Open Access
Add to BibSonomy
Abstract
This paper investigates if the color appearance model accurately predicts the lightness of a display whose surface features specular reflection such as a mirror display. Eight observers participate in psychophysical experiments and the results are compared with model prediction value. It is found that the model overestimates the lightness under the average surround condition where there is an intense specular reflection on the display surface. Based on the measurement results, we figured out that the conventional measurement method cannot measure practical specular reflection value. Thus we propose a new measurement method and modify the background parameter to include the specular reflection value. The proposed CIECAM02 considering specular reflection improves the coefficient of variation (CV) value under average surround condition from 20.56 to 12.40 while maintaining CVs under dim and dark surround conditions. In conclusion, we found that people perceive the specular reflection image as a background.
© 2020 Optical Society of America under the terms of the OSA Open Access Publishing Agreement
1. Introduction
There have been many attempts to analyze human color vision. After the International Commission on Illumination (CIE) conducted psychophysical experiments, CIE proposed XYZ color space based on experimental results in 1931. But if the viewing condition of two stimuli are different, identical XYZ tri-stimulus values may cause different color appearance. Thereafter several color appearance models including viewing conditions were proposed to solve these differences [1–3].
CIECAM02, one of the color appearance models, was published in 2002 by CIE Technical Committee [3]. Because of its simplicity and reversibility, it has been widely used as an international standard [4–7]. Designed for both surface colors and self-luminous ones, it has also been used for display devices. However, several issues have occurred as the performance of the display has dramatically improved. One of them is Helmholtz-Kohlrausch (H-K) effect referring to the color appearance phenomenon that colored objects (or light) appear brighter than achromatic objects (light) of the same luminance [8–12]. The higher the saturation of the color, the brighter it appears. CIECAM02 does not compensate for H-K effect. It has not been a big issue to use CIECAM02 without considering the H-K effect because the display devices at that time had a narrow color gamut. But, these days this phenomenon can be easily observed on display devices with wide color gamut [8]. Thus, an amendment of CIECAM02 was proposed by M. Kim et al. to compensate for the H-K effect [13].
Most of the displays used in mobile phones, monitors, and televisions have glare surface recently. That means people see their reflected face when they watch the displays. Furthermore, a new display device called a mirror display has emerged. Mirror displays are used for a special purpose in which reflected images intentionally overlap with displayed images. Thus, the effects of reflected images should be considered simultaneously along with displayed images to properly analyze the color appearance of display devices. There has been attempt to figure out the color appearance phenomenon In this paper, we show psychophysical experiments to investigate if CIECAM02 can properly predict color appearance of displays with the specular reflection. In addition, we propose a new appropriate method to include the specular reflection in CIECAM02.
2. Experimental setup
2.1 Hardware setup
A 15-in. active-matrix LCD (AMLCD) monitor panel was used for this study, and its specifications are listed in Table 1. To make a mirror display, we attached a mirror film which reflects some of the ambient light and transmits the light from the monitor. Figure 1 shows the photos of the AMLCD panel with and without the mirror film.
Table 1. Specifications of the display panel used in the experiment.
Characteristics of the mirror film used in our experiment was investigated by spectrophotometer (Cary 5000, Agilent) in the wavelength range of 300nm to 800nm. The transmittance is calculated by measuring the intensity of light before and after passing through the sample. However, the principle and the required accessories to measure specular and diffuse reflectance are completely different. Results as shown in Fig. 2(a) and (b) were measured independently. The sum of transmittance and total reflectance (diffuse reflectance + specular reflectance) shown in Fig. 2 does not seem equal to 100%. We guess rest of light was absorbed in the mirror film. Figures 2(a) and 2(b) show the reflectance and transmittance of the mirror film, respectively.
Color gamut of the AMLCD panel with and without the mirror film was measured by a Minolta CS-1000 tele-spectroradiometer in a dark room. The measurement results are listed in Table 2 and plotted in x-y chromaticity diagram (CIE 1931) as shown in Fig. 3. The mirror film used in our experiment blocks some of the light emitted from the display and transmits the others. And it has non-uniform transmittance in the visible light range. Thus, it is reasonable to contain the luminance information. When comparing the gamut of the display, tri-stimulus value XYZ should be normalized relative to the luminance of white. And this makes the white luminance 100. Relative luminance with a maximum value of 100 (Y100) also listed in Table 2.
