U T T E R W O F~ 1 H I N ET M k '% N

Mu*cnm 'vbw<aecm~n~ wul ( urazor~'hIp, Vol. 14. No. 3. pp. 253 281, 1995 Copyright © 1995 Elsevier Science Lid

Printed in Great Britain. All rights reserved I)260-4779/95 $1(I.(X) + 0,(Ill

0 2 6 0 - 4 7 7 9 ( 95 )00052-6

Evaluation of Colour Fidelity for Reproductions of Fine Art Paintings

LINDSA'~ MA(:[)~N<\t i), JAN MOROVI(AN[) [)AVID SAUNDERS

Overview

This paper describes the results of experiments carried out at the National Gallery in London, comparing original paintings with printed colour reproduc- tions made by processing digital images captured by a high resolution digital camera. A pairwise comparison method was used for three paintings under several different illumination conditions with a panel of observers. The results indicate that the optimum choice of colour gamut mapping procedure depends on the pictorial content of the painting.

1. Introduct ion

In the Esprit II research project VASARI, technology was developed at the National Gallery in London and at the Doerner Institute in Munich to capture high-resolution colorimetric digital images directly from paintings, with no intermediate photographs. The objective was to produce for each painting a digital 'mother image', which could be stored permanently on optical media and be used for both scientific and educational purposes. Application areas of particular interest for art conservation were the detection of changes in crack patterns and colour (Saunders & Cupitt, 1993). The Esprit III project 'Methodology for Art Reproduction in Colour' (MARC) is the successor to VASARI. The project has further developed the image acquisition technology in an ultra-high-resolution digital camera, which combines the principle of piezo- controlled microscanning of a two-dimensional CCD sensor array with macro- scanning over the image plane to generate a digital image of up to 20,000 pixels square (MacDonald & Lenz, 1994).

The application domain of the MAR(: project differs from VASARI. Instead of scientific analysis, the primary objective of MARC is the reproduction of paintings in print (Derrien, 1993). The project aims to develop a system that can produce high quality facsimile reproductions from fine-art paintings, using direct digital image acquisition and improved methods of image processing, as per Figure 1. Each resulting print should visuallv match the original painting as closely as possible, without any visible defects in tone, colour or spatial characteristics. Another aspect of the project has therefore been the development of an optimal inkset for a seven-colour printing process to increase the colour gamut of the printed reproduction (MacI)onald, 1994).

254 [ valu,~t~on ot Colour t:zdekt v/or ReproductJons of Fine Art Paintings

Digital Image Printing camera store press

1. Simplified M,,\R(~ image processing path.

Ideally a reproduction should be the same size and have the same surface texture as the original, and at each point the tone and colour should be identical when viewed under any type of illumination. In practice the reproduction is generally less than a perfect facsimile, not only because of differences in surface texture but also because the reproduction colorants (printing inks) have different spectral absorption characteristics, leading to metameric differences under different illuminants, and because their tonal and colour gamut is generally less than the gamut of artists' pigments. Nevertheless, with careful processing of tone and colour it is possible to achieve a very good match between the original and reproduction under a specified illuminant (MacDonald, 1993a).

Various techniques may be employed in the tone and colour correction and colour gamut mapping from origina] to reproduction as part of the colour separation processing, and these can result in subtle differences in the appearance of the reproductions. Of particular significance is the handling of out-of-gamut colours, depending on whether they are clipped to the gamut boundary or scaled in some manner. The effect of suc[1 changes on the final image cannot easily be predicted by computation or instrumental measurement. The best method remains visual assessment by a panel of observers, providing a suitable set of criteria for making judgements about colour fidelity.

In order to test the effectiveness of the MARC colour separation algorithms and of various methods of gamut mapping, we conducted a series of experiments at the National Gallery in London. Three paintings were selected and digitised via the VASARI scanning system, then reproduced as four-colour Cromalin proofs at Crosfield Electronics. Eight different reproductions of each painting were then compared side-by-side ,~.ith the original under controlled viewing conditions. The pairwise preferences of observers were recorded and analysed to determine which gamut mapping methods produced the most pleasing overall reproduction.

2. Assessment Procedures

Before any consideration of assessment techniques, one must first understand the objectives of colour image reproduction in general. What is to be matched: the original scene; a picture of the original scene (such as a photographic transparency or print); an image on a display; or another printed reproduction? In most cases the original scene will no longer be available for reference, either because it has changed with the passage of time or because it is geographically distant. The photographic representation of the scene, captured at a certain time and place, is then the only tangible reference by which one can judge the reproduction (MacDonald, 1993a). The MAR(; scenario differs in this respect from normal graphic arts practice because the original is available for comparison

255

and the reproduction should as far as possible be a facsimile rather than just a pleasing representation.

Two different assessment procedures may be used for comparing the fidelity of the MARC printed reproduction with the original painting:

(1) Instrumental measurement

The colorimetric match can be established by spot measurements with a reflection spectrophotometer. Comparison of the CIELAB values at correspond- ing locations of the original and the print allows colour difference metrics such as AE~AB or ~E(:x4(: to be calculated and averaged to give a quantitative measure of reproduction fidelity. Alternatively the MARC camera itself could be used to scan (parts of) the print and the resulting C1ELAB data compared with corresponding points of the original. This method may suffer, however, from metameric differences between the colorants of original and print because the MARC camera does not 'see' them with exactly the same spectral response as the human eye.

(2) Visual assessment

Visual assessment is performed by a panel of human observers, who look at both the original and the print and give their opinions about the faithfulness of the reproduction. The evaluation procedure consists of placing the original and the print side by side under the same conditions of illumination and comparing the two. For gallery viewing situations the print should be mounted in an identical frame and hung alongside the original at the same height, with the observers standing midway between the two. For more critical assessment, the original should be removed from its frame and positioned alongside the print in a standard graphic arts viewing cabinet with neutral grey walls and controlled lighting with a correlated colour temperature of 5000K and illuminance of 500+ 125 lux (Field, 1988).

Inter-Comparison Issues

Issues that can affect the comparison of the original and print for both instrumental and visual assessment include the following (Johnson, 1992):

Cop), size

It is assumed that original and reproduction have the same dimensions and are assessed at same viewing distance, so that differences in visual field size are not a consideration.

Illumination

It is assumed that the intensity and diffuseness of the illuminant is identical for both original and print. The intensity level is critical in the viewing of pictorial subjects such as printed matter, photographs and paintings because both contrast and colour saturation will appear to increase with the level of illumination. The illumination should ideally have no ultra-violet component (to exclude pigment

256 t c,duazw,~ q~ Colour t:idek O, fin Rwroducgzons of Fine Art Paintings

or ink fluorescence as well as preventing deterioration of the original) and be unpolarised.

