Polynomial texture mapping: a new tool for examining the surface of paintings Joph Padfield and David Saunders Scientific Department The National Gallery Trafalgar Square London WC2N 5DN United Kingdom E-mail: Joph.Padfiuk; David.uk Tom Malzbender Hewlett Packard Laboratories 1501 Page Mill Road Palo Alto CA 94306, USA E-mail: tom_ Introduction Although many of the photographic methods ud in the technical examination and documentation of paintings during conrvation treatment now have improved digital analogues, raking light photography has received relatively little attention. A raking light photograph, made by casting light across the surface of a painting at a very low angle, highlights any surface texture or irregularities,
including incisions, impasto, raid or flaking paint, and damages or deformations
of the canvas or panel.
There are two major problems with this form of examination. As raking light
images are normally made with the lights in only one, or perhaps two, positions,
the information captured depends strongly on the choice of lighting position; a
raking light photograph designed to accentuate one area may not reveal features
in another part of a painting. Second, becau it is difficult to document the
position of the lights accurately enough to allow them to be returned to exactly鸣人日雏田
the same location, a comparison between images made before and after a painting
has undergone some physical change, perhaps as a result of conrvation
treatment, is rarely possible.
There are veral methods available to look at the surface of a painting in a
more quantitative and reproducible manner, including lar scanning (Beraldin et
al. 2003) or imaging under structured light (Guidi et al. 2004). Both the
techniques work well and show great promi, but the equipment needed is both
specialized and very expensive, and it is difficult to store the large quantity of data
generated. In the recent study of Leonardo da Vinci’s Adoration of Magi , the size
of the data files meant that the whole painting (240cm ×240cm) could only be
examined at low resolution, with a high-resolution study rerved for a central
(58cm ×77cm) portion (Guidi et al. 2004).
Polynomial texture mapping
To overcome some of the drawbacks, the authors have been investigating the
application of polynomial texture mapping (PTM) to the improved examination
of surface features of paintings. PTMs were first developed for rendering
applications in three-dimensional (3D) computer graphics, but also have been ud
in 2D image processing applications (Malzbender et al. 2001). PTMs are compact
reprentations of the appearance of a surface under varying lighting conditions,
which can be produced either from images of a ‘real’ surface or analytically from
基质育苗a mathematical model; the approach described here is image-bad. The technique
has previously been applied successfully to render the surface detail on three-
dimensional muum objects, including an Egyptian Ushabti , cuneiform tablets
(Malzbender et al. 2001), and fossils (Hammer et al. 2002).
The advantage of a PTM is that it allows the viewer to simulate on the
computer screen the effect on the appearance of the painting of moving a point
source of light to any position between the viewer and the object. The painting can be en under ‘normal’ illumination, but can also be viewed as if lit by a
504
P UBLISHED IN THE 14TH T RIENNIAL M EETING T HE H AGUE P REPRINTS V OL I Abstract
Polynomial texture mapping (PTM)
offers advantages over traditional raking
网名怎么取light photography for examining and
documenting the surface texture and
shape of paintings. It is less expensive
and computationally intensive than
structured light or lar-bad scanning
techniques. Several small paintings and
mock-up paintings have been
examined and it is shown that PTMs
are able to record surface features
including craquelure, planar distortion,
wood grain, canvas weave and
pentimenti . By making images before
and after physical change, the PTM
technique can monitor and map change
to the surface texture and shape of
paintings. Future development of the
system is described briefly.