Table 2. Color gamut of the AMLCD panel with and without mirror film compared to the sRGB standard.
2.2 Setup for psychophysical experiment
Figure 4 shows a schematic diagram of our experimental setup. A viewing cabinet was used in the psychophysical experiment to maintain a consistent experimental environment. The experimental setup employed D65, a standard light source for various illuminance conditions. Also, we used a chin rest to fix the observer’s eye location as shown in Fig. 4. 2. The position of observer, display, and the viewing cabinet was carefully arranged so that direct lighting toward the eyes of the observer can be avoided.
In a dark room, we measured the luminance of display white (LDW) and surround white (LSW) to set up surround conditions defined in CIECAM02 [3]. A white reference material that provides accurate and consistent reflected white color (Δx and Δy < 0.001) under D65 was used to measure surround white. Also, we selected 28 colors for the psychophysical experiment. Tri-stimulus values of these were measured under three surround conditions. Classifications of the three conditions are listed in Table 3 and the tri-stimulus results are listed in Table 4.
Table 3. Classifications of the three surround conditions defined by CIECAM02 (Ref. [3]).
Table 4. 28 color sets used in the psychophysical experiment and their tri-stimulus values under the three surround conditions.
Table 5. Comparison of XYZ tri-stimulus values between the displays with and without mirror film measured by our proposed method.
Figure 5 shows a test pattern displayed during the psychophysical experiment. The background was black and three color patches were placed in the middle of the screen. The lower left and right patches were the reference white and the reference color, respectively. And the upper one was a test color patch. Two reference patches were fixed during the experiment and the test patch was randomly selected among the 28 colors. The size of each color patch was determined to have a 2° field of view and 40 cm viewing distance from the observer, as shown in Fig. 4.
In the psychophysical experiment, observers were asked to evaluate the lightness and colorfulness of the test colors using a magnitude estimation method. Although there are many scaling methods to evaluate stimuli, the magnitude estimation method is the simplest one if the observers are familiar with it [12–15]. In the magnitude estimation method, observers evaluate various color attributes as numerical values compared to the reference white or reference color. In this experiment, the observers evaluated the lightness of the test colors compared with the reference white. They were told that the lightness of reference white was 100 and the lightness of test patch must not exceed it. They evaluated the lightness of the test patch between 0 and 100, which corresponds to the lightness of black and the reference white respectively. Unlike lightness which is relative (lightness of reference white is always 100), colorfulness is absolute attribute. Thus, we need a colorfulness standard pad that we already know the colorfulness value under a specific standard illumination. Before starting a new session with a new surround condition, the colorfulness standard pad was given to observers under the specific standard illumination. They were told that its colorfulness was 40. They memorized its value and stimulus with their eyes closed. After the experimenter changes the illumination of viewing cabinet to one of three surround conditions (average, dim, and dark), the observers evaluated colorfulness of the reference color having (R, G, B) = (11, 185, 57) on the display screen comparing it with the memorized value. Finally, they were asked to estimate the colorfulness of test color patches by comparing them with the reference color. Thus, the reference color does not affect the lightness estimation of test color patches. But we are going to deal with only lightness in this paper. The data regarding colorfulness are not shown hereafter.
For this experiment, eight female observers were carefully selected. They were qualified as Industrial Engineer Colorists and had the certificates issued by HRDK (Human Resources Development Service of Korea). They were working at Ewha Color Design Research Institute as researchers. They were also familiar with the magnitude estimation method. Thus, we believe that they had accurate color vision and were the best candidates that we could find. For each condition, 28 colors repeated four times in random order. Thus, each observer answered 336 times (28 color sets × 4 repeats × 3 surround conditions) in total. Between the sessions, observers took a break for 10 minutes for adaptation under the next surround condition. In addition, the experiments of the three surround conditions were conducted in the order of increasing illuminance. Because light adaptation is much faster than dark adaptation, we designed the order of sessions to be dark, dim, and average surround conditions, which shortened time for the experiments.
2.3 Measurement of specular reflection
Figure 6(a) shows the measurement method. In dark room, 28 test patterns were measured by this method.
Fig. 6. (a) Schematic diagram of the conventional measurement method and (b) the image taken by the spectroradiometer.
However, measurement results show little difference between the three surround conditions as listed in Table 4. These results conflict with the characteristics of mirror display which has a highly reflective surface. It resulted from the mechanism of the measurement method. There must be an object to reflect a light source. But, with this method, we cannot obtain the mirror image of the observer but only that of the spectroradiometer as shown in Fig. 6(b). Thus, unpractical results were caused by the spectroradiometer which hardly reflects light. In this reason, we propose a new method to measure practical specular reflection.
