CSD—A New Unified Threshold Metric
of Evaluating LCD Viewing Angle
by Color Saturation Degradation
Szu-Fen F. Chen, Wei-Chung W. Cheng, and Han-Ping D. Shieh
Abstract—Lower luminance contrast ratio and chromatic
changes affect the visual performance (i.e. color shift) of a thin-film transistor liquid-crystal device (TFT-LCD) at large viewing angles. The de facto method of defining viewing angle, contrast ratio of luminance, fails to represent the substantial visual performance viewed at a larger angle. We found the degra-dation of color saturation, , to be an appropriate metric to aid the conventional viewing angle definition ( 10). We empirically determined the threshold for defining the color
viewing angles of TFT-LCDs, = 0 03 ,
which reflects the variation not only in chromaticity but also in luminance. The proposed metric was evaluated by psychophysical experiments, whose results validate the efficacy of the proposed metric.
Index Terms—Color saturation degradation (CSD) metric ,
con-trast ratio metric, just-noticeable difference (JND), liquid-crystal device (LCD) viewing angle evaluation, psychophysical evaluation.
I. INTRODUCTION
A
S THE RAMPING liquid-crystal device TV (LCD-TV) market rapidly replacing the existing CRT TV sets, color performance has become one of the key factors for potential buyers to make their move. Currently, the major challenge of LCD-TV is the visual quality degradation at large angles such as color shift and contrast decrease. Unfortunately, there is no prac-tical and representative metric to describe such quality degrada-tion for the end users. The focus of this paper is to find an ideal metric for evaluating color viewing angle of thin-film transistors LCDs (TFT-LCDs).In the TFT-LCD industry, the conventional metric, which is used to evaluate the viewing angle of LCDs, is luminance con-trast ratio (CR), but it is not sufficient to represent the visual performance at large angles, especially color shift. However, the metric (i.e., CR) is simple and easy for judging the LCD’s viewing angle performance. In practical applications, we still have not found a new metric with unified threshold to evaluate the color viewing angle as CR did. In academia, the most of academics emphasize to use perfect models for color viewing angle evaluation [1]–[4]. Some metrics were even created by
Manuscript received January 9, 2006; revised February 16, 2006.
S.-F. F. Chen is with the Department of Photonics and Display Institute, Na-tional Chiao Tung University, Hsinchu 30010, Taiwan, R.O.C., and also with R&D Design Division, Central Research Institute, Chung-Hwa Picture Tubes, Ltd., Taoyuan 33444, Taiwan, R.O.C. (e-mail: fenny@mail.cptt.com.tw).
W.-C. W. Cheng and H.-P. D. Shieh are with the Department of Photonics and Display Institute, National Chiao Tung University, Hsinchu 30010, Taiwan, R.O.C.
Digital Object Identifier 10.1109/JDT.2006.874504
Fig. 1. Visual effects of color shift and contrast degradation at large viewing angles. (Color version available online at http://ieeexplore.ieee.org.)
using 3-D models, although, they included both luminance vari-ance and chromatic changes simultaneously. These metrics are not convenient to use in TFT-LCD industry. The new unified threshold metric is necessary to enhance the deficiency of CR for evaluating TFT-LCDs.
The variation in CR with viewing angle is a well known phenomenon in TFT-LCDs, adversely affect image, viewing angle, color shift, etc. For color TFT-LCDs, the shift in both chromaticity and luminance can be dramatic with changes in viewing angle. An example is shown as Fig. 1. Conventionally, the viewing angle of a TFT-LCD is determined by a threshold of the contrast ratio of luminance such as
(1) which indicates the range of viewing angles that the luminance
of white is at least a factor of 10
higher than the luminance of black . This popular metric, however, is not representative of the color per-formance of a TFT-LCD, because it does not capture the chro-matic changes. In practice, when a TFT-LCD is examined at a larger angle, the color shift caused by the retardation value variance of the liquid crystals is more pronounced than the lu-minance contrast ratio.
We propose a new unified threshold metric, which is par-ticularly suitable for evaluating the visual performance of
Fig. 2. Viewing angle definition. (Color version available online at http://ieeexplore.ieee.org.)
TFT-LCDs at large viewing angles. Our approach is using the degradation of color saturation to determine the acceptable view angles based on a just-noticeable difference (JND) threshold [6], [7]. We fabricated and thoroughly characterized three TFT-LCDs for analyses, which have very similar structures but perform noticeably differently at larger viewing angles, to verify our proposal. Based on the measured colorimetric data, we derived a numerical metric for viewing angles. We also discuss the other metric, , defined by color difference for viewing angles [3], [4], [6], [7]. To evaluate the proposed metric, we used the other three commercial TFT-LCD monitors for experiments, which have completely different designs and characteristics, and conducted psychophysical experiments to collect subjective preferences. The experimental results show that the proposed metric is a mathematically well-defined, rep-resentative, reproducible, and objective metric for evaluating the view angles of TFT-LCDs.
