Color Break-up Examination

A relative contrast sensitivity (RCS) index was used to evaluate the color break-up phenomenon [23]. The RCS values were computed following the procedures in the S-CIELAB with some modifications, as shown in Fig. 3-8. Firstly, transform the digits of an input image and CBU image into the corresponding tri-stimulus values of each primary channel. Then, tri-stimulus values, XYZ transform into opponent-color space, AC1C2, where A denotes the luminance-related channel signals; C1 as the red-green channel signals and C2 is the blue-yellow channel signals. Third, each channel of opponent-color signals convolute the contrast sensitivity function (CSF) to simulate the spatial blurring by the human visual system. Forth, subtract the filtered channel images between each channel to extract the CBU fringe. Finally, the subtracted result of each pair of filtered channel images are summed together in different weights. The outcome is relative contrast sensitivity, RCS. The CBU phenomenon will be evaluated by RCS index to reflect the stimulus of CBU fringes in human visual system.

29

Fig. 3-8 The RCS flowchart

3.4 Summary

The two-color-field sequential method proposed to further reduce the field number which can result in longer time for LC response, which commercial LC modes can achieve. In this method, we incorporate local color dimming backlight technique to substitute for the equivalent function of the special color filters, and keep the whole optical throughput.

The S-CIEDE2000 index combined a spatial modulation of human visual system before calculating CIEDE2000 values. The RCS value reflects humans’ response for CBU edge stimulus. Therefore, the two indexes, S-CIEDE2000 and RCS will be used to verify the experimental accuracy in follow sections.

30

Chapter 4

Experimental Results

A verification of color reproduction accuracy was based on a 37” HDR display with 8 by 8 locally controlled color backlight divisions. The light spread function (LSF) of each division and display characteristic were measured to predict the LED and LC driving signals. Color difference values between the target image and the reproduced image were evaluated for color reproduction accuracy. Moreover, CBU examination will be presented in follow sections.

4.1 Experiments

The experimental apparatus are illustrated in Fig. 4-1. The LED and the compensated LC signals were calculated by using computer. The driving signals input to the HDR panel, and then the colorimeter measured tri-stimulus. The experimental flowchart is shown in Fig. 4-2.

First, the light spread functions per color per divisions were measured since the HDR display is a spatial shift-variant system, as shown in Fig. 4-3. Each LSF had different profile, especially at the display edges. Therefore, the measurement of light spread function was used to predict accurate LED signals. Secondly, LCD characteristics were measured including tri-stimulus values of maximums and flare terms in each primary color, also the OETF of LC cells in the panel. Then, measured parameters were used to establish a suitable color model for this panel, as mentioned in section 3.2. The following step, depend on a two-color-field algorithm to predict the LED and LC signals. Next step, input the calculated signals into the panel, and

31

measured the tri-stimulus values of the test image. Finally, calculate the color difference between target image and reproduced image. Therefore, color difference values evaluated the accuracy of color reproduction.

37” HDR display with 64 BL segments

Colorimeter Computer

Fig. 4-1 The experimental apparatus.

Measurement of LSF in each B/L segment

Measurement of display characteristic Establishment of color model

Two-color-field sequential method

Pure color verified Spatial frequency variation

Fig. 4-2 The experimental flowchart.

32

Fig. 4-3 The spatial shift-variant system.

4.1.1 Light Spread Function Measurement

The experimental results of light spread function measurement are illustrated in Fig. 4-4. In the two-color-field sequential method algorithm, the backlight distribution was gathered using the superposition method. Comparing the conventional convolution method with the superposition method, the convolution method measured one light spread function to simulate each light spread function and used convolution algorithm to get backlight distributions. On the other hand, the superposition method measures every light spread function and adds each light distribution to obtain backlight distribution. For example, in the full on white backlight; the superposition method has a higher correlation coefficient of 95% with measurement result than convolution methods of 83% as shown in Fig. 4-5. Therefore, the light spread function of 64 backlight divisions of three primary colors were measured to predict accurate LED signals and compensated LC signals.

33

Fig. 4-4 The experimental results of light spread function measurement.

83% 95%

(a) (b) (c) Fig. 4-5 The comparison of (a) measurement full on white backlight and (b) the convolution method and (c) the superposition method.

34

4.1.2 Panel Characteristic Measurement

In order to obtain device-independent color, the color model was established. As section 3.2 mentioned, a suitable color model needs three 3-dimensional look-up tables and the tri-stimulus of the panel. Therefore, 37” HDR display parameters were measured to build the color model.

