, analysis and processing a
system composed by the following elements has been
One of the analytical techniques most frequently used for used: a portable microscope, DinoLite AD4113T-I2V
in situ pigments identification is X-rays fluorescence spec- with IR illumination (940 nm), a visible (VIS) light ring
trometry. It is well known that this technique provides illuminator, located at near 7 mm distance from sample
data concerning the chemical elements, but cannot give plane, with a color temperature of 6500 K, an optical fil-
information about the nature of the pigments. Thus, the ter (polariser), a computer and a MATLAB algorithm (in-
utilization of complementary techniques, able to complete house code created with Matlab version 7.12, R2011a,
the information obtained by X-rays fluorescence, is a The MathWorks). The microscope was mounted in the
must. Recently, multispectral, hyperspectral imaging, and middle of VIS illumination system, at near 3 mm distance
from sample. A ColorChecker passport (X-rite) was used
*Correspondence to: Olimpia-Hinamatsuri Barbu (e-mail: barbuolimpia@
for color correction, using only the chart with 24 patches.
gmail.com)
Additional Supporting Information may be found in the online version of The color profile of the microscope was determined using
this article. L*, a*, b* averages values for each color of Color-
Checker chart and recomposing the image with the 24
C 2016 Wiley Periodicals, Inc.
V patches, image utilized in the application RoughProfiler,

Volume 41, Number 3, June 2016 321

Fig. 1. VIS and IR digital microscopy images of 82 reference pigments (1 3 0.5 mm) with black underlying layer (a); the
reconstructed 24 color patches of the ColorChecker passport before (b) and after (c) color correction.

version 1.7. For the reconstruction of colors of Color- cence technology (PXRF), tungsten anticathode, Si–PIN
Checker chart, but also for decreasing the image process- detector, 35 KV acceleration voltage, 40 lA, with 30-s
ing time, the images have been reduced using bicubic acquisition time.
interpolation. For infrared images, we chose the automatic
calibration of images, interposing a thin plastic foil (L*
RESULTS AND DISCUSSION
11, a* 3, b* 0 in VIS and L* 31, a* 0, b* 0 in IR) in
front of the microscope. To determine the color differ- Images captured with a portable microscope set at right
ence, the delta E was calculated using the formula angle to the surface of painting samples, at about 2203
CIEDE2000 (notated DE00). The pigment analyzes were magnification, with IR and VIS light, are used as referen-
performed using a portable analyzer (Innov–X Systems, ces, to create a database. VIS images are combined with IR
Alpha Series) based on energy dispersive X-ray fluores- images as proposed by Mazzeo et al.7 in order to obtain

Fig. 2. Fresco painting from Doamnei church (altar); investigated areas are marked with numbers (a); photomicrographs
(VIS) of the painting areas, 2203, the distribution maps of the matching colors for VIS and IRFC images (with DE00 limit
set to 3) and the RGB square images corresponding to the average values L*, a*, b* of the matching colors; black was
used for all the pixels outside the matching colors in the VIS and IRFC images (b).

322 COLOR research and application

Fig. 3. Color specification for the references shown in Supporting Information Table SI (marked with “x”) and the samples
(marked with “o”) extracted from VIS (a) and IRFC (b) images; see Table I for the identification of samples.

false color infrared images (IRFC) to be used for pigment source, nonuniform illumination, small defects on the Col-
identification. The process of calibration consists first, in orChecker chart surface, visible only under the microscope,
computing the average color of the X-rite ColorChecker wrong reconstruction of the ColorChecker chart, incorrect
Passport patches viewed through the microscope in VIS images calibration and color modification at images resiz-
light, then creating a color profile of the microscope to cor- ing are some of the errors that may appear. In the present
rect images.8 Using the MATLAB software, R, G, B values case, the accuracy and the precision are acceptable for pig-
from color images taken for pigments (82 pigments are pre- ment identification, the color difference DE00 between the
sented in Fig. 1) are transformed into L*, a*, b* parame- ColorChecker chart target and those estimated taking an
ters. The average L*, a*, b* of the matching colors for the average value of 2.6 (Supporting Information Table SII and
pigments in Fig. 1 are reported in the Supporting Informa- Fig. S1) and the precision is better than 4%. It must be
tion Table SI, as well as the main element(s) detected by mentioned that for pigment samples different granulation
the portable XRF instrument. The area of interest is can lead to different L*, a*, and b* values. The pigments
selected for each pigment and then the software finds the with granules transparent in VIS and/or IR can have differ-
matching colors with a color difference between 0 and the ent values as well, function of the support onto which the
average DE00 of the selected area plus the standard devia- pigment is applied. Once the pigments database built, the
tion. The user can choose the DE00 corresponding to the identification is fast, taking on average 5 min for image
maximum number of pixels, on the histogram of the DE00 acquisition and processing, searching and listing the pig-
for the selected area, or can lower the DE00 limit, thus ments of the same color, and nearly 1 min for PXRF analy-
extracting the colors closer to the average L*, a*, b* of the sis, to have confirmation for the presence of one or more
selected area. In order that the color determinations, made pigments based on the chemical elements identified.
on photos taken under the microscope, be useful in pig- The unknown pigments are identified comparing L*,
ments identification it is necessary to know and minimize a*, b* parameters of the matching colors corresponding
the possible errors. Errors could come from images acquisi- to the selected area within the photomicrographs (VIS
tion process, as well as from their subsequent processing. and IRFC), with the L*, a*, b* parameters of the referen-
Different settings of the microscope and/or illumination ces in the database. As an example, we present in the

