Microsc. Microanal. 22, 448–457, 2016doi:10.1017/S1431927616000076

© MICROSCOPY SOCIETYOF AMERICA 2016

Pigment Degradation in Oil Paint Induced by IndoorClimate: Comparison of Visual and ComputationalBackscattered Electron ImagesKatrien Keune,1,2,3,* Rick P. Kramer,4 Zara Huijbregts,4 Henk L. Schellen,4 Marc H.L. Stappers,5

and Margriet H. van Eikema Hommes1,6

1Laboratory of Materials Science, Materials in Art and Archaeology, Delft University of Technology, Mekelweg 2,2628 CD Delft, The Netherlands2Van’t Hoff Institute for Molecular Sciences, University of Amsterdam, Science Park 904, 1098 XH Amsterdam,The Netherlands3Conservation and Restoration, Rijksmuseum Amsterdam, PO BOX 74888, 1070 DN Amsterdam, The Netherlands4Department of the Built Environment, Eindhoven University of Technology, Den Dolech 2, 5612 AZ Eindhoven,The Netherlands5Cultural Heritage Agency of The Netherlands, Smallepad 5, 3811 MG Amersfoort, The Netherlands6Cultural Heritage Agency of The Netherlands, Hobbemastraat 22, 1071 ZC Amsterdam, The Netherlands

Abstract: For the first time the degradation of lead white pigment in mature oil paint has been used as an internalmarker for the degree of saponification and hence chemical degradation of oil paint. Computational imageanalysis of the backscattered electron images quantified the degree of the intact lead white pigment versus thenonpigmented and lead-rich areas (degraded lead white) in the paint layers. This newmethodology was applied toa series of paint samples taken from four painted wall hangings (dated 1778), which makes it possible to study theinfluence of indoor climate on chemical degradation of aged oil paintings. The visual interpretation andcomputational image analysis of the backscattered electron images revealed clear trends. The highest degree oflead white degradation in the room was found in samples from the north wall close to the windows, whereasdegradation diminished further away from the window. Lead white from the south wall was less degraded, butshowed a similar trend as in the paintings on the north wall. These results imply a strong relationship betweenchemical degradation of paint and location of the paint in the room.

Key words: lead soap, indoor climate, pigment degradation, computational image analysis, scanning electronmicroscopy

INTRODUCTIONIn order to preserve old master oil paintings for futuregenerations, it is crucial to understand the impact of theindoor climate on the paintings’ condition. Currently,museum guidelines for climate conditions are subjects ofdebate. There is a tendency toward allowing greater fluc-tuations in temperature and relative humidity (Ankersmit,2009). However, it is not known to what extent the ranges ofthese parameters can be expanded before they have a detri-mental effect on preservation of the oil paint. A study of TheYoung Mother by Gerrit Dou (1658, Mauritshuis, inv. no. 32)suggests that larger fluctuations in a typical museum climateinduce degradation of the paint. A white haze formed on thedark paint areas during a period of just a few months(between March and July 2007) when the picture hung on anexterior southwest wall of the museum (Noble & Van Loon,2010). The temperature on the external wall was found tohave fluctuated between 15.8 and 30.5°C, whereas inside theroom acceptable values of 21°C and 50% room humidity

were constantly measured. It is hypothesized that the whitehaze was caused by free fatty acids that evaporated from thedark paint. This evaporation was induced by higher andfluctuating temperatures on the exterior wall. This exampleillustrates that chemical and physical processes are stilltaking place in mature oil paintings. Furthermore, fluctua-tions in temperature seem to induce these processes.

One of the most important degradation phenomenaencountered in old master oil paintings is the formation oflead soaps (Boon et al., 2002; Higgitt et al., 2003; Keune &Boon, 2007; Noble et al., 2008). These are the result of alead-containing pigment that has reacted with fatty acidsoriginating from the oil. This process is found to have takenplace in almost all old master paintings because lead white(2PbCO3·Pb(OH)2) was the most commonly used pigment.The lead soaps are formed at the surface of the lead whiteparticles. This is an ongoing process where particles areslowly reacting away. Therefore, the smallest particlesdisappear first, while the coarser particles or lumps are stillpresent, but are reduced in size and do finally disappear(Keune & Boon, 2007; Keune et al., 2011).

