Content uploaded by Robert Sablatnig
Author content
All content in this area was uploaded by Robert Sablatnig
Content may be subject to copyright.
4th International Symposium on Virtual Reality, Archaeology and Intelligent Cultural Heritage
(2003), pp. 1–9
D. Arnold, A. Chalmers, F. Niccolucci (Editors)
Computer Aided Analysis of Underdrawings in
Infrared Reflectograms †
Paul Kammerer, Ernestine Zolda, Robert Sablatnig
Pattern Recognition and Image Processing Group
Vienna University of Technology
Favoritenstrasse 9/183/2, A-1040 Vienna, Austria
{paul, zolda, sab}@prip.tuwien.ac.at
Abstract
Recent developments in computer vision are providing powerful tools for the evaluation of data gathered by art
historians and archaeologists. New camera hardware allows new insights into cultural heritage, especially if
infrared cameras are concerned, since they allow the of study structures that are visually hidden. In this paper
preliminary results of developing a system for automatic analysis of infrared reflectograms are presented. We
concentrate on an algorithm for the automatic segmentation of strokes in underdrawings - the basic concept of
the artist - in ancient panel paintings and the removal of cracks in infrared images. The purpose of the stroke
analysis is the determination of the drawing tool used to draft the painting. This information allows significant
support for a systematic stylistic approach in the analysis of paintings. Stroke segmentation in paintings is related
to the extraction and recognition of handwriting, therefore similar techniques to segment the strokes from the
background incorporating boundary information are used. Results of the algorithms developed are presented for
both test panels and real reflectograms.
1. Introduction
Europe’s rich cultural heritage is one of its important assets.
Recovering more of this heritage and making it accessible
to the public must be a concern of research work also in the
future. For example panelpaintings from 1400 to 1520 have
been of great influence to European culture during this pe-
riod and beyond. To learn more about the unknown draw-
ing technique below the colored surface can give new in-
sights into the working of famous artists or painting schools
and so increase significantly cultural knowledge and aware-
ness. Interdisciplinary projects between the field of art his-
tory and computer based image analysis have brought new
aspects in both of the fields to save, protect and extend cul-
tural heritage. While art historians benefit from new objec-
tive analysis methods and improved efficiency due to com-
puter based solutions, for technicians a new field of applica-
†The project was supported by the Austrian Science Foundation
(FWF) under grant P15471-MAT
tion was opened, which requires the adaptation and develop-
ment of algorithms to the specific needs of art history.
A current project develops a computer based analysis sys-
tem of underdrawings in medieval paintings. Underdrawings
are the basic concept of an artist when he starts the cre-
ation of his work of art. Therefore the art historians and
restorers are interested in investigations of these underdraw-
ings. Moreover, a systematic analysis, starting with medieval
paintings, over a longer period will bring insights into the
practice in painting schools which is still rarely examined
up to now.
Normally the underdrawing is hidden by covering paint
layers and is invisible to the observer in the visible light
spectrum. Using sensors that are sensible in the near in-
frared, especially in the spectral range from 1000 nm to 2400
nm, underdrawings of paintings can be visualized, even be-
low the hardly penetrable colors blue and green in the paint
layer. Figure 1 (a) shows an image of apanel painting and the
visualization of the underdrawing by an IR-reflectogram (b)
taken from a detail of the painting as outlined in (a).
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
(a) (b)
Figure 1: "Adoration of the Kings", master of the Schotten-
stift (1470): (a) color image (b) image taken in the near IR
range (700nm–900 nm)
To be able to investigate image processing methods for the
analysis of underdrawings a specific acquisition system with
a high resolution infrared camera is necessary. In contrast
to other infrared projects 1,2,3our goal is not only to digi-
tize, visualize, and improve images of underdrawings, but to
analyze the structure of the underdrawings with methods of
image processing and pattern recognition to obtain insights
from the unknown working procedure in medieval painting
schools of famous artists.