II. METHODOLOGY
The methodology of this study includes the following steps: A) three types of TFT-LCDs were fabricated as test panels; B) the test panels were characterized by using de facto test methods; C) the measurement data were analyzed to identify the shortcoming of conventional metrics; and D) a new metric was proposed.
A. Design of Test Panels
We designed and fabricated three test panels, whose common specification is listed in Table I. For fair comparison, all test panels possess similar optical and electrical properties (e.g. same drivers, same color filters, and same backlight modules), except for their wide-viewing-angle designs. The first panel, , is a typical twisted nematic type TFT-LCD panel with
super-wide viewing angle films. The rest two are optical com-pensation bend (OCB) type. One, OCB-90, is OCB-type rubbed in vertical direction. The other, OCB-45, is also OCB-type but rubbed at 45 deg. Despite of their different rubbing directions, the two OCB-type panels have very similar visual performance, which are far superior to the TN panel. This predetermined difference in visual performance will be used to evaluate the efficacy of viewing angle metrics.
B. Photometric and Colorimetric Parameters of Test Panels We chose the most adopted Swedish Confederation of Profes-sional Employees (TCO) certification [8], [9] for Visual Display Units (VDUs) to characterize the test panels. The TCO certifica-tion originated from the end-users’ perspective rather than that of the manufacturers. We used the test methods in TCO’03 to measure the optical properties of the three test panels [9]. The standard viewing angle definition is shown in Fig. 2.
By definition, lightness is
(2) Chroma is defined by
(3) Color Saturation is defined as the ratio of chroma to light-ness
(4) In these equations, and ( , ) denote the measured luminance and the CIEXYZ chromaticity coordinates of the target color,
Fig. 3. The iso-contrast contour plots of three 7" TFT-LCDs. (a)T N (b) OCB 0 90, and (c) OCB 0 45. (Color version available online at http://ieeexplore. ieee.org.)
TABLE II
VIEWINGANGLESDEFINED BYLUMINANCECONTRASTRATIO
respectively. ( , , ) denotes the same parameters of the reference white. The reference white was chosen to be
measured at normal angle near the center of the panel.
We measured the luminance, contrast ratio, viewing angles and color tri-stimuli ( , , ) every 10 degree. The , , and denote the color tri-stimuli defined in the CIEXYZ system [9], [11], [12]. We converted ( , , ) in CIEXYZ to ( , , , , ) in the CIELUV color space [1], [11], [12] and calculated the colorimetric parameters including lightness, chroma, color difference, saturation, etc.
The conventional viewing angles determined by the lumi-nance contrast ratio are listed in Table II as well as their iso-contrast contour plots in Fig. 3. denotes the mea-sured luminance at viewing angle . For example, indicates normal viewing direction while indicates hori-zontal and vertical viewing direction at 30 . From Table II, it is difficult to tell the visual difference between the and panels by comparing the horizontal (9 o’clock–3 o’clock) and vertical (12 o’clock –6 o’clock) viewing angles, because the lu-minance is angular-dependent. In other words, the -value of white-state decreases as the viewing angle increases. In con-trast, the -value of black-state becomes higher and higher. Consequently, the contrast ratio becomes lower and the color is shifted at large viewing angles as revealed by the iso-contrast contour plots in Fig. 3. These plots convey only the viewing angles defined by contrast ratio in the whole azimuth, but not the visual difference caused by color shift. To sum up, the data
may mislead the substantial visual differences between the three panels.
C. Prior Methods for Evaluating Color Viewing Angle In order to evaluate the color shift, two metrics were proposed in [1], [2]. The first metric used a 3-D model including lumi-nance variance and chromatic change to evaluate the color shift [1]. The metric represents color shift by using a figure of 3-D CIE1931XYZ chromaticity diagram with respect to viewing an-gles. The second one proposed a metric using the iso-luminance concept to calculate the color difference, defined by CIELAB and CIE1976UCS [6], [11], [12], between two target colors [2]. Both metrics are not suitable for our purpose, but they induce us to evaluate color viewing angle by using color difference.