Artificially aged oil paint reconstructions with leadwhite and kinetic studies of the saponification of lead salt*Corresponding author. k.keune@rijksmuseum.nl

Received June 11, 2015; accepted January 11, 2016

mailto:k.keune@rijksmuseum.nl

with oil have shown that higher temperatures and relativehumidity induce the formation of lead soaps (Keune et al.,2005; Cotte et al., 2006). The question now is to what extenthigher temperatures and relative humidity, and their rangesof fluctuations, promote degradation of lead white pigmentin mature oil paintings. This knowledge is crucial inmaking informed decisions about safe display and storageconditions.

In order to get a better understanding of the influence ofindoor climate on lead white pigment degradation processesin mature oil paintings, we investigated a series of paintedwall hangings (signed and dated 1778 by Andries Warmoes)hanging in a salon (about 7.5 × 6m) on the first floor of aprivate house in Almelo called the Hofkeshuis (Fig. 1;(Bakker & van Eikema Hommes, 2015; Koldeweij, 2011).The wall hangings are 2.5m high and depict a Romantriumphal procession painted in imitation relief. They aresurrounded by painted profiled architectural frames. Thepaintings are done in shades of brown, containing lead white,chalk, black, and brown pigments. The wall hangings aresituated above a wainscoting, and cover all the walls apartfrom the window wall on the east, interrupted only by thefireplace in the south wall and the door to the room inthe west wall. The procession is thus divided into four. Thepaintings have never been lined or removed from the walland have had only a few cleaning interventions. The wallhangings consequently provide a unique opportunity to linkthe degree of lead white pigment degradation in mature oilpaint systems to the local climate conditions.

This paper discusses a new methodology for determin-ing the degree of lead white pigment degradation. The out-come presented here will make it possible to correlate thechemical condition of wall hangings with the indoor climate,which was accurately monitored for 2 years and will be thesubject of a forthcoming article.

In order to study the lead white degradation, we took 18paint samples from the painted profiled frame (Fig. 2),of which the pigmentation and paint layer build-up isidentical for the four wall hangings. Only the paint layerthicknesses vary somewhat. The sample locations are

regularly distributed over the wall hangings and are repre-sentative for different climate conditions in the room. Thenorth canvas, which is 7.5m wide, is adjacent to a hallwayand on its right side is exposed to direct sunlight in themorning, whereas the left side remains unexposed. Eightsamples were taken from this wall hanging. Four sampleswere taken from the canvas on the west wall, an inner wall,

Figure 1. North wall hanging of The Triumph of the Roman General Quintus Fabius Maximus (1778) by AndriesWarmoes, Hofkeshuis, Almelo (photograph: Rik Klein Gotink, 2013).

Figure 2. Detail of the trompe l’oeil relief frame around thecentral representation in all four canvases. An arrow indicates thepainted shadow line from which the 18 samples were taken.

Pigment Degradation Visualized in BSE Images 449

which is adjacent to the kitchen. This wall is not exposed todirect sunlight. Six samples were taken from the two can-vases on the south wall, which is an outer wall. At noon thesun heats the left side of this external wall, whereas the rightside is shielded from the sun by a neighboring house, whichis only 0.5m from the Hofkeshuis.

Fourier transform infrared spectroscopy (FTIR) is asuitable technique for identifying lead soaps formed in a paintsystem because of the characteristic asymmetric andsymmetric stretch vibrations of the lead carboxylate group(Robinet & Corbeil, 2003). The drawback to this technique,though, is that it is difficult to identify small differences inconcentration of lead soaps in the paint. More informative, inthis case, are scanning electron microscopic (SEM) back-scattered electron images (BSE images). BSE images of paintcross-sections clearly show lead white pigmentation and paintmorphology. In these contrast images, the lead soap-rich areasappear darker in the BSE image than intact lead white particlesdue to their higher organic content, but brighter than theorganic medium and the chalk and earth pigment particles.

In this study we showed that the degree of intact leadwhite pigment in cross-sections can be used as an internalmarker for the degree of saponification (i.e., chemicaldegradation) of the oil paint. In order to do this the BSEimages of the 18 samples were classified according to theirdegree of lead white pigmentation compared by visualobservation to the relative amount of lead soaps formed inthem. Standard image analysis quantifying the amount ofintact lead white is not applicable here: this analysis cannotdistinguish between residues of partially degraded lead whiteparticles and small unaffected lead white particles, as bothparticles show up bright white in the BSE image.

