A cross-context matching task is employed to investigate the type of transformation resulting from the change of the color of the context. Some of the colors presented to the subjects have to be highly saturated to investigate whether for these colors always a match within the changed context can be obtained. If this is true, only a projective transformation can predict the effect of changing the context.
On the other hand, not all stimuli must be highly saturated in order to be able to estimate the parameters of the transformation caused by changing the context: If highly saturated colors do not change under different contexts, the effect of the context has to be shown with other stimuli. To be able to present as well highly saturated stimuli as colors with a low saturation, a new experimental setup described below was designed and realized.
Four out of 28 subjects tested are selected based on their ability to produce reliable color matches. All of these subjects exhibit good or average ability to discriminate colors when performing a Farnsworth-Munsell 100-Hue-Test. They take part in the experiment comprising 21 sessions lasting for approximately one hour.
Figure 3:
Color-discrimination of subject EIM: This figure shows the ability to discriminate colors as measured by the Farnsworth-Munsell 100-Hue-Test (Farnsworth, 1957).
Figure 4:
Color-discrimination of subject SCA: This figure shows the ability to discriminate colors as measured by the Farnsworth-Munsell 100-Hue-Test (Farnsworth, 1957).
To qualify for participation in the experiment, the subjects furthermore participate in five sessions lasting approximately one hour. During that phase they learn to operate the experimental setup by producing more than 20 cross-context-matches and more than 100 matches within the original context. When starting the experimental trials, they have enough practice to be able to produce color matches reliably. The results of Brown (1957) suggest, that extensive practice reduces the variability of the colors matched to the target (see figure 5). No further improvement of the subjects skills to perform color-matches should occur during the experimental phase.
Figure 5:
The effect of extensive practice on color-discrimination (Brown, 1957) The figure on the left shows the color-discrimination ellipses at the beginning of the experiment of Brown (1957), the figure on the right after extended practice (17 matches).
Since some of the stimuli presented are highly saturated and thus cannot be displayed by a CRT-monitor. Therefore a new experimental setup based on a liquid crystal tunable imaging filter (LCTF) is used to produce the infield, which is projected onto a reflecting ``white'' area fixed in the middle of a CRT-monitor. The surrounding context simply is produced by that monitor. The subjects sit in front of the monitor. The distance between them and the stimulus is restrained to 53 cm by a headrest.
The light finally forming the infield are created the following way (see figure 6): A 100 W quartz tungsten lamp emits two beams of light on opposing sides, which are bundled using a condenser each. One of these beams is filtered by the LCTF after being cooled down so that the spectrum of the light passing the LCTF is reduced to a nearly monochromatic one. The use of the LCTF is motivated by the fact that it can produce monochromatic spectra with high efficiency compared to a conventional monochromator.
Figure 6:
The apparatus used: This figure shows the integral parts of the experimental setup. The subject looks at a circular target-stimulus fixed in front of a larger, also circular context. The context is produced by a CRT-monitor, while the target stimulus is projected onto a uniformly reflecting area fixed on that monitor.
This monochromatic light can be desaturated by mixing it with the other beam whose intensity can be reduced by the use of a variable neutral density filter. Finally the intensity of the united rays can be reduced by the use of a second variable neutral density filter. Since the intensity of the light can only be reduced, the luminance of the stimuli resulting is limited (by the luminous intensity of the lamp used and by the transmittance of the optical components, especially the LCTF).
The characteristics of the essential parts of the experimental setup are as follows: The light source used, a 100 W QTH-Lamp, has a smooth spectrum of radiant intensity which contains no peaks and increases at higher wavelengths (see figure 7). This type of lamp produces a very stable light whose intensity does not vary in time.
Figure 7:
Spectrum of the radiant intensity of a 100 W Quartz-Tungsten-Halogen-Lamp: This figure shows the radiant intensity of the lamp used for the experiment as quoted by the manufacturer of the device (Oriel Corporation, 1994).
In order to regulate the luminance of the stimulus produced and to decrease the intensity of the white light used to desaturate it, two variable neutral density filter are employed. These circular filters are mounted on the axis of a stepping motor that is controlled by a computer. Depending on the degree of rotation of the filter, its transmittance decreases (see figure 8).
Figure:
Transmission of the neutral density filter 1: This figure shows the relative transmission of the neutral density filter varying with the degree of rotation (solid line) as measured by a photometer and its idealized transmission (dashed line), starting with the maximum transmission. The irregularities of the transmission measured resluts from unprecise measurements taken by the photometer used after an automatic change of the measurement-range.