In TFT-LCD industrial applications, some companies ini-tially attempt to use to define the unified threshold metric for evaluating color viewing angles, because the color difference is a well-known parameter in colorimetry to describe the variance in luminance and chromaticity simultane-ously [3], [4]. First, we used it to characterize the three 7-inch TFT-LCDs.
(5) It is necessary to consider luminance and chromaticity simul-taneously at large viewing angles [1], [2], [5]. After analyzing the data shown in Fig. 4, we found that we could not distin-guish the significant difference between the three panels. The luminance variance was too large and misled the color differ-ence calculation results. In other words, the color shift effect was overshadowed by color difference calculation. In Fig. (4a) and (b), the color differences are almost the same when the viewing angle is smaller than 30 deg and 40 deg, respectively. The is about 40 in horizontal direction (9–3 o’clock direction), and is about 60 in vertical direction (12–6 o’clock direction). Even the threshold of the metric is difficult to be decided, especially, when we evaluate a TFT-LCD with different luminance levels. For instance, we compare two TFT-LCD monitors with different luminance levels (e.g. one is 400 cd m of luminance and the other is 200 cd m ). The between different viewing an-gles of the two LCDs are quite different and result in difficulty for deciding a general threshold of color viewing angle.
Fig. 4. Color difference(1E ) vs. viewing angle in white-state (RGB = h255 255 255i). (a) In horizontal—9–3 o’clock, and (b) in vertical—12–6 o’clock. (Color version available online at http://ieeexplore.ieee.org.)
Fig. 5. Measured color saturation vs. viewing angle ofT N. (a) Color saturation (S ) versus viewing angles, and (b) radar diagram of S versus viewing angles. Color viewing angle range: 9–3 o’clock:040 40 (blue dominates); 12–6 o’clock: 050 50 (blue dominates). (Color version available online at http:// ieeexplore.ieee.org.)
The metric, yet, is not suitable to be a unified stan-dard criterion, as luminance contrast ratio , to define viewing angle. The criterion perhaps needs to be changed when evaluating different type TFT-LCDs (e.g. different applications: monitor, notebook or TV). Consequently, to improve the over-shadowing effect of in the metric is necessary.
D. Proposed Unified Threshold Metric: CSD
Considering the chromatic changes and luminance degra-dation effects at large viewings, we propose a new unified threshold metric for defining color viewing angles. The pro-posed metric employs color saturation degradation in the red, green, and blue channel.
The color saturations of red, green, and blue sub-pixels versus viewing angle of the three panels are shown in Figs. 5, 6 and 7, respectively. In these three panels, the blue and red sub-pixels start to degrade earlier than the green sub-pixels as the viewing angle increases. By setting a threshold on the color saturation degradation, we can find the acceptable viewing angle range. The threshold is defined as the slope of the curve (i.e., derivative
of color saturation as a function of viewing angle), and the Color Saturation Degradation (CSD) is defined by
(6) The worst degradation among the three channels determines the Color Viewing Angle (CVA), which is defined as the range of acceptable viewing angles.
(7a) (7b) According to the empirical JND defined by CIE1976UCS and ISO-endorsed 0.004 [2], [6], [7], the color difference is
Fig. 6. Measured color saturation vs. viewing angle ofOCB 0 90. (a) Color saturation (S ) vs. viewing angles, and (b) radar diagram of S vs. viewing angles. Color viewing angle range: 9–3 o’clock :080 70 (blue dominates); 12–6 o’clock: 030 30 (red dominates). (Color version available online at http://ieeexplore.ieee.org.)
Fig. 7. Measured color saturation vs. viewing angle ofOCB 0 45. (a) Color saturation (S ) vs. viewing angles, and (b) radar diagram of S vs. viewing angles. Color viewing angle range: 9–3 o’clock:070 70 (blue dominates); 12–6 o’clock: 070 70 (blue dominates). (Color version available online at http://ieeexplore.ieee.org.)
We chose the threshold as
(9) The threshold was chosen based on the empirical JND mentioned above. We converted the criteria of CSD into JND, and found that the color difference between the target color and reference white was about 5 JND. In practice, the human eyes can not distinguish the color difference which is smaller than 5 JND. We use the threshold to define the color viewing angle instead of the conventional viewing angles defi-nition determined by luminance contrast ratio .
