Zone map of color categories

Participants were instructed to sort each stimulus into one of the given color categories. A custom keyboard with 12 tags of Mandarin color vocabulary was used to input the sorting results. The initial 40 trials were for practice and were not recorded. The practiced observers were familiar with the position of color terms on the keyboard and were able to produce rapid and accurate sorting actions. All stimuli were presented in random succession. During the experiment, the observers could use a pause key and a resume key to break and restart the experiment. The flow of the experiment was controlled by Presentation^® (Neurobehavioral System, Inc.).

5.3 Results

5.3.1. Zone map of color categories

The participants have produced 20,284 color category judgments via the force-choice sorting task. These judgments are submitted to render color zone maps connecting semantics with perception. Table 5-1 provides a descriptive overview of categorical sorting results in conditions of different L levels. In this table, the rank order for the sum of frequency counts is G (green), Br (brown), P (purple), B (blue), O (orange), Y (yellow), Gr (gray), R (red), Pk (pink), Dpk (deep pink), Bk (black) and W (white). There were very few unreasonable judgments—precisely two votes for Bk in 170 and one in 100cd/m² conditions, and one for W in 5 and 10 cd/m². These should be ignored because they may easily have been keyboard input errors. The input key for Bk was close to the key for W.

Table 5.1. The descriptive overview of categorical sorting results in different L level conditions

R O Y G B P Pk Dpk Br Gr W Bk

5 cd/m² 88 23 12 671 336 578 9 19 805 104 1 304 10 cd/m² 203 47 34 1089 367 791 34 43 982 235 1 90 25 cd/m² 348 437 39 1143 399 794 38 140 798 211 0 9 50 cd/m² 248 786 265 1459 464 680 297 415 497 213 0 0

100 cd/m² 15 284 331 1077 329 224 301 17 101 127 9 1

170 cd/m² 4 20 300 255 124 23 87 2 22 39 46 2

54

sum 906 1597 981 5694 2019 3090 766 634 3205 929 57 406

Figure 5.2. The stack histogram for presenting normalized frequency distribution of each condition. The x-axis shows the color categories, and the y-axis shows the (accumulated) ratio. The various filled grey levels are used to denote six luminance conditions.

A cross-L comparison of the normalized frequency distribution is shown on the stack histogram in Figure 5-2. The x-axis lists the given color categories, while y-axis presents the ratio of original counts to stimulus numbers of each condition, with various grey-level fills to differentiate the six L levels. The histogram presents a rough structure of the frequency distribution across color categories and L levels. The green, blue and grey categories give relatively even frequency ratios, implying that these three color concepts are luminance invariant; that is, they exist in all L conditions. In contrast, the other color categories are perceived at a limited number of L levels. Red, purple and brown are more frequently perceived in medium to low L conditions, orange and deep pink are recognized in middle L conditions and yellow and pink are apparent in high L conditions. It is particularly noteworthy that the chosen color terms (except for the achromatic terms) are often referred to as ‘hue’

terms, suggesting that these should be more or less independent from luminance and

55

saturation. However, some ‘hue’ terms, such as yellow and deep pink, seems to be typical of specific luminance levels. This luminance-dependent phenomenon in color category sorting has been addressed(Shinoda, et al., 1993), and will be discussed further in the following section.

The interaction between the L condition and the recognized color category is illustrated in Figure 5-3. The upper six x-y chromaticity diagrams with colored circles represent the demarcated color zones in six L conditions. The coordinates of each circle’s center correspond to those of the stimulus. The colors of the circles intuitively symbolize the category that gained maximum votes, the mode, except for the light grey in the L170 condition symbolizes white. In addition, the circle sizes correspond to the maximum number of judgments in order to better visualize the degree of consensus under each stimulus condition. Generally, larger circles symbolize the focal color of the category spread over the peripheral districts of high purity in colorimetry, whereas the smaller circles are found in the common border between distinct color zones and in the central area of low purity surrounding the reference white W.

The composition of color zones appears to be diverse across the six L conditions. In the lowest L condition (5cd/m²), there are only five perceptually dominant categories: green, blue, brown, purple and black. As luminance levels increase, the other color categories gradually become apparent. Specifically, red and grey become recognizable from L=10, orange and deep pink from L=25 and pink and yellow from L=50. Certain color categories become less apparent in higher L conditions; specifically, red, deep pink and brown are seldom identified from L=100. In the highest L condition (170cd/m²), only yellow, pink, blue, green and white remain visible. These results suggest that a common concept of color, labeled with a specific color term, is not merely an idea of a hue independent from brightness and saturation information. The results suggest that some color (hue) terms, such as red, yellow, pink and others are strongly associated with luminance.

The lower diagram in Figure 5-3 combines the above six diagrams. It reveals the spatial

56

changes of color category compositions along with L level conditions. The color of the open circle also represents the mode of color sorting, and the size of the circle corresponds to the L level. The largest outer circles denote color zones designated in L=170, and the inner smaller circles that decrease in size denote gradually decreasing L conditions. The color consistency of the concentric circles is an index of the degree of luminance dependency in the sorting results. Stimuli located around the borders between categories appear more ambiguous, and naturally are designated into different categories when luminance changes. The top and lower-left corner of the gamut triangle, demarcated as green, blue and partial purple, show strong consistency across all L levels. However, color zones in the area from the center to lower-right corner of the triangle are strongly influenced by luminance conditions. It is important to note the superseding pattern of some groups in that area, such as a warm color group (yellow, orange and brown) and another group (pink, deep pink and purple). Members in these two groups seem to displace each other as the luminance conditions change. It is remarkable that the brown category overlaps a large range of color categories according to the fluctuation of luminance. A stimulus fixed in a chromaticity coordinate is recognized as brown in lower L but would be called yellow, orange or even pink as the L gets higher. In addition, the gray category also demonstrates a similar but weaker effect; it can substitute for many other colors as the luminance condition changes. This effect is related to the so-called

‘wild-card’ phenomenon.(Greenfeld, 1986)

57

Figure 5.3. Upper six x-y diagrams of different L levels using circle color and size to present color category and mode size, respectively. The lower diagram combines all results and differentiates the modes of six conditions with open circles that decrease in size. The larger outer circle represents the mode of the 170cd/m² condition, and the smallest circle represents the mode of the 5cd/m² condition.

