CHAPTER 4 TWO-FIELD DRIVING SCHEME AND SYSTEM
4.5 D EMONSTRATION
A. Image Reproduction
A further comparison is made by simulation, using the proposed driving scheme, between the current platform and the pseudo-platform with the optimized backlight parameters. Four images, Butterfly, Color Balls, Lily, and Parrot, as shown in the middle of Fig. 4-10(a) through Fig. 4-10(d), are the testing images representing natural scene, global color variation, high contrast, and detailed color, accordingly.
The reproduced image and the S-ΔE00 color difference map, resulted from 8-by-8 backlight division, are in the left side while those, resulted form 80-by-45 backlight division, are in the right. Generally, both the average and the standard deviation of S-ΔE00 of the reproduced images, as shown in Fig. 4-10(e), are all below 1.0, which confirms the applicability of the proposed method. Moreover, the optimized backlight parameters, more backlight division and localized LSF, are indeed effective to improve color reproduction accuracy.
L B P CB
8x8 80x45 1
0.8 0.6 0.4 0.2 0
S-ΔE 00(ave.) (a)
(b)
(c)
(d)
(e)
8 x 8 Target Image 80 x 45
L B P CB
8x8 80x45 1
0.8 0.6 0.4 0.2 0
S-ΔE 00(ave.)
L B P CB
8x8 80x45 1
0.8 0.6 0.4 0.2 0
S-ΔE 00(ave.) (a)
(b)
(c)
(d)
(e)
8 x 8 Target Image 80 x 45
Fig. 4-10 Target images, Butterfly (B), Lily (L), Parrot (P), and Color Balls (CB), are reproduced and compared in terms of in (a)-(d). (e) Comparison of the average S-ΔE00 from the two color difference maps of each image
According to the proposed method, the reproduced first and the second primary information are basically the same as those of the targets. That is, color difference in principle arises from the inexact reproduction of the third primary information. This deduction is observed easily in the reproduced images based on 8-by-8 backlight division. Those images are de-saturated and blurred in comparison with the other reproduced images because more blue information than necessary is introduced to achieve global target luminance. On the contrary, the stamens of the flowers in Butterfly and Lily resulted from 80-by-45 backlight division, for example, become saturated towards the targets because of rather accurate blue information involved.
This observation conforms that the proposed method works well for large backlight division amount and narrow LSF.
Some strategies can further improve the color reproduction accuracy. The global analysis, firstly, is to compile primary information statistically to decide the least significant one, which is chosen as the third primary redistributed into the two fields.
Fig. 4-11, with 51% of blue information greater than the other two primaries, is an example illustrating that blue is improper to be the third primary. Errors from reproducing blue information easily occur when the LC signals, based on red information, tend to be minimized in Fig. 4-11. The local analysis, furthermore, decides the third primary dependent on the image detail. Take for example the reproduced images of Color Balls and Parrot based on 80-by-45 backlight division, as shown in Fig. 4-10(b) and Fig. 4-10(d). In color difference maps of these reproduced images, larger color deviation occurs at the cyan-like areas, such as the cyan balls and parrot head around the eye. It is difficult to manipulate blue information in those areas since the LC signals here approach zeros because of less red and green information to be displayed. In such case, three combinations of two primary LEDs, R-G, G-B, or B-R, may be turned on at different location in each of
the two fields. The least significant primary selected by either the global or the local analyses is expected to achieve minimum color reproduction errors.
R-17%
G-B- 32%
51%
Target image Reproduced image
Color Difference Map (S-ΔE00)
R-17%
G-B- 32%
51%
Target image Reproduced image
Color Difference Map (S-ΔE00)
Fig. 4-11 An image in which blue is inadequate to be the third primary
B. Color Break-up (CBU) Examination
CBU visibility is compared between two images formed by the conventional three-field (R/G/B) and the proposed two-field methods. A high-speed camera moves horizontally in front of an LCD, with refreshing rate 120Hz, to simulate eye movement to capture CBU images, as shown in Fig. 4-12(a). The capturing time of the camera in both cases is well-arranged to ensure equal CBU width. The CBU images arise from the R/G/B three-field and the proposed two-field methods are shown in Fig. 4-12(b) and Fig. 4-12(c), respectively. The extracted pink ball in Fig.