We therefore developed a computational image analysisprotocol for the BSE images to quantify the amount ofintact lead white pigments in relation to the degree ofnonpigmented and lead-rich areas. This protocol will bevalidated against the visually observed classified BSE images.

MATERIALS AND METHODSSamplesAll samples were taken from the painted brown cast shadowof the inner relief line of the profiled frame at the bottom ofthe paintings (Fig. 2). Samples A–H were taken from thecanvas on the north wall [72, 137, 238, 349, 358, 494, 600,and 725.5 cm from the right (east) edge of the painting,respectively], samples I–L from the west wall hanging (42.5and 170 cm from the right edge of the painting starting fromthe door and 169.5 and 24 cm from the left edge of thepainting), and samples M–R from the canvas on the southwall to the right of the fireplace (28, 105, and 207.5 cm from theright edge of the painting) and the south canvas to the left of thefireplace (266, 135, and 61 cm from the left edge of the paint-ing). The samples were embedded in a polyester resin (Polypol,polyester resin Poly-pol PS230: Poly-service kunsthars tech-nieken, Amsterdam, The Netherlands) and dry polished with

Micro-mesh® (Scientific Instruments Services Inc., Minnesota,USA) polishing cloths (final step 12,000 mesh).

Light MicroscopyThe paint cross-sections were examined with a ZeissAxioplan 2 microscope (Carl Zeiss Microscopy, LLC, NewYork, USA) with both incident polarized light and incidentultraviolet (UV) light (from a Xenon lamp and a mercuryshort-arc photo optic HBO lamp, respectively). The filter setUV H365 used for examination in UV light consisted of thefollowing filters: excitation BP 365/12, beam splitter FT 395,and emission LP 397.

Attenuated Total Reflectance (ATR)-FTIR ImagingMicroscopyFTIR spectral data from paint cross-sections were collectedon a Perkin Elmer Spectrum 100 FTIR spectrometer(PerkinElmer Office, Groningen, The Netherlands) com-bined with a Spectrum Spotlight 400 FTIR microscope(PerkinElmer Office, Groningen, The Netherlands) with a16 × 1 pixel linear mercury cadmium telluride array detector.A Perkin Elmer ATR imaging accessory consisting of agermanium crystal was used for ATR imaging.

SEMSEM in combination with energy-dispersive X-ray analysis(SEM-EDX) studies of the Hofkeshuis samples was per-formed on a Verion high-vacuum electron microscope (FEI,Eindhoven, The Netherlands) equipped with an Oxford EDXsystem with spot analysis and elemental mapping facilities.BSE images of the cross-sections were taken at 20 kVaccelerating voltage, 5mm eucentric working distance, andcurrent density of ~130 pA. Before SEM-EDX analysis,samples were gold coated (3 nm thickness) in a SC7640 goldsputter coater (Quorum Technologies, Newhaven, EastSussex, UK) to improve surface conductivity.

Computational Image AnalysisTexture contrast of the BSE images was used to determinethe degree of degraded lead white pigment. To quantify thetexture contrast, an image analysis procedure was developedusing MATLAB release 2013b (technical computingsoftware of The MathWorks, Inc., Natick, Massachusetts,USA). A grayscale image, containing n×m pixels, is repre-sented in MATLAB by an [n×m] matrix. In this matrix anumerical value is assigned to every pixel ranging from 0(black) to 255 (white).

Preprocessing the original BSE images was the initialstep in image analysis. First, the RGB-BSE images wereconverted into grayscale images. Second, the images did notall have the same brightness. Because brightness is the levelof whiteness of the pixels, it influences the outcome of theanalysis. The image brightness level was equalized bycalculating the median and then subtracting the bias to

450 Katrien Keune et al.

obtain the desired median. Third, the imadjust function wasused. This stretches the histogram to improve the contrast(the lead white pigments’ numerical values are increasedwith respect to the other pixels). Fourth, calcium wasexcluded from the analysis (calcium appears as dark regionsin the original images) by determining the maximumnumerical value of calcium in the original images and usingthis threshold value to find all pixels that are darker. Thesepixels were excluded from the analysis by setting them to 0(black).

