Chapter 3 Stencil Field Sequential Color Method
3.2 Algorithm Optimization
3.2.1 Backlight colorful method
The field images generated from the algorithm mentioned in section 3.1.2 are shown in Fig. 3-6. Unfortunately, the multi-color field is not colorful enough.
Therefore, the most color and luminance information are displayed at red, green, and blue field, and the CBU suppression is limited. Therefore, the backlight colorful method is proposed to overcome this issue. In order to make the first multi-color field be more colorful, it is important to reduce the difference between compensated LC signal of red, green, and blue, which the backlight colorful method is based on the concept. The original locally controlled backlight signal is gotten by the backlight process as show in Fig. 3-7(a). In order to make the backlight more colorful, the minimum backlight is dimmed by a ratio (ex. 10%), and new locally controlled backlight signal is gotten, like Fig. 3-7(b). By a fundamental equation of displaying, Eq. 3-3, Iimage and IBL denote the emitting intensity of image and backlight
respectively; GLLC denotes the gray level of LC signal.
Iimage = (GLLC/255)γ × IBL (3-3) If the backlight intensity is reduced, the LC signal will be enhanced. Therefore, a larger minimum LC signal can be generated, and the multi-color field will be more colorful by the method, as shown in Fig. 3-8. Thus, the most color information is displayed in the first multi-color field, so the field images of red, green, and blue will be less colorful, and the CBU suppression will be more effective.
Furthermore, by utilizing the backlight colorful method to get the more colorful multi-color field, the determination of the dimming ratio is a critical work. In Table.
3-1, a comparison of different dimming ratio is made to discuss the effect on the CBU and the distortion ratio (D) of image, and the distortion ratio is defined by Eq. 3-4.
Distortion ratio (D) = number of distorted pixels/ number of total pixels (3-4) There are three dimming ratios are presented, and some summaries can be made. The smaller dimming ratio can get more colorful multi-color field, and it is more helpful for suppressing CBU. The green components of CBU with different dimming ratios are shown in the brackets of second row. However, the smaller dimming ratio may cause more image distortion because the LC and backlight cannot display enough luminance. Thus, the dimming ratio of backlight dimming method is needed to be optimized, and the optimization will be discussed in the Chapter 5.
Fig. 3-6 The field images of Stencil-FSC algorithm mentioned in last section
Fig. 3-7 The locally controlled backlight signal of (a) original algorithm and (b) algorithm with the backlight colorful method
Fig. 3-8 The field images of Stencil-FSC algorithm with the backlight dimming method
Table. 3-1The CBU and the distortion ratio (D) of image with different dimming ratio
(225,161,65) (225,161,7)
3.2.2 Backlight distribution generation by Fourier transformation process
In order to get the compensated LC signal, the backlight intensity distribution of the locally controlled backlight must be known. The usual method is utilizing convolution process. By measuring the light intensity distribution of an individual region, a light spread function can be gathered, like Fig. 3-9(a). Then, a backlight distribution can be generated by making a convolution of the light spread function and locally controlled backlight signal, like Fig. 3-9. The convolution method is a straight way to get the backlight distribution. However, the computation complexity is shown in Eq.3-5, and the X and Y mean the size of light spread function; the X1 and Y1 stand for the size of image size, and No. means the number of backlight regions.
Computation complexity