Colorant mctamerism

The printing inks will in general have absorption spectra that are quite different from the artist's pigments in the original painting. The MARC reproduction process aims to achieve a visual match (for normal human colour vision) under specified illumination conditions but under different illumination a mismatch may occur duc to metamerism.

Gloss

Gloss is a property ot reflective samples that influences the perceived lightness of the sample. If a sample has high gloss, the surface reflection will be very directional and, if the viewing geometry is correct (such as light source perpendicular to surface and eye of observer at 45 degrees), will not influence the perceived colour of the sample. Where the surface has a low gloss (matte finish) the light is reflected diffusely at all angles and has the effect of diluting the sample colour, reducing both the apparent density, of dark areas and the colourfulness of strongly cotoured areas. A painting with a glossy textured surface, resulting from the thick application of oil or acrylic paints, may show small specular reflections (catchlight points) a~ many places over the surface, which cannot be reproduced on a print.

Instrumental Quality Metrics

The acceptability of the match between the original and the print needs to be quantified to be rneaningful. Some metrics can be determined from instrumental measurement and calculation (Hunter & Harold, 1987):

Det2slly ?a?l~'

The minimum and maximum density values of both original and reproduction can be measured directly by a reflection densitometer.

Average ,t,lour diffcrence expressed in kEi~Al~ units between corresponding locations of original and reproduction can be measured directly by a reflection spectrophotometer.

Dimenszon,d a~curao~

The dimensional accuracy between the original and reproduced image is normally expressed as geometric distortion (e.g. pincushion or barrel distortion) measured as % error from the true position of a point on the original. Both short range (less than 4ram) and long-range (greater than 4ram) measures may be required.

l.~lx.I)',,\~ M,\{ [)<~N,\i 1), ,1,~,5; Mt>l<~ ,\1< ,,\'-~l) I)A\'II) SAUNI)ERS 257

Spatial resolution

Normalised contrast ratio as a function ot spatial frequency (cycles/mm) can be measured both vertically and horizontally on image at a number of correspond- ing locations of both original and print. This may be determined from Fourier transform analysis of the digital images.

Signal-to-noise r, ttzo

Signal-to-noise ratio (SNR) across the full area of the image is calculated as the RMS value of pixel differences between the two digital images.

Visual Quality Criteria

For quantification ot the visual match between the original and the print it is necessary to formalist the procedures for observation and recording of results. Quality criteria for visual assessment can be defined as subjective measures of spatial, tonal and colour attributes of the image, such as the following:

Sharpness

Edge quality and sharpness throughout the imagc.

Detail

Rendition of fine detail in highlight, midtonc and shadow regions of image.

Smoot/2ness

Amount of noise "texture" at all densities and enlargements.

Spatial artq/,~ct~,

Any visible pattcrnm~ ~,r disturbances in the reproduction, such as streaks, asymmetry of solid lines, blocking, iitter, moir6 and aliasing.

Tonal rendite~JTe

Tone reproduction o~cr the full dcnsit~ range of original.

Hue renditio~

Accuracy of rcndititm ot hues m c r the lull gamut of original.

Co[ol~ 5atbt~~Htotl

Degree of colour saturation, i.e. neither too much nor too little, over all hues and t o n e s .

258 [ u,du,lm,n . / Colour I'~delitr /,,r Re/.oductwns q/Fine Art Paintings

7bne and colour arte/acts

Any visible disturbance or distortion m the colour reproduction, such as contouring, colour casts in certain tonal ranges or imbalance between differing hues .

hnages are normally reproduced on different media from the originals and, because of spectral differences between the colorants, differing illumination levels and the influence of viewing conditions, a visual colour match is the best that can be expected. In this case human judgement is the only proper basis of assessment, at least until the advent of a fully comprehensive model of colour appearance. Typically a panel of observers is asked to assess the colour fidelity of the reproduction, perhaps with a checklist of particular colours or areas of the image to examine. No two experts e~er agree totally on colour fidelity, but by anah'sis ot results a trend can bc established.

3. Processing MARC Images for Reproduction

MARC images are captured and encoded colorimetrically in CIE L"'a":b": colour coordinates, referenced to D65 white, which define precisely the colour stimulus at each point (pixel) of the image (Figure 1). The primary objective of the reproduction software is to generate colour separation values (%dot) for the printing inks, which when printed will produce the identical colour stimulus at each point of the Print- Considerable effort has been applied within the MARC project to achieving a high standard of colour calibration accuracy, so that any desired I:: a::b ; value can be accurately reproduced in print over the full gamut of the printing inks (MacDonald, 1993b).

For tlle MARC reproduction of fine art paintings it is theoretically possible to achieve an exact colorimetric match, provided that all colours of the original are within the gamut of the printing process. In general there will also be colours present in the original which, although they are captured by the MARC camera and arc represented accurately in CIEI,AB coordinates in the digital image, lie outside the colour gamut attainable by the target printing process. Even the seven-colour process with six chromatic primary inks plus black cannot reproduce ccrtai, extreme pigment colours such as cadmium yellow or ultramarine, unless special inks are made up from the actual pigments in question. Thus in general there will exist out-of-gamut colours in the image which must bc modified (or mapped) to produce the nearest colour within the printing gamut and this inevitably leads to differences in colour appearance between original and reproduction.

To compensate for the unavoidable differences between the original and the reproduction, various correction techniques can be applied when processing the digital imagc. As changes in hue are more perceptible than changes of other perceptual attributcs, all but one of thc following techniques preserve hue and alter only lightness and colour saturation:

(a) Paper cast ~emoval: ~Ib account for the colour of the printing substrate (normally white paper), the a and b* coordinates of the reproduction's substrate arc subtracted from all colour coordinates in the original image. This is the only correction algorithm that alters hue, albeit to a minor e x t e n t

IANt)'~A; M\( I)t~NAI I), JAN M~H<~\I(ANI) DAvit) SAUN1)kRS 259

(b) Tonal mapDmg: To preserve the relative lightness values in the image, the larger range of densities in the original (typically up to density 3.0) is mapped onto the more limited density range of the reproduction (typically 2.0). This is in effect a one-dimensional transformation along the L:: axis, whereby the original's white point and black point arc mapped onto the reproduction's white point and black point respectively.