Figure 7 describes the practical situation when the observer watches the mirror display. Figure 7(a) shows how observers see their reflected image. And Fig. 7(b) shows the actual image seen by observers.
To obtain practical results corresponding to real situation, we propose a new method. Figure 8 illustrates the top view of the proposed measurement method. In this method, the spectroradiometer does not face straight the display device. Instead, we slightly rotate it to face the reflection image of the observer. The angle of rotation is set to the minimum angle where the observer does not block the spectroradiometer. It was 12.5° in our experiment. In addition, the display device is turned off to exclude the effect of light emitted from the display. We measured XYZ tri-stimulus values of the display with and without mirror film under three surround conditions and the results are listed in Table 5. To avoid confusion, these values are named XRYRZR.
Results of comparison indicate that practical specular reflection values can be obtained by our proposed method.
3. Testing performance of CIECAM02
3.1 Observer accuracy
A coefficient of variation (CV) was calculated as given in Eq. (1) where ${x_i}$, ${y_i}$, n, $\bar{y}$, and k are the ${i_{th}}$ samples of the x and y data sets, the number of samples, the mean of the y data set, and a scaling factor obtained by the least square method, respectively. The CV value indicates quantitative agreement of two data sets. For example, a CV value of zero indicates two data sets are identical and a CV value of 15 roughly indicates 15% variation between two data sets [16–18].Observer accuracy was investigated by comparing each individual observer’s data with the mean of all the observers. This can be used to evaluate reliability of observers. CV values about observer accuracy (CV-Observer) were listed in Table 6. Note that each individual observer’s data for the ${i_{th}}$ sample correspond to xi of Eq. (1) and the mean of all the observers’ data for the ${i_{th}}$ sample corresponds to yi of Eq. (1).
Table 6. Observer accuracy.
3.2 Testing performance of CIECAM02
Although the mirror film was attached to the display, the color gamut was little different from that of the display without it as shown in Fig. 3. This color gamut is wide enough to cause the Helmholtz-Kohlrausch (H-K) effect. Thus, instead of conventional CIECAM02, we used the amended version for H-K effect which was proposed by our research team [13]. The input parameters listed in Table 7 and XYZ tri-stimulus values listed in Table 4 were used to compute the model.
Table 7. Input parameters for CIECAM02.
Figure 9 shows the results of comparing the estimated value of the psychophysical experiment with the calculated value by the amended CIECAM02 under three surround conditions.
Fig. 9. The results of comparing the estimated value of the psychophysical experiment with the calculated value by the amended CIECAM02 under (a) average, (b) dim, and (c) dark surround conditions.
CV values for model accuracy (CV-Model) were calculated for evaluating agreements between the predicted values by the amended model [13] and the ones collected in the psychophysical experiment. In this case, the former corresponds to xi of Eq. (1) and the later corresponds to yi of Eq. (1) for the ${i_{th}}$ sample. As a result, the CV values of three surround conditions (average, dim, and dark) were 20.56, 12.41, and 13.80, respectively. The results under dim and dark surround conditions can be interpreted that the model predicts the lightness very well because the experimental data were scattered very close to the line of y = x. However, the model overestimated the lightness under the average surround condition as shown in Fig. 9(a). We thought that the overestimation is caused by the specular reflection since it occurs only under a bright illumination. Thus, more investigation on specular reflection is necessary.
4. Amendment of CIECAM02 to include specular reflection
In the previous section, we proposed a practical measurement method to obtain a specular reflection. In this section, we are going to investigate two ways of including these values into the CIECAM02.
4.1 Re-amended CIECAM02 version 1
Figure 10 shows the performance of the re-amended CIECAM02 under three surround conditions. Since observers see reflected light and emitted light simultaneously, we added specular reflection data (XRYRZR) to XYZ tri-stimulus data measured by the conventional method. The CV values of three surround conditions (average, dim, and dark) were 37.48, 12.40, and 13.80, respectively. In average condition, CV value increased to 37.48 from 20.56 and the trend line has poor linearity. The performance of the average condition got worse than the previous model.
Fig. 10. The results of comparing the estimated value of the psychophysical experiment with the predicted value of the amended CIECAM02 using the sum of specular reflection and original XYZ tri-stimulus values under (a) average, (b) dim, and (c) dark surround conditions.
