Applying CIECAM2 for Mobile Display Viewing Conditions YungKyung Park*, ChangJun Li*, M.. Luo*, Youngshin Kwak**, Du-Sik Park **, and Changyeong Kim**; * University of Leeds, Colour Imaging Lab, UK*, ** Samsung Advaned Institute of Tehnology, Yongin, South Korea** Abstrat Small displays are widely used for mobile phones, PDA and Portable DVD players. They are small to be arried around and viewed under various surround onditions. An experiment was arried out to aumulate olour appearane data on a 2 inh mobile phone display, a 4 inh PDA display and a 7 inh LCD display using the magnitude estimation method. It was divided into 12 experimental phases aording to four surround onditions (dark, dim, average, and bright). The visual results in terms of lightness, olourfulness, brightness and hue from different phases were used to test and refine the CIE olour appearane model, CIECAM2 [1]. The refined model is based on ontinuous funtions to alulate different surround parameters for mobile displays. There was a large improvement of the model performane, espeially for bright surround ondition. Introdution Many previous olour appearane studies were arried out using household TV or PC displays viewed under rather restrited viewing onditions. In pratie, the olour appearane of mobile displays is affeted by a variety of viewing onditions. First of all, the display size is muh smaller than the other displays as it is built to be arried around easily. Seondly, the portability allows the display to be viewed under surround onditions varying from suh as dark night to bright sunlight, and indoor and outdoor onditions. In many ases, large amount of flare are inluded in the displayed images. The aim of this study is to model the hange of olour appearane on mobile displays under a wide range of viewing onditions. Three different sizes of small mobile displays were used: a 2 mobile phone display, a 4 PDA and a 7 LCD display. A haraterisation model was derived to transform between the tristimulus values and the monitor s GB values for all the 3 displays studied. Twelve experimental phases were onduted under 4 surround onditions (dark, dim, average, and bright) with the luminane levels ranged from to 55 d/m 2. Eah test olour was estimated by 5 to 1 observers in terms of lightness, olourfulness, brightness and hue appearane attributes. The brightness and olourfulness are onsidered to be important to represent the absolute appearane under different viewing onditions. The visual results were used to evaluate the CIE olour appearane model, CIECAM2 [1]. The model was then modified speifially for mobile display viewing onditions. Experimental Setup A Minolta CS1 spetroradiometer (TS) was used to measure all the olours in the experiment with a /45 geometry and viewing distane of 3 m. Figure 1 shows the olour gamuts of the 2 (dashed line), 4 (solid line) and 7 (double dashed line) displays. It an be seen that the gamut of the 4 transrefletive display is muh smaller than that of the 2 LCD display. v'.7.6.5.4.3.2.1.1.2.3.4.5.6.7 Figure 1. The olour gamut of the three displays studied. Figure 2. The pattern used in the experiment. u' The viewing and illumination geometry was 45/, whih is typial in viewing portable displays and the viewing distane was 3 m. The test olour to be judged was displayed in the middle of display together with the white referene and the olourfulness referene. The former is the display peak white having a lightness value of 1. The referene olourfulness is always referred to a partiular physial sample viewed in a viewing abinet. It is memorised by eah observer. Twenty deorative olours on the edge of display were used to form a omplex image. The physial size of the three middle olours was 1 viewing field for all displays. Table 1 summarises the experimental onditions used in the twelve experimental phases aording to four different surrounds (dark, dim, average, and bright) and 3 sizes of displays. Different olours were seleted in different phases to give a reasonable overage in CIELAB spae. Different neutral grey bakground olours were used in eah phase to learly show the olours on the monitor with right ontrast. The surround luminane was defined by measuring a referene white beside the display using the TS 15th Color Imaging Conferene Final Program and Proeedings 169