The preprocessed images were computationallyanalyzed. Images of samples containing a lot of intact leadwhite show high texture contrast, whereas images of sampleswith severe saponification show a high amount of grayishmass. The gray-level co-occurrence matrix (GLCM) isused to calculate the texture contrast of the image. Thegraycomatrix function creates the GLCM by calculatinghow often a pixel with gray-level (grayscale intensity)value i occurs adjacent to a pixel with the value j, see Figure 3for an illustration of the 4 × 5 image I. Element (1,1) inthe GLCM contains the value 1 because there is only oneinstance in the image where two adjacent pixels havethe values 1 and 1. Element (1,2) in the GLCM contains thevalue 2 because there are two instances in the image wheretwo adjacent pixels have the values 1 and 2. Graycomatrixcontinues this processing to fill in all the values in the GLCM.

Each element (i,j) in the GLCM therefore specifies thenumber of times that the pixel with value i occurred adjacent toa pixel with value j. The adjacent pixels that are taken intoaccount can be freely defined by the pixel spatial relationshipusing the Offset parameter. The pixel spatial relationship asused in themethod is shown in Figure 4, in which k is defined as

k=length of image

500:

For example, an image of 1,500×3,000 pixels results in a k valueof 6 pixels. The factor 500 was found to be most appropriate. Inintact lead white pigment, k pixels provide just enough distance

to compare the pixels inside the pigment near the edge andpixels directly adjacent to the pigment. Graycomatrix calculatesthe GLCM from a scaled version of the image. The GLCM inFigure 3 is scaled to eight gray levels. The GLCM in the imageanalysis is scaled to 20 gray levels using the NumLevels para-meter. The first row and first column of the GLCM are set to 0in order to eliminate the influence of black pixels, whichrepresent regions of the image that need to be excluded from theanalysis. The number of occurrences where black pixels areadjacent to other pixels is set to 0.

Of the various properties that are included in the GLCM,only the texture contrast was of importance in this study.

RESULTS AND DISCUSSIONVisual Analysis of Two Paint SamplesWe compared two paint cross-sections H and A, 725.5 and72 cm, respectively, from the right edge of the canvas of thepainting on the north wall (Fig. 5). They have a similar paint

Figure 3. Illustration of how the image (left) is transformed into the gray-level co-occurrence matrix (GLCM) (right).

Figure 4. The adjacent pixels that are observed are at a distanceof k pixels from the pixel of interest.

Pigment Degradation Visualized in BSE Images 451

structure: first, a silicate glue ground (20–50 μm) (layer 1)with a chalk glue ground (100–200 μm) (layer 2) on top(Fig. 5, layer 1 is not shown). Two paint layers were applied,both composed of lead white, chalk, fine earth pigmentand very fine black pigment (layers 3 and 4). Technicalexamination showed that the lower paint layer (layer 3,~20 μm), a light beige-gray, was applied over the whole paintsurface of all canvases to provide a base color for the repre-sentation. The brown top paint layer (layer 4, ~10 μm) wasused to depict the cast shadow of the relief frame (Fig. 5).Finally, there is a thin (1–2 μm) varnish layer that wasapplied in 1956 (layer 5).

Visual comparison of the BSE images of the two samplesfocused on the relative amount of white areas of lead whitepigment and light gray nonpigmented areas (representativeof the lead soap-rich area) in the paint. The BSE imagesrevealed a striking difference in morphology in layers 3 and 4in both samples (Figs. 5c, 5f). In sample A (close to thewindow), the amount of lead white particles (brightest par-ticles) is very low in contrast to the light gray nonpigmentedareas present (Fig. 5c). Elemental analysis revealed lead inthe light gray nonpigmented areas. Furthermore, the inter-face between layers 3 and 4 was barely visible. This contrastsstrongly with sample H taken from the far left side of thecanvas (close to the door). This BSE image reveals compactand intact lead white granules. The light gray nonpigmentedareas, predominantly present in sample A, are almost absenthere (especially in layer 3) (Fig. 5f).