In this paper we present a system that will use the IR-
reflectography technique to obtain digital images of under-
drawings and that will apply image analysis methods to ex-
tract objective and reproducible information to support the
experts in studying underdrawings. Although parts of the
system are well-known algorithms and have been published
elsewhere 4,5, this paper will introduce a new challenging
field for image analysis on works of art. The paper is orga-
nized as follows. First a motivation will show the need for
a computational support for the analysis of paintings. Sec-
tion 3 gives art historic facts, necessary for the development
of algorithms. The components of the system are described
in Section 4, with emphasis on the image preprocessing for
crack elimination, stroke segmentation and stroke feature ex-
traction. Section 5 presents and discusses the results of ap-
plying the algorithms developed to IR-reflectograms as well
as test images. Section 6 will conclude with a brief overview
of work in progress and future work.
2. Motivation
Since the late 1960s examination of paintings with IR pho-
tography and IR-reflectography opened a new window for
the art historian, restorer, and conservator into the working
process of artists 6,7,8,9,10,11. It helped to visualize the under-
drawing on the ground of a painting and offered so far totally
(a) (b)
Figure 2: Color instructions: (a) the original painting (b)
color instruction "W" in the right-top part of the underdraw-
ing
unknown types and forms of composition design from the
14th to the 16th century. During the last twenty years tech-
niques of IR-reflectography have developed further world-
wide and have been more and more adapted to the needs of:
•Examination of paintings 12,13
•Restoration and conservation 14
•Fake detection 1
Meanwhile, many leading museums and centers for art re-
search and conservation (i.e. Victoria and Albert Museum,
Harvard Museum) have installed their own IR equipment for
regular use and specialized research projects are on the way
in several countries. But due to the lack of state of the art
digital IR cameras, the potential of digital image analysis
for this field of research is not used. Therefore a significant
support for a required systematic stylistic approach in the
analysis of medieval and Renaissance paintings using un-
derdrawings is still missing 15.
From the conservator’s point of view answers on individ-
ual parts of the working process will be given concerning:
•Execution between the first concept and the final result
•Visualization of paint instructions in terms of written
color names (Figure 2b shows an example of a paint in-
struction visible in the infrared image "W = white")
•Differentiation between freehand drawings and drawings
applied with different kinds of stencils 14.
Aside from the large number of images acquired and the
improvement of these images – this is already state of the
art 16 – the innovation of this project is the computer-based
analysis of the structure of the underdrawing. Analysis of
IR-reflectograms is performed primarily by visual inspec-
tion only. It is visual, that the analysis of a large number
of images has been made by naked eye examination only.
The restricted human optical retentiveness complicates the
comparison of different underdrawings concerning drawing
tools, drawing materials, and stroke characteristics.
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
3. Art Historic Background
In conservation and art history three prominent questions are
of particular interest. The first question deals with the devel-
opment of underdrawings and their relations to other draw-
ings and between underdrawings and the covering painting.
Secondly, art historians and restorers are interested in the
style of the underdrawing, and whether the underdrawing is
sketchy, freehand or a copy from a template. Finally an im-
portant question is, what kind of materials and drawing tools
are used in an underdrawing 3.
The system presented in this paper will contribute to an-
swering the last question, while providing answers concern-
ing the style or developments of underdrawings will be part
of future research. In order to analyze the strokes with re-
spect to the drawing tools used, the visual appearance is in-
vestigated. The following section gives a characterization.
3.1. Characterizing Drawing Tools / Materials
Drawing tools used in medieval panel paintings can be cate-
gorized into two different types, into those that are fluid and
into a group consisting of dry drawing material 3. In Fig-
ure 3 six examples of a stroke for both of the groups are
depicted. Three strokes represent the class of drawing tools
using fluid materials (a,c,e) and three strokes represent dry
materials (b,d,f). These examples have been taken from a
panel prepared for our experiments by a restorer.
(a) (b)
(c) (d)
(e) (f)
Figure 3: Stroke details showing tools using fluid materials
on the left, brush (a), quill (c), reed pen (e) and dry mate-
rial tools on the right, black chalk (b), silver point (d) and
graphite (f).
Our analysis approach is based on the observation that
prominent characteristics of drawn strokes are variations of
shape and variations of the intensity in the drawing direction.