under the viewing onditions for eah phase. The results from the 2 display under dark, dim and average were reported earlier [2]. Table 1. Summary of the viewing onditions of different experimental phases. Display Surround No of Colour Ambient Lighting Luminane (d/m 2 ) Dark 4 2 Dim 4 5 Average 4 1 Bright 4 55 Dark 3 4 Dim 3 5 Average 3 1 Bright 3 55 Dark 2 7 Dim 2 5 Average 2 1 Bright 2 55 For the bright surround, the display was plaed inside a viewing booth equipped with a strong spot light. Ten normal olour vision observers, aording to the Ishihara test, partiipated all experimental phases exept bright surround pahses, for whih the 5 best observers from the previous 1 were used. All observers were familiar with the magnitude estimation method. There were 3 male and 7 female observers. Eah was asked to estimate test olours in terms of brightness, lightness, olourfulness and hue losely following the method used by Luo et al [2]. Lightness was saled against the referene white having a lightness of 1 and an imaginary blak,. An anhor path that was assigned a olourfulness of 4 and brightness of 1 was first shown in a viewing booth. Eah observer had to judge the olourfulness of the referene path in the beginning of eah phase regarding to the anhor path. The brightness was judged aording to the anhor path having a value of 1 with an open end. The hue was judged by reporting the perentage of the two olours from the four psyhologial hues (red, yellow, green, and blue). Observer variations Observer variations in terms of observer auray were examined. The former was ompared between eah observer s repeated judgments. The auray was ompared between eah individual observer and mean visual results. The measure of Coeffiient of Variation (CV) given in equation (1) was used to indiate the disagreement between two sets of data. It is a measure of the distane of the points from the 45 line in the y diretion. The more the points are sattered about the line, the poorer the agreement. A perfet agreement, CV should be zero and larger the value, the poorer the agreement. ( yi k xi ) / n CV 1 (1) y where x i and y i are the i sample in the x and y data sets; n is the number of samples; y is the mean of the y data set and k is a saling fator, whih is obtained by the least square method. For alulating observer auray, the saling fator (k) in equation 1 was set to one. The mean CV values of the 12 phases were 19, 28, 31 and 9 units for the lightness, olourfulness, brightness and hue, respetively. The auray results were slightly worse than those found in Luo et al s study [3] due to the property of the viewing onditions involved, i.e. small sreen and Table 2. The performane of CIECAM2 in terms of CV Colour attributes Surround 2 inh 4 inh 7 inh Mean Total mean Observer auray Lightness Dark 38 13 29 27 Dim 22 13 24 2 Average 27 12 24 21 Bright 65 55 46 55 31 19 Colourfulness Dark 31 39 26 32 Dim 3 29 28 29 Average 37 27 32 32 Bright 48 49 31 43 34 28 Brightness Dark 22 2 19 2 Dim 8 25 21 18 Average 17 25 22 21 Bright 27 43 51 4 25 31 Hue Dark 14 12 8 11 Dim 13 11 6 1 Average 14 9 7 1 Bright 13 9 7 1 1 9 17 Copyright 27 Soiety for Imaging Siene and Tehnology