Keywords
paintings, surface texture, impasto,
treatment, alteration, PTM, raking light
raking light that can be moved through 360°around the object. Becau the PTM rendering is indepe
ndent of where the lights were located during the image acquisition pha (as it interpolates between the positions) comparisons can easily be made between PTMs made before and after changes or treatments.Finally, by extrapolating the data, it is possible to render the image as if it were lit from a lighting direction that is more ‘raking’ than is physically realizable; the resultant incread contrast is often helpful for visualizing surface detail, as described in one of the examples that follow.The PTM software developed at Hewlett Packard Laboratories in Palo Alto requires a ries of images of the painting made under varying, but well-defined,lighting conditions; normally 24–50 images are taken, each with the surface illuminated by a point light source from a unique direction. The images, along with information about the position of each of the lights, are pasd to the software and rendered into the model of the painting surface described above. A description of the mathematical basis of the PTM technique is not included here;interested readers are referred to the paper describing its development (Malzbender et al. 2001).This study aims to show that PTMs can help to overcome the two disadvantages of simple raking light photography using relatively inexpensive equipment and modest computer power. First, through the examination of veral small paintings from the collections of the National Gallery and Tate,London, the authors t out to demonstrate that typical surface features, including impasto and planar distortion, can be rendered easily visible and documented. A cond part of this study examined a ries of mock-up paintings on canvas or panel
before and after they had been subjected to physical alteration, for example dropping or flattening of impasto, to asss the reproducibility of the PTMs and their value in documenting change.Making polynomial texture mappings of paintings It is possible to produce a PTM by supporting a painting on a standard eal in front of a digital camera and to acquire the t of images required for the PTM software by moving a single photographic light between each exposure. It is necessary to measure the position of the light accurately before each acquisition,which is time-consuming and prone to error. While the National Gallery was evaluating the ufulness of the PTM software for examining paintings, this approach was tested, but was quickly rejected in favour of a system comprising a camera and a t of lights in fixed positions.A prototype system was designed to make PTM images of small paintings (Figure 1). This compris an octagonal wooden ‘dome’ supporting 24 lights in three tiers, each of which can be lected individually. The 12V, 50W tungsten halogen lamps illuminate the painting at angles of about 60, 30 and 2°;the last corresponds to raking light. To reduce reflections, the reflectors of the lamps not in u were shrouded with matt black material and the structure of the dome was painted black.The images for the PTM are made with a digital camera mounted at the apex of the dome; originally a Nikon Coolpix 995 camera was ud, but this has been replaced by a higher-resolution Sony DCS-F828 camera. Before examining a
painting, a t of reference white target images is made by illuminating a matt
white Teflon sheet with each lamp in turn. The images of the painting are then
made; the acquisition process takes around 10 min, during which time the
painting has been expod to a total of 800lux h, equivalent to around 4h on
display in the Gallery.昙花一现只为韦陀
At this stage adjustments can be made to the t of files to account for slight
rotation of the image with respect to the camera, to rotate paintings with a
‘portrait’ format through 90°, to reduce the resolution if required, and to crop
the images. The colour of each image is corrected by reference to the
corresponding image of the white target illuminated with the same lamp. Finally
the images and a data file (defining the number of images, the filenames for each
怎样清理滚筒洗衣机
image file, and x , y and z coordinates for the light source at the time of capture)are pasd to the rendering software to produce the completed PTM file.
V OL I
Paintings: scientific study, conrvation and restoration
505Figure 1. The prototype lighting dome
created at the National Gallery
506ICOM C OMMITTEE FOR C ONSERVATION, 2005V OL I
V OL I
Paintings: scientific study, conrvation and restoration 507
Examination of paintings The PTM renderings from the paintings detailed in Table 1 clearly displayed
the ability of the technique to detect and display features including impasto, cracks,general and point surface deformations, canvas weave, wood grain and even
pentimenti . The scale of the features captured varies from fine canvas weave or wood grain and individual brush strokes, to areas of thick impasto and distortion of the surface of a panel painting caud by desiccation of the wooden support.Figure 2 shows a rendering of the Portrait of Jean de la Chambre at the Age of 33by Frans Hals (National Gallery, NG 6411) from the PTM visualization of this painting, as if illuminated by raking light from the left. The brush strokes in the young man’s cheek, ear and forehead are evident, as is the creamy paint in the sitter’s ruff. However, the image of this painting is also characterized by two other features. First, the strong wood grain is visible in the PTM visualization becau of the unusually thin paint layers in this painting, which allow the grain of the wooden panel to become evident in low-angle light. The cond feature is the liberal scattering of dust on the surface of the panel, particularly visible in the dark passages of the painting. Although this emphasizes the need to ensure the painting surface is free from dust before imaging, a clor examination of the image revealed that not all the material first thought to be dust was easily displaced. The gritty texture of the surface in some places is very like that reported in paintings containing lead (or other metal) soap protrusions (Higgitt et al. 2003, Van der Weerd et al. 2002).The rendering of Willows, with a Man Fishing by Jules-Louis Dupré (NG 2634)shown in Figure 3 provides a good example of how a PTM can show structural effects in canvas paintings. In the sky it is possible to e the raid edges of the craquelure. In the sky above the right side of the central tree,
a circular crack can be en which is typical of a point impact, suggesting that the painting has been subject to this type of shock at some stage in its history.Canvas deformation of a different type can be en in the rendering of a detail from a third painting, Figure 4, where cusping can be en at the left side of the canvas. This image, which has been made by extrapolating the lighting position as described above, also provides some information about the nature and depth of the paint layers and the technique employed by the artist. The background paint, to the left of the image shows a raid lip of paint, where the artist has painted the background up to the edge of a figure, which was probably painted earlier.Pentimenti, or ‘changes of mind’ by the artist between the initial design stages of the painting and its finished state are often visible using other imaging techniques such as infrared reflectography or X-radiography. However, the PTM
网络作家Figure 2. Rendering of the Portrait of Jean
de la Chambre at the Age of 33 from the
PTM, showing the painting lit as though
by raking light from the left
Figure 3. Rendering of Willows, with a Man Fishing from the PTM, showing the painting lit as though by raking light from the top right Figure 4. Rendering, showing a painting lit
as though by raking light from the extreme
left; the lighting position has been
extrapolated using the procedure described in
the text
visualization technique has also proved to be a uful additional tool for this type of study. Figure 5 shows one rendering from the PTM file for The Seine en from La Grande Jatte by Georges Seurat (NG 6558). To the left of the painting a triangular t of brushstrokes is easily visible, which reprents the sail of a boat on the Seine that does not feature in the final painting.
