See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/5427937

Multispectral camera and radiative transfer equation used to depict

Leonardo’s sfumato in Mona Lisa

Article in Applied Optics · April 2008

DOI: 10.1364/AO.47.002146 · Source: PubMed

CITATIONS READS

59 4,559

2 authors:

Mady Elias Pascal Cotte

Sorbonne University 15 PUBLICATIONS 177 CITATIONS
48 PUBLICATIONS 928 CITATIONS
SEE PROFILE
SEE PROFILE

All content following this page was uploaded by Mady Elias on 18 February 2015.

The user has requested enhancement of the downloaded file.

 Multispectral camera and radiative transfer equation
used to depict Leonardo’s sfumato
in Mona Lisa

Mady Elias1,2,* and Pascal Cotte3

1
Institut des NanoSciences de Paris, Université Pierre et Marie Curie, Unité Mixte de Recherche 7588,
Centre National de la Recherche Scientifique, Campus Boucicaut, 140 rue de Lourmel, 75 015 Paris, France
2
Université d’Evry Val d’Essonne, boulevard François Mitterrand, 91000 Evry, France
3
Lumiere Technology S.A.S., 215 bis Boulevard St. Germain, 75 006 Paris, France
*Corresponding author: [email protected]

Received 8 January 2008; revised 13 March 2008; accepted 14 March 2008;

posted 17 March 2008 (Doc. ID 91529); published 18 April 2008

The technique used by Leonardo da Vinci to paint flesh tints—the sfumato—has never been scientifically
depicted until now. From 100,000,000 reflectance spectra recorded on Mona Lisa, a virtual removal of the
varnish is first obtained. A unique umber pigment is then identified in the upper layer and an exceptional
maximum of the color saturation is underlined, both characteristics of a glaze technique. The modeling
calling upon the radiative transfer equation confirms this maximum of saturation, the identification of an
umber in the upper layer, and moreover underlines a mixture of 1% vermilion and 99% lead white in the
base layer. Finally, the modeling, using the auxiliary function method, explains the spectacular maxi-
mum of saturation by the multiple scattering. © 2008 Optical Society of America
OCIS codes: 300.6550, 110.4234, 030.5620, 290.4210, 330.1690.

1. Introduction phase function for each identical pigment, plane in-

This study is the first direct application of previous terfaces and a Lambertian scattering base layer have
theoretical research on art glazes, already published been perfectly validated by comparison between the
in the same review [1,2], to a real painting. In this modeling and the bidirectional spectrometric mea-
artistic technique, the pigments are the only scatter- surements on school samples [1]. Especially the color
ing centers. The pigments are randomly spread with variations obtained with this technique lead to an ex-
a very low volume concentration (∼2%) in a transpar- ceptional maximum of saturation that no other tech-
ent binder (mainly linseed oil). The artist only varies nique can reach and that can be considered as the
the number of glaze layers to modulate the lightness specificity of a glaze. This property has been ex-
and the color saturation without adding any white or plained by the calculation of the different light flux
black pigments, and thus creates a typical excep- in the pictorial layer showing the predominance of
tional rendering [3]. Previous research has explained the multiple scattering, which is magnified by the
this special visual appearance by studying the inter- low concentration of scatterers [2].
action between light and matter, calling upon the Here we intend to implement the inverse process,
radiative transfer equation (RTE) [4] and using i.e., to guess if a glaze technique has been used in a
the exact solving of the auxiliary function method real painting and with which components, starting
(AFM) [5]. A very simple model with an isotropic from experimental reflectance spectra. We chose to
study the Joconda, painted by Leonardo da Vinci
0003-6935/08/122146-09$15.00/0 in Florence between 1503 and 1506, for its interna-
© 2008 Optical Society of America tional importance, but also because the rendering of