The particularized formation within each color category’s luminance condition is presented in Figures 5-4 to 5-7. Each figure contains 6ͪ3 (the number of L levels by the number of color categories) diagrams of smoothed-out contour maps, and all diagrams share the equivalent x-y unit and scale. The contour-smoothing algorithm was provided by

58

OriginPro 8.0 by OriginLab. The denotation of the conditions (cd/m²- color category name) is shown in top-left corner of each diagram. The small black dot in the midlle of diagram marks the position of reference white. Different fills of grey level between contour lines represent the frequency ratio of the observers’ judgments. Zones filled with black indicate that these obtained over 90% of the votes for the corresponding color terms, which means they are focal zones with the least controversy. The other grey fills gradually increase in lightness according to their respective percentage of votes, with a decreasing interval of 10%. Consequently, the darker the zone fill, the higher the frequency, and the more representative the stimulus. The darkest grey fill indicates 80%-90% votes, while the lightest fill (white) indicates 10-20%

votes. A frequency ratio below 10% is ignored and filled with slash lines to mark the gamut.

These countour line maps reflect the noticeable transformation of each color zone involving luminance variation. Each color category shows distinct pattern of emergence, congregation and lapse on the color space.

Figure 5-4 presents zones of red, orange and yellow categories that show prominent lumiance-dependent features in their distribution. The red is recognizable below 50cd/m², and its focal zone (the zone with highest ratio) is relatively small, reaching only 70-80% ratio level. The covered area completely overlaps with the brown zone in 5cd/m², although the probability of seeing red at this L level is quite low. The orange zone is also designated at restricted luminance levels, mainly in 50 and 100cd/m². The formation of the orange contour map reveals a very concentrated pattern; it contains a recognizable focal zone of over 90% in 50cd/m² level, and then the zone diminishes drastically in 100 and 25cd/m²levels. The range of orange and brown categories also overlap considerably, and brown also overlaps with yellow. The yellow zone is recognizable in conditions above 50cd/m², and the focal zone of over 90% can be found in 100 and 170cd/m² level. The first three color categories discussed thus far contain colors of long wavelength range. Their territories are all luminance-dependent, and overlap with the brown zone in low luminance conditions. This result is consistent with

59

the familiar perception that so-called warm colors (red, orange and yellow) would shift into brown as they become darker.

Another cluster of color category zones—green, blue and purple—is shown in Figure 5-5. Apparently, viewers were able to percieve these three colors across all six luminance levels, except that purple was infrequently indentified in 170cd/m² condition. The green category might be considered as a unique color concept that is particularly easy to define, given its large focal zone of over 90% votes and sharp border. Its dense peripheral contour lines reflect a steep fall in frequency ratio. The loaction of the focal zone remains constant, rather than shifting with luminance changes. Moreover, the overlapping area between the green zone and neighboring purple, green and brown zones is very limited in size. A similar pattern of contour lines can be found in the blue zone, although its covered range is much narrower than that of the green zone. The purple category is also a easy-to-identify color, as revealed in the concentrated pattern in the map, typically in conditions below 50cd/m².

However, the perceptual definition of purple seems not as distinct as that of blue or green. In the darker conditions, its zone overlaps patially with those of brown and red, while in lighter conditions it overlaps with deep pink and pink. Generally, when compared with the previously discussed warm color cluster and the other color categories, the three colors in Figure 5-5 demonstrate the notable characteristic of being preceptable across every L level.

Furthermore, the overlapping zone between these and neighboring colors is relatively small, especially for the green and blue zones. All of these observed features suggest that the psychological quality of these colors is more universal, stable and less ambiguous when compared with other colors in the study.

Figure 5-7 shows the contour map of the achromatic categories of gray, white and black. The Gray zone distributes around the lower-left area in all luminance situations, close to the intersection of the blue, purple, green and brown zones. It also appears more clearly in the middle luminance levels, and switches to black in the lowest luminance conditions. The

60

term of white was used only in 170cd/m² condition, and its zone encircles the W point. White and black categories gained very low votes, perhapes due to the fact that observers were provided with thin outlines of white and black for reference with each stimulus. Literally, gray and black should be neutral color concepts that do not involve any hue information. However, the results show gray as a category that represents the ‘cold’ cluster of colors, typically blue and purple, in very low saturation conditions, and black corresponds to cold colors in very low saturation and luminance conditions. The actual neutral exemplar in any luminance level should be located around the reference white point, just as in the white zone. Based on the present results, the ideal neutral point lies on the border between the brown and gray/black zones.

Figure text.

e 5.4. Contouur line map shhowing the for

61

rmation of redd, orange andd yellow. A deetailed descripption is in thee

Figu text.

ure 5.5. Contoour line map sshowing the fo

62

ormation of green, blue andd purple. A deetailed descripption is in thee

Figure the tex

e 5.6. Contou xt.

ur line map shhowing the for

63

rmation of pinnk, deep pinkk and brown. AA detailed desscription is inn

Figure text.

e 5.7. Contouur line map shhowing the fo

64

ormation of grray, white andd black. A deetailed descripption is in thee

65