When designing the look-up tables, the LC driving signals were divided ten parts from 0 to 255 gray levels in full-on backlight. Measuring the tri-stimulus, X, Y, Z, of these eleven levels, and the correlation between the digital input signals used to drive a display and the radiant output produced through LC cells by accounting for non-linear optoelectronic transfer functions were obtained. The 3-dimensional look-up tables of the three primary colors are illustrated in Fig. 4-6.

In the second part of the measurement process, the maximum and flare terms of tri-stimulus values in the HDR panel were measured by using the colorimeter. The maximums of tri-stimulus were measured with full-on backlight and largest LC driving signals in each primary color individually. Thus, Eq.3-2 was transformed to Eq.4-1. Consequentially, the result color model was used to predict the accurate LED and compensated LC signals.

35

Fig. 4-6 Three 3-dimentional LUTs of (a) red, (b) green, and (c) blue.

r

36

4.2 Results

The results of colorimetric reproductions are illustrated in 4.2.1. The test images include pure color images with different gray levels and random color patch images with differing spatial frequency. Moreover, the results of color break-up examinations will be given.

4.2.1 Color Differences

The color reproduction verifications include two parts, pure color image reproductions and the color patch images reproductions. The twelve pure color images of 10, 90, 180, and 250 gray levels in three primary colors were chosen as the test images. Moreover, the average CIEDE2000 values evaluated the accuracy of colorimetric reproduction. The colorimetric reproduction results of pure color images are illustrated in Fig. 4-7. As the results shown, the color differences of average CIEDE2000 values were all of less than 1. The results demonstrated colorimetric reproduction accurate and verified the suitable color model.

Red (GL) 10 90 180 250

∆E₀₀ 0.02 0.09 0.06 0.08

Green (GL) 10 90 180 250

∆E₀₀ 0.03 0.12 0.08 0.69

Blue (GL) 10 90 180 250

∆E₀₀ 0.41 0.29 0.97 0.38

Fig. 4-7 The results of colorimetric reproductions in pure color images.

37

In spatial frequency verification, there were five color patch images with different image contents, 1x1, 2x2, 4x4, 8x8, and 16x16 were chosen as the test images. The test images were used to analyze the correlations between spatial frequency variations and colorimetric reproduction accuracy. The color difference values of the test images are shown in Fig. 4-8. The results indicated the color difference increased in the images with more complex image contents. The color difference values in higher spatial frequency were larger than three, whose values were distinguished for human eye. The unsuitable colorimetric reproductions were caused by the limitation in backlight divisions.

Test image

ΔE₀₀ 0.06 1.02 3.8 5.4 7.23

0 2 4 6 8

1x1 2x2 4x4 8x8 16x16

ΔE00

Spatial frequency

Fig. 4-8 The spatial frequency variation results.

38

4.2.2 Color Break-up Examination

A color break-up examination based on five test color patches with light-skin, Asia-skin, strong red, sky blue, grass green, and white colors as shown in Fig. 4-9.

The three primary colors were chosen as the third primary in two-color-field algorithm alternately. Moreover, the third primary component was divided into ten sections separating into two field images in terms of specific ratio. Then, analyze the correlations between color arrangement and color break-up suppressions by using RCS evaluation index.

The CBU examination results are illustrated in Fig. 4-10. In the light-skin test image, the thirty RSC values described the CBU suppression efficiency normalized with the maximum of these thirty values, as shown in Fig. 4-10. (a). In the diagram, the horizontal axis was the ratio of color component arrangement into two field images, where 2 meant the separated ratio of each component between the first and the second field images was two eighth. As a result, the minimum value appeared at the fifth point in red component which divided into two field images. The minimum value of RCS index meant the effective CBU suppression. Fig. 4-10 (b) demonstrated the CBU examination result of Asia-skin test image, the minimum value of RCS showed at sixth point of red component. Refer to Fig. 4-10 (c), the good CBU suppression value of strong-red image was the fifth point of red divided component.

In sky-blue test image shown in Fig. 4-10 (d), the second to sixth points in blue component separation have the same slight RCS values. Based on the results shown in Fig. 4-10 (e), the optimal point appeared at the sixth of green component. In the last test white image Fig. 4-10 (f), the minimum RCS value was shown at the fifth point in green part. Summarily, the CBU suppressions were based on the color arrangement which separated the main color component into two field images.

