Abstract :
[en] Determining the composition of works of art is of utmost importance to adequately
choose their conservation conditions, understand potential degradation processes and
suitably treat the damages they suffer. In addition, a thorough analysis also helps
to unravel the successive steps carried out during the works of art production and
to better understand the evolution of the ideas and the execution of the artists. A
common approach in the field of cultural heritage is to deploy a range of complementary
non-invasive analytical techniques, usually with a mobile setup. In particular,
hyperspectral imaging is a tool of choice to study the composition of works of art, and
especially paintings, since it provides a spectrum of the reflectance as a function of the
wavelength for each pixel of the recorded image, therefore combining the advantages of
imaging and spectroscopy. The spectral features are directly linked to the underlying
specific chemicophysical processes implied in light-matter interactions, which in turn
enables pigments identification.
This thesis is devoted to the study of paintings and, particularly, to the development
of analytical methods to identify and quantify the composition of mixtures of
pigments in large regions. Experimental investigations are conducted on an oil painting
and fresco samples with a combination of techniques, namely photography under
various kinds of illumination, microscopy, X-ray fluorescence, Raman spectroscopy and
hyperspectral imaging. While some of them directly probe the molecular composition
of colors, other are sensitive to the elements present locally or in the scanned area.
These complementary techniques are applied on the case study of La violoniste, an oil
painting by Kees van Dongen, and the successive steps in this work of art production
are highlighted, in particular the presence of a hidden paint underlayer representing
a woman. Moreover, the pigments present in the painting are identified: lead white,
cadmium yellow, vermilion, Prussian blue, titanium white, ultramarine, carbon black
and viridian. While the identification of single pigments is quite direct, the results
obtained for La violoniste emphasize the complexity of isolating the components of a
mixture.
Among the models describing the interaction of light with paint layers, the Kubelka-Munk model, based on linear combinations of the K/S functions calculated from the
reflectance spectra, stands out as a tool of choice for routine analysis of paintings.
Indeed, it has the advantage of being easy to implement in practice, while enabling
pigments identification and quantification. We successfully apply it to hyperspectral
imaging data acquired on a mixture in La violoniste to highlight the presence of cobalt
blue, cobalt chromite and ultramarine pigments. To establish a procedure to quantify
the components of pigment mixtures, 88 calibrated mixtures of green earth and
Egyptian blue pigments are produced with coat colored in the mass and according to
ancient Roman wall painting recipes of a fresco and a secco techniques. Hyperspectral
imaging data acquired on these mock-up samples are analyzed with two methods, relying
on the Kubelka-Munk model or on principal component analysis (PCA). On the
one hand, we find that the coefficients obtained from fittings of the first derivative of
the K/S functions are correlated with the proportion of blue and green pigments in the
samples. A reference function, based on an exponential, is proposed to characterize
samples with unknown pigment concentrations, with an error of less than 6%. On the
other hand, PCA is a tool of choice to rapidly identify similarities between mixtures
in the principal components space and compare the concentrations of a large number
of samples. The comparison of the localization of mixtures with pure pigments in
this new representation is directly related to the proportion of their constituents. The
error on the concentrations determined with PCA are in general on the order of 7.5%,
but the additional knowledge of the application technique and the properties of the
white component present in the mixture reduces it to 3%.
Finally, a hundred fragments from mural paintings of the Domus dei Bucrani, dating
from the first century BC, are analyzed in the archaeological site of Ostia Antica.
The pigments found in scenes of the oecus decoration are identified and mapped: green
earth, Egyptian blue, iron-based pigments, carbon black, cinnabar. . . The data processing
techniques we implemented, relying on hyperspectral imaging data, enable a
rapid comparison of the green earth and Egyptian blue proportions in various zones
from the oecus, providing a different insight into the work of artisans in ancient Rome.
They are also applied to the identification of the components of mixtures of other
colors. Common characteristics are shared between several parts of the decor: the superimposition of colors matching the pigments affinity with the medium, the presence
of blue grains in unexpected colors, the white highlights as a final stage of the painting
production and the multipurpose utilization of cinnabar, to name a few.