4-12(b) displays clear reddish and bluish bands; that in Fig. 4-12(c) shows de-saturated magenta bands which are perceived less color separation. Similarly, the extracted white ball in Fig. 4-12(b) incurs more apparent contrast of color bands than that in Fig. 4-12(c). The result agrees with the innate feature of the two-field driving scheme in reducing CBU visibility [58]. Namely, the contrast sensitivity between the magenta-like and cyan-like fields is generally lower than that between the opponent color, red and green, fields [59,60,61].
(a)
(b) (c)
(a)
(b) (c)
Fig. 4-12 (a) The apparatus of capturing CBU image by a high-speed camera moving horizontally. The CBU images arise from (b) the R/G/B three-field and (c) the proposed two-field methods, respectively. (An example video of synthesizing the image, Color Balls, can be browsed via the hyperlink: http://adolab.ieo.nctu.edu.tw/app/news.php?Sn=32.)
Some derivative issues are worth review on the examination of CBU visibility.
First, the capture of CBU image is dependent on the viewing condition. Fig. 4-12(b) and Fig. 4-12(c) are photographed under specific conditions, at a position close to the LCD and with rather fast speed of the camera movement, which are suitable for capturing apparent CBU fringes but are not the real ones. The discrimination of CBU fringes is the second debatable point. The visibility of CBU fringes, e.g. Fig.
4-12(b) and Fig. 4-12(c), varies with observers because the judgment relates to psychophysical response rather than simple colorimetric tri-stimulus difference.
Those considerations induce the demand for a metric, based on psychophysical nature of the HVS, introduced in Chapter 5.
4.6 Conclusion
A two-field driving scheme is proposed to facilitate the field-sequential-color LCD. The color reproduction achieves average CIEDE2000 color difference of less than 3. Moreover, the average CIEDE2000 values are well-below 1.0 by simulation when the low- and band-pass features of the human visual system are taken into account. In addition, the color breakup visibility is suppressed by reducing the contrast sensitivity between the two color fields. Some comparisons on the system properties with the other driving schemes are summarized in Table 4-1. The promoted optical efficiency and the reduced material cost can be realized by removing the CF film. Power consumption is also reduced since a spatially- modulated backlight is utilized. Particularly, some commercial LC modes, e.g. MVA (Multi-domain Vertical Alignment) mode, can be utilized without extra cost to perform two fields sequentially in displaying color images on LCDs. Therefore, the proposed driving scheme, accompanied by the corresponding system configuration, is an applicable candidate to large-size, green display.
Table 4-1 Comparisons between the proposed two-field and the other driving schemes
Full-on B/L RGB-field Prior Two-field [53]
NCTU Two-field [64]
Color Filters 3 0 2 0
Field / Frame 1 3 2 2
Refresh Rate (Hz) 60 180 120 120
Light Source 1x CCFL 3x LED 3x LED 3x LED
BL System Global Global Global Local
Optical Throughput 100% 300% 150% 300%
Resolution 100% 300% 150% 300%
LC Response Time (ms) < 8 < 2 [15] < 5 < 5
Color Breakup [58] None High Medium Medium
Chapter 5
Relative Contrast Sensitivity as CBU Index
5.1 Introduction
Although many driving schemes for color break-up (CBU) suppression have been proposed [62,63,64], no concrete evaluation metric can assert the effectiveness of those methods. Amongst the prior works, the device-dependent method, fitting empirical formula based on the display parameters, is commonly applied to the evaluation of CBU effect [65,66,67,68]. The advantage is that the fitted formulae are useful to predict the width of CBU fringe or to derive the cut-off frequency of the field rate, etc. However, the utility is limited due to the lack of generality and efficiency. That is, the formula deduced from one display under certain viewing condition may not be effective in other displays under different viewing conditions.
As for the device-independent method, the well-developed CIE color differences, CIEDE2000 (ΔE00) for instance, are the candidates. Nevertheless, the applicability of CIE color difference to CBU phenomenon must be reviewed carefully. The CIE color difference formulae are developed to measure the color difference between uniform color patches with small color difference, e.g. Fig. 5-1(a), after moderate chromatic adaptation [69,70]. CBU phenomenon, on the contrary, occurs at any image contents, simple or complex, with large color difference, red-green as shown Fig. 5-1(b), for example. Although color differences of the opponent colors, e.g.
red-green and yellow-blue, still can be calculated by formulae, the real perception is generally non-linear to large color difference values. Moreover, CBU stimuli are always glanced in a short period, which is insufficient for chromatic adaptation [71].