39

Asia Skin Light Skin

Sky Blue Grass Green White

Strong Red

Fig. 4-9 Six test images were used to analyze the CBU suppression.

0

Color arrangement ratio Color arrangement ratio

RCS RCS

Color arrangement ratio Color arrangement ratio

RCS RCS

40

0 0.2 0.4 0.6 0.8 1 1.2

0% 20% 40% 60% 80% 100% 120%

blue red green

Color arrangement ratio Color arrangement ratio

RCS RCS

0 0.2 0.4 0.6 0.8 1 1.2

0% 20% 40% 60% 80% 100% 120%

blue red green

(e) (f)

Fig. 4-10 The CBU examination results of (a) light-skin, (b) Asia-skin, (c) strong-red, (d) sky-blue, (e) grass-green, and (f) white images.

Furthermore, the comparisons of RCS values between two-color-field method and other sequential methods are illustrated in Fig. 4-11. The RCS value of conventional RGB sequential method was defined as 100%. Comparing the two-color-field method with four kinds color sequences, RGB, RGBCY, RGBW, and RYGB, the RSC value of proposed method was 37.9% in light-skin image (Fig. 4-11 (a)). In Asia-skin test image, as shown in Fig. 4-11(b), the CBU suppression of two-color-field method was 53.5%. Fig. 4-11 (c) shows the RCS value of two-color-field method in strong-red color is 0.53. In the next test image, sky-blue color shown in Fig. 4-11 (d), demonstrated the RCS value in proposed method was 53.8% less than RGB method. Additionally, the CBU suppressions ratios of two-color-field method in grass-green and white images were 21.1% and 45.6%

respectively, as shown in Fig. 4-11 (e) and (f). Consequently, the CBU suppressions ratios of two-color-filed method were about 37.8% than conventional RGB method.

41

RGB RGBCY RGBW RYGB 2-field

RCS 37.9%

RGB RGBCY RGBW RYGB 2-field

RCS

RGB RGBCY RGBW RYGB 2-field

RCS

RGB RGBCY RGBW RYGB 2-field

RCS 46.2%

RGB RGBCY RGBW RYGB 2-field

RCS 46.2%

RGB RGBCY RGBW RYGB 2-field

RCS ^45.6%

(e) (f)

Fig. 4-11 The comparisons of RSC values between two-color-field method and other sequential methods in (a) light-skin, (b) Asia-skin, (c) strong-red, (d) sky-blue, (e) grass-green, and (f) white images.

42

4.3 Discussions

The verifications of colorimetric reproduction were based on an LCD with locally controlled backlight module of 8 by 8 divisions. As the experimental results showed, the color differences were increased along with complex image contents. The CIEDE2000 values in images with high spatial frequency were larger than three, which were unacceptable values of colorimetric reproduction accuracy since the limitation of presented backlight divisions. Therefore, the essential backlight parameters, number of divisions and light spread function (LSF) size, must be optimized to improve accuracy of colorimetric reproduction on images with complex contents.

Color arrangements in two-color-field method play an important role. The CBU examination results demonstrated that the minimum RCS value appeared when the main color components of test images separated into two field images. However, the accuracy of colorimetric reproduction was based on dividing the least component into two field images because of less reproduction errors. Therefore, the color arrangement of CBU suppression and colorimetric reproduction was tradeoff.

43

Chapter 5

Optimizations of Two-color-field Method

To improve accuracy of colorimetric reproduction in images with complex content, the backlight parameters, number of segments and light spread function (LSF) size, must be optimized. Moreover, color difference maps were used to describe the accuracy of colorimetric reproductions. Finally, results and discussions will be presented.

5.1 Backlight parameters

The two main backlight parameter factors affected color reproduction accuracy were the number of backlight divisions and light spread function size. Considering implementation complexity and thermal effect, the backlight parameters optimizations were essential. The two-dimensional Gaussian profile was simulated as the LSF in optimal process, where σ size was used to adjust the LSF width, as shown in Fig.

5-1. The optimal process and results are detailed below.

σ

Fig. 5-1 Gaussian profile

44

5.1.1 Backlight Divisions

There were five test color patch images with different image content were chosen as test images as shown in Fig. 5-2. An LSF, formulated by a two-dimensional Gaussian profile with horizontal and vertical standard deviations σx = 60 pixels and σy

= 60 pixels, was applied to initiate the optimization. The Gaussian profile was adopted for simplicity; σx and σy were selected because of the largest backlight segment size.