Table 1 gives an overview of the characteristics of the two
groups of drawing tools. The first characteristic we analyzed
is the boundary of a stroke. It can be observed that there
are variations in smoothness depending on the drawing tool
used. While strokes applied with a pen or brush using a fluid
medium show a smoother boundary, the boundary of strokes
applied with a dry material, e.g. black chalk or graphite is
less smooth.
Table 1: Characteristics of different drawing tools and ma-
terials
Tools/Materials Characteristics
fluid materials fluid lines
- paint or ink applied by - continuous and smooth
pen or brush - vary in width and density
- pooling of paint at the edges
- droplet at the end
- different endings (brush/pen)
dry materials dry lines
- charcoal - less variation in width
- chalks - less continuous
- metal points - more granular
- graphite
4. System Overview
The analysis system, according to the standard process of
digital image analysis 17, consists of an acquisition step, a
preprocessing step, an image processing step and finally a
classification step. A schematic overview is given in Figure
4. Since the system is still under development, we can only
present the image processing part in more detail. In the fol-
lowing sections the processing steps and the subtasks will be
discussed.
IR-
reflectogram
Color-
image
Crack-
removal
Acquisition
(Input)
Pre-
processing
Mosaicing
Image-
Improvement
Registration
Image-
processing
Stroke-
segmentation
Tool
classification
Visualization
Feature
extraction
System
Output
Figure 4: System overview
4.1. Acquisition
The quality of an image acquired and therefore the quality
of the information of the images has a great influence on the
success of the image processing and analysis phase 18 . Van
Asperen De Boer 19 showed that the range of optimal trans-
mittance for many visually opaque paint layers is located in
the region around 2 µm, which is only accessible by spe-
cial electronic imaging devices. For our purpose we use a
Focal Plane Array Camera (FPA) with a PtSi sensor, which
has a sensitivity range from 1.0 µm to 5.7 µm that can be
adapted by using a band pass filter. FPA cameras have higher
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
thermal stability, higher resolution and less geometric dis-
tortions than Vidicon cameras. The digital images captured
fulfill the requirements with respect to radiometric resolu-
tion, geometric distortions, pixel-resolution and sensitivity
constancy over time for the application of further processing
steps 18.
4.2. Preprocessing
When acquiring an image with an infrared array, noise from
the detector as well as from the illumination source is to be
expected 18. Equally distributed detector noise is reduced by
calculating the average image from a series of images, taken
with the same camera setup 20.
Due to non-uniform illumination or inhomogeneities in
the sensitivity of the sensor array, the intensity values of the
digital image are vary (radiometric distortions). The devia-
tions of the intensity values are measured in images taken
from a uniform colored test-plane. This allows corrections
of sensor response in already acquired images 21 .
An important preprocessing step for building IR-
reflectograms of larger paintings is mosaicing. Panel paint-
ings can have sizes of 2mby 2mor even larger, but the res-
olution and the pixel number of the camera is limited. In or-
der to get a complete IR-reflectogram of a painting, smaller
sub images are stitched together into one larger image. The
alignment of images depends on the geometry of the acqui-
sition setup, i.e. how the camera is moved with respect to
the object. In the simplest case these are pure image-plane
translations. This can be obtained if the camera is mounted
on an XY-shift unit. This acquisition setup further allows the
combination of the two overlapping parts by simple averag-
ing. Finally changes of the brightness in different images,
which is usually a result of automatic gain control have to be
corrected. Using a positioning unit, which shifts the camera
and thus the image plane within a virtual plane, planar image
mosaicing methods 22 can be applied.
4.3. Image Processing
One major goal of the project is to identify the drawing tools
used by the painter to create the underdrawing from the ap-
pearance of the strokes in the IR-reflectogram. A step to-
wards the identification is the segmentation of the individ-
ual strokes. From the segmentation point of view, cracks are
treated as structural noise and will produce artifacts in the
segmentation step. To overcome this problem, our intention
is to eliminate the cracks while keeping the boundaries of
the strokes as accurately as possible for further analysis. The
following sections will present (1) a mathematical morphol-
ogy based method for detection and elimination of cracks,
(2) an edge-based method for segmentation of the strokes,
and (3) finally the detection of features to differentiate be-
tween drawing tools.