2146 APPLIED OPTICS / Vol. 47, No. 12 / 20 April 2008

her flesh tint is one of the present crucial subjects of 2. Mona Lisa’s Reflectance Spectra
research in art history and Leonardo never wrote any
description of his method. A particular word, the A. Multispectral Camera
“sfumato” (coming from a past participle of the Ita- The high-definition multispectral camera [21] con-
lian verb “sfumare”) has been coined [6] to describe ceived by Lumiere and Technology uses a 12,000 pix-
Leonardo’s technique, expressing fine shading and el CCD sensor. A stepper motor moves the sensor on
soft transitions between colors and tones. Much his- 20,000 vertical lines in order to obtain a maximum
torical and scientific information can assume a glaze definition of 240,000,000 pixels on each channel.
in Leonardo’s sfumato [7], a technique devised by the The recording process uses 13 filters with 40 nm
Primitive Flemish painters, when the Italian artists bandpass, 10 filters in the visible range, and 3 filters
painted the faces at the same time with a mixture of in the infrared range. The correction process is ap-
pigments, currently made of vermilion and lead plied to take into account the focusing on each chan-
white. Vasari [8] in 1550 explained that the Italian nel. The lighting is obtained with two elliptic
painter Antonello da Messina, after his long Flemish projectors made of eight halogen lamps synchronized
stay, taught the glaze technique to his Italian friends with the sensor moving and providing a 100; 000 lx
such as Leonardo da Vinci. Several art historians illumination of the 53 cm × 78 cm painting. The non-
then suggested a micropointillism [9] or a stratifica- uniformity and anisotropy of the illumination is cor-
tion of translucent paint layers [10]. However, most rected by spectrometric measurements of a white
board [21]. The obtained configuration is close to
of them focus on the effect produced by the technique
57°=0°, and allows neglecting the light reflected by
[11–14]. With the same aim, numerous scientists
the upper surface in the collected beams. For each
have taken part in a better knowledge of the master-
of the 100,000,000 used pixels, 13 reflectance factors
piece, mostly focusing on the understanding of the are recorded using a white reference for the normal-
emotion expressed by Mona Lisa’s face [15–18]. Fi- ization and smoothed with the pseudoinverse
nally, a measurement campaign took place in 2004 method to get the corresponding diffuse reflectance
using only nondestructive methods in order to ex- spectra [22]. Figure 1(a) presents the result of the
plore the used materials and their locations. A scanning.
high-resolution three-dimensional (3D) scanner
using a color coded depth image has underlined a B. Mona Lisa Virtually Unvarnished
succession of thinly applied layers with different To analyze the spectra and to depict the composition
thicknesses in Mona Lisa’s face [19]. X-ray fluores- of the sfumato, the aged varnish present on the
cence has identified mercury, lead, iron, and manga- painting, and inducing an important yellowing, must
nese in different locations of the face, certainly due to necessarily be taken into account.
the presence of vermilion (mercury sulfide), lead A first method can calculate the modification of the
white (lead carbonate), and umber (iron oxide with light scattered by the paint due to the varnish. The
a small quantity of manganese) pigments [20]. Final- latter induces a light absorption according to Beer’s
ly, multispectral imaging has been implemented and law and a multiplicative factor reducing the reflec-
the recorded spectra are analyzed in this paper. tance factors, which depends on the wavelength
The huge advantage of a multispectral camera is to [23]. It also induces a leveling of the paint surface
simultaneously record 100,000,000 reflectance spec- involving a downward translation of the spectra
tra on the whole painting in a short time. These raw [24]. These physical properties depend on the nature
data are analyzed in Section 2. They allow us to pro- of the varnish and on its state of degradation. Unfor-
vide a virtual unvarnishing of the painting to identify tunately, no varnish identification has been imple-
the pigments embedded in the superficial layers, and mented during the measurement campaign and
to calculate the trichromatic coordinates for each this method cannot be used here.
So, a second statistical procedure of virtual var-
spectrum and thus for each location. The modeling
nish removal has been used from the reflectance
is then implemented in Section 3, calling upon
spectra initially recorded. For this purpose a color
RTE and AFM, and using the pigment identified
chart of varnished and unvarnished paints, made of
in Subsection 2.C with their optical properties and classical pigments used in Italy in the 16th century
different assumptions for the base layer. The and different kinds of artificially aged varnishes, has
modeling leads to results in good accordance with been realized. After comparison of the varnished and
the experimental spectra only for a given base layer, unvarnished spectra of the chart, a mean multiplica-
which is then identified. Thus Section 3 presents a tive factor was deduced for each wavelength and ap-
new nondestructive method to depict the composi- plied to Mona Lisa’s spectra [25].
tion of stratified layers and moreover it confirms Figure 1(b) presents the results of this virtual var-
the use of a glaze technique in Mona Lisa’s sfumato. nish removal, showing the possible original colors as
Subsection 3.B explains the exceptional maximum they were when Leonardo offered the painting to
of color saturation of the sfumato by the separation King François I. In particular the lapis lazuli identi-
of the different light flux here applied to Mona fied in the sky has lost its green color due to the yel-
Lisa’s face. low old varnish and has recovered its blue hue. It is