(c) Gamut mapping: The original's colour gamut is in most cases larger than that of the reproduction, and it is therefore necessary to devise some rules according to which out-of-gamut colours are to be reproduced. The two gamut mapping algorithms used in this experiment were orthogonal mapping and chord mapping. Orthogonal mapping reproduces an out-of-gamut cotour as the nearest colour on the reproduction's gamut boundary. In Figure 2 c2 is the colour resulting from cl being orthogonally mapped onto the gamut boundary. (both colours are in a plane having constant hue). This gamut mapping technique provides the smallest AELA B colour difference. Chord mapping reproduces the out-of-gamut colour as the colour which lies on the intersection of the gamut boundary with the line joining that colour's coordinates to the lightness coordinate of the colour with the highest saturation at the given hue angle. Using chord mapping cl is mapped in Figure 2 onto c3. Chord mapping has a lesser effect on saturation, but results in a larger overall colour difference. These two techniques are in some sources referred to as gamut clipping, since the>' change only the coordinates of colours that are out of gamut (Robertson, 1988).

(d) Gamut compression: It is generally regarded as more important to preserve the relative differences in colour saturation (or colourfulness) between all colours, rather than to preserve the absolute coordinates of those in gamut (Hunt, 1987). ~Ii~ achieve this, gamut compression can be applied to all colours in the original image by radial scaling of all coordinates in L":-a*b": towards the 'gamut centre', assumed to be the point on the L:: axis with value L* = 50. Another way to think of this process is applying an expansion to the gamut of the reproduction process, whereby all colours are moved outward

m ~

C'I~ C1

~-- C *ab 2. Gamut mapping.

260

i 9~

t

I ;ah¢,tnotl of Colour Fideh 0 ./3~ Reproductions of Fine Arg Paintings

3. (;amut compression.

"'-... i-~b

5O

'C 4

, C1

~-- C *aL

from the central point on the L:: axis (see Figure 3). The original's colours are then mapped onto the expanded gamut. For these experiments a fixed compression factor of 20% was used for the two cases of gamut compression tested.

To determine what influence the above algorithms have on the extent to which the reproductions match the original, eight different combinations of the correction factors were tested. For each combination the original L:'a*b* image was processed to produce four colour separations for cyan, magenta, yellow and black (CMYK) printing inks and a colour proof was made using the DuPont Cromalin process. To sitnulate the surface characteristics of the original oil paintings more closel), all proofs had a matte coating applied ([)uPont 'lop (]oat Matt). These ~eproductions will be referred to by their labels, as defined in Table 1

]able 1 l~gl~t c~,nbinati~u> o t , , , rTecuon tactors usedin the experiments

l.abe! o/ Paper (.a,z Reprodu(.z,~. Removal

A 5 cs NL,

A( "los N~, 1 > "~cs h c, PC "~ cs "~ c~

P(; "~ cs 5 c~

lima/ Gamut Gamut U,tppmv Mapping Compression

C) rthogonal No Chord No C)rthogonal No Chord No Orthogonal No Chord No N/A Yes N/A Yes

1,1\I> ' , \5 M \( l ) t ~NAI.I), JA,~ M, , l ¢ , , \ ~c A \ I ) [)AVID ~AUNI)IIRS 261

4. The Experiment

The objective of the experiment was to compare the visual appearance resulting from different combinations of correction factors (i.e. to give information about their relationships) and not to ascertain the 'absolute' goodness of match of the reproductions. When setting up a comparison, the influences of variables other than the one under study need to be reduced. To prevent the results being biased bv, an individual painting's characteristics and subject matter, a set of three different paintings was chosen for the experiments.

Paintings

The first painting was a fragment of an altarpiece by the Master of Liesborn called The Adoration of the Kings believed to have been painted between 1470 and 1480. This painting has a wide range of hues, areas of high saturation, neutral areas and a well balanced tonal range. Analysis of the digital image showed that over 30% of all pixels had colours which were out of gamut. The second painting was the Portrait qfa Woman from a painter of the Cologne School painted in 1495. It has large areas of dark, saturated red and considerable shadow and highlight detail. Its characteristics and the ~act that 68% of the pixels had colours out of gamut, make it particularly' difficult to reproduce. The third painting was an 18th century 'Claudesque' landscape with colours predominantly in the desaturated shadow region, h is typical ot many old oil paintings in that its original appearance is partially obscurcd bx a laver of yellow-brown varnish, resuhing in the desaturation and darkeningof all the colours.

4. Viewing the Master of Liesborn painting in situ on the wall of the Sainsbury Wing of the Natinnal Gallery.

262 I: :',dl~aeu,n ,d" Colour Fzdeht v/rn Reproductions Of Fine Art Paintings

Vie~'ing u)ndzllons

Since the first painting (Master ot Liesborn) could not be removed from its normal wall hanging location in the Sainsbury Wing at the National Gallery, the comparison was carried out under the illumination present in the Gallery (Figure 4). The room was illuminated by a limited amount of daylight admitted through overhead skylights, augmented by tungsten lights with filters to give them correlated colour temperatures of 4100K and 3000K. The two types of tungsten lights were mixed in a ratio of 1:1. An added obstacle was the varying level of illumination on tlne wall where the painting and the reproductions were compared, because of the proximity of a doorway.

The second and third paintings were compared under standard viewing conditions (Johnson & Scott-Taggart, 1993) with three different illuminants: D50 (the graphic arts standard), D65 (the illuminant used as a reference white for the image capture and ink colorimetry) and A (the equivalent of a domestic tungsten light). These comparisons were carried out in a viewing booth at the National Gallery', in which the original painting and the reproduction proofs could be arranged side-by-side (Figures 5, 6).

Observer,

To reduce the influence of the observer as an individual, a number of observers took part in the comparison. All observers were male with ages ranging from sixteen to fiftx:-two and had normal colour vision. The first painting was assessed

5. Experimental comparison of the Cologne School painting with two reproductions (left and right) in a standard viewing cabinet.

I.IT~L~,A~ ~v~.',(:l)t)NAl I), JAx b, , loR~l< ANt) DAVID SALINI)t, RS 263

6. Experimental comparison of the Claudesque painting with two reproductions (left and right) in a standard viewing cabinet.

by six differcnt observers (resulting in six sets of data). The second painting was assessed by 10 observers, each making a different number of comparisons for the three different illuminants, resulting in a total of 24 sets of data. The third painting was assessed by only two observers for all three illuminants, giving a total of six sets of data.