4.3.1. Crack Removal
During the aging of the paintings, climactic fluctuations
cause changes in the dimensionality of the panels. While
younger pigment layers are elastic enough to follow con-
tractions, a network of fine cracks (craquelé) may cover the
whole painting during the aging process. The example of an
IR-reflectogram depicted in Figure 1(b) shows several dark
thin horizontally aligned lines representing the cracks in the
ground layer. The pattern of the cracks is determined by the
background used. In the case of wooden panels, the cracks
are primarily oriented perpendicular to the grain 23.
Willingen et al. 24 have studied the appearance of cracks
and determined features to classify different types of cracks.
They differentiate between features of individual cracks
(smooth, jagged, depth, thickness etc.) and features of crack
patterns (distance between cracks, type of junctions). A sim-
ilar problem has been treated by Giakoumis and Pitas 25 .
They use a three step process which first detects the cracks
using a top-hat operator, separates them from brush strokes
using color information, and then fills them in. In contrast
to this work, we are working on greyscale images (recorded
in the infra-red region). We therefore have no color infor-
mation for separating the cracks from the brush strokes. The
information we start with is that cracks are usually thinner
than the brush strokes, and that they have a favored orien-
tation. To take this information into account, we make use
of a morphological opening 26 with a viscous reconstruction
step, which detects the cracks and fills them in in one step.
Abas and Martinez 27, on the other hand, are interested in the
structure of the crack network. They use a top-hat operator
to extract the cracks, and then extract descriptive information
about the crack network so as to classify it.
In the context of this paper, we are interested in the ability
of the viscous morphological reconstruction to reconstruct
small details while preventing certain elements from being
reconstructed. Viscosity is added to standard morphological
reconstruction by including an opening after each geodesic
dilation step. The challenges faced in the art history applica-
tion are illustrated schematically in the simple binary exam-
ple shown in Figure 5a. In this image, we wish to preserve
the thick line and all its details as accurately as possible,
while removing the thin lines which intersect it. The thin
lines are known to have a diameter of less than 10 pixels,
but intersections can sometimes result in thicker regions. An
opening of Figure 5a with a disc-shaped structuring element
of radius 5 is shown in Figure 5b. As expected, the details on
the thick line have been smoothed, however, not all the thin
lines have been successfully removed. Using a larger struc-
turing element would smooth the thick line even more, and
reconstruction cannot be used as the thin lines intersect the
thick line, and would therefore be reconstructed too.
The use of viscous reconstruction is a good solution to
this problem. We begin by creating the marker image shown
in Figure 5c by eroding Figure 5a by a disc of radius 5 pix-
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
(a) (b)
(c) (d)
Figure 5: (a) Initial image. (b) Opening of image (a) with a
disc of radius 5 pixels. (c) Marker image obtained by eroding
image (a) with a disc of radius 5 pixels. (d) Viscous recon-
struction for mask (a) from marker (c). Images are of size
256 by 256 pixels.
els. To reconstruct the initial image, we use a 3 ×3 pixel
square structuring element for the geodesic erosion and a
disc of radius 5 pixels as structuring element for the asso-
ciated opening. In order to reconstruct the small details, we
append an extra geodesic dilation onto the reconstruction al-
gorithm. The result of this reconstruction is shown in Fig-
ure 5d.
4.3.2. Stroke Segmentation and Feature Extraction
Stroke segmentation in paintings is related to the extrac-
tion and recognition of handwriting 28. Letters and words
in Western languages and symbols or signs in Chinese or
Japanese languages are built of manually drawn strokes or
lines. Many approaches start with thresholding and thinning
methods. While these methods are fast and save resources,
valuable information for a more detailed analysis of strokes
requires an approach that also incorporates boundary infor-
mation 29. We used Doermann’s segmentation algorithm in
the segmentation part of our approach, since it provides both
the boundary of a stroke and its intensity profiles, which will
be used to characterize strokes. Figure 6 gives an overview
of our approach consisting of three basic steps, segmenta-
tion, boundary refinement and feature extraction.