TABLE I. Chemical elements identified by PXRF, the estimated values L*, a*, b*, DE00 and standard deviation
(std) of the matching colors from VIS and IRFC images of the samples (shown in Fig. 2b); DE00 is calculated
here using as standard the average values L*, a*, b*.
VIS IRFC
Elements Number in
Painting area (PXRF) Selected color Fig. 3 L* a* b* DE00 std L* a* b* DE00 std

1 Garment, red partially Hg, Pb, Ca Red (from 83 52 38 23 3 1 71 232 41 2 1

covered by orange lacuna)
Orange 84 57 28 38 2 1 72 231 32 2 1
2 Aureole, yellow Fe, Ca Yellow 85 60 12 43 2 1 68 226 16 2 1
3 Garment, blue with black Cu, Fe, Ca Blue 86 57 214 231 1 1 28 25 240 2 1
underlying layer
4 Garment, green Fe, Ca Green 87 48 24 6 3 1 43 216 29 2 1

Volume 41, Number 3, June 2016 323

following some of our results obtained in studying the pigments from a 17th century historical monument, previ-
fresco from Doamnei church (Fig. 2), a 17th century ously investigated using a multianalytical approach, com-
monument situated in the centre of Bucharest, Romania. bining PXRF with FTIR and SEM-EDX analysis.
The average L*, a*, b* parameters obtained for red,
orange, yellow, blue, and green pigments are shown in 1. Verri G, Comelli D, Cather S, Saunders D, Pique F. Post-capture Data
Analysis as an Aid to the Interpretation of Ultraviolet Induced
Table I. Searching in the database, for a DE00 limit set to Fluorescence Images. In: Stork DG, Coddington J, editors. Proceedings
3, cinnabar and red ochre were the estimated pigments of SPIE 6810, Computer Image Analysis in the Study of Art. San
for both VIS and IRFC images for the red areas (See Fig. Jose, California, USA; 2008. p 681002-1–681002-12.
3 for the position of the references and the samples in the 2. Toque JA, Sakatoku Y, Ide-Ektessabi A. Pigment identification by ana-
lytical imaging using multispectral imaging. 16th IEEE International
CIELAB a*b* space). Also, red lead was estimated in the Conference on Image Processing (ICIP), Cairo; 2009. p 2861–2864.
orange area, above the red layer. As the PXRF spectra doi:10.1109/ICIP.2009.5414508.
show the presence of Hg and Pb on the analyzed area, 3. Cornelis B, Dooms A, Leen F, Munteanu A, Schelkens P, Multispectral
the pigments could be indeed cinnabar and red lead. In imaging for digital painting analysis: A Gauguin case study. Proc. of
SPIE, Applications of Digital Image Processing XXXIII, Vol. 7798,
the same manner, the other pigments were identified: yel- San Diego, 2010.
low ochre, azurite, and green earth. FTIR-ATR and SEM- 4. Rosi F, Miliani C, Braun R, Harig R, Sali D, Brunetti BG, Sgamellotti A.
EDX analyses on small samples taken from lacunae in Noninvasive analysis of paintings by mid-infrared hyperspectral imaging.
Angew Chem Int Ed 2013;52:5258–5261. doi:10.1002/anie.201209929.
similar areas, confirmed the presence of these pigments.
5. Cosentino A. Identification of pigments by multispectral imaging: A
flowchart method. Herit Sci 2014;2:8. doi:10.1186/2050-7445-2-8.
6. Mounier A, Le Bourdon G, Aupetit C, Belin C, Servant L, Lazare S,
ACKNOWLEDGMENTS Lefrais Y, Daniel F. Hyperspectral imaging, spectrofluorimetry, FORS
and XRF for the non-invasive study of mediaeval miniatures materials.
We are grateful for the opportunity to study the paintings
Herit Sci 2014;2–24. doi:10.1186/s40494-014-0024-z.
from Doamnei church and for the fruitful discussions and 7. Mazzeo R, Palazzi CE, Roccetti M, Sciutto G. Computer-assisted pig-
support from professor Dan Mohanu. ment identification in artworks. In: Roccetti M, editor. Proceedings of
the IASTED European Conference on Internet and Multimedia Systems
and Applications (EuroIMSA 2007). Calgary: ACTA Press; Chamonix,
CONCLUSIONS France, 2007. p 266–271.
8. Charrière R, Hebert M, Tremeau A, Destouches N. Color calibration of
The calibrated digital microscope was used successfully an RGB camera mounted in front of a microscope with strong color
to measure the color of reference pigments and to identify distortion. Appl Opt 2013;52:5262–5271. doi:10.1364/AO.52.005262.

324 COLOR research and application