The results, as shown in Fig. 5-3, show that color differences of the five test images are decreased along with increasing amount of backlight divisions. The reason is that more independent backlight segments provide higher resolution of output backlight distribution, which is more adequate to be compensated by the LC module. Under current LSF setup, backlight division 80*45 is the optimized value since color difference only varies slightly exceeding this value.

1-patch 4-patches 12-patches 48-patches 144-patches

Fig. 5-2 Five test images with different image contents

5.1.2 Light Spread Function Size

Color reproduction accuracy dependence on LSF size was performed under 80*45 backlight segments. Similarly, color difference reduces when the LSF size shrinks, as shown in Fig. 5-4. LSF must be concentrated to adapt to increased spatial frequency of images. With σx*σy = 31*31 pixels, the color difference is lowered to an average of ΔE00ave< 3, which is generally regarded as an acceptable color difference.

45

0 2 4 6 8 10 12 14 16

16*9 32*18 48*27 64*36 80*45 96*54

Δ E

00

(ave.)

B/L divisions

16*9 8*6 4*3

2*2 1

Fig. 5-3 The correlation between number of backlight divisions and color differences.

0 0.5 1 1.5 2 2.5 3 3.5 4 4.5

34 31 27 23 20 17

Δ E

00

(ave.)

LSF size (σ pixels)

16*9 8*6 4*3

2*2 1

Fig. 5-4 The optimal results with the σ size of Gaussian profile against color difference.

46

5.2 Results

The backlight parameter optimizations were accomplished in prior sections. The reproduction images with optimal results are presented below. Moreover, color difference maps evaluated colorimetric reproduction accuracy. Finally, demonstration results will be given.

5.2.1 Color Difference Maps

Color difference maps were used to evaluate the accuracy of colorimetric reproductions, where S-CIEDE2000 was the evaluation index. The four test images, (a) Lily, (b) Butterfly, (c) Parrot, and (d) Color-ball with different detail and color complexities are shown in Fig. 5-5. The optimal results, which were simulated by using Matlab, are illustrated in Fig. 5-6. Comparing the target image with the optimal Lily test reproduction image (Fig. 5-6 (a)), the human eye can hardly differentiate between the two images. Moreover, the average S-CIEDE2000 value in the Lily image is 0.07, which is an acceptable color difference value. The reproduction result in the Butterfly image is shown in Fig. 5-6 (b), the color difference map shows the maximum S-CIEDE2000 value is lower than 4, and the standard deviation value is 0.14, that indicate the acceptable colorimetric reproduction. Similarly, the results of Parrot and Color-ball images are illustrated in Figs. 5-6 (c) and (d). The average S-CIEDE2000 values were all of less than 1, which means the human eye could not distinguish difference between the target image and the reproduction image.

(a) (b) (c) (d)

Fig. 5-5 Four test images: (a) Lily, (b) Butterfly, (c) Parrot, and (d) Color-ball images.

47

2-field image Target image

Difference image

S-ΔE₀₀

min max ave std 0 2.71 0.07 0.10

(a) Lily image

2-field image Target image

Difference image

S-ΔE₀₀

min max ave std 0 3.19 0.10 0.14

(b) Butterfly image

48

2-field image Target image

Difference image

S-ΔE00

min max ave std 0 3.93 0.23 0.28

(c) Parrot image

2-field image Target image

Difference image

S-ΔE₀₀

min max ave std 0 4.91 0.16 0.15

(d) Color-ball image

Fig. 5-6 The reproduction results of (a) Lily, (b) Butterfly, (c) Parrot, and (d) Color-ball images.

49

5.2.2 Demonstration Results

A digital light processing (DLP) LCD was used to simulate a FSC-LCD using the two-color-field method. The two-color-field method could be verified using the experimental demonstration. The simulated field images with optimal parameters, 80*45 backlight divisions, and σx*σy = 31*31 pixels of LSF, were displayed sequentially at 120Hz field rate on the DLP LCD. The resulting images are illustrated in Fig. 5-7. The simulated backlight images of the first field (Fig. 5-7 (a)) and the second field (Fig. 5-7 (b)), and LC images in the first field (Fig. 5-7 (c)) and the second field (Fig. 5-7 (d)) were captured by using a Canon D60 digital camera. Then, displayed two field images ( Figs.5-7 (e) and (f)) with 120Hz field rate, and the vivid color image was generated by using temporal color mixing, as shown in Fig. 5-8 (b).