Segmentation In the Step I, first edgels Ei(x,y)located at
the stroke contour are detected by a Canny edge detector.
Second, based on the hypothesis that the gradient vectors
of the edgels point in opposite directions, the set of edgels
Figure 6: Schematic diagram of our approach
are grouped into distinctive pairs (cross sections). Finally,
neighboring cross sections are grouped into sets and rep-
resent a stroke segment. Figure 7(a) shows the cross sec-
tions grouped into one stroke segment and the polygonal
boundary. For further algorithmic details of we refer to 5.
Boundary refinement In Step II the approximation of the
stroke boundary by a closed polygon is refined by
"snakes", a method based on active contours 30. After de-
termining the principal component of the edgel distribu-
tion, the contour is split into two sides ("top" and "bottom"
boundary) that are treated separately. A set of gray value
profiles, perpendicular to the axis, represent the domain
for the snake algorithm. Figure 7(b) shows the equidistant
profiles in the original image, and arranged to form an
image (c). The snake moves through thisdomain to mini-
mize an energy functional determined by inner parameters
controlling rigidity and tension of the snake and an exter-
nal energy influenced by a gradient vector flow in order to
provide accurate and fast convergence to boundary con-
cavities.
Feature extraction Contour estimates with different levels
of elasticity provide descriptive information by means
of deviation against each other. We used two succeed-
ing snakes. The first rigid snake was initialized on the
coarse contour estimate. The second, more elastic snake
proceeds from this position. Figure 7(d) shows the con-
verged rigid and non-rigid snakes. MEAN of the deviation
and standard deviation (SDV) of the deviation between
the two snakes are used as descriptive features. For more
details refer to 31.
4.4. System Output
The system will provide objective support for the interpreta-
tion of underdrawings with high quality visualizations of IR-
reflectograms combined with color images on the one hand,
on the other hand a description and classification of details
of the underdrawing with respect to drawing tools and mate-
rials.
When the objects of interest are detected and described
by features (e.g. boundary, shape, orientation, color, and the
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
(a) (b)
(c) (d)
Figure 7: Segmentation and Refinement (a) cross sections
and polygonal boundary (b) edgels with axis and gray
value profiles (c) initial "top" and "bottom" boundary (d)
converged "rigid" (black) and "non-rigid" snakes (white-
dashed)
like) these features may be the input to the classification
stage. Classification basically consists of two tasks 17:
•investigation of the relation between the image features
and the object classes
•the actual classification task, that selects an optimal set
of features which allows the different object classes to be
distinguished with minimum effort and minimal errors
Results of the methods described above have to be pre-
sented in a visual form. The registration of images from dif-
ferent sources will allow the provision of a visual overlay of,
e.g. the image of the paint layer and the segmented strokes
of the underdrawing.
5. Experimental Results and Discussion
The methods developed will be applied to IR-reflectograms
and test panels. Since the acquisition of original panels with
an IR camera is ongoing work, IR-reflectograms with dif-
ferent drawing tools are not available at present. We there-
fore tested the segmentation and feature extraction algo-
rithm on test panels and the crack removal algorithm on IR-
reflectograms.
(a) (b) (c)
Figure 8: (a) Initial image. (b) Erosion of (a) by a vertical
line of length 10 pixels. (c) Viscous reconstruction of (a) us-
ing (b) as a marker.
5.1. Crack Removal
To show the application of the crack elimination algorithm
we use Figure 8a, which corresponds to the lower sub-region
from the IR-reflectogram in Figure 1b. For the erosion step
we took a priori information into account, namely that a
large majority of the cracks have a preferred orientation, as
discussed in the introduction. For the image under consider-
ation, this preferred orientation is horizontal. We therefore
take as our marker image an erosion of the initial image
by a vertical line of length 10 pixels, shown in Figure 8b.
The viscous reconstruction from this marker image, using
a 3 ×3 pixel square for the geodesic reconstruction, and a
disc-shaped structuring element of radius 6 for the opening
step, is shown in Figure 8c. While the cracks are eliminated
efficiently, the structure in the strokes remains. For more de-
tails we refer to Hanbury et al. 4.