(a) (b)
(a) (b)
Fig. 5-1 Pairs of color patches with (a) small and (b) large color differences
An example is given to further exemplify the disagreement between the visual response of CBU and the calculated ΔE00 value. A series of synthesized CBU images, as shown in Fig. 5-2, are the simulated results of a white bar moving horizontally from left to right in black background at speeds of 200mm/s, 400mm/s, 600mm/s and 800mm/s, respectively. Observers report unanimously that the awareness of CBU effect increases along with increasing moving speed. However, the averaged ΔE00 values, calculated by Eq. (5-1), of the four CBU images are the same, 68.03, in the experiment platform. The disagreement results from the fact that the accumulated average ΔE00 value of CBU fringe is proportional to the moving speed, thus the fringe width, so the common factor is cancelled by the division.
fringe CBU of number pixel
total
fringe CBU of pixel each of value CIEDE2000 .)
00(ave =
∑
ΔE (5-1)
200 400 600 800
200
200 400400 600600 800800
Fig. 5-2 Synthesized CBU images of a white block moving horizontally;
numbers at the bottom-left corner are the moving speeds in unit of mm/s.
The objective of this study is to establish a robust method, which can closely reflect the real response of the human visual system to evaluate the CBU effect effectively. An overview of the procedures of the proposed method and the verification by psychometric experiments are described. An application of this proposed index will be given as well.
5.2 Evaluation Method and the Index
Visual sensitivity based on the pattern-color-separable appearance pathways [61]
is proposed to correlate with the sensibility of CBU. In our hypothesis, CBU stimulus is presumed to be strongly related to the neural signals, just leaving the lateral geniculate nucleus (LGN) cells, as shown in Fig. 5-3 [72], with incomplete chromatic adaptation because of very short existing time period. According to the theory of pattern-color separability, the raw neural signals are proportional to the product of three terms: the stimulus strength, the pathway’s sensitivity to the pattern color, and the pathway’s sensitivity to the spatial pattern [59,60]. The procedures of reckoning sensitivity, as shown in Fig. 5-4, are modified from S-CIELAB [54,57]
and described below.
Eye
Optic nerve
Optic chiasam
Optic tract
LGN cell
Optic radiations Primary visual cortex
Eye
Optic nerve
Optic chiasam
Optic tract
LGN cell
Optic radiations Primary visual cortex
Fig. 5-3 Visual pathway from the eye ball to the visual cortex, in which the LGN cell is the intermediate node [72]
(d
rd
gd
b)
Fig. 5-4 The computation flowchart of relative contrast sensitivity: (a) from input digits (dr dg db) to spatially-blurred opponent signals (A’ C1’ C2’), and (b) weighting sum performed on the subtracted CBU fringe signals, where K is the scale ratio between the luminance and the chrominance channels.
The procedures of the proposed method start at tri-stimulus conversion, from the device-dependent input digits to the device-independent tri-stimulus values, CIE
1931 XYZ. Tri-stimulus values XYZ are then transformed into opponent-color space, AC1C2, where A denotes the luminance-related channel and C1, C2 the chrominance-related ones. Sometimes, AC1C2 representation is referred to as luminance, red-green, and blue-yellow channel signals, respectively. Each channel of opponent-color signals is then spatially filtered by the corresponding spatial contrast sensitivity function (CSF) to simulate the spatial blurring by the human visual system. Finally, subtraction is performed to extract the filtered signals of CBU fringe, followed by the weighting sum between the opponent channels pixel by pixel.
The outcome is termed as relative contrast sensitivity, RCS.
A. Tri-stimulus Conversion
Since CBU effect is studied on LCDs, a mapping rule of LCDs is required for the tri-stimulus conversion. In current situation, all the CBU images, based on different driving schemes, are processed, synthesized, and then displayed on an FSC LCD.