20 April 2008 / Vol. 47, No. 12 / APPLIED OPTICS 2147

Fig. 1. (Color online) Mona Lisa (a) Today, image obtained with the multispectral camera. (b) After virtual removal of the varnish. (c) Ele-
ven selected areas in the unvarnished face. (d) Diffuse reflectance spectra of the 11 locations.

the same for the face that finds again its predomi- which is the first condition to assume a glaze
nant pink hue. technique.
To study the sfumato, we now focus only on Mona The identification of the pigment has been done
Lisa’s face, where about 10,000,000 spectra have first using a spectral database made of more than
been recorded and virtually unvarnished. Only 200 reference pigments and dyes [26] with a perfect
11 locations have been selected in the face [see plane surface. The best matches are numerically se-
Fig. 1(c)], and the corresponding unvarnished spec- lected by the lowest least-squares distance ε between
tra are presented [see Fig. 1(d)], allowing a clear the unknown spectrum and the reference spectra,
discrimination between the spectra. All the other taking into account an upward translation due to
spectra stand between the presented ones and can the roughness of the studied surface. Figure 2 pre-
be deduced one from the other by a vertical trans- sents the four best matches corresponding to the
lation. lightest location (j) of Mona Lisa’s face. This compar-
ison allows the identification of an umber pigment in
the upper layer perhaps burnt and coming from
C. Pigment Identification of the Upper Layer
Cyprus, i.e., an iron oxide with a low percentage of
All the spectra show the same characteristics: the manganese oxide, which is in accordance with x-ray
slope breaks occur for the same wavelengths and fluorescence results [20].
the slopes are similar. This reveals the presence of To check the assumption of a mixture made of ver-
the same pigment composition in the upper layer, milion and lead white, as classically used by Italian

2148 APPLIED OPTICS / Vol. 47, No. 12 / 20 April 2008

 sen using the D65 standard illuminant (color tem-
perature equal to 6504 K corresponding to a
moderated daylight at European latitude) and the
2° standard observer defined by the C.I.E 1931. In
the Cartesian representation, L
is the lightness,
a
is the green–red axis, and b
is the blue–yellow
axis. In the polar representation a color is also
defined by the lightness L
, the hue h ¼
−1
tan ðb
=a
Þ,
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
ffi and the chroma or saturation C ¼
a
þb
). The hue is essentially due to the pig-
2 2

ment absorption property and thus presents little

variation in the face due to the same pigment. The
colors of the previous 11 selected areas are then pre-
sented by crosses in the plane C
–L
in Fig. 4. This
representation shows the presence of a very distinct
maximum of the color saturation in the face. This
maximum and the presence of a unique umber pig-
ment are two arguments underlying the use of a
Fig. 2. Pigment identification of the upper layer corresponding to glaze technique in the sfumato.
the unvarnished spectrum of location (j). The four best matches are
presented with the smallest least-squares distance ε between the 3. Modeling Using the Radiative Transfer Equation
unknown spectrum (solid line) and those of reference pigments and the Auxiliary Function Method
(dotted line) supplied by OKHRA (Provence).
A. Pigment Identification of the Base Layer
painters, another spectral database made of only
these mixtures with different proportions was used
for the numerical comparison and did not give rele- 1. Validity Domain of the Radiative Transfer
vant results as shown in Fig. 3. Equation
Finally, the recent commercial color matching soft-
The radiative transfer is relevant to describe the in-
ware Ciba COLIBRI, developed by Ciba Konika
teraction between light and scatterers if the light re-
Minolta, which can recognize pigment and pigment
mains incoherent. This condition requires an
mixtures, was implemented and gave the same iden-
incident incoherent light and a very low concentrated
tification of an umber. The presence of a unique pig-
medium. This is the case of a glaze paint where the
ment satisfies the first criteria of a glaze technique in
pigments are the scatterers illuminated by white
this upper layer.
light. Scalar variables can then be used and a flux
D. Colorimetric Variations in Mona Lisa’s Face balance is calculated with the RTE [4] for a given
The previous unvarnished spectra also allow the cal- wavelength in a given direction defined by its angle
culation of the trichromatic coordinates of the θ with the normal to the interface or by μ ¼ cos θ. The
selected locations. The colorimetric color space flux balance is recalled in Eq. (1) for an elementary
CIE-Lab defined by the C.I.E. 1976 [27] has been cho- layer with an optical thickness dτ:

Fig. 3. (Color online) Reflectance spectra of the mixture made of Fig. 4. Colorimetric variations in the chroma C
and lightness L

vermilion and lead white with various volume concentrations in planes of the CIE-Lab space of Mona Lisa’s face deduced from
vermilion (shown above each spectrum) compared to the previous the 11 previous experimental unvarnished spectra presented in
spectrum of location (j) in Mona Lisa’s face. Fig. 1(d).

20 April 2008 / Vol. 47, No. 12 / APPLIED OPTICS 2149

 dw ðμ; τÞ w ðμ; τÞ q calls upon the two-flux method for two glaze samples
¼∓  ½ðf ðτÞ þ gðτÞ; ð1Þ with the same thickness and composition applied on
dτ jμj 2
a white and a black background [28]. Both methods
where wþ and w− are the scattered flux, respectively, lead to the same results, but the second one is easier
to implement. The variations of kðλÞ and sðλÞ of the
R z the upstream and downstream directions. τ ¼
in
burnt umber are presented in Fig. 5. It must be no-
x¼0 ðk þ sÞdx is the optical thickness. The first term ticed that these coefficients are proportional to the
on the right-hand side of Eq. (1) corresponds to the
pigment volume concentration of the samples used
attenuation of the diffuse flux due to absorption
for their determination. This remark induces that
and scattering by the pigments with corresponding
only relative optical thickness will be able to be later
k and s coefficients gathered in the albedo q ¼
deduced by comparison between experimental and
s=ðk þ sÞ. The second term corresponds to the multi-
modeled spectra or trichromatic coordinates.
ple scattering in the μ direction of the scattered flux
coming from all directions μ1 . f ðτÞ is the auxiliary
function
R1 in the AFM and writes f ðτÞ ¼ 4. Modeling for Various Optical Thickness
½w þ ðμ ; τÞ þ w− ðμ ; τÞðdμ =μ Þ. The third term
μ1 ¼0 1 1 1 1
The intensity of the diffuse flux calculated in the
corresponds to the single scattering where a colli- modeling will depend on the optical thickness of each
mated flux W þ coming from the μ0 direction is scat- layer. To implement the modeling for different opti-
tered in the direction μ: gðτÞ ¼ W þ ðτÞ=2μo . cal thicknesses, we choose 13 layers made of 1–13
The exact solving by the AFM leads to a Fredholm identical films each with a thickness equal to 5 μm
integral equation on f ðτÞ [5]. The boundary condi- and a pigment volume concentration of 2% as pre-
tions for the air–glaze interfaces at τ ¼ 0 (taking into viously implemented [1]. These parameters are arbi-
account the refractive index n ¼ 1:5 of the medium) trarily chosen to obtain large variations of the optical
and glaze–base layer at τ ¼ h (taking into account thickness and get closer to the artist’s experience.
the reflectance factor ρ of the base layer) are then The intensity of the diffuse flux calculated in the
introduced. The diffuse flux coming out of the glaze modeling will also depend on the base layer. For this
layer is calculated for each wavelength and with the interface we choose mixtures of lead white and ver-
same configuration 57°=0° as that used for the milion and the boundary condition calls upon the re-
measured spectra (θ ¼ 0° for τ ¼ 0 and θ0 ¼ 57°). flectance factor of a given mixture of lead white and
The calculation is repeated for all the wavelengths vermilion (Fig. 3). The same mixture is obviously
of the visible range and the reflectance spectrum, used for all 13 glaze samples. Different simulated
then the corresponding trichromatic coordinates spectra are then obtained for 1–13 glaze layers ap-
are finally deduced. plied on the same base layer. The whole process is
repeated for different concentrations of vermilion
2. Need of a Model for the Glaze Layer and lead white.
The principle of the RTE is the substitution of the
real layer by an homogeneous one made of a unique 5. Comparison of Simulated and Experimental
kind of scatterer characterized by its absorption and Results
scattering coefficients kðλÞ and sðλÞ and its phase
function. Direct calculations of these parameters The simulated reflectance spectra are in good agree-
are complex for nonspherical scatterers. A model of ment with the spectra presented in Fig. 1(d) when
a glaze layer is then necessary. We have previously the reflectance factor of a red under layer made of
validated such a model [1] on different glaze layers
with various optical thicknesses applied on different
colored base layers by comparison between mea-
sured reflectance spectra and results of the modeling.
The simple model of a glaze layer with plane inter-
faces, isotropic scatterers, applied on a Lambertian
base layer characterized by their reflectance factor
ρðλÞ, is completely satisfying considering a 1% accu-
racy for the measurements.