Comparing the target image (Fig. 5-8 (a)) with reproduction image (Fig. 5-8 (b)) using the two-field-color method, the results demonstrated accurate colorimetric reproduction. Therefore, the two-color-field method was successfully verified by the experiment.

Furthermore, CBU demonstrated results are detailed. CBU visibility was compared between two images formed by the conventional three-field (RGB) and the proposed two-color-field sequential methods. A high-speed camera moved horizontally to simulate eye movement (Fig. 5-9 (a)) in capturing CBU images, shown in Fig. 5-9 (b) and (c). For example, the color band edges of the white ball induced by the primary colors (red, green and blue), are shown in Fig. 5-9 (b), incurring higher sensitivity than the mixed colors (red with partial blue and green with remaining partial blue), as shown in Fig. 5-9 (c). The result displayed an innate advantage of the two-field driving scheme in reducing CBU visibility.

50

(a) (b)

(c) (d)

(e) (f)

Fig. 5-7 The demonstrated results of (a) 1^st B/L image, (b) 2^nd B/L image, (c) 1^st LC image, (d) 2^nd LC image, (e) 1^st field image, and (f) 2^nd field image.

51

Fig. 5-8 Comparison of (a) target and (b) reproduced image

46” 120Hz MVA LCD

X-Y move table Camera

(a)

(b) (c)

Fig. 5-9 Target image was input to a 46” 120Hz MVA LCD (a), the corresponding CBU images, synthesized by three-field (b) and two-field (c), were captured by a high-speed camera moving horizontally.

52

5.3 Discussions

By optimization works presented in this chapter, optimal hardware parameters using the two-color-field method were collected. The number of back divisions 80*45, and light spread function size with σ= 31*31 pixels were used to obtain an accurate colorimetric reproduction using the two-color-field method. Comparing the optimal results with the experimental results of the four test images, color differences were reduced to an average of 30% in the optimal process, as shown in Fig. 5-10. Therefore, the optimal two-color-field method successfully reduced color difference in complex images.

0 0.1 0.2 0.3 0.4 0.5 0.6 0.7

L B P CB

8x8 80x45

29%

35%

29%

24%

S-ΔE00( ave.)

Lily Butterfly Parrot ColorBall

Fig. 5-10 Comparisons of optimal results and experimental results.

53

As the experimental results show, the two-color-field method reduced CBU visibility. The reason was that our method’s color arrangement reduced opponent colors in each field. When the image and human eye had relative motion, the color band edges at the image fringes can get slight color difference sensitivity for human eye, as shown in Fig. 5-11. Thus, based on the concept, two-color-field method can suppress both static and dynamic CBU phenomenon.

1

^st

2

^nd

R G

B

Y

Fig. 5-11 The slight CBU sensitivity yielded by reducing opponent colors in each field.

The color-mixing band edge of two-color-field image incurred less color different sensitivity.

54

However, the colorimetric reproductions of some specific test images do not have good performance. For example, the Blue Hill test image shown in Fig. 5-12.

The average color difference value is 2.08, but the maximum value is larger than 4.

The percentage of unacceptable values which means the values larger than 3 is 25%.

Comparing the target image with reproduced image as shown in Fig. 5-12. (a) and (b), the difference between these two images can distinguish by observer’s eye. Therefore, the third primary option may have another choice since the less colorimetric reproduction error is generated.

S-CIEDE2000

min max std ave >3

0 4.9483 0.1247 2.0844 25.8%

(a) (b) (c)

Fig. 5-12 (a)Target image (b)reproduced image (c) reproduced image S-CIEDE2000 values

55

5.4 Comparisons

A comparison between the proposed method and other methods are shown in Table. 4. The two-color-field sequential method has some advantages such as color filter free (0), higher luminance and resolution (both 300%), lower field rate (120Hz), and less field number (2). Moreover, the experiment demonstrated the two-color-field

A comparison between the proposed method and other methods are shown in Table. 4. The two-color-field sequential method has some advantages such as color filter free (0), higher luminance and resolution (both 300%), lower field rate (120Hz), and less field number (2). Moreover, the experiment demonstrated the two-color-field

在文檔中 雙色場色序法應用於無彩色濾光片之液晶顯示器 (頁 40-0)