Elemental analyses of both samples revealed the pre-sence of calcium in the darkest gray particles, which are

interpreted as chalk particles. The brown and black pigmentsobserved in the light microscopic image are barely noticeablein the BSE image due to their minuscule size (

stronger in the spectrum of sample A than the same band inthe spectrum of sample H.

Both the BSE images and the ATR-FTIR spectra indi-cated that the paint on the right side of the north paintinghas a lower degree of intact lead white particles and a higherdegree of lead soaps than the paint on the left side. Hence, weconcluded that the paint on the right side has a higher degreeof saponification and is thus in a higher state of chemicaldegradation than the paint on the left side.

Variance in Lead White Pigmentation inHandmade Paints: Visual ObservationBased on the large differences found between samples A andH, it was important to verify whether this is due to theirposition in the room or to differences in the original paintcomposition. These historical paints are handmade and sothey are not expected to be completely homogeneous. In orderto evaluate the variance of lead white pigmentation, BSEimages were taken from eight embedded and polished paintfragments of one paint chip, sample D (Fig. 7). Themorphology in all eight BSE images appeared to becomparable. In all cases, layer 3 has a relatively low degree ofintact lead white and a high degree of light gray nonpigmentedareas, whereas the degree of light gray nonpigmented areas inlayer 4 is low. The relative amount of lead white pigment inlayer 4 seems to be the same for all eight BSE images.

If we compare the paint morphology of sample D withthose of samples A and H, it is clear that they differconsiderably (Figs. 5c, 5f, 7). This reproducibility in the eight

images of sample D and the large difference between samplesA andH led us to conclude that we can take the degree of intactlead white particles as an internal marker for the degree ofchemical degradation of oil paint. The average value for therelative amount of intact lead white and its standard deviationof the eight images of sample D will be given in theComputational Image Analysis section of this paper.

Visual Analysis of the 18 Samples from FourPaintingsFigure 8 shows a floor plan of the room on which thelocations of the 18 samples are indicated: A–R, from thenorth, west, and two south canvases. The corresponding BSEimages of these samples are depicted around the floor plan.In these BSE images, paint layers 3 and 4 revealed differencesin the quantity of fine lead white pigment. We visuallyestimated the degree of intact lead white pigment relative tothe amount of light gray nonpigmented lead-rich areas.Based on these observations, we classified the degradation oflead white in the paint layers into three levels: low, medium,and high, which are indicated by the colors green, orange,and red, respectively (Fig. 8).

In general the degree of lead white degradation in thenorth canvas is relatively high compared with the canvaseson the south and west walls. In the north canvas, a clear trendis visible in both layers 3 and 4 in the increase of lead whitedegradation, from the left to the right side of the canvas. Thecloser the paint is to the window, the more the lead whitepigment is degraded.

Figure 7. Eight different backscatter images were taken from eight embedded and polished paint fragments of onepaint chip, sample D. The graph shows the relative amount of white in each paint sample with an average value of 16.2and a SD of 1.2.

Pigment Degradation Visualized in BSE Images 453

Figure 8. Floor plan of the room with schematic indication of the locations of samples of A–R. The correspondingbackscatter images of these samples are depicted around the floor plan. The degree of lead white degradation of layers 3and 4 is divided into three classes: high degree of degradation in red, medium degree in orange, and low degree ingreen. The degree of lead white degradation is indicated with these colors both in the backscatter images and at thesample locations. The measurements obtained from layer 3 are indicated by squares and those from layer 4 by circles.The red dotted line drawn on the backscatter images indicates the interface between layers 3 and 4, as based on obser-vation of the samples with light microscopy (images not presented).

454 Katrien Keune et al.

A comparable trend was observed in layer 3 in the southpaintings, although to a lesser extent. The lead white particlesin this layer are better preserved in the painting to the rightof the fireplace (far from the window) compared with thepainting to the left of the fireplace (close to the window).Notably, the lead white in the upper paint layer (layer 4) is wellpreserved in both paintings. No large differences in pigmen-tation were found here at the different sample locations.

The lead white particles in layer 4 also appear well pre-served in the samples from the west paintings. This contrastswith the lead white particles in layer 3, which are highlydegraded on the right side of the canvas (close to the door)and better preserved toward the south side.