3. Optical Characteristic of the Scatterers

The coefficients kðλÞ and sðλÞ of the previously iden-
tified burnt umber (see Fig. 2) have been deduced
from two methods, both using reflectance measure-
ments with a Cary 5 spectrometer and its integration
sphere. The first one calls upon the RTE solved by Fig. 5. Spectral variations of absorption k and scattering s coeffi-
the AFM for two glaze samples with different known cients (in m−1 ) of the umber identified in the upper layer of Mona
thicknesses and the same composition. The second Lisa’s face.

2150 APPLIED OPTICS / Vol. 47, No. 12 / 20 April 2008

1% vermilion and 99% lead white (see Fig. 3) is used, can then be assumed that the mixture of 1% of ver-
but too many simulated spectra are provided to be milion and 99% of lead white corresponds to the base
clearly presented on a single figure. layer on which the glaze is applied. This identifica-
The simulated reflectance spectra are used to cal- tion is in accordance with the detection of lead and
culate the corresponding trichromatic coordinates. A mercury by x-ray fluorescence [20] and the present
spline method is then implemented for getting a study adds the composition of the mixture and its lo-
curve smoothing of the results from the 13 different cation in the base layer.
glaze layers applied on the same base layer. The col- In order to definitively exclude the case of a pig-
orimetric results deduced from this modeling are pre- ment mixture in the upper layer, we have redone
sented in Fig. 6 by a full line and are in excellent the modeling for the case of an umber mixed with
agreement with the previous results (cross) directly a lead white in various proportions. The colorimetric
deduced from the unvarnished spectra. The excep- results are presented in Fig. 7 and, as expected,
tional maximum of saturation of a glaze is definitely clearly show that the maximum of saturation is
confirmed. For other proportions of the pigment mix- smaller than the one obtained with a glaze
ture of the base layer or for a lead white base layer, technique.
the simulated curves L
¼ f ðC
Þ never go through The present modeling confirms the existence of a
the experimental results. Unfortunately, it is not pos-
glaze technique by the exceptional maximum of
sible to deduce the real thickness of the glaze for the
saturation. It confirms the pigment identification
different locations (a)–(f) in Mona Lisa’s face because
of the upper layer. It adds the pigment identification
only the relative optical thickness is available from
modeling related to the pigment concentration of of the base layer. Used with a multispectral camera,
the samples used to determine k and s, as previously the RTE solved by the AFM proves to be an
noticed. innovating and powerful tool for nondestructive
recognition of artistic stratified materials and tech-
niques.
6. Last Validations
B. Explanation of the Exceptional Maximum of the
The extrapolation of the modeled curve for small
Saturation
lightness goes satisfactorily through the colorimetric
coordinates of the pure umber, which corresponds to This exceptional maximum value of the saturation
an infinite glaze layer and confirms the pigment explains the visual appearance of such a painting
identification of the upper layer. with a color that seems to be built in the bulk, as
The extrapolation of the curve for larger lightness, opposed to the case of a pigment mixture where the
corresponding to no glaze layer, actually goes color seems to come from the surface. This excep-
through the color of a pigment mixture of 1% vermi- tional saturation is due to the unique kind of pigment
lion and 99% lead white measured separately on a in the glaze and can be explained by the modeling
dry pigment mixture. A pure lead white, classically using the RTE. Its solving by the AFM allows the se-
used for a base layer in Italy in the 16th century, has paration between the different contributions to the
colorimetric coordinates (L
100 and C
¼ 0) that do diffuse flux coming out of the paint layer. The three
not fit with the simulated curve and thus are not sa- fluxes due to the single scattering, the multiple scat-
tisfactory. It is the same for other mixtures of lead tering, and the scattering by the base layer can then
white and vermilion, more concentrated in red pig- be separately calculated. This decomposition cannot
ment, they do not lead to a relevant modeling. It be directly obtained by experiments, but only by the
use of modeling. These calculations have already
been realized for school glaze samples made of chro-
mium oxide. We apply here the same process to Mona
Lisa’s face and we only present the results of the