From the above-mentioned criteria, we can derive the color viewing angle ranges of the three panels. The color viewing
angle ranges of derived from Fig. 5 are in 9–3 o’clock direction and in 12–6 o’clock direction. The viewing angle ranges are narrower than the conventional ones defined by shown in Table II. The blue channel dominates the color viewing angle at both directions in the panel. The results of are also shown in Fig. 6. The
color viewing angle range of is and
dominated by the blue channel in 9–3 o’clock direction. In 12–6 o’clock direction, the range is and dominated by the red channel. We observed that the viewing angle range of is narrower than that of in 12–6 o’clock di-rection, which is opposite to the results listed in Table II. How-ever, the color viewing angle ranges of are almost the largest in the three panels in both directions. The ranges are
users can perceive the color viewing angle range visually by the shape (roundish or not) of the radar-diagram. Considering the above mentioned with the gamut value (NTSC ratio), we can recognize the visual performance of different LCD modes with similar optical specification easily (e.g. compare various moni-tors). In the meantime, we can identify which color dominates the viewing angle range by using the radar-diagram as well.
To compare the viewing angle range defined by
shown in Fig. 5 that defined by in Table II, the sub-stantial visual differences were investigated. Using the conven-tional viewing angle to evaluate the three panels, their viewing angles are larger than 85 degree in horizontal. The conventional metric lacks the visual differences. The proposed CSD metric can recognize that the type is superior to the type in terms of color viewing angle. Moreover, the vertical color visual quality of is worse than , which is opposite to the results in Table II.
III. PSYCHOPHYSICALEVALUATION
To objectively evaluate the performance of the CSD metric, we applied it to another three TFT-LCDs and conducted psy-chophysical experiments [10]–[13]. The new metric, justified by psychophysical experiments, is workable and superior to prior metrics in distinguishing the visual difference between TFT-LCDs at large viewing angles.
A. Methodology
We prepared three commercial TFT-LCD monitors, which have different designs for wide viewing angles and vary sig-nificantly in color performance. Without revealing information of the monitors, we asked the subjects to judge the color per-formance subjectively. The subjects’ task was to rank the three monitors in order. Then we applied the signal detection theory to analyze the experimental data and obtained a quantitative es-timate of color performance of the three monitors, which was compared with the results from the CSD metric.
B. Experiments
The three commercial TFT-LCD monitors were 19 twisted nematic (TN)-type, multi-domain vertical alignment (MVA)-type, and patterned vertical alignment (PVA)-(MVA)-type, which have completely different designs and characteristics. The color per-formance of MVA- and PVA-type, however, is assumed to be su-perior to that of TN-type. The monitors were positioned side-by-side at an angle of 30 to the subject, who was 2 m away from the monitors. A standard test pattern consisting of color stripes
Fig. 8. Color performances as probability distributions.
was used as the target. The 38 subjects were Asian in the age of 22 to 28. Most of them had normal vision after lens correction. The subjects were given enough time to adapt the dark surround while the experimenter explained the task, which was “Please sit still and observe these three monitors from a specific angle; eval-uated their color performance and determined their order.” The experimenter did not give hints about how to define color per-formance such that the subject could determine it subjectively. C. Experimental Results
The subjects’ comparison data are shown in Table III, where a number in the cell means that there are subjects who think monitor is better than monitor in terms of color per-formance vs. viewing angle. For example, all 38 subjects agree that monitor is better than monitor , but only 7 con-sider monitor is better than . Notice that “ is better than " is a consensus among 38 subjects, but the ranking of is undeterminable—25 subjects think” is worse than ," 6 think “ is worse than but better than ,” and 7 think “ is better than .”
D. Data Analysis
According to the signal detection theory [10], we can rep-resent the color performance of each monitor as a probability distribution based on the above “votes.” The results are shown in Fig. 8. The -axis represents the color performance. The area under each curve represents the number of votes. All subjects agree that is superior to , so these two distribu-tions, each enclosing 38 votes, are disjoint. The area under can be divided into three portions: greater than (7) be-tween and (6), and less than (25). Based on Fig. 8, qualitatively, we can conclude that the visual perfor-mance ranking of the three monitors is . In addition, and perform similarly and better than
, which receives quite diverse opinions.
For comparison, we measured the colorimetric parameters of the three monitors and calculated their CSDs shown in Table IV. The domination CSDs of , , and at 30 are 0.0025, 0.0049, and 0.0273, respectively. Intuitively speaking,
TABLE IV
COLORSATURATIONDEGRADATION AT30 OFTHREEMONITORS
the former two have similar color performance, which is supe-rior to the latter. In other words, the proposed CSD metric is compliant with the psychophysical results.
IV. DISCUSSION
Using CSD to be metric, the visual angular dependence of TFT-LCD is quantified. We can recognize the visual perfor-mance of different TFT-LCDs modes with similar optical spec-ifications. The CSD also achieves our objective to improve on the conventional metric by proposing a unified threshold metric and evaluating color shift at large viewing angles. The advan-tages of CSD are summarized as follows.