Examination of mock-ups
The surface features of the test canvas and panels are summarized in Table 2,along with a description of the changes made to the mock-up paintings and the results en using the PTM visualization. All the PTMs for the mock-up paintings can be viewed in detail on the web-site given earlier (uk/ptm/); here we describe, and illustrate in two dimensions, a few of the results recorded.
A first t of PTMs was made for the bare canvas or panels; the main features are the weave of the canvas, which appears rather fuzzy due to the nap of the canvas, and the grain and saw marks on the wood panels. The t of PTMs made after priming show slight cockling near the corners of some of the canvas, due to the changes in tension when a wet ground is applied and subquently dries.The PTMs of painted canvas clearly show the weave of the support, the brushstrokes or impasto and, in some cas, deformation of the canvas at the inside edge of the stretcher. The PTMs for the painted panels still show wood grain and saw marks, in addition to the brushstrokes or palette knife marks.Figure 6 shows renderings from the PTMs made for canvas sample C1 before and after the high impasto was flattened against a heavy weave canvas. The impression of the canvas weave and deformation of the impasto can be en particularly clearly around the edges of the painting in the rendering to the right.508ICOM C OMMITTEE FOR C ONSERVATION , 2005V OL I
Figure 5. Rendering of The Seine en from La Grande Jatte from the PTM, showing the painting lit as though by raking light from the right
Figure 6. Canvas sample C1 before (left) and after (right) flattening of the impasto. Renderings from the PTM, showing the painting lit as though by raking light from the top: note the crushed impasto and impresd canvas weave
Renderings from one of the two overpainted mock-ups are shown in Figure
7. To the left is the top left corner of panel P4 before it was overpainted,rendered as if lit by raking light from the top. The centre image shows the same area after overpainting, again as if lit from the top. In the overpainted area, the brushstrokes of the new layer are more clearly visible than tho for the original paint. However, if the lighting position is changed so that the area is lit as if from the top left, the brushstrokes and impasto from the original layer become more prominent (Figure 7, right). When viewing the PTM on the computer screen it is possible to move slowly between the two light
ing states interactively to obrve the gradual change in the prominence of the two paint layers.
A third example is prented in Figure 8, which shows renderings for canvas sample C3 before and after the painting was scratched from behind. The distortion is clearly visible in the rendering at centre of the figure, but even more evident in the image to the right, which was produced by subtracting the renderings made before and after the canvas was damaged. In addition to the local distortion caud by the scratch, it is possible to e cracking of the paint around the deformation and slight shifts in the prominent brushstrokes to the right of the scratch. The latter are probably rendered visible becau of a slight overall horizontal stretching of the canvas in respon to the force applied during scratching.
The subtraction to the right of Figure 8 is only possible becau the before and after PTMs can be precily superpod and the lighting direction lected to be exactly the same in each rendering; such a comparison would be extremely difficult using raking light photography.
Conclusions
The results show that a system bad on a standard PC, an inexpensive digital camera, free software and a lighting system costing less than €500 can be ud to V OL I Paintings: scientific stud漂亮英文
y, conrvation and restoration 509
描写绿色的成语Figure 7. Rendering from the PTMs of panel sample P4. Left, before overpainting as though lit by raking light from the top; centre, after overpainting as though lit from the top; right, after overpainting as though lit from the top left
Figure 8. PTM renderings for canvas sample C3 before (left) and after (centre) scratching. The image to the right is the result of subtracting one of the renderings from the other