Fig. 6. (Color online) Colorimetric variations in the chroma C

and lightness L
planes of the CIE-Lab space, deduced from the
modeling of an umber glaze (solid line) from 1–13 layers (number
in parentheses) and applied on a mixture of 1% vermilion and 99%
lead white. The coordinates of a pure dry lead white (L), a dry um- Fig. 7. (Color online) Comparison between the simulated color
ber (U), and a dry mixture (M) of 1% vermilion and 99% lead white variations of the previous glaze and those of a pigment mixture
pigments have been added. made of umber and lead white with varying proportions .

20 April 2008 / Vol. 47, No. 12 / APPLIED OPTICS 2151

calculations, which can be found in our previous when the number of layers increases because the
paper [2]. base layer is less and less reached by the incident
The diffuse flux coming from the paint layer in the light. The shape of the spectrum corresponding to
μ direction for τ ¼ 0, which is the solution of Eq. (1), a mixture of 1% vermilion in lead white (see Fig. 3)
can be written as the sum of three terms: is recognizable only for one and two glaze layers. The
large values of the outcoming flux in Fig. 8(c) com-
w− ðμ; 0Þ ¼ w− ss ðμ; 0Þ þ w− ms ðμ; 0Þ þ w− B ðμ; 0Þ: ð2Þ pared with those in Figs. 8(a) and 8(b) show that
most of the light reaches the base layer according
The meaning of these three terms is schematized in to the transparency of the glaze layers.
the second column of Fig. 8. The corresponding colors calculated in the CIE-
The first term corresponds to the light undergoing Lab space are presented in Fig. 9 in the C
–L

only a single scattering before coming out of the plane. Unfortunately, the trichromatic coordinates
paint. The second term corresponds to the light are not additive and the sum of these three curves
undergoing multiple scattering before coming out cannot be used to find again the global color varia-
of the paint. In both terms the light is only scattered tions presented in Fig. 6. When the light scattered
by pigments and has never reached the base layer. by the pigments [see Figs. 8(a) and 8(b)] is completely
The third term corresponds to the light scattered separated from the light scattered by the base layer
by the base layer. This light can undergo single or [see Fig. 8(c)], it is clearly shown that the maximum
multiple scattering before or after reaching the base of the saturation is due to the multiple scattering and
layer. The global flux, outcoming in the half upper not to the single scattering. The last component also
hemisphere and normalized by the incident flux, is shows a less pronounced maximum value of the
finally calculated for the three cases. saturation once more due to the multiple scattering
The third column of Fig. 8 presents the three by pigments before and after the scattering by the
global flux scattered by the 13 glaze layers made base layer.
of an umber pigment applied to a mixture of 1% ver- These results explain that the multiple scattering
milion and 99% lead white base layer. The by the low concentrated pigments contained in the
single scattering [see Fig. 8(a)] increases with the upper layer is the main origin of the exceptional
number of scatterers and then with the number of maximum of the saturation and that the single scat-
layers. The multiple scattering [see Fig. 8(b)] in- tering does not contribute to this maximum. This
creases up to three layers and then decreases. Its underlines the spectral filtering of each pigment,
variation undoubtedly underlines the origin of the which improves with the number of interactions be-
maximum of saturation. Moreover, one finds again tween light and pigment. The multiple scattering
the shape of an umber spectrum. Finally, the light
scattered by the base layer decreases [see Fig. 8(c)]