5.2. Stroke Segmentation Results
In our experiments we studied the differences of three types
of drawing tools - brush, chalk and graphite. Test panels
(21cm x 30cm) containing sets of the mentioned strokes
have been prepared by a restorer. The test panels were dig-
itized using a flat-bed scanner with an optical resolution of
1200 dpi. Details from images, as depicted in Figure 9 have
been cropped manually. Figure 9 (a) shows a series of brush
strokes, (c) chalk strokes and (e) graphite strokes, all applied
in bottom up direction.
The result of the segmentation step is illustrated in Fig-
ure 9 (b),(d) and (f) respectively. The boundary of the stroke
segments, consisting of at least 20 cross sections are de-
picted. The segmentation algorithm works well for most of
the brush strokes and graphite stokes. Problems arise e.g.
at left stroke in Figure 9(a), which is not segmented com-
pletely, since the stroke width parameter was set too narrow.
The segmentation algorithm still has problems with overlap-
ping strokes like the "arrow top " in the left most stroke of
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
(a) (b)
(c) (d)
(e) (f)
Figure 9: The left column shows details from the test panel
with strokes used in our experiments: brush strokes (873x729
pixel) (a), chalk strokes (992x631 pixel)(c) and graphite
strokes (989x729) pixel)(e). The right column shows an over-
lay of the detected boundaries of the segmentation step
Figure 9(f) and (d). Problems occur with the chalk strokes in
Figure 9(d) which are segmented into many small segments
due to the inhomogeneity of the strokes. This necessitates a
further processing step, that will be handled together with
the overlapping problem.
5.3. Feature Extraction Results
For the refinement and feature extraction step, the stroke
segments shown are used. First, the refinement step is ini-
tialized by the boundary of the segmentation step. Figure
10(a,c,e) shows the detected boundary of the segmentation
step for three example strokes. The refinement algorithm,
i.e. the adaptation of the two snakes with different rigidity, is
applied separately to the "top" and "bottom" boundary of a
stroke. Figure 10(b,d,f) shows the example strokes together
with an overlay of the more elastic (dotted bright line) and
more rigid snake (underlying black line). It can be observed
that the deviation of the rigid and elastic snake is smaller
from the brush stroke then those from the black chalk and
graphite strokes.
To show the differences calculated, the SDV- and MEAN-
values of the deviations of the two snakes, i.e. two values,
(a) (b)
(c) (d)
(e) (f)
Figure 10: Details from the test panel showing stroke used
in our experiments: brush strokes (a), chalk strokes (c) and
graphite strokes (e). The right column shows an overlay of
the snakes to corresponding stroke sample (b,d,f)
one for the "top" and one for the "bottom" boundary, are
plotted in the diagram of Figure 11. The MEAN values of
the brush strokes (denoted as circles) are concentrated near
zero, while there is a higher variation of the MEAN graphite
strokes (denoted as "x") and brush strokes (denoted as stars).
Similarly, the standard deviation SDV of brush strokes is
below 0.2 for all but two of the stroke borders. The SDV
values for chalk and graphite is between 0.2 and 1.6 in our
samples. So using the SDV feature will allow to distinguish
between brush, i.e. a fluid drawing tool, and graphite and
chalk respectively as dry drawing tools. Using a combina-
tion of SDV and MEAN the data of our samples can be used
to differentiate between graphite and chalk, since most of
the chalk values are positioned right and above the graphite
values. Still, these results are preliminary and experiments
with more samples are necessary. Furthermore the reliability
of this differentiation can be improved if a set of strokes is
considered. As can be observed in underdrawings, in certain
regions of a drawing, a group of strokes are applied withthe
same drawing tool, e.g. as hatches or cross hatches.