To investigate the cause of the poor performance, visual lightness values under the average surround were compared with the values under the dark surround, and model prediction values were also compared in the same way. As shown in Fig. 11, visual lightness values have a linear trend while model prediction values do not. Based on this, we thought that observers recognize the displayed image and the reflected image separately. If the observer perceives the image as a sum of them, like model prediction, there should be a minimum value saturated in the low lightness region. Because specular reflection is much stronger than the displayed light for the test pattern of low lightness.
Fig. 11. The results of comparison of visual lightness and model prediction values between the dark and the average surround condition.
4.2 Re-amended CIECAM02 version 2
In the previous section, we found that the observers recognize the displayed image and reflective image separately. Choi et al. concluded that people discount the reflected images in their judgment of lightness [6]. This is because the effect of the reflected image is not high enough to see an obvious tendency. On the other hand, we used a mirror display that has a highly reflective surface. As a result, in our experiment, observers tended to evaluate lightness higher under the dark surround than the average surround for the same pattern.
We supposed that it resulted from the observer’s perception of reflected image as a background. To verify supposition, background luminance value (Yb) was modified by using specular reflection data which were collected before. Equation (2) shows how to get the modified background luminance (Yb’):
We use only YR among the XRYRZR tri-stimulus values of specular reflection. We multiplied YR by YW/LW to have relative value to the reference white. Note that YW is the relative luminance of the reference white and LW is the absolute luminance of the reference white.
Figure 12 shows the performance of the CIECAM02 with modified background luminance (Re-amended CIECAM02 v2). The CV values of three surround conditions (average, dim, and dark) were 12.39, 12.11, and 13.80, respectively. The performance comparison between models was listed in Table 8. As listed in the table, the CV value under average surround decreased to 12.39 from 20.56 and the CV value under dim surround decreased to 12.11 from 12.41 which means the proposed model has better performance, while the CV remained the same under the dark surround where specular reflection does not exist. According to the results, the amended model can successfully predict the lightness even when the screen surface features the specular reflection.
Fig. 12. The results of comparing the estimated value of the psychophysical experiment with the predicted value of amended CIECAM02 v2 under (a) average, (b) dim, and (c) dark surround conditions.
Table 8. Performance comparison between color appearance models
We concluded that although a specular reflection image is in a different visual field from the real background, people perceive it as an equivalent background.
5. Discussion
There were several variables of specular reflection. For better reproducibility and generality, these variables must be controlled throughout the experiment. In this section, we are going to discuss how to control variables and how to standardize them.
The reflectance of the observer varies. Also, it is hard to standardize a specific numerical value. Although we measured specular reflection of one observer’s face as a representative, we can demonstrate the validity of this measurement. As shown in Fig. 13, we investigated the change of CV-Model value according to the luminance of specular reflectance (YR) under average surround condition. Also, YR of black paper and N5 paper (medium gray in Munsell color system) was measured and used for reference. The figure below shows that one representative data that we used in our experiment has little difference (ΔCV-Model = 0.014) compared with ideal data which makes CV-Model minimum. As you can see, CV-Model values for YR = 10 or 30 (-/+ 50% variation in YR) were smaller than 15. This means that even dramatic change in skin tone in our experiment could not cause any big trouble. Therefore, we can conclude that our proposal to include YR will work well even for mixed-race observers if they are all good colorists.
For standardization, we propose an experimental setup as shown in Fig. 14. As shown in the Fig. 14(a), a reference reflecting plate is located between the observer and the display. The plate has a uniform surface and two holes through which the observer sees the display. The role of the plate is to block various skin tones of observers and to fix the reflected light from the display to observer’s eyes. Then, we can control YR throughout the experiment. Since the observer sees the reflected light from the plate, a reflectance variation of the observer's face hardly affects the specular reflection. Furthermore, a variation of the mirror film could be compensated by changing the reflectance of the plate until the specular reflection becomes a reference value. We believe that this method could standardize our experiment.