Comparison Method

For comparing the individual reproductions a pairwise comparison method was chosen. The individual reproductions were presented to the observers in pairs alongside the original and they were asked to chose the reproduction which represented the closer match. So as not to force the observers into a decision, they were also given the option to indicate that both reproductions matched the original to the same extent. This technique was chosen over other methods which require the observers to award points to the objects under comparison, or to rank them. The reason is that pairwise comparison only asks the observer to make a choice, and doesn't require him or her to specify or quantify the response, which means that less subjective judgement is involved. Although the observer is only required to pick one reproduction out of a pair, the individual choices can later be combined to provide the same type of result as techniques asking observers to rank the reproductions (Gescheider, 1976).

Making a decision about which reproduction is the closer match involves the consideration of a range of variables, which are consciously or subconsciously given different weight when forming the overall choice. Since asking questions based on a single criterion involves less of this process, a number of specific questions were asked alongside the question about an overall match. In the specific questions observers were asked to make a comparison based only on closely defined regions, each having a dominant characteristic (e.g. hue,

264 t ;',duazzon o/' Colour Fidektv fo~ Rwroductmns Of Fine Art Paintings

saturation, lightness). Even in these specific questions, however, the observer's choice may be influenced by variables other than the main one (according to observers, questions about hue were often influenced by the amount of detail preserved in a particular region). The effect of these influences is nevertheless smaller than in the case of asking one overall question for a reproduction.

Questionnaires

Since the three chosen paintings had very different characteristics, primarily in terms of gamut and tonal range, a different questionnaire was designed for each painting as shown in Appendix B. All the questionnaires consisted of the following general components:

(a) Questions about the observer

Name, age and sex

(b) tS.vplam~t.m o/ the observaticm '_, am~s

Here the <d~servcrs wcre told that they had to choose the reproduction that better matches the original in onc of the areas pointed out in the next component

(L) Ha!fionc ~vp~od.ction q[ the pam.ng

The individual areas to bc compared were highlighted in this image, which was taken from the Microsoft Art Gallery (}I) ROM.

(d) lnst .1 arc,~: t . Dc compared

Each followed bx two boxes, representing the pair of reproductions. One box was to bc ticked {o indicate the preferred choice, or both boxes were to be ticked if no difference was perceived. For the first painting questions about red, green, blue, veiN>w, saturated, highlight, neutral and shadow areas were included alongside questions about overall contrast and saturation and a question asking tor an mcrall choice. For the second painting questions about red, green, blue, skin tone. highlight, neutral and shadow areas were asked alongside questions about overall contrast and saturatioi~ and a question about the overall match. The third painting was compared in the red, green, blue, yellow, highlight, neutral and shadow areas as well as in terlns Of contrast; and again a question was asked about the overall match.

The Assessment Procedure

To carry out a lull pairwise comparison ot the eight reproductions meant that 28 pairs oi: paintings had to be compared. With nine to eleven questions per pair (depending on the painting) the time required to go through the questionnaire proxed u> be the main drawback of the pairwise comparison technique. When conducting tile comparison the individual reproductions were labelled in a different way for each painting and the pairs were also presented to the observers

[ , i x l ) ~ Mx~ [)~>'~Ai l), .]A'~ Mc) l~ \ i ( ,,\~1) [)Ax, lI) SAUNDERS 265

Original

7. ['hc ~ctup used tot comparing tilv Master of Liesborn painting.

in a random order. The Master of Liesborn painting was assessed on the wall of the Gallery, so that it was not possible to have one reproduction either side. The pair of reproductions in this case had to be hung on the wall to the right of the painting as shown in Figures 4,7. The assessment of this painting was carried out simultaneously by six observers standing at a distance of approximately two metres from tile painting (mainly for security reasons).

The Cologne School and Claudesque paintings were placed in a standard viewing booth with the original in the centre and one reproduction on either side (Figures 5,6). The comparison of all 28 pairs was then conducted under each of the three illuminants. For practical reasons, three observers performed the experiment at the same time. With ten questions per pair, the full series of comparisons under one illuminant took between sixty and ninety minutes.

SCOrlYI~

Slllce the observers were given the choice t~, declare both reproductions in a pair as being equal, the following scoring system was used. The chosen reproduction was given two points and the other one zero; when both were selected, they were awarded one point each. For evaluation the scores were transcribed into a spreadsheet (Microsoft Excel) in the f()rm of an 8x8 matrix for all pairings of the rcproductions for each question, where tile points given to each reproduction were placed in its column and in the row of the reproduction to which it was compared (e.g. if A is chosen over C then two is written into column A row C and minus two in column C row A).

5. E v a l u a t i o n o f R e s u l t s

The following procedure was used to obtain a ranking of the reproductions. For each questionnaire a matrix was set up containing the summary of

266 t: :'a/u,ztzcm <~/" ('olour hdelitv fi~r Reproductzons of Fine Art Paintings

answers to all questions. Each score in this 8x8 matrix was obtained by summing the scores for that particular combination and dividing by the maximum possible score (i.e. the number of questions times two). All questions were given an equal weight, since the introduction of a weighting system for the questions would only add another subjective influence and thus distort the results. An average score was then calculated from all observers' questionnaires under the same viewing conditions. To determine the statistical significance levels, standard deviations were calculated for each reproduction's total score, namely the sum of all scores in its column. For the differences between two reproductions to be statistically significant at a 95% confidence level, the mean total score of one reproduction had to be outside the other reproduction's 95% confidence interval. The 95% confidence interval is delimited by the mean of the total score plus and minus 1.96 standard errors. Standard error is equal to the standard deviation divided by the square root of the number of observers, from which it follows that the precision of the results increases with the square of the number of observers. To obtain the final scores for a given painting the results from all illuminants were averaged, which also increased the precision of the results since they were obtained from a higher number of observations.

Some sources (Bristow & Johansson, 1983) suggest the use of a coefficient o f consistence, which is given by subtracting the percentage of illogical triads made by the observer from one (an illogical triad is if an observer's results suggest for example that A > B > C but also C > A). This coefficient is than used to weight the scores from the individual observers when a total score is computed. Obtaining this coefficient is time consuming because the number of illogical triads would have to be obtained manually. This is because the formula for its computation would treat all cases where two reproductions are declared equal as illogical (except when all members of the triad are declared equal). This approach is mathematically correct, but does not take into account situations like A = B, B = C but A > (~, which do not contradict the way human perception works. In this case there might be no noticeable difference between stimuli A and B nor between B and C, but the two differences could add up to a noticeable difference between A and C. Considering this difficulty and seeing that the results from the individual observers agreed with each other to a substantial extent, the coefficient of consistence was not used in this experiment.