Computational Image Analysis of Two PaintSamplesAlong with visual inspection of the BSE images, computa-tional image analysis was used to determine the relativeamount of white areas of lead white pigment and light graynonpigmented areas, representative of the lead soap-richarea, in the paint by using the MATLAB Image AnalysisToolbox. This enabled us to measure the degree of chemicaldegradation of the lead white in a more objective way.

We assessed paint layers 3 and 4 in the samplesseparately by isolating each layer from the BSE image. The twolayers were subjected separately to image analysis. The imageanalysis method was demonstrated using layer 3 in samples Aand H. The original and preprocessed BSE images are shownin Figure 9. The relative amounts of white in layer 3 of samplesA and H were 11.4 and 25.7. These values show that sample Acontains roughly half the amount of intact lead white particlesin layer 3 than sample H. A low value of the relative amount ofwhite is representative of a high degree of lead white pigmentdegradation. A high degree of lead white pigment degradationwas also visually observed in layer 3 of sample A, whereas insample H low degradation was visually observed (Fig. 8). Thetwo methods, visual analysis and computational image analy-sis, were in agreement regarding the degree of intact lead whitepigment in the mature paint.

Variance in Lead White Pigmentation inHandmade Paints: Computational Image AnalysisComputational image analysis was applied to each of theeight BSE images of sample D (Fig. 7). Here, the image of thesum of layers 3 and 4 was taken to calculate the relativeamount of white. The average value for the relative amountof white was 16.2 with an SD of 1.2 (Fig. 7). This standarddeviation implies that the variance in the relative amount ofwhite found in samples A and H is significant for the degreeof lead white degradation.

Computational Image Analysis of the 18 Samplesfrom Four PaintingsThe demonstrated method of computational image analysiswas applied to all 18 samples as shown in Figure 8. As two

paint layers (3 and 4) were analyzed separately, 36 imageswere subjected to computational analyses. Figure 10 showsthe results. The texture contrast (y axis) reflects the relativedegree of intact lead white. The results of the computationalanalysis were compared with the classification based onvisual examination (red, orange, and green) (Figs. 8, 10). Thecomputational image analysis shows a similar trend in thenorth wall hanging, i.e., the decrease of intact lead white bothin layers 3 and 4 nearer the window and in the visualobservations. The values for intact lead white in sample Awere low, whereas they were high in sample H. The samplesin between (B–G) showed a decreasing amount of intact leadwhite toward the right side (window side). The onlydiscrepancy between the visual and computational imageanalyses was found in layer 4 in sample F. Visually this layerwas classified as orange, whereas the computational imageanalysis gave values corresponding to samples that werevisually classified as green.

The computational data generally corresponded to thevisual observations in the south paintings too. The compu-tational image analysis showed that the lead white in layers 3and 4 is better preserved in the painting to the right of thefireplace than in the one on the left. The only exception wassample Q, taken in the middle of the left-hand painting,where the lead white pigment is essentially intact.

Similarly, the computational image analyses confirmedthe visual observations in the west painting. The lower layer(layer 3) has a higher degree of lead white degradation close tothe door, whereas the degree of intact lead white in the upperpaint (layer 4) remains constant over all sampled areas.

Critical evaluation of the BSE images with both visualand computational image analyses shows similar trends inthe relative amount of intact lead white pigments in relation

Figure 9. Samples A and H (layer 3) are used to demonstrate theeffect of preprocessing of the image. The upper figure shows theoriginal image and the lower figure shows the preprocessed image.The preprocessed image clearly shows the excluded regions (blackpixels).

Pigment Degradation Visualized in BSE Images 455

to the degree of nonpigmented and lead-rich areas. Thecomputational image analyses do showminor fluctuations inthe degree of intact lead white. However, for our study, thegeneral trends are most relevant, and strongly suggest thatthere is a relationship between degradation of the lead whitepaint and climate conditions in the room. For example, it isnoteworthy that the highest degree of degradation was foundin the paint of the north canvas close to the window, wherethe painting is exposed to direct sunlight in the morning. Atthe other edge of this 7.5-m-long painting, which is notexposed to direct sunlight, the lead white pigment is muchbetter preserved. Interestingly, on the south wall, an outerwall, the lead white in the paint from the painting to the left ofthe fireplace is more degraded than that from the painting tothe right of the fireplace. This difference may relate to the factthat the left side of this wall is warmed by sunlight, whereasthe right side is shielded from the sun by a neighboring house.Notably, the lead white paint from this right-hand painting isthe best preserved in the whole room. The west painting is onan inner wall, which is adjacent to the kitchen. We suspectthat the trends in degradation of this paint may be influencedby radiation from various appliances, such as a combi oven orcoffee machine, located on the other side of the paintings.

