Creagh D., Bradley D. (Eds.) Physical Techniques in the Study of Art, Archaeology and Cultural Heritage. Volume 2

Подождите немного. Документ загружается.

The second Nolan to be tested was Cove at Hydra, with much darker blue paint present

in the sea (Fig. 41). This painting was interesting as the paint was solidly applied at the

horizon line, and gradually washed out towards the bottom using a diluent, becoming paler

in tone as the white ground was allowed to show through. This permitted testing on differ-

ent concentrations of the same paint, which is similar in effect to mixtures with white. As

with Gallipoli Landscape, the dark blue was found to be phthalocyanine blue (Fig. 42).

242 M. Kubik

100

400

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

Fig. 41. Sidney Nolan (1956) Cove at Hydra.

Fig. 42. Phthalocyanine blue reflectance spectrum and distribution.

Distribution for this reference spectrum, however, was limited to the dark blue in the back-

ground, and not the diluted areas in the foreground. A similar effect was observed with

diluted cerulean Blue.

The third Nolan painting to be tested was Gallipoli Landscape with Gun (Fig. 43). The

blues in this third painting appear to consist solely of indigo blue, seen distributed in the

foreground and left sky (Fig. 44).

Thus, it may be concluded that Nolan used a combination of indigo and phthalocyanine

blues in the three paintings studied. Indigo and phthalocyanine could easily be identified

Hyperspectral Imaging: A New Technique for the Non-Invasive Study of Artworks 243

100

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

Fig. 43. Sidney Nolan (c1963) Gallipoli Landscape with gun.

Fig. 44. Indigo blue, distribution and reflectance spectrum.

separately, based on their reflectance spectra, which also matched those collected during

blue reference studies. As before, the spectrum of indigo shows a much higher reflectance

in the NIR range, whereas phthalocyanine is practically absorbent throughout the

400–1000 nm range.

4.1.7. Application 5: Ivor Hele and a copy

Also, at the Australian War Memorial, it was possible to compare a recent copy against the

original painting by Ivor Hele–Private John Growns (1952) (Fig. 46). As the artist of the

copy, John Ballan, was still in contact with the AWM, it was possible to ask which

244 M. Kubik

Fig. 45. Imaging system set up at the Australian War Memorial’s Treloar offices, with data

capture of Ivor Hele.

Fig. 46. Ivor Hele (1952) Private John Growns, and Ballan (2005) Copy of Private John

Growns.

pigments he had used for his copy. According to Ballan, his palette consisted of the following

pigments:

–Lead white

–Yellow ochre

–Mars red

–Alizarin crimson

–Raw umber

–Payne’s grey

–Cadmium yellow

–Cadmium red

The captured images of the Ballan copy produced a good-quality data cube, with a flat

response from the Spectralon cube, and therefore a good-quality spectra for analysis and

comparison. For identification, a Gamblin target containing most of the above pigments

was captured under the same lighting conditions to replicate pigment spectra more accu-

rately. This permitted searching for distribution of known pigment spectra and possible

identification of paints used within the two target images.

Spectra were extracted from different areas of both paintings and compared (Figs. 47–50).

From this, it may be seen that similarly coloured regions revealed dissimilar spectra and

distributions, easily highlighting the use of different paints and techniques, e.g. facial

colour, seen in Figs. 47 and 48, where Ballan has extended the use of this paint into the

forehead and right shoulder. The two spectra could not be identified against standards, but

are possibly a combination of yellow ochre and cadmium red, with a yellow peak at

600 nm evident in some of the more yellow areas (Fig. 51), and a flat, high reflectance

between 750 and 1000 nm, characteristic of cadmium paints (Fig. 24). In the copied

version by Ballan, raw umber was identified in the dark brown regions of the hat, with a

small reflectance maximum at 700 nm (Fig. 49). The distribution of this pigment on its

own appears limited. When testing similarly coloured dark areas of the Hele painting

(Fig. 50), distribution of this colour is more widespread. The spectrum of this region has a

Hyperspectral Imaging: A New Technique for the Non-Invasive Study of Artworks 245

100

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

Fig. 47. Ballan results: facial colour distribution and spectrum.