Fig. 8. (Color online) Three contributions to the global diffusion Fig. 9. (Color online) Colorimetric variations in the chroma C

coming from the light flux, according to the number of glaze layers and lightness L
planes of the CIE-Lab space as a function of
for an umber embedded in an oil binder and applied to a 1% ver- the glaze layer number for each contribution to the diffuse light:
milion and 99% lead white base layer: (a) single scattering, (b) mul- (a) single scattering, (b) multiple scattering, (c) scattering by the
tiple scattering, (c) scattering by the background. background.

2152 APPLIED OPTICS / Vol. 47, No. 12 / 20 April 2008

clearly favors the purity of the color, the single scat- check the presence of a sfumato before Mona Lisa.
tering being widely insufficient to do it. This spectral The results could generalize the term “sfumato” cur-
filtering by colored pigments is less effective for rently used to describe Leonardo’s technique.
layers made of pigment mixtures where multiple
scattering is less important due to the presence of We thank Chuck Siewert, specialist of radiative
white pigments in the medium, and this explains transfer at the North Carolina State University;
the smaller maximum value of the saturation in René de la Rie, head of scientific research at the
this case. National Gallery of Art, Washington, D.C.; and Yves
Borensztein at Institut des Nano Sciences de Paris
for their interest in the subject and their constructive
4. Conclusion proofreading of the paper.
The diffuse reflectance spectra recorded by a multi-
spectral camera after a virtual varnish removal
prove that the sfumato of Mona Lisa is made of a sin- References
gle pigment with a small volume concentration, an
1. L. Simonot and M. Elias, “Special visual effect of art glazes
umber, very common in Italy in the 16th century. explained by the radiative transfer equation,” Appl. Opt.
The same spectra allow the calculation of the trichro- 43, 2580–2587 (2004).
matic coordinates in the CIE-Lab space and empha- 2. M. Elias and L. Simonot, “Separation between the different
size a large maximum value of the color saturation in fluxes scattered by art glazes: explanation of the special color
the global face. These criteria attest to the use of a saturation,” Appl. Opt. 45, 3163–3172 (2006).
glaze technique inspired by the Primitive Flemish 3. E. Panofski, Les Primitifs Flamands (Hazan, 1992).
painters such as Van Eyck and Van Der Weyden, 4. S. Chandrasekhar, Radiative Transfer (Dover, 1960).
where the modulations of the color are realized by 5. M. Elias and G. Elias, “Radiative transfer in inhomogeneous
varying the number of layers without adding any stratified media using the auxiliary function method,” J. Opt.
white pigment. The maximum of saturation is con- Soc. Am. A 21, 580–589 (2004).
6. C. N. Cochin, Voyage Pittoresque d’Italie (Ch.-Ant. Jombert,
firmed by the modeling using the RTE solved by 1773).
the AFM. The satisfactory agreement between mod- 7. V. Pomarède, La Joconde (Prat / Europa, 1988).
eling and experiments also proves that the base layer 8. G. Vasari, The Lives of the Artists (Penguin, 1987), English
is made of a mixture of 1% vermilion and 99% lead translation.
white. Finally, the AFM explains the exceptional 9. J. Franck, “The Unrestorable Sfumato,” in Academia Leonardi
maximum of the saturation by separating the differ- Vinci (UCLA , 1993), Vol. VI, p. 238.
ent origins of light scattering. The typical rendering 10. M. Kemp, Leonardo da Vinci (Oxford, 2006).
of the sfumato is then optically explained by the pre- 11. D. Arasse, Léonard de Vinci (Hazan, 2003).
dominance of the multiple scattering favored by the 12. E. H. Gombrich, The Story of Art (Phaidon, 1999), p. 300.
13. K. Clark, “Mona Lisa,” The Burlington Magazine 115, 144–