The LCTF filters the light passing through it to a narrow spectrum approximately 5 nm broad whose maximum can be controlled by a computer. The transmittance of the LCTF depends strongly on the wavelength it is adjusted to (see figure 9). In order to produce a stimulus with a well-defined luminance, this variation has to be compensated for. To do this, the transmittance is measured weekly and stored in a database that is used each time a stimulus is produced. However, the values stored in this database do not specify the transmittance exactly, since the transmittance of the LCTF does vary irregularly in the course of time (see figure 10). The extent of these variations is dependent on the wavelength the LCTF is adjusted to. The transmittance is very unstable with some settings (e.g. 500 or 555 nm), while it remains quite fixed with others (e.g. 600 nm). Only stimuli whose dominant wavelength can be produced by the LCTF with minor variations are used for the experiment. Since the transmittance of the LCTF changes further depending on its temperature, the experimental setup was started at least one hour before the first trial of the day. So its temperature reaches a stable level before the experimental sessions get started.
Figure 9:
Transmission of the LCTF: This figure shows the absolute spectral transmittance curve of the LCTF measured by placing a photometer directly into the beam passing the LCTF.
Figure 10:
Variation in the transmittance of the LCTF in the course of time: These figures show 7300 measurements taken within one hour while the parameters of the experimental setup remained fixed. The neutral density filter 1 was set to maximal transmittance, the neutral density filter 2 to minimal transmittance and the maximum of the transmittance of the LCTF was set to a wavelength of 500 nm (upper figure), 555 nm (center) and 600 nm (lower figure).
The stimulus produced by the optical apparatus is projected onto a circular white paper (diameter 16 mm) which is fixed in front of the CRT-monitor displaying the context. The spectral reflectance of this paper is very homogenous and does not change much in the course of time (see figure 11). So this material has no influence on the hue of the stimulus reflected by it.
Figure 11:
Spectral remission of the reflecting area the stimulus is projected onto: This figure shows the spectral remission of the paper which is fixed to the CRT-monitor and onto which the target stimulus is projected. The spectrum labeled ``old'' (x = 0.3264, y = 0.3236; solid line) was measured from a probe exposed to sunlight for three months; the spectrum labeled ``new'' (x = 0.3269, y = 0.3241; dashed line) was taken from a new probe.
Integrating the parts described above, the experimental setup can produce many colors, including highly saturated ones. Since the intensity of the monochromatic light and of the light added to desaturate it can only be decreased by the neutral density filters, the luminance of the resulting stimulus is limited. Some colors can be produced only at low intensities if the LCTF has a low transmittance at their dominant wavelength. Figure 12 shows the gamut of the apparatus for a luminance of 
.
Figure 12:
Gamut of the experimental setup: This figure shows the colors which can be projected onto the reflecting area with a luminance of measured at the position of the subjects eyes. A diamond symbolizes the colors that can be produced.
The subjects are presented a 
small circular stimulus surrounded by a homogenous context with 
visual angle. The adaptation-context A is slightly reddish ( 
), while the changed context B appears slightly bluish ( 
). The ten stimuli (their coordinates are shown in table 2 and in figure 13) presented within that contexts were selected because they minimize the difficulties occuring at the production of certain other colors using the apparatus described above.
Table 2:
xyL-coordinates of the stimuli presented to the subjects
Figure 13:
The target stimuli and contexts used: This figure shows the chromaticity locus of the ten stimuli used for the experiment ('+') and of the original and changed context (' 
') in the CIE 1931 chromaticity diagram.
After extensive practice in the pre-experimental phase (approximately 120 color-adjustments from 20 cross-context-matches) the subjects have to produce cross-context matches for ten different colors the following way employing the so-called memory method: First the target stimulus is presented as the infield within the first context A. After having pressed a reaction button, the subjects can vary hue, brightness and saturation of a randomly changed infield until it looks the same as the previously presented target. Afterwards the original stimulus is presented and the subject is asked about the quality of its match on a scale of five items (how satisfied?).
This procedure is repeated five times. After the sixth presentation of the target stimulus within context A, the context is changed to context B and the subjects has to adjust the infield until it looks the same as the infield presented as target before. After finishing these adjustments, the target stimulus is presented again within context A and the subject shall rate the quality of this match again.
For each trial, the color-coordinates of the adjustments are recorded as well as the number of actions performed and the time used.