large variation of surround onditions. esults and Disussions Testing CIECAM2 The CIECAM2 model is the olour appearane model reommended by CIE. The model s performane is evaluated using the present experimental data in terms of CV alulated between the model s predited and visual results. The CV values are summarised in Table 2. Table 2 shows that the model gave a reasonable predition to the brightness and hue visual results, but performed badly for prediting lightness and olourfulness visual results, i.e. the total mean values are 31, 34, 25 and 1 omparing with 19, 28, 31 and 9 of observer auray results for lightness, olourfulness, brightness and hue respetively. Figures 3 to 5 are the plots of lightness, olourfulness and brightness visual results against the orresponding CIECAM2 preditions, respetively. Eah figure inludes four diagrams whih orrespond to dark, dim, average and bright surround onditions from left to right respetively. Note that the hue results are not plotted here beause the model predits well to the hue visual results, i.e. most of the data points are lose to the 45 o line. Figure 3 shows that CIECAM2 lightness is predited well to the visual results under dim surround onditions but is overestimated those of the other surrounds and espeially bright surround onditions. Figure 4 show that CIECAM2 predits the olourfulness reasonably well for all surrounds exept average surround onditions. Figure 5 shows that the brightness predited poorly by the CIECAM2 under all surround onditions, espeially in bright and average surround. Overall, CIECAM2 gave the poorest predition to all the visual results under bright surround onditions. efinement of CIECAM2 In the last setion, it was found that the performane of CIECAM2 is somewhat dissatisfatory, i.e. the results in terms of CV in prediting urrent results are muh worse than those in prediting LUTCHI data [2]. Hene, various trials were made to improve the model s performane for prediting visual data. The general strategy was to modify CIECAM2 model as little as possible. The modifiation was made only for the surround parameters under various viewing onditions:, F and N. Lightness Dark-grey Lightness Dim-grey Lightness Average-grey Lightness Bright-grey 1 1 1 1 8 8 8 8 CIECAM2 6 4 CIECAM2 6 4 CIECAM2 6 4 CIECAM2 6 4 2 2 2 2 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 Figure 3. CIECAM2 predited lightness results plotted against visual results for four 4 surround onditions (from left to right: dark, dim, average, and bright). The symbols of square, ross and irle represent the data for 7, 4 and 2 displays, respetively. Colourfulness Dark-grey Colourfulness Dim-grey Colourfulness Average-grey Colourfulness Bright-grey 1 1 1 1 8 8 8 8 CIECAM2 6 4 CIECAM2 6 4 CIECAM2 6 4 CIECAM2 6 4 2 2 2 2 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 Figure 4. The same as Figure 3 exept that the olourfulness results are plotted. Brightness Dark-grey Brightness Dim-grey Brightness Average-grey Brightness Bright-grey 3 3 3 3 25 25 25 25 2 2 2 2 CIECAM2 15 CIECAM2 15 CIECAM2 15 CIECAM2 15 1 1 1 1 5 5 5 5 5 1 15 2 25 3 5 1 15 2 25 3 5 1 15 2 25 3 5 1 15 2 25 3 Figure 5. The same as Figure 3 exept that the brightness results are plotted. 15th Color Imaging Conferene Final Program and Proeedings 171
F N 1.6 1.2.8.4. 1 2 3 S F 1.4 1.2 1.8.6.4.2 1 2 3 S N 3 2.5 2 1.5 1.5 1 2 3 S Figure 6. The optimized, F and N values are plotted against the S. The best fitted lines are also plotted. The symbols of square, ross and irle represent the data for 7, 4 and 2 displays, respetively Table 3. Surround parameters in CIECAM2 Surround C F N Average,69 1, 1, Dim,59,9,9 Dark,525,8,8 The main viewing parameter studied in the present experiment is surround. Table 3 gives the three surround parameters defined in CIECAM2. They are divided into three surrounds: average, dim and dark. These are determined by S (Surround ratio), in equation (2). LSW S (2) LDW Where L the luminane of the surround is white and SW L DW is the luminane of the devie white. If S is then a dark surround is appropriate. If S is less than.2 then a dim surround should be used while an S of greater than or equal to.2 orresponds to an average surround. Table 4. The S values for eah display under eah surround ondition. Surround 2 inh 4 inh 7 inh display display display Dark Dim.3.3.6 Average 6 7 13 Bright 22 25 25 CIECAM2 defines only three surrounds whih are insuffiient for the use of portable displays. For this reason, it was deided to develop a ontinuous funtion to define surround parameter for eah phase of viewing onditions. Firstly, S values for the 12 phases were