A1] results obtained so far can be used only to set up a ranking of the reproductions, but it is also possible to convert these results into an interval scale, where the differences between individual reproductions can be com- pared. This conversion is suggested by Thurstone's law of comparative judgement which is 'based on the notion that the proportion of times stimulus A will be judged greater than stimulus B is determined by the degree to which sensation A and sensation B differ' (Gescheider, 1976). According to Thurstone the influence of a stimulus on the organism's sensory receptors results in a discrimination process which has a certain value on the psycho- logical continuum. Due to instabilities ot the organism, the same stimulus is capable of producing a range of discriminai responses, which form a normal distribution. Therefore to determine the difference between two stimuli, it is necessary to determine the distance between the means of the distributions of their discriminal processes. To do this, the standard formulae for calculating

1.1%I)~,%'~ ~'~.%<[I)t>~A[ I), JAN ~'/t*lZt ,\IC .%NI) 1.)A~[|) SAUNI)ERS 267

the standard deviation of differences between pairs of numbers (1) and for calculating z-scores (2) are combined:

'r,I ..... 1,-, : \ or;.,. + ,r:~,,. - ,r~, ~ ,, or,, , or ( 1 )

X - X Z . . . . (2)

In these equations ~A and ~bu are the two discriminal processes, o'% *A is the standard deviation of the differences between ~A and qJB, or, and ~,% are the standard deviations of the two discriminal processes and or )` or or is the correlation between pairs of t~ A and ~ values. Z is the distance of a particular value from the distributions mean on a scale with the unit of one standard deviation. After combining these two formula:, Thurstone's complete law of comparative judgement can be expressed as:

* ~ - ~,,x = zBA ~ , r ; , + ,r ; , , - o r ~ , # , % or,,~ (3)

This can be further simplified, since in our case there is no correlation between the individual stinmli. The formula used for this experiment is therefore:

2 ' ~ l ~ - ~ A = z~, \ ,(~,~,, + or;i,,, (3)

whereby the z-scores were obtained from normal distribution tables relating p-scores to z-scores. The p-scores represent the area under the standard distribution curve on the interval of (-~-, z], which is equivalent to the percentage of the population being in that interval. In this case the p-scores are the values in the matrices produced for the ranking, which represent the percentage of times that a given reproduction was chosen over another reproduction.

6 . R e s u l t s

The results of the comparisons (see Appendix A for details) will be discussed individually for each of the three paintings:

(a) Master of Lzesborn

The reproductions of this painting can bc divided into three significantly different groups, as shown by the scores in Figures 8 and 9. Those obtaining the highest scores were P and PC, which have both had tonal mapping and paper cast removal applied.

Coming last was PG, being the reproduction where gamut compression was applied in addition to tonal mapping and paper cast removal. The remaining reproductions were not significantly different from each other and were in between the other two groups. It is interesting to note the disagreement between observers regarding the reproduction NG. This was possibly caused by some observers being more influenced by the tonal characteristics of the reproduction, whereas others based their choice on the colour of the area under comparison, which was what they were asked to do. This discrepancy

268,

2"r'3 I C

I valuaz.m o/Colour Lzdek:~ /m Reproductions of Fine Art Paimings

8 . Scores for Master of Liesborn.

: 7-

I f ) I t t I I

between observers was probably not due their age, professional background or other characteristic, since such a correlation is not apparent from their responses. A more likely reason is the difficulty experienced by an observer to base his judgement solely on the criteria he was asked to judge. In this case part of the task given to the observers was to compare certain areas of a characteristic colour, wherc it was difficult not to be influenced by the amount of tonal detail maintained in those areas. This influence migl~t have been reduced, but not eliminated, by giving this potential problem more emphasis when the task was explained to the observers.

Since this painting has a wide tonal range, the corrections modifying tonal reproduction were preferred most often. The failure of gamut compression techniques seems to be caused bx the fact that the painting's gamut did not exceed the reproduction's gamut significantly and there was therefore no real need for this correction.

, v' % m !

. ~ +

4 . . : '~

Z. 0 O

. 3 0

2.3 D

4.2,5

.2, 0

9. Z-scores for Master of Liesborn.

5.00-

4.50-

4.00-

3.50-

3.00-

2.50-

2.00-

1.50-

1.00

0.50.

0.00

l.l-,i)>,.\~ M \(;l.)~>N,",l i), .]..\N N'h~l<t,x w .\NI) D.,\',:u) SAUNI)FRS

t t i t 10. Scores for Cologne School.

I I I I I t I I rS m 0 'b' 21 < 0 U

269

(b) Cologne School

This painting differs from the first, since 68% of its pixels have colours out of the CMYK gamut. The need for gamut compression was confirmed by the results, shown in Figures 10 and l l , indicating that the two reproductions with gamut compression (NG and PG) had the highest scores, alongside reproduction P. These three reproductions are followed by reproduction PC, which differs from P only in the gamut mapping algorithm. The wide tonal range of this painting explains why the two reproductions with tonal mapping are among the most preferred in this case. The least preferred reproductions were those without correction or o n h ' with paper cast removal.

Since restdts o{ the comparisons under the different illuminants do not differ significantly, only their average is shown here (results for the individual

3.00-

2.00"

i . 0 O"

0 . 0 0

-1.08"

2.00

- 3 . 0 0 '

{

I

0 Z

I I I I I I

{

'1 b

11. Z-scores for Cologne School.

270 £~aluatzon of Colour t'7dekLy lor Reproductions of Fine Art Paintings

illuminants are given in Appendix A). The differences in results for the different illuminants can be explained in terms of metamerism between the pigments of the painting and the dyes of the Cromalin proofs, and also changes in appearance due to chromatic adaptation. Some differences between the results from different illuminants can also be accredited to the small number of observations made under the individual illuminants.

(c) Claudesqu~"

Due to the limited contrast and small tonal range of this painting, and the fact that the colours were predominantly in gamut, the reproductions with the highest scores were NC, N and A, which either make no correction or only remove the paper cast. The second group of reproductions is made up of PC, NG and A closely followed by P, as shown in Figures 12 and 13. The least preferred reproduction for this painting was PG, which affects both the gamut and the tonal characteristics of the image.

.L~

• O)

,: . {

/

i ,

I I I

L I I

12. Scores for Claudesque.

t : - ' ~ 13. Z-scores for Claudesque.