6. Conclusion and Outlook
In this paper we presented a first step towards a system
for automatic analysis of IR-reflectograms. We have demon-
strated the application of viscous morphological reconstruc-
tion to eliminate thin lines (cracks), while retaining as much
detail as possible in the thicker lines (the brush strokes). The
suggested approach works well except in more complicated
regions of a painting where the brush strokes have a similar
width to the cracks. Further work on separating strokes and
cracks based on their smoothness remains to be done. The
results from the boundary analysis algorithm show that the
contour feature extracted to initialize the snakes that model
the contour are a promising way to obtain satisfactory re-
sults, although improvements in the segmentation are needed
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
−0.4 −0.2 0 0.2 0.4 0.6
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6
MEAN
SDV
p1−brush.tif
p1−chalk.tif
p1−graphite.tif
Figure 11: Standard deviation (SDV) and MEAN of the
snake deviations. The deviations are measured on the "top"
and "bottom" boundary of the individual brush, chalk and
graphite strokes.
to cope with sudden variations of the contour, as shown in
Figure 9. The first results show, that the visual appearance of
the boundary of a stroke can be used for discrimination. Fur-
ther experiments with more samples are necessary to valitate
our method.
We further plan to incorporate additional features, like the
texture of the different types of strokes, to get a measure for
granularity of a stroke. Furthermore we have noticed, that in
some cases, there is a difference between the "top" and "bot-
tom" boundary of a stroke in dry drawing tools. This obser-
vation has to be prooved and evaluated. As reported, some
problems occur in the segmentation step if the strokes are
interrupted. One of our goals is therefore to improve the ro-
bustness of the segmentation step and to extend the approach
to segment overlapping and crossing stroke formations as
e.g. reported in 32.
7. Acknowledgement
We would like to thank Prof. Mairinger for his valuable input
concerning IR-reflectography, Allan Hanbury for his support
and contribution to mathematical morphology, and Georg
Langs for providing us his tools for adaptive contour models.
Thanks to the Center of Art Conservation and the Austrian
Galleries Belvedere for providing paintings and test mate-
rial.
References
1. P. Le Chanu. Scientific examination and analysis in
the detection of forgeries of old master paintings. In
W.C. McCrone, D.R. Chartier, and R.J. Weiss, editors,
Proceedings of SPIE, Scientific Detection of Fakery in
Art, pages 62–73, 1999.
2. J.R.J. van Asperen de Boer, J. Dijkstra, and R. van
Schoute. Underdrawing in paintings of the rogier van
der weyden and master of flémalle groups. Nederlands
Kunsthistorisch Jaarboek, 1990.
3. D. Bomford, editor. Art in the Making, Underdrawings
in Renaissance Paintings. National Gallery, London,
2002.
4. A. Hanbury, P. Kammerer, and E. Zolda. Painting
crack elimination using viscous morphological recon-
struction. appears in 12th Intl. Conf. on Image Analysis
and Processing, ICIAP2003.
5. P. Kammerer, G. Langs, R. Sablatnig, and E. Zolda.
Stroke segmentation in infrared reflectograms. 2003.
The 13’th Scandinavian Conference on Image analysis,
SCIA2003.
6. F. Mairinger and A. Papst. Die infrarotreflektographis-
che Untersuchung von Gemälden und die Erstel-
lung von Bildmosaiken mittels des Programmpaketes
IREIKON. In 4th International Conference Non-
destructiveTesting of Works of Art, volume 1, pages
175–182, 1994.
7. C. Périer-D’Ieteren. Methodes scientifiques déxamen
à mettre en œuvre pour améliorer les connaissances de
la technique pictorale de primitifs flamands. In ICOM
Comm. f. Cons.,Triennial Meeting, pages 1–107, Ot-
tawa, 1981.
8. J.R.J. Van Asperen de Boer. Recent developments in
infrared reflectography of paintings and its applications
in art history. In ICOM Comm. f. Cons., 3rd Plenary
Meeting, Madrid, 1972.
9. M. Faries. Discovering Underdrawings, A Guide to
Method and Interpretation. 1997.
10. J. Taubert. Scientific examination of early netherlan-
dish paintings. Nederlands Kunsthistorisch Jaarboek,
26:41–72, 1975.
11. S. Ebadollahi, S.-F. Chang, and J. Coddington. Multi-
spectral image analysis and its applications in art image
classification. Technical report, Columbia University,
New York, 1999.
