Experimental Set-up

2.1.1. Stimuli and Viewing Conditions

The two computer-controlled colour monitors used were the following: a diagonally 54 cm Hewlett Packard P1100 CRT monitor and a Panasonic TH-42PHD plasma display panel (PDP) having a diagonal of 106 cm. The two devices were set up next to each other on a table. The angle between their screens was 110° (see Figure I/3).

The PDP served to show the large test field. To achieve a quasi-immersion by filling the whole visual field of the observers, a wooden viewing booth was set up in front of the PDP. The inner surface of the booth was covered by mirrors (2 walls, floor, and ceiling).

The depth of the viewing booth was 50 cm. The visual field of the screen of the PDP itself subtended 85° (horizontal) and 55° (vertical), from the viewing distance of 50 cm (the edge of the booth). In order to achieve a more uniform viewing field, and because of the short viewing distance, the screen of the PDP was covered by a neutral density diffusion filter.

The standard size stimuli to be matched to the large stimuli were displayed on the CRT monitor. The entire screen of the CRT monitor (with the achromatic background) subtended a 39°×30° (horizontal × vertical) viewing angle. The viewing distance for the CRT was 60 cm. To exclude light reflections from either device, a separator was mounted

between the two displays, see Figure I/3. The separator was a board covered by black matte paper. To ensure natural binocular viewing, there was no chin rest for the observers but positioning their head to the right side of the separator helped them maintain the correct viewing distance from the matching stimuli to view it under the desired angles on the CRT monitor.

The CRT and the PDP displays were driven by two different video controllers in two separate computers. The large stimuli were pre-set throughout the whole experiment, and only the matching stimuli presented on the CRT were adjustable. Measuring the luminance of a full-screen white at 9 evenly distributed points resulted in a mean value of Y=68.6cd/m² (STD=1.8 cd/m²) and this homogeneity was considered acceptable for the purpose of this research. Although, the CRT proved to be colorimetrically stable, in the evaluation of the results, instead of a model-predicted output, only in-situ measured values of the displayed test stimuli and the observer’s matching answers were considered. A calibrated PhotoReserach™ 705 spectroradiometer^* was used to carry out these in-situ spectral measurements. Specifications of this instrument are listed in Table I/1.

Table I/1. Specifications of the PR-705 spectro-radiometer (manufacturer’s data) Specifiction Value

Spectral range 380 – 780 nm Angle of view 2°

Spectral bandwidth 20 nm Spectral accuracy ±2 nm Luminance accuracy

(Standard illuminant A) ±2%

Chromaticity accuracy (Standard illuminant A)

x ±.0015, y ±.001

In previous examinations and in the course of several measurements⁵¹, though the plasma monitor turned out to be a stable display, no characterization model^52,53 was developed for it, but the stimuli were set experimentally. It only displayed static stimuli repeatedly during the experiments.

* Calibration of the instrument was performed at the Ilmenau University of Technology using a PTB (Physikalisch-Technische Bundesanstalt, Germany) calibrated FEL type incandescent lamp. Based on these comparative measurements, after correction of the systematic errors of the instrument, a maximum uncertainty of one to two digits in the third decimal of the chromaticity co-ordinates could be obtained. For wavelength calibration different (Argon and Mercury) gas-discharge pencil lamps were used.

In this part of the experiments, the effect of two separate parameters was focused on regarding the investigated phenomenon. First, the question was if, assuming that observers would not reproduce the colorimetric values of the large stimuli due to the size effect, the use of 2° or 10° corresponding samples yield better results. In other words: Does the employing of 10° samples instead of 2°samples help predict the appearance of the immersive colours? On the other hand, the photopic luminance of the background of the small sample was also of interest: if keeping it on a constant mid value or making it equal to that of the test colours’ luminance for every match yielded less error in reproducing the immersive colour perception (The colour size effect may be assigned to a contrast effect due to the indispensable consecutive brightness contrast between the often significantly brighter immersive test colour and the background of the corresponding small stimulus to be adjusted during memory matching).

As noted previously, 16 test colour stimuli were selected to experiment with. The test colour stimuli covered the whole PDP screen but the corresponding matching colour stimuli subtended only 2° or 10° on the CRT monitor. The matching stimuli were of square shape (side length: 10.5 cm – 10° and 2.1 cm – 2°) and appeared in the middle of the CRT display. The chromaticities of all test colour stimuli were away from the colour gamut boundaries of the CRT monitor, in order to be able to display the expected magnitude of the colour size effect on the CRT, according to earlier results.42,43,46,47 The CIELAB a*-b*

coordinates of the test colour collection are plotted in Figure I/4. These CIELAB values were determined from the measured spectral data by using the CIE 1931 colorimetric observer and the measured tristimulus values of the maximum achromatic output of the CRT monitor as white point^* (Y=59.69 cd/m², x=0.2956, y=0.3087, CCT=7874 K;

r=g=b=255). These values are also listed in Table I/2 together with numbering and approximate colour names to make them easier to imagine. The numbers (as next to each dot in Figure I/4) will refer to the stimuli in the rest of Section 2.