1,i>,l~s~ M.x( 1)~ ~N,,\t l), ]A'< M~ ~l<< ,\ic ,X~<l~ D,x\ u) SAUNDt~WS 271

7. Conclusions

Since the comparisons of each painting suggest a different ranking for the preferred types of corrections under examination, it is not possible to suggest one correction method that is optimal for all images. Rather, the results of this experiment suggest that the success of a particular correction algorithm depends primarily on the characteristics of the image to which it is applied.

As can be seen, gamut compression achieves good results if the original painting's gamut significantly exceeds the gamut of the reproduction process, but it degrades the quality of images having gamuts comparable to the gamut of the reproduction process. Further experiments could be conducted, comparing a series of paintings with a gradually increasing number of out-of-gamut colours, to determine a threshold level above which gamut mapping should be applied.

The tonal mapping techniques improved the match of paintings having a wide tonal range, and analogically paintings with large gamut and a wide tonal range achieved the closest match when a combination of gamut compression and tonal mapping was applied to them. In most cases the type of gamut mapping used (i.e. either orthogonal or chord) did not have a significant influence on the success or failure of a particular algorithm.

From the results of this experiment it is clear that a particular correction should be only used if there is need for its effect. The results also confirm that if the correct algorithm is used it can produce a positive influence on the reproduction's visual match with the original painting. These results therefore support the approach of providing the user of the MARC image processing software with a choice of correction methods, *o be selected according to the pictorial content of the painting to be processed. There is no substitute for human judgement in making a choice ot this kind.

Acknowledgements

We would like to thank Dr. John Cupitt (National Gallery) for assistance and valuable suggestions in setting up the experiments. Also thanks are due to Dr. Bob Thompson (London College of Printing) and Dax Rughani (Crosfield Electronics Ltd.) for their contributions, and to all the other observers including Dr. Andrew Manning and Peter Morovit; for their patience.

References

Bristow, ]. and Johansson, P.A. (1983). Subjective Evaluation by Pair Comparison: Pitfalls to Aw)id and Suggestions for the Presentation of Results, in Advances in Printing Science and Technology 17, pp. 278-291.

Derrien, H. (1993). MARC--A New Methodology for Art Reproduction in Colour Proc Electronic Imaging and the Visual Arts, Fourth Summer Conf., London

Field, G.G (1988). Color and its Reproduction, GNI'F Press, Pittsburgh PA. Gescheider, G. A. (1976). Psychophysics, Method and Theory,, Lawrence Erlbaum Associates,

pp. 84-102. Hunt, R.W.G. (1987). The Reproduction q[ Colom m Photography, Printing and Television 4th

Edition, Fountain Press, Tolworth UK, p. 43. Hunter R.S. and Harold R.W. (1987). The Measurement of Appearance, Second Edition, John

Wiley, New York. Johnson A.J. (1992). [)e~icc lndependcnt Colour ... Is It Real? Proc. TAGA Conference,

Vancouver, April I992.

272 1 :,~luat~m ~!/ Co/our F~d~'lit) /Or Reproductions of Fine Art Paintings

Johnston, A. and Scott Taggart, M. (1993). Gb~ldelincs jbr Choosing the Correct Viewing Condition, s/i~r (2olour Publishing, P1RA Press, UK.

MacDonald, L.W. (1993a). Colour Fidelity issues in Image Reproduction for Print, Proc. EOS/SP1E Europto Syrup., SHE Vol. 1987, pp. 77-87.

MacI)onald, L.W. (1993b). Image Representation and Colour Separation in the MARC Project, Pr~)c. Electronic Imaging and the Visual Arts, Fourth Summer Conf., London.

MacI)onald, I,.W., I)eane, J.D. and Rughani, D.N. (I994). Extending the Colour Gamut of Printed Images, ]. Phot. Science, 42 (3), pp. 97-99.

MacDonald, [,.W. & Lenz, R. ([994). An Ultra-I-tigh Resolution Digital Camera, J. Phot. Science, 42 (2), pp. 49-51.

Robertson, P.K. (1988). Perceptual Color Spaces, IFEl Computer Graphics & Applications, 8 (5), pp. 50--64.

Saunders, D. and Cupitt, J. (1993). Image l>rocessing at the National Gallery, The VASARI project, N~tzon,d Gallery Technical Bulletin, 14, pp. 77 85.

[,[NI)~,'\'~ NI\¢ [)',',N~\l D, J . \N ~'ll',l~.l }\l~ \ N I ) i )A\ ' i l )SAUN|)ERS

A p p e n d i x A

Master of Liesborn

Sum 0[ Scores.liar Master o¢ Lieshorn

A NC PG N P PC NG AC

~\ 0.38 0.24 0.48 0.70 0.84 0.58 0.44

N(" 0.62 0.20 0.45 0.70 0.74 0.54 0.51

P(; 0.76 0.80 0.80 0.97 0.95 0.90 0.84

N 0.52 0.55 0.20 0.83 0.85 0.50 0.67

P 0.30 0.30 0.03 0.17 0.69 0.48 0.27

P ( 0.16 0.26 0.05 0.15 0.31 0.30 0.23

NG 0.42 0.46 0.10 0.50 0.52 0.70 0.61

~\(" 0.56 0.49 0.16 0,33 0.73 0.77 0.39

sum 3.33 3.24 0.98 2.89 4.77 5.55 3.70 3.56

,;Idc~ (.~7 ().60 0.43 0.37 ().54 0.48 1.18 0.61

273

A NC PG N P PC NG AC

+2crT 3.62 3.72 1.33 3.19 "5.20 5.93 4.64 4.05

,urn 3.33 3.24 0.98 2.89 4.77 5.55 3.70 3.56

2err- 3.(13 2.76 0.63 2.59 4.34 5.16 2.76 3.07

Sum o! Z-s('oreLfor Master of Lieshorn

A NC PG N P PC NG AC

A -0.22 -0.40 -0.02 0.35 0.60 0.26 -0.11

NC 0.22 - 0 . 6 2 - 0 . 1 0 0.42 (/.50 0.13 0.02

PG 0.40 0.62 0.49 1.27 1.10 1.62 0.75

N 0.02 0.10 -0.49 0.64 0.63 0.00 0.31

P -0.35 -0.42 -1,27 -0.64 0.35 -0,05 -0.51

PC -0.60 -0.50 - l ,10 -0.63 -0.35 -0,69 -0.56

NG -0.26,-0.13 -1.62 0.00 0.05 0.69 0.36

AC 0.11 -0.02 -0.75 -0.31 0.51 0.56 -0.36

sum - 0 . 4 6 - 0 . 5 7 - 6 . 2 3 - 1 . 2 0 2.89 4.42 0.91 0.25

A NC PG N P PC

~2crr 1.50 1.39 -4.27 0.76 4.85 6.38

sum -0.46 -0.57 -6.23 -1.201 2.89 4.42

- 2 e r r - 2 . 4 2 - 2 . 5 3 -8.19,-3.16 0.93 2.46

NG AC

2.87 2.21

0.91 0.25

-1.05 -1.71

274

Cologne .~choo]