6. Conclusion
CIECAM02 has been widely used for an international standard to analyze the color appearance. But the conventional model cannot predict the color appearance of a mirror display or mirror-like display with a specular reflection. This paper discussed the lightness considering the specular reflection which occurs in the mirror display. The magnitude estimation method was used, and its results were compared with the predicted value of the model. And the CV value was used to evaluate how identical the two data sets are. Comparing the measurement results under the three surround conditions, we figured out that the conventional measurement cannot measure practical specular reflection. Thus, a new measurement was proposed and CIECAM02 was amended to include it. We concluded that although a specular reflection image is in a different visual field from the real background, people perceive it as an equivalent background. These days, display devices such as automotive and mobile phone displays are used not only indoors but also outdoors. There is always a reflected image outdoors because of the ambient sunlight. Therefore, we believe that our findings can be applied not only to a mirror display but also to the displays with glare surface.
Funding
National Research Foundation of Korea (2016R1D1A1B03935421); Ministry of Education (21A20130000018).
Disclosures
The authors declare that there are no conflicts of interest related to this article.
References
1. Y. Nayatani, H. Sobagaki, K. Hashimoto, and T. Yano, “Lightness dependency of chroma scales of a nonlinear color-appearance model and its latest formulation,” Color Res. Appl. 20(3), 156–167 (1995). [CrossRef]
2. R. W. G. Hunt, “A model of colour vision for predicting colour appearance,” Color Res. Appl. 7(2), 95–112 (1982). [CrossRef]
3. . International Commission on Illumination Technical Committee TC 8-01 Technical Report (Draft 11), A color appearance model for color management systems: CIECAM02, 2003.
4. International Commission on Illumination, “A colour appearance model for colour management systems: CIECAM02,” CIE Publ., vol. 159 (2004).
5. C. Li and M. R. Luo, “Testing the robustness of CIECAM02,” Color Res. Appl. 30(2), 99–106 (2005). [CrossRef]
6. S. Y. Choi, M. R. Luo, M. R. Pointer, C. Li, and P. A. Rhodes, “Changes in colour appearance of a large display in various surround ambient conditions,” Color Res. Appl. 35(3), 200–212 (2010). [CrossRef]
7. K. Xiao, M. R. Luo, and C. Li, “Colour size effect modelling,” Color Res. Appl. 37(1), 4–12 (2012). [CrossRef]
8. S. Oh and Y. Kwak, “Investigation of the Helmholtz-Kohlrausch effect using wide-gamut display,” Color Imaging: Displaying, Processing, Hardcopy, and Applications, 2015, 93950X (2015).
9. R. L. Donofrio, “Review paper: the Helmholtz-Kohlrausch effect,” J. Soc. Inf. Disp. 19(10), 658–664 (2011). [CrossRef]
10. M. Withouck, K. A. G. Smet, W. R. Ryckaert, M. R. Pointer, G. Deconinck, J. Koenderink, and P. Hanselaer, “Brightness perception of unrelated self-luminous colors,” J. Opt. Soc. Am. A 30(6), 1248–1255 (2013). [CrossRef]
11. Y. Nayatani, “Simple estimation methods for the Helmholtz—Kohlrausch effect,” Color Res. Appl. 22(6), 385–401 (1997). [CrossRef]
12. M. Withouck, K. A. G. Smet, W. R. Ryckaert, G. Deconinck, and P. Hanselaer, “Predicting the brightness of unrelated self-luminous stimuli,” Opt. Express 22(13), 16298–16309 (2014). [CrossRef]
13. M. Kim, J. Jo, Y. Park, and S. Lee, “Amendment of CIECAM02 with a technical extension to compensate Helmholtz-Kohlrausch effect for chromatic characterization of display devices,” Displays 56, 1–10 (2019). [CrossRef]
14. M. D. Fairchild, “Color Appearance Models,” Wiley & Sons, Ltd (2013).
15. M. Withouck, K. A. G. Smet, W. R. Ryckaert, and P. Hanselaer, “Experimental driven modelling of the color appearance of unrelated self-luminous stimuli: CAM15u,” Opt. Express 23(9), 12045–12064 (2015). [CrossRef]
16. B. Koo and Y. Kwak, “Color appearance and color connotation models for unrelated colors,” Color Res. Appl. 40(1), 40–49 (2015). [CrossRef]
17. M. R. Luo and C. Li, “CIECAM02 and its recent developments, Advanced Color Image Processing and Analysis,” Springer, New York, 19–58 (2013).
18. C. Fu, C. Li, G. Cui, M. R. Luo, R. W. G. Hunt, and M. R. Pointer, “An investigation of colour appearance for unrelated colours under photopic and mesopic vision,” Color Res. Appl. 37(4), 238–254 (2012). [CrossRef]