* It seems to be a contradicting procedure to derive the CIELAB values for the immersive stimuli of the PDP, too, using a reference white that is not present on the screen, since, it is the white of the CRT but for a first approach and for practical reasons it was accepted to do so. Since, a model will be set up to predict the colour appearance of the immersive colour, a colour difference measure was needed to be able to express the goodness of the model and compare it to that of the related models found in literature. It will be reflected on in the forthcoming sections that this approach does not radically decrease the nature and predictability of the size effect (see the model performance tests at the end of Section 3.2.1 where the model derived from an experiment performed adequately applied to the result of another experiment using another white point).

Also, in the General Discussions (Section 4), besides further anatomizing the legitimate use of CIELAB, even comparing its performance in the present experiments to that of CIECAM02, experimental data will be converted into absolute measures (luminance and CIE u’-v’) and Hunt model attributes for further discussions.

13

Figure I/4. The 16 test colour stimuli used in the experiment in a CIELAB a*-b* diagram, for L^* values refer to Table I/2.

Table I/2. List of the CIE L*, a*, b* values of the 16 test colour stimuli together with their naming

The luminance of the background of the matching stimuli on the CRT had two values.

In the further part of the paper the two background luminance conditions will be referred to as fix background (the same for all stimuli) and changing background (the same luminance as that for the immersive test colour stimulus). For the fixed condition this grey background was X=29.35, Y=30.89 (cd/m²) and Z=38.72, which is half the luminance of the peak white of the CRT; for the changing background criterion it was the same chromaticity grey as for the fixed case (x=0.297, y=0.312) with the same luminance as the actual test stimulus being observed.

2.1.2. Procedure

The 16 test colours (see Figure I/4 and Table I/2) were organized in a sequence of five repetitions each and additional 20 randomly selected colours were added, not being evaluated later, just to make the observations more varied. The resulting set of hundred colours (5×16+20=100) was then randomized and divided into four groups (25 matchings in each). Observers had to assess the appearance of the immersed stimuli of one group at a time. They were requested to adjust the colour of the matching stimuli on the CRT monitor by the aid of three sliders, each representing a colour attribute (lightness, hue, saturation) in HSV colour space,⁵⁴ until the colour perception of the large stimulus matched that of the matching sample. Instead of the classical tristimulus mixing procedure (i.e. using the display’s RGB channels directly) this and other methods based on adjusting the correlates of perceptual attributes of colours (e.g. hue, chroma and lightness) give significantly better performance in terms of both accuracy and time.⁵⁵ Though regarding the HSV system, the words hue, saturation and value are not so reasonable colorimetrically, HSV itself is a mathematically equivalent representation of the RGB colour space (a simple reformulation) used frequently in the digital media for the digital representation or computer-controlled display of colours.

Each matching started with the observation of the large field presented to the observer on the PDP for two seconds only (the reason for the short appearance of the large field is described later), and then, the grey background appeared (evidently, its luminance corresponded with the background condition of the experimental session). Then, subjects moved their head and looked at the CRT monitor to adjust the matching colour. Initially, the matching colour was set to achromatic mid-grey as close as possible to the grey background of the PDP both in chromaticity and luminance (Y=29.9 cd/m²) for each matching. The task was to modify the hue (H), saturation (S) and value (V) sliders in the

HSV colour model, until the colour appearance of the standard size matching colour matched the colour appearance of the memorized immersive scene. The large stimulus appeared repeatedly at fixed time intervals (8 seconds) automatically but always for 2 seconds only, and observers had unlimited occasions to look back and forth between the two situations, until a perfect match was found between their perceptions. After storing the adjustments, the next large stimulus appeared from the current group of stimuli and a new matching was requested. The matching procedure for one group of stimuli lasted about 45 minutes.

All observations were carried out in an experimental room where the only light sources were the monitors. The subjects were provided a 5 minute adaptation period in the experimental room with the displays on before starting their adjustments. To complete the experiment, each subject made 5 repeated matchings for each of the 16 test colours and for both of the two background situations (fixed and changing, see above) and for both matching sample sizes (2° or 10°), hence there were 16x5x2x2=320 matchings per observer altogether (or 400 with the additional 4×20 random colours). For the four different situations (2 background types and 2 matching stimulus sizes) four different series were conducted, these parameters were fixed during a matching session.

2.1.3. Subjects

Three trained female colour-normal observers participated in the experiments, aged between 22 and 27. They have been involved in many colour experiments of our Laboratory for several years. They were aware of the perceptual attributes of colours and familiar with the HSV colour system. Right before the experiment observers received detailed instructions on the task to perform. An English translation of the handout can be found in the Appendices.