low concentration of one kind of pigment in a trans-
150 (1973).
lucent medium. 14. M. Gelb, The How to Think Like Leonardo Da Vinci Workbook
Associated with a multispectral camera, the RTE (Dell, 1998).
solved by the AFM proved to be an innovative and 15. M. S. Livingstone, “Is it warm? Is it real? Or just low spatial
powerful tool for a nondestructive recognition of frequency,” Science 290, 1299 (2000), doi: 10.1126/science
artistic stratified materials and artistic techniques. .290.5495.1299b.
Until now all the results could be obtained only by 16. D. Queiros-Conde, “The turbulent structure of sfumato within
destructive sampling. In the case of Mona Lisa, Mona Lisa,” Leonardo 37, 223–228 (2004).
the combination of both techniques leads to the re- 17. J. F. Asmus, “Mona Lisa symbolism uncovered by computer
cognition of a glaze technique and to the pigment processing,” Mater. Charact. 29, 119–128 (1992).
18. L. L. Kontsevich and C. W. Tyler, “Making Mona frown,”
identifications of the upper layer and the base layer.
Science 304, 1900 (2004), doi: 10.1126/science.304.5679
It is the first time that modeling by use of radiative .1900b.
transfer and the solving of the inverse problem is 19. F. Blais, J. Taylor, L. Cournoyer, M. Picard, L. Borgeat, G. God-
applied to the field of works of art, even when it is in, J. A. Beraldin, and M. Rioux, “More than a poplar plank:
now applied in atmospheric optics, astrophysics, the shape and subtle colours of the masterpiece ‘Mona Lisa’ by
oceanography, and climatology. In the near future Leonardo,” Proc. SPIE 6491, 649106 (2007).
we intend to systematize the process by use of large 20. J. P. Mohen, M. Menu, and B. Mottin, Mona Lisa: Inside the
databases of absorption and scattering coefficients of Painting (Abrams, 2006), p. 92.
reference pigments and of reflectance spectra of clas- 21. P. Cotte and M. Dupouy, “CRISATEL: a high resolution multi-
sical base layers. spectral system,” Proceedings of PICS’03 Conference, Roche-
ster, USA (2003), pp. 161–165.
The previous and numerous assumptions of a glaze
22. P. Cotte and D. Dupraz, “Spectral imaging of Leonardo da
technique used in Leonardo’s sfumato are now ex-
Vinci’s Mona Lisa: an authentic smile at 1523 dpi,” Proceed-
perimentally and theoretically attested. It remains ings of IS&T Archiving’06, Ottawa, USA (2006), p. 228.
to carry out similar studies on other Leonardo paint- 23. M. Elias, L. Simonot, L. M. Thoury, and J. M. Frigerio,
ings such as the portrait of Cecilia Gallerani painted “Bi-directional reflectance of a varnished painting part 2:
in 1490 (Fine Arts Museum of Cracovia) or the por- influence of the refractive indices, surface state and absorp-
trait of Ginevra de Benci painted in approximately tion—experiments and simulations,” Opt. Commun. 231,
1479 (National Gallery of Art, Washington, D.C.) to 25–33 (2004).

20 April 2008 / Vol. 47, No. 12 / APPLIED OPTICS 2153

24. M. Elias, M. R. De La Rie, J. Delanay, and E. Charron, “Level- 27. G. Wyszecki and W. S. Stiles, Color Science: Concepts and
ing of varnishes over rough substrates,” Opt. Commun. 266, Methods, Quantitative Data and Formulae 2nd ed. (Wiley,
586–591 (2006). 1982).
25. P. Cotte and D. Dupraz, “Spectral imaging of Leonardo da 28. G. Latour, M. Elias, and J. M. Frigerio, “Color modeling of stra-
Vinci’s Mona Lisa: a true colour smile without the influence tified pictorial layers using the radiative transfer equation
of aged varnish,” Proceeding IS&T CGVI ’06, 311, Leeds, solved by the auxiliary function method,” J. Opt. Soc. Am.
UK (2006). A 24, 3045–3053 (2007).
26. G. Dupuis, M. Elias, and L. Simonot, “Pigment identification
by fiber-optics diffuse reflectance spectroscopy,” Appl. Spec-
trosc. 56, 1329–1336 (2002).

2154 APPLIED OPTICS / Vol. 47, No. 12 / 20 April 2008

View publication stats