As the general trends are clearly visible in the results, adetailed explanation for the minor fluctuations is hard togive at this point. As we are investigating real paintings, wenever can be fully sure about the original paint composition,painting technique used, and their history. We assume thatit is the same for all paintings, but minor differences cancertainly not be excluded.

In a subsequent study we will discuss in more detail thecorrelation between degradation of the paint and tempera-ture and relative humidity at various positions in the room.We will also address the observed difference in degradationbetween layers 3 and 4 in the four paintings.

CONCLUSIONThe image analysis procedure was developed to quantify thedegree of lead white degradation in oil paint systems basedon BSE images. The method proved to be qualified and is inagreement with visual observations. It eliminates the sub-jective character of classification based on visual inspection.

Our study shows that the amount of intact lead whitepigment compared with the lead soap-rich regions is indi-cative of the degree of saponification of the lead white paintlayer and can thus be used as an internal marker for itschemical degradation. The lead white containing paint fromthe painted profiled frame on all four paintings exhibitsstrong differences in chemical degradation, depending on itsposition in the room. These differences are not random,but show clear trends. Detailed indoor climate studies arecurrently being carried out to link these trends to differencesin temperature and relative humidity in the room.

ACKNOWLEDGMENTSThis research is part of the 5-year (2012–2017) researchproject: From Isolation to Coherence: An Integrated Technical,Visual and Historical Study of 17th- and 18th-Century Dutch

Figure 10. Results of the image analysis applied to all samples (A–H), layers 3 (black bars) and 4 (gray bars). Thetexture contrast proves to be a good measure for the degree of lead white degradation and is in good agreement withthe results of visual inspection.

456 Katrien Keune et al.

Painting Ensembles (supervised by Dr. Margriet van EikemaHommes), supported by The Netherlands Organization forScientific Research (NWO: Innovational Research IncentivesSchemes Vidi Grant). This project is based at Delft University ofTechnology. The Cultural Heritage Agency of the Netherlands(RCE) and the Rijksmuseum are partners in the project.

The authors wish to thank Henk van Keulen (RCE) forgas chromatography-mass spectrometry (GCMS) analysis,Eloy Koldeweij (RCE) for sharing his broad knowledge ofthe Hofkeshuis, the Stichting Hofkeshuis, Almelo, andHans Niers, and André Hovink, owner and renter of theHofkeshuis, respectively.

REFERENCESANKERSMIT, B. (2009). Klimaatwerk: Richtlijnen voor het museale

binnenklimaat. Amsterdam: Amsterdam University Press.BAKKER, P. & VAN EIKEMA HOMMES, M. (2015). The coarse painter and

his position in seventeenth- and eighteenth-century Dutchdecorative painting. In Technology & practice: studying 18th-century paintings and works of art on paper, Evans, H. & Muir,K. (Eds.), pp. 70–82. Postprints of the conference Centre for ArtTechnological Studies and Conservation (CATS), Copenhagen.

BOON, J.J., VAN DER WEERD, J., KEUNE, K., NOBLE, P. & WADUM, J.(2002). Mechanical and chemical changes in old masterpaintings: Dissolution, metal soap formation andremineralization processes in lead pigmented ground/intermediate paint layers of 17th century paintings. In ICOM-CC Preprints of the 13th Triennial Meeting, Vontobel, R. (Ed.),vol. 1. London: James and James.

COTTE, M., CHECROUN, E., SUSINI, J., DUMAS, P., TCHORELOFF, P.,BESNARD, M. & WALTER, P. (2006). Kinetics of oil saponificationby lead salts in ancient preparations of pharmaceutical leadplasters and painting lead mediums. Talanta 70, 1136–1142.