Summa~iv Of Scores for Portrait ~¢a Woman

NG P PC NC N A PG AC

NG :0.57 0.45 0.23 0.33 0.28 0.58 0.28

P 0.43 0.43 0.28 0.32 0.27 0.53 0.27

PC 0.55 0.57 0.33 0.35 0.36 0.61, 0.34

NC 0.77 0.72 0.67 0.44 0.40 0.66 0.45

N 0.67 0.68 0.65 0.56 0.41 0.62 0.42

A 0.72 0.73 0.64 0.60 0.59 0.73 0.47

PG 0.42 0.47 0.39 0.34 0.38 0.27 0.29

AC 0,72 0.73 0.66 0.55 0.58 0.53 0.71

sum 4.29 4.47 3.90 2.90 2.97 2.52 4.43 2.52

cy 0.72 0.72 0.68 0.35 0.39 0.39 0.75 0.66

t: ;',du,aum of Colour Fideht) /i, Reprodz¢ctzons of Fine Art Paintings

NG P PC NC N A PG AC

+2err 4.58 4.75 4,17 3.04 3.13 2.67 4.73 2.79

sum 4.29 4.47 3.90 2.90 2.97 2.52 4.43 2.52

-2err 4.00 4.18 3.63 2.76 2.82 2.36 4.13 2.26

Summary (¢ Z-scores fi)r Portrait O[a Woman

NG P PC NC N A PG AC

NG 0.18 -0.12 -0.74 -0.45 -0.58 0.18 -0.58 i

P -0.18 - 0 . 1 8 - 0 . 5 8 - 0 . 4 5 - 0 . 5 8 0.07 -0.58

PC 0.12 0.18 - 0 . 4 3 - 0 . 4 0 - 0 . 3 6 0.28 -0.43

NC 0.74 0.58 0.43 -0 .16-0 .26 0.41 -0.12

N 0.45 0.45 0.40 0.16 -0.23 0.31 -0.19

A 0.58 0.58 0.36 0.26 0.23 0.60 -0.07

PG -0,18 -0.07 -0.28 -0.41 -0,31 -0,60 -0.56

,\C 0.58 0.58 0.43 0.12 0.19 0.07 0.56

sum 2.11 2.48 1.04 - 1 . 6 2 - 1 . 3 5 - 2 . 5 4 2.41 -2.53

NG P PC N(" N A PG AC

+2err 2.51 2.88 1.44 -1.22 -0 .95-2 .14 2.81 -2.13

sum 2.11 2.48 1 . 0 4 - 1 . 6 2 - 1 . 3 5 - 2 . 5 4 2.41 -2.53

-2err 1.71 2.08 0.64 - 2 . 0 2 - 1 . 7 5 - 2 . 9 4 2.01 I-2.93

i,IINI)S,V~ MA{:[)<)NAII), ]/qN N'I~W,¢*\I(,",NI) I)AkI[) SAUNI)ERS

Cologne School (Illuminant A)

NG P PC NC N A PG AC

NG 0.59 ,0"5,1 0.2 0.22 0.21 0.65 0.28

P 0.41 0.48 10.21 0.28 0.23 I 0.53 0.29

PC 0.49 0.52 0.18 0,2 0.23 0.66 0.26

NC 0.8 0.79 0.82 0.36 0.36 0.73 0.4

N 0.78 0.72 0.8 0.64 0.43 0.76 10.43

A 0.79 ,0.77 0.77 0.64 0.57 0.76 0.49

PG 0.35 0.47 0.34 0.27 0.24 0.24 0.25

AC 0.72 0.71 0.74 0.6 0.57 0.51 0.75

sum 4.34 4.58 4.46 2.74 2.44 2.19 4.84 2.4

stdev 0.75 0.56 0.74 0.37 0.44 0.33 0.66 0.73

NG P PC NC N A PG AC

+2err 4.9 4.99 5.01 3.01 2.77 2.43 5.32 2.94

sum 4.34 4,58 4.46 2.74 2.44 2.19 4.84 2.4

2err-3.79 [,4.16 3.92 2.47 2.12 1.95 4.35 1.86

275

I 1 - i I l '+

NG F' PC NC N A

Scores for Cologne School (A)

I

PG AC

276 t: :,ah¢azzo,e o/Colour Flldcllt.1 ~07" Reproductions of Fine Art Paintings

Cologne Mbool Illuminant (D50)

NG P PC NC N A PG AC

NG ] 0.54 0.47 0.26 0.36 0.28 0.49 0.23

P 0.46] 0.49 0.3 0.27 0.31 0.56 0.2

PC 0.53 0.51 0.44 0.36 0.41 0.64 0.36

NC 0.74 i 0.7 0.56 0.47 0.39 0.63 0.52

N 0.64 0.73 0.64 0.53 0.43 0.57 0.51

A 0.72 0.69 0.59 0.61 0.57 0.73 0.45

PG 0.51 0.44 0.36 0.37 0.43 0.27 0.32 I

AC 0.77 0.8 0.64 0.48 0.49 0.55 !0.69

s u m 4.38 4.42 3 .76 2 .99 2.93 2 .64 4 .32 2 .57

stdev 0.74 0.87 0.68 0.76 0.76 0.62 0.98 0.37

NG P PC NC N A PG AC

+2err 4.86 4.98 4.2 ]3.49 3.43 3.05 4.96 2.82 i

sum !4.38 4.42 3.76 2.99 2.93 2,64 4.32 2.57

2err- 3.9 !3.85 3.31 2.49 2.44 2.24 3.68 2.33

4

. c i

I

, , J

' 7

i t

t I' t t ~o ~; Z <2 (D

Scores for (iolognc School (D50)

I O ,<

[A"d l )~ , , \ '~ ~ . ' I A ( : I ) t ) N A [ . I ) , J A N ~"lt >Rt , \ I ( , ~ , N I ) l ) A ' v l l ) S A U N I ) E R S

Cologne School (Illuminant D65)