12. M.A. Faries. Underdrawing in the workshop production
of Jan van Scorel – a study with infrared reflectogra-
phy. Nederlands Kunsthistorisch Jaarboek, 26:89–228,
1975.
13. B. Corley. Conrad von Soest, Painter among Merchant
Princes. Harvey Miller Publishers, London, 1996.
14. M. Gallagher. The passion scenes of the Wurzacher
Altar: Restauration and painting technique. Jahrbuch
der Berliner Museen, pages 201–213, 1996.
c
The Eurographics Association 2003.
Paul Kammerer / Computer Aided Analysis of Underdrawings in Infrared Reflectograms
15. J.R. Mansfield, M.G. Sowa, C. Majzels, C.Collins,
E. Cloutis, and H.H. Mantsch. Near infrared spec-
troscopic reflectance imaging: supervised vs. unsuper-
vised analysis using an art conservation application. Vi-
brational Spectroscopy, 19:33–45, 1999.
16. J.R.J. van Asperen de Boer. Infrared reflectography and
computer image processing. New alternatives. In Le
dessin sousjacent dans la peinture, Coll. IX, pages 267–
273, 1993.
17. B. Jähne. Digital Image Processing : Concepts, Al-
gorithms, and Scientific Applications with CD-ROM.
Springer, 1997.
18. M.W. Burke. Image Acquisition. Handbook of Machine
Vision Engineering, volume 1. Chapman & Hall, 1996.
19. J.R.J. Van Asperen de Boer. Infrared Reflectography. -
A Contribution to the Examination of Earlier European
Paintings. PhD thesis, Univ. Amsterdam, 1970.
20. William K. Pratt. Digital Image Processing. John Wiley
& Sons, Inc., 1991.
21. Julio Sanchez and Maria P. Canton. Space Image Pro-
cessing. CRC Press, 1998.
22. P. Jaillon and A. Montanvert. Image mosaicing applied
to three-dimensional surfaces. In 12th. International
Conference on Pattern Recognition, volume I, TrackA,
pages 253–257, Jerusalem, 1994.
23. F. Mairinger. Strahlenuntersuchung an Kunstwerken.
E.A.Seemann, Berlin, 2003.
24. P. de Willigen. A mathematical study on craquelure and
other mechanical damage in paintings. Technical re-
port, Delft University of Technology, Faculty of Infor-
mation Technology and Systems, Department of Math-
ematics and Computer Science, 1999.
25. Ioannis Giakoumis and Ioannis Pitas. Digital restora-
tion of painting cracks. In Proceedings of the IEEE Int.
Symposium on Circuits and Systems (ISCAS ’98), 1998.
26. Jean Serra and Pierre Soille. Mathematical Morphol-
ogy and its Applications to Image Processing. Kluwer,
1994.
27. Fazly S. Abas and Kirk Martinez. Classification of
painting cracks for content-based analysis. In Proceed-
ings of IS&T/SPIE’s 15th Annual Symposium on Elec-
tronic Imaging: Machine Vision Applications in Indus-
trial Inspection XI, 2003.
28. R. Plamondon and S.N. Srihari. On-line and off-
line handwriting recognition: A comprehensive survey.
Trans. on Pattern Analysis and Machine Intelligence,
22(1):63–84, 2000.
29. D.S. Doermann and A.Rosenfeld. Recovery of tempo-
ral information from static images of handwriting. In-
ternational Journal of Computer Vision, 52(1-2):143–
164, 1994.
30. Chenyang Xu and Jerry L. Prince. Snakes, shapes and
gradient vector flow. IEEE Transactions on image Pro-
cessing, 7(3):359–369, March 1998.
31. G. Langs, H. Bischof, and P.L. Peloschek. Auto-
matic quantification of destructive changes caused by
rheumatoid arthritis. Technical Report 79, Vienna Uni-
versity of Technology, Pattern Recognition and Image
Processing Group, 2003.
32. E. L’Homer. Extraction of strokes in handwritten char-
acters. Pattern Recognition, 33(7):1147–1160, 1999.
c
The Eurographics Association 2003.