1) Lightness independent: The degradation of lightness overshadows the color difference calculation, which makes the color shift effect be concealed at large viewing angles. The CSD is independent of lightness and resolves the issue.
2) Central reference white: It is necessary to compare the color at large angles with the central white ( , , ) of the panel. In other words, the central white shall be the reference white and be employed in the color difference calculations.
3) Analyze R, G, and B individually: We have to analyze red, green, and blue individually, because the CVA is de-termined by the worst degradation among the three chan-nels. By considering the CSD metric and the gamut value, we can recognize the visual performance of different LCD modes with similar optical specification easily.
4) Provide a visual assistant tool “Radar-Diagram”: The radar-diagram is a visual evaluation tool for assisting end-users to judge TFT-LCD’s color performance. The rounder shape and larger radius of the almost round curve in the radar-diagram imply better chromatic angular uniformity and higher color saturation in TFT-LCDs, respectively.
V. CONCLUSION
We have proposed a new unified threshold metric, CSD, for evaluating color viewing angle of TFT-LCDs by using the degradation of color saturation . It is easy to define acceptable color viewing angle ranges by
as conventional viewing angle definition of luminance contrast ratio . The threshold of CSD was chosen based on
the empirical JND and ISO-endorsed factor 0.004. From the mentioned criteria, we can set the color viewing angle (CVA) instead of conventional defined viewing angle (VA) to evaluate visual performance of TFT-LCDs at large angles. To combine the new metric with gamut value (the NTSC ratio), we can distinguish the visual performance from different LCD modes with similar optical specification easily. We also proposed the radar-diagram, a visual assistant tool for end-users, to judge the uniformity of color saturation across viewing angles from to 88 . In order to evaluate the performance of the proposed metric for practical applications, we conducted psychophysical experiments using three commercial TFT-LCDs and the exper-imental results also validated the efficacy of the CSD metric.
REFERENCES
[1] T. G. Fiske and L. D. Silverstein, “Characterization of viewing-angle dependent colorimetric and photometric performance of color LCDs,” in Soc. Inf. Display Tech. Dig., 1993, pp. 565–568.
[2] M. H. Brill and L. D. Silverstein, “Iso-luminance color difference metric for application to displays,” in Soc. Inf. Display Tech. Dig., 2002, pp. 722–725.
[3] K. C. Ho et al., “Colorimetric characterization of wide-viewing angle technologies,” in Soc. Inf. Display Tech. Dig, 2000, pp. 184–187. [4] ——, “Colorimetric characterization of ridge fringe field multi-domain
homeotropic LCDs,” in Soc. Inf. Display Tech. Dig, 2000, pp. 253–256. [5] M. E. Becker, “Display metrology for liquid-crystal-television
screens,” in DISPLAYS, 2005, vol. 26, no. 4–5, pp. 197–207. [6] G. Wyszecki and W. S. Stiles, Color Science. New York: Wiley, 1982. [7] W. N. Sproson, Color Science in Television and Display Systems.
New York: Adam Hilger, 1983.
[8] The Swedish Confederation of Professional Employees, in TCO’99
Re-port 2 Edition 3—Flat Panel VDUs, 2003.
[9] —— in TCO’03 Displays—Flat Panel Display, 2004, vol. 2. [10] R. Sekuler and R. Blake, Perception, 3rd ed. New York:
McGraw-Hill, 1994.
[11] R. G. W. Hunt, The Reproduction of Colour, 5th ed. Blackpool, U.K.: Fountain Press, 1995.
[12] P. Green and L. MacDonald, Color Engineering. Hoboken, NJ: Wiley, 2002.
[13] M. D. Fairchild, Color Appearance Models. Hoboken, NJ: Wiley-IS&T, 2005.
Szu-Fen F. Chen received the M.S. degree from the
Institute of Electro-Optical Engineering, National Chiao Tung University, Hsinchu, Taiwan, R.O.C. in 1994, where she is currently working toward the doctorate.
She is also with the R&D Design Division of Central Research Institute, Chung-Hwa Picture Tubes, Ltd., Taoyuan, Taiwan, R.O.C.
Wei-Chung W. Cheng received the Ph.D. degree
from Electrical Engineering-Systems Department, University of Southern California, Los Angeles, in 2003.
He joined the faculty of Department of Photonics and Display Institute, National Chiao Tung Univer-sity, Hsinchu, Taiwan, R.O.C., as an assistant pro-fessor in 2005. His research interests include color science, perception-oriented display design, and low-power system-level design.