The color model expressed by Eqs. (2-3) and (2-4), except for some modifications [73], can be applied to the conversion of tri-stimulus value from the input digits. The color model has to be implemented correspondingly once if other type of LCD is the testing platform.
B. XYZ to AC1C2
The opponent-color sensitivity signals, AC1C2, are determined through psychophysical experiments for pattern-color separability [61]. Mathematically, A set of AC1C2 is linearly transformed from a set of XYZ by Eq. (5-2) [54,57] since the application is focused on transformation of XYZ tri-stimulus values on LCDs.
⎥⎥
C. Spatial Filtering
The spatial filtering on the opponent-color signals can be operated in the frequency domain. A, C1, and C2 signals are transformed into frequency domain by Fourier transformation. Multiplications are then performed channel by channel between the opponent signals and the corresponding CSFs. The frequency version of CSFs is expressed by Eqs. (5-3) and (5-4) [54], where csflum and csfchrom
respectively denote CSFs of luminance and chrominance channels, f the spatial frequency in unit of cycle per degree. The numerical values of the fitted parameters are summarized in Table 5-1 [54], and the normalized one-dimensional CSFs are depicted in Fig. 5-5. Fig. 5-5(a) represents the 1-D cross-section profile while Fig.
5-5(b) corresponds to the blurring masks as shown in Fig. 5-4(a). The CSF of channel A exhibits band-pass nature, while the other two are low-pass filters.
( )
c bflum f a f e
csf = ⋅ ⋅ − ⋅ (5-3)
( )
1 b1 fC1 2 b2 fC2chrom f a e a e
csf = ⋅ − ⋅ + ⋅ − ⋅ (5-4)
Table 5-1 Numerical values of the coefficients of Eqs. (5-3) and (5-4) [54]
Parameters Values
(Channel A) Parameters Values (Channel C1)
Values (Channel C2)
a 75 a1 109.1413 7.0328
b 0.2 b1 0.00038 0.000004
c 0.5 c1 3.4244 4.2582
a2 93.5971 40.6910
b2 0.00367 0.10391
c2 2.1677 1.6487
0 10 20 30 40
Fig. 5-5 Normalized frequency filters in (a) 1-D cross-section and (b) 2-D distribution to approximate the spatial contrast sensitivity functions in unit of cycle per degree
D. Single-valued Index: Subtraction and Weighting Sum
CBU fringe is the object for the RCS computation. As shown in Fig. 5-2 for example, the separate color fields form the colored bands at both sides of the white bar in the horizontal direction. These two color bands in both sides are the CBU fringe here. The filtered A, C1, and C2 signals are extracted by subtracting the signals at the white area channel by channel, followed by weighting sum of the difference values among three channels, as shown in Fig. 5-4(b). The scale ratio, K, is an empirical value deduced from the psychophysical experiments.
Before finishing the summation, two more considerations must be dealt with. First, the values of opponent signals, C1 (red-green) and C2 (blue-yellow) must use the absolute values. The positive and the negative signs in the representation of opponent signals just show the concept of opposite polarity. Second, the summation between the luminance channel and the chrominance channels requires a suitable scale ratio to account for the fact that the human visual system has different modulation responses to the two kinds of channels [59].
5.3 Experiment
The experiments are divided into three steps. The first is aimed at retrieving the scale ratio between the luminance and the chrominance channels. The second is to evaluate the proposed RCS index. Finally, a useful application is demonstrated.
Apparatus
A 32”, optically-compensated-mode (OCB) FSC LCD is constructed as the testing platform. Three-in-one light emitting diodes (LEDs) are the light sources, supporting the red, green, and blue field rate at 180Hz, as shown in Fig. 5-6.
D
Observer
Dark room FSC Display FSC Display
D
Observer
Dark room FSC Display FSC Display
Fig. 5-6 Apparatus of the psychophysical experiments viewed in a dark room, where D denotes the viewing distance
Observers
There are undergraduate and graduate observers (three females and eight males) of ages ranging from 23 to 36 years old. All observers, some wearing their correcting lenses, perform normally on Snellen acuity pattern test for visual acuity and on Ishihara test for color blindness. All the experiments are performed in a dark room.