HIGGITT, C., SPRING, M. & SAUNDERS, D. (2003). Pigment-mediuminteractions in oil paint films containing red lead or lead-tin yellow.Natl Gallery Tech Bull 24, 75–96.

KEUNE, K. & BOON, J.J. (2007). Analytical imaging studies of paintcross-sections illustrate the oil paint defect of lead soapaggregate formation. Stud Conserv 52, 161–176.

KEUNE, K., FERREIRA, E. & BOON, J.J. (2005). Characterisation andlocalisation of the oil binding medium in paint cross-sectionsusing imaging secondary ion mass spectrometry. In ICOMCommittee for Conservation, 14th Triennial Meeting, TheHague, September 12–16, 2005 (Preprints), Verger, I. (Ed.),vol. II, pp. 796–802. London: James and James.

KEUNE, K., VAN LOON, A. & BOON, J.J. (2011). SEM backscattered-electron images of paint cross-sections as information source forthe presence of the lead white pigment and lead-relateddegradation and migration phenomena in oil paintings.Microsc Microanal 17, 696–701.

KOLDEWEIJ, E.F. (2011). Het Hofkeshuis, Grotestraat 62 te Almelo.De achterkamer met het geschilderde behangsel. Naderewaardestelling. Amersfoort: Aandachtspunten voorrestauratie.

NOBLE, P. & VAN LOON, A. (2010). Evaporation of fatty acids andformation of whitish deposits on the inside of the glass/microclimate boxes: A case study in the Mauritshuis. In TheEU Project FP6: PROPAINT: Improved Protection of PaintingsDuring Exhibition, Storage and Transit, Final Activity Report,Dahlin, E. (Ed.), pp. 154–169. Norway: NILU.

NOBLE, P., VAN LOON, A. & BOON, J.J. (2008). Selective darkening ofground layers associated with the wood grain in 17th-centurypanel paintings. In Preparation for Painting: The Artist’sChoice and its Consequences, Townsend, J., Doherty, T.,Heydenreich, G. & Ridge, J. (Eds.), pp. 68–78. London:Archetype Publications.

ROBINET, L. & CORBEIL, M.C. (2003). The characterization ofmetal soaps. Stud Conserv 48, 23–40.

Pigment Degradation Visualized in BSE Images 457

Pigment Degradation in Oil Paint Induced by Indoor Climate: Comparison of Visual and Computational Backscattered ElectronImagesIntroductionFigure 1North wall hanging of The Triumph of the Roman General Quintus Fabius Maximus (1778) by Andries Warmoes, Hofkeshuis, Almelo (photograph: Rik Klein Gotink,2013).Figure 2Detail of the trompe l’oeil relief frame around the central representation in all four canvases.Materials and MethodsSamplesLight MicroscopyAttenuated Total Reflectance (ATR)-FTIR Imaging MicroscopySEMComputational Image Analysis

Results and DiscussionVisual Analysis of Two Paint Samples

Figure 3Illustration of how the image (left) is transformed into the gray-level co-occurrence matrix (GLCM) (right).Figure 4The adjacent pixels that are observed are at a distance of k pixels from the pixel of interest.Figure 5Light microscopic image of sample A under white (a) and ultraviolet (UV) (b) light and sample H under white (d) and UV (e) light.Figure 6Part of a sum attenuated total reflectance Fourier transform infrared spectroscopic (ATR-FTIR) spectrum extracted from paint layers 3 and 4 in the ATR-FTIR image of samples A (solid line) and H (dottedline).Variance in Lead White Pigmentation in Handmade Paints: Visual ObservationVisual Analysis of the 18 Samples from Four Paintings

Figure 7Eight different backscatter images were taken from eight embedded and polished paint fragments of one paint chip, sample D.Figure 8Floor plan of the room with schematic indication of the locations of samples of A–R.Computational Image Analysis of Two Paint SamplesVariance in Lead White Pigmentation in Handmade Paints: Computational Image AnalysisComputational Image Analysis of the 18 Samples from Four Paintings

Figure 9Samples A and H (layer 3) are used to demonstrate the effect of preprocessing of the image.ConclusionAcknowledgmentsACKNOWLEDGEMENTSFigure 10Results of the image analysis applied to all samples (A–H), layers 3 (black bars) and 4 (gray bars).References