A

B 0.43

C 0.61

D 0,77

E 0.6

F 0.65

G 0.41

H O.68

sum 4.14

sider 0.88

B C D E F G H

0.58 0.39 0.23 0.4 0,35 0.59 0.33

0.32 0.34 0.42 0,28 0.49 0,33

0.68 0.38 0,48 0.44 0.54 0.39

0.66 0.63 0.49 0.45 0.61 0.44

0.58 0.52 0.51 0.37 0.54 0.34

0.73 0.56 0.55 0,63 0.68 0.48

0.51 0,46 0.39 0.46 0.32 0.31

0.68 0.61 0.56 0.66 0,52 0,69

4 . 4 3 . 4 8 2 . 9 6 3 . 5 4 2 . 7 2 4 . 1 4 2 .61

0.8 0.32 0.49 0.8 0.79 1.13 0.5

+2er

NG P PC NC N A PG AC

4.75 4.96 3.7 3.31 4.1 3.26 4.92 2.96

sum 4.14 4.4 3.48 2.96

2er- 3.54 3.84 3.26 2.62

[Jl~IM I~ll i lm mlWlll m

277

4 . ~

4

: +

i

I I , I I I I I a, L0 :_, Z < (5, 0

c~ Z c~ <

Scores for Colognu School (I)65)

278

Claudesque

Sum qf Scores tc~r Claudesque painting

NC PC N P NG PG A AC

NC 0.50 0.62 0.22 0.36 0.03 0.36 0.55

P(" 0.50 0.77 0.48 0.46 0.08 0.56 0.69

N 0.38 0.23 0.28 0.33 0.135 0.38 0.40

P [).78 0.52 0.72 0.57 0.23 0.65 0.78

NG 0.(~ 0.54 0.67 0.43 0.06 0.58 0.69

PG 0.97 0.92 0.95 0.77 0.94 0.94 0.95

A 0.64 0.44 0.62 0.35 0.42 i 0.06 0.64

AC 0.45 0.31 0.60 0.22 0.31 0.05 0.36

s u m 4.36 3.45 4 .95 2.75 3.40 0.56 3.82 4 .70

cr (~63 0.51 0.32 0.39 0.52 0.53 0.36 0.59

t:valuatw~ o] Colour t'ideht) for Reproductzons of Fine Art Paintings

NC PC N P NG PG A AC

+2err 4.87 3.87 5.21 3.06 3.81 0.98 4.12 5.17

sum 4.36 3.45 4.95 2.75 3.40 0.56 3 . 8 2 4.70

-2er 3.85 3.04 4.70 2.44 2.98 0.13 3.53 4.23

Sum :~/Z .s(ores for Claudesque painting

NC PC N P NG PG A AC

NC 0.00 0.22 -0.55 - 0 . 3 0 - 1 . 5 7 - 0 . 2 6 0.10

PC 0.00 0.45 -0.05 -0 .09-1 .02 0.09 0.40

N -0.22 -0.45 -0.29 -0.28 -1.04 -0.15 -0.17

P 0.55 0.05 0.29 0.12 -0.48 0.20 0.54

NG 0.30 0.09 0.28 -0.12 -I.18 0.13 0.40

PG 1.57 1.(12 1.04 0.48 1.18 0.97 1.33

A 0.26 -0.09 0.15 - 0 . 2 0 - 0 . 1 3 - 0 . 9 7 0.25

AC 4 ) .10 -0 .40 0.17 - 0 . 5 4 - 0 . 4 0 - 1 . 3 3 -0.25

s u m 2.36 0.22 2.60 -1 .27 0.11 -7 .60 0.74 2.84

NC PC N P NG PG

+2err 3.16 1.02 3.40 -0.47 0.91 -6.80

sum 2.36 0.22 2.60 -I.27 0.11 -7.60

2err 1 .56 -0 .58 1.80 - 2 . 0 7 - 0 . 6 9 - 8 . 4 0

A AC

1.54 3.64

0.74 2.84

-0.06 2.04

LIND~,,',~ N'I.,k(:I)~NAID,,JAN M{~R{~\I(" ANI) D,'\~[I) SAUNDERS

Appendix B

279

MARC Quest ionnaire Master of Liesborn page 1 of 3

Name: Age: Sex: M F

You will be asked to choose the reproduction that matches the original better in a given area (described by its characteristics and shown in the image below) or a given overall criterion (points 9 to 11 ). Please tick the box showing your choice in the following points; tick both boxes to indicate that both reproductions are equal.

highlight

Iuration

reen

J blue

Pair No, I(AB) 2 (AC) 3 (AD) 4 (AE) Top Bottorr Top Bottom Top Botton Top Bottom

1: RED area

2: GREEN area

3: BLUE area

4: YELLOW area

5: SHADOW area

6: HIGHLIGHT area

7: NEUTRAL area

8: SATURATED area

9: OVERALL SATURATION

10: OVERALL CONTRAST

11 : OVERALL

280 1 ~,aluatmn qf (.'olc>ur l"td~'htv /.> Reproductions q/Fine Art Paintings

MARC Quest ionnaire Cologne School page 2 of 3

Name: Age: Sex: M F

You will be asked to choose the reproduction that matches the original better in a given area (described by its characteristics and shown in the image below) or a given overall criterion (points 8 to 10). Please tick the box showing your choice in the following points; tick both boxes to indicate that both reproductions are equal.

I L L U M I N A N T D50

:d

skin tone

h=ghlight snaeow

Pair No. I (AB) 2 (AC) 3 (AD) 4 (AE) Left Right Left Right Left Right Left Right

1: RED area 2: GREEN area 3: BLUE area 4: SHADOW area 5: HIGHLIGHT area 6: NEUTRAL area 7: SKIN TONE area 8: OVERALL SATURATION 9: OVERALL CONTRAST 10: OVERALL

i

[.INI ~,,\'~ ~k'~,\( ])~'~NA[ [~, .[AN M~ ~1<~ ~\ [ ( , \ N [ ) [)AVH) gAUNI)}~RS 29l

MARC Questionnaire Claudesque page 3 of 3

Name: Age: Sex: M F

You will be asked to choose the reproduction that matches the original better in a given area (described by its characteristics and shown in the image below) or a given overall criterion (points 8 and 9). Please tick the box showing your choice in the following points; tick both boxes to indicate that both reproductions are equal.

red blue

Pair No. I(AB) 2 (AC) 3 (AD) 4 (AE) Left Right Left Right Left Right Left Right

1: RED area

2: GREEN area

3: BLUE area

4: YELLOW area

5: SHADOW area

6: HIGHLIGHT area

7: NEUTRAL area

8: OVERALL CONTRAST 9: OVERALL