246 M. Kubik

100

400

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

100

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

100

400

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

Fig. 48. Hele results: facial colour distribution and spectrum.

Fig. 49. Ballan results: limited umber distribution and spectrum.

Fig. 50. Hele results: shadows distribution and spectrum.

reflectance peak at 800 nm, so that this colour must be attributed to a different pigment or

mixture.

Other colours tested, such as the green uniform and purple background, could not be

identified. In fact, few pigments could be spectrally matched due to the complexity of

spectral mixtures. Instead, spectra revealed working methods and allowed comparison

between the two paintings to see whether the same mixtures had been used. Dark brown

areas varied between the two artists, and similar areas or spectra could not be found. The

raw umber regions also did not match those of the shadows, meaning a different dark

pigment had been used. While similarly shaped spectra were produced by both portrait’s

face colours (Figs. 47 and 48), a slight variation in reflectance peak maxima was observed,

possibly due to the contributing effects of a mixture.

The technique is therefore useful to discriminate similarly coloured areas by distribution

and reflectance spectra, but the complexity of mixture analysis is evident. An unexpected

result was found by comparing reflectance images at 1000 nm. Here, the two paintings

were easily differentiated as Ivor Hele had resorted to extensive underdrawings before

commencing painting (Fig. 60), while Balan had not make preparatory sketches.

4.1.8. Application 6: documenting and monitoring varnish

UV light is not useful for pigment identification as it does not penetrate below any surface

coating such as dirt, dust, or varnish. Instead, it offers the opportunity to detect and moni-

tor the yellowing of such varnish layers present. Detecting varnish on a painting has typi-

cally been achieved using UV light fluorescence and documented using a normal camera

with UV cutoff filter (such as the Kodak 2E). This may similarly be achieved using the

hyperspectral imaging system with only three interference filters. A colour composite is

created using the three wavelengths 450, 550, and 650 nm to create a true colour replicate

of any fluorescence detected (Fig. 52). The various features typically detected in this

manner are thus easily revealed, including breaks in the varnish layer caused by a tear

Hyperspectral Imaging: A New Technique for the Non-Invasive Study of Artworks 247

100

400

450

500

550

600

650

700

750

780

800

830

850

880

900

930

950

980

1000

Wavelength (nm)

% Reflectance

Fig. 51. Hele results: highlights distribution and spectrum.

repair, retouchings, a test varnish removal section, and several areas of losses. To achieve

correct white balance, equal reflectance must be achieved for the three colours using the

Spectralon plate. This is achieved by varying the exposure times.

Also, the yellowing of varnish may be monitored by collecting reflectance spectra. To

test this, a partially cleaned painting with aged dammar varnish (Fig. 53) was tested. The

central yellowed stripe represents the remaining varnish, while the areas to the left and

right of this have been cleaned using solvents. Reflectance spectra collected from the two

areas could then be compared (Fig. 54).

248 M. Kubik

Fig. 52. A colour composite image of varnish fluorescence is created by combining the

three images at green, blue, and red wavelengths.

Fig. 53. Partially removed varnish with tested areas (blue dot = cleaned area, yellow

dot = aged varnish).

Comparing the two reflectance curves, it can be seen that the varnish causes not only a

drop in reflectance, but also a shift in the peak at 500–550 nm, describing the yellowed

effect. A peak shift is also noted from 800 down to 780 nm.

Thus, the gradual yellowing of a varnish layer may be monitored objectively using

hyperspectral imaging. This application demonstrates the advantage of this method that, in

taking measurements under exactly the same experimental conditions, comparison against

reference standards is straightforward, and relative measurements and differences become

much more significant than with other methods.

4.2. Discussion

Based on the application and results of the new hyperspectral system, it was found to be a

successful technique that was easy to apply. It is shown that a wide range of colours and

pigments may be identified in this manner.

4.2.1. Ambient light

The effect of ambient light on measurements was negligible; to check, measurements were

taken with projector lights only, with projector lights in combination with fluoro, and/or

tungsten. This did not cause a significant change in intensity of the Spectralon plate

throughout the region 400–1000 nm (Fig. 55). The system may thus be considered to be

immune to the effects of ambient lighting, provided calibration using the Spectralon refer-

ence plate has been performed.