Test Images
A series of CBU images are synthesized as the testing images. The original images are square, uniform blocks of different colors in the center of the black ground. The width of the blocks is 2cm, subtending around 2-degree viewing angle at viewing distance of three-fifths the diagonal size. Color fringes appear in the horizontal direction to simulate the bar moving horizontally, like the case of white bar shown in Fig. 5-2. Color fringe width is above 0.1 degree, depending on the velocity applied to the synthesis.
Experiment (I): Mean Opinion Score and Absolute Threshold
Experiments of mean opinion score (MOS) and absolute threshold are performed to retrieve the scale ratio between the luminance and the chrominance channels and to examine the proposed index. In this experiment, color fringes are synthesized, based on three-field scheme, onto a white block; the color fringe width extends along with simulating velocity, ranging from 200mm/s to 800mm/s, separated by every 100mm/s. The blocks are of four luminance levels, 18, 28, 42, and 56cd/m2.
MOS is one of the ordinal scales, category scaling [74]. Synthesized CBU images are rated by assigning a score, summarized in Table 5-2. The sensibility of CBU fringe increases from “Imperceptible” to “Vary annoying” of the CBU fringe along with the decrease of MOS from 1 to 5. Particularly, an anchoring CBU image, with the widest color fringe arising from the velocity 800mm/s, is scored 5 since it is regarded most annoying. Each CBU image pops up on the LCD and vanishes after three seconds to imitate closely to the real situation. A full dark image is inserted between any two succeeding CBU images to reduce after-image effect. Each test round for an observer, the CBU images are shown in random sequence. The comparison will be made between the MOS and the computed RCS.
Table 5-2 Mean opinion score definition
Score Evaluation result
1 Imperceptible 2 Perceptible, but not annoying
3 Slight annoying
4 Annoying
5 Very annoying
The absolute threshold is determined manually based on method of adjustment [74]. The same series of CBU images, as those in MOS experiment, are displayed on the LCD. Observers are asked to adjust the viewing distance till the CBU fringe is imperceptible; this distance is termed as indistinguishable viewing distance, denoted as DInD. The resultant indistinguishable CBUA (CBU angle), as shown in Fig. 5-7, denoted as InD-CBUA can be obtained by substituting DInD into Eq. (5-5), where D denotes the viewing distance, T the target width, V the moving speed, and F the field rate, respectively [75]. Cross comparisons are then made between InD-CBUA and the computed RCS.
2 ) ( tan 2 )
(
tan 1 1
D T FD
V D
CBUA= − T + − − (5-5)
D
T V/F
CBUA D
T V/F
CBUA
Fig. 5-7 Illustration of the geometric relation of CBUA with other parameters
Experiment (II): Optimization
The RCS index is applied to the comparison between various CBU suppression driving schemes: RGB, RYGB, RGBWmin, and RGBCY [76,77], as shown in Fig.
5-8. For those multi-primary methods being tested, the CBU images are synthesized according to the method and then are decomposed to the signals that can be used in the testing platform. Specific colors are used for comparison: White (255, 255, 255), Asia Skin (213, 139, 112), Light Skin (205, 163, 144), Strong Red (186, 70, 73), and Sky Blue (65, 75, 163), as shown in Table 5-3. The input digits, White (255, 255, 255) for example, follow the definition of that shown in Eq. (2-3). The color fringes for all color blocks are synthesized based on velocity 800mm/s because of the widest color fringe width. The way to decide the MOS here is the same as that performed in the MOS decision of Experiment (I). The anchoring CBU image, scored 5, here becomes the white block with color fringes resulted from RGB
5-8. For those multi-primary methods being tested, the CBU images are synthesized according to the method and then are decomposed to the signals that can be used in the testing platform. Specific colors are used for comparison: White (255, 255, 255), Asia Skin (213, 139, 112), Light Skin (205, 163, 144), Strong Red (186, 70, 73), and Sky Blue (65, 75, 163), as shown in Table 5-3. The input digits, White (255, 255, 255) for example, follow the definition of that shown in Eq. (2-3). The color fringes for all color blocks are synthesized based on velocity 800mm/s because of the widest color fringe width. The way to decide the MOS here is the same as that performed in the MOS decision of Experiment (I). The anchoring CBU image, scored 5, here becomes the white block with color fringes resulted from RGB