alulated as given in Table 4. It an be seen that the values vary from to 38. The CV values alulated between the visual and the CIECAM2 preditions were Table 5. CV values for testing the refined CIECAM2 Colour attributes Surround 2 inh 4 inh 7 inh Mean Lightness Dark 28 22 18 23 Dim 28 13 17 19 Average 22 26 17 22 Total mean Observer auray Bright 32 22 17 24 22 19 Colourfulness Dark 29 33 28 3 Dim 27 23 28 26 Average 4 26 39 35 Bright 46 4 37 41 33 28 Brightness Dark 11 15 15 14 Dim 16 36 16 23 Average 9 17 24 17 Bright 12 33 14 2 18 31 Hue Dark 14 12 8 11 Dim 13 11 6 1 Average 14 9 7 1 Bright 13 9 7 1 1 9 172 Copyright 27 Soiety for Imaging Siene and Tehnology
Lightness Dark-grey Lightness Dim-grey Lightness Average-grey Lightness Bright-grey 1 1 1 1 8 8 8 8 6 4 6 4 6 4 6 4 2 2 2 2 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 Figure 7. The same as Figure 3 exept that the refined CIECAM2 lightness preditions are plotted. Colourfulness Dark-Grey Colourfulness Dim-Grey Colourfulness Average-Grey Colourfulness Bright-Grey 1 1 1 1 8 8 8 8 6 4 6 4 6 4 6 4 2 2 2 2 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 2 4 6 8 1 Figure 8. The same as Figure 4 exept that the refined CIECAM2 olourfulness preditions are plotted. Brightness Dark-grey Brightness Dim-grey Brightness Average-grey Brightness Bright-grey 3 3 3 3 25 25 25 25 2 2 2 2 15 15 15 15 1 1 1 1 5 5 5 5 5 1 15 2 25 3 5 1 15 2 25 3 5 1 15 2 25 3 5 1 15 2 25 3 Figure 9. The same as Figure 5 exept that the refined CIECAM2 brightness preditions are plotted. minimised by optimising the, F and N variables. This was performed for all 12 phases. Eah set of the optimised, F and N values were then plotted against the S in Figure 6. It an be seen strong linear relationships between the S and surround parameters. A line was fitted to eah of the three surround parameters as defined by equations 3 to 5 respetively. '.23S +.7887 (3) F '.3S + 1.1474 (4) N '.23S + 1.2369 (5) The new equations are named, F and N respetively. These were used to replae the fixed parameters in Table 3 to be used by CIECAM2. Its preditions and visual results are plotted in Figures 7 to 9 for lightness, olourfulness and brightness results respetively. Also the CV values are summarised in Table 5. Table 5 shows that there is a good predition of the refined CIECAM2. It an be seen learly in Figures 7 to 9 that the largest improvement an be found in brightness, followed by lightness and olourfulness the smallest. The improvement is large marked for bright surround onditions. The findings an also be shown in Table 5, for whih the refined CIECAM2 was tested in terms of CV unit using the present data. In onlusion, there are marked improvements from the original CIECAM2 i.e. from CV values of 31, 34, 25, 1 to 22, 33, 18, 1 for lightness, olourfulness, brightness and hue, respetively. It performs equal to or better than the typial observer auray from the panel of observers. Conlusion In this study, a refined version of CIECAM2 was developed for mobile displays viewed under different surround onditions. A set of equations based on S was derived to be able to aurately define surround parameters. This greatly improves the performane of CIECAM2 in prediting the visual results. eferene [1] CIE Publiation 159-24: A olour appearane model for olour management systems: CIECAM2 (24). [2] YungKyung Park ChangJun Li, M.. Luo Wonhee Choe, Seongdeok Lee and Changyeong Kim, Colour Appearane Modelling for Mobile Displays, CIC 14th, sottsdale, arizona (26) [3] M.. Luo, A.A Clarke, P.A. hodes, S. A.. Srivener, A. Shappor, C. J. Tait, Quantifying Colour appearane. Part1. LUTCHI data, Colour es. Appl., Vol 16, pp 166-18 (1991). Author Biography YungKyung Park reeived her BS and Masters in physis from the Ewha Womans University (Korea) and her Master in olour siene from Derby University (UK). Sine then she is a PhD student in Leeds University (UK). Her researh has been foused on Colour Appearane modelling for Mobile Displays. 15th Color Imaging Conferene Final Program and Proeedings 173