Hyperspectral Imaging: A New Technique for the Non-Invasive Study of Artworks 249

Fig. 54. Reflectance spectra of cleaned and varnished areas.

4.2.2. Reducing the number of filters

While in an optimal system even more bands than were used are desirable in order to cover

all possible colours, an increase in filters also increases the associated cost and processing

time (Day, 2003a). It also dramatically increases the difficulties in ease of use in terms of

handling and flat-fielding (Slawson et al., 1999). For basic analysis work, it may be possi-

ble to reduce the number of filters required. As seen in the comparison between

UV/Vis/NIR spectrometer results and the IS spectra, only a few points are required to char-

acterise the smooth spectral curves presented by artists’ pigments. This is confirmed by

other researchers, who were able to reduce the number of filters to 12 (Saunders and

Cupitt, 1993) or even 7 (Day, 2003a).

By plotting a number of different blue pigments in a variety of media, it may be seen

that discrimination between blues is already possible based on only two wavelengths, such

as 780 and 1000 nm (Fig. 56). These two points represent the start and end points of

increased spectral reflectance in the near infrared, which were found to be the most char-

acteristic points in the blues tested. While this is an oversimplification, it illustrates the

possible reduction in cost, capturing, and processing time. While two filters present a

viable system for discriminating between most blues, other colours would require filters at

wavelengths specific to them. For completely unknown pigments, many filters are

required, but the system may be tailored to certain recognition tasks, selecting those band-

widths where there is most difference.

4.2.3. Mixture analysis

When the size of the measured surface is large, it is likely that more than one material

contributes to the overall spectrum measured, and the mixed surface is overlapped onto a

single pixel. The result is a composite or mixed spectrum, where the energy reflected

from the materials combines additively (Adams et al., 1993; Ichoku and Karnieli, 1996;

250 M. Kubik

Fig. 55. The effect of ambient light on the Spectralon plate.

Roberts et al., 1998; Berns et al., 2005). Spectral mixture analysis (SMA) is used in remote

sensing applications to calculate land cover fractions within a pixel, and involves model-

ling a mixed spectrum as a combination of pure land cover spectra (Bell et al., 2002; Wu

and Murray, 2003). The software SMA models a hyperspectral image as a linear combina-

tion of end-member spectra, and anomalous materials that do not fit the model are detected

as model residuals. While opportunities for mixture analysis are offered by the

Kubelka–Munk equation, a simple demonstration is presented by the simple mixture of

Prussian blue and lead white seen in the Australian War Memorial’s small painting of plane

wings (Fig. 57).

The contribution of lead white to the lighter blues changes the spectrum in such a way

that identification is not possible using only the Prussian blue reference target.

The contribution from lead white causes not only an increase of percentage reflectance

in the lighter blue mixtures’ spectrum, but also changes the shape of the curve (Fig. 58).

This is particularly visible in a reflectance peak at 450 nm, and an upward slope in

reflectance at 1000 nm. The two lighter blue mixtures thus lie in between the two pure

spectra of Prussian blue and lead white, and would need deconvolution using the

Kubelka–Munk equation or other mixture analysis.

4.2.4. Opportunities and interference from reflected UV and NIR

Ultraviolet light reflectance studies present an extension to the objective method to moni-

tor colour changes such as fading. However, the presence of an aged, yellowed varnish

may mask the spectra of underlying paints and interfere with their identification. As can

be seen in Fig. 59, excitation light at 380 nm was completely absorbed by the surface

varnish layer, preventing any studies of the layers below. Therefore, while UV light

reflectance provided an extension of the reflectance curves used for pigment identification,

Hyperspectral Imaging: A New Technique for the Non-Invasive Study of Artworks 251

Ratio of 780 to 1000nm

0.2

0.4

0.6

0.8

1.2

1.4

Ultramarine

Cobalt Blue light

Cobalt Blue deep

Prussian blue

Phthalo Blue

Cerulean Blue

Indigo

Phthalo Turquoise

Cobalt Turquoise

Pigment

Ratio

Water colour

Tempera

Oil

Fig. 56. Discriminating blues using only two datapoints (ratio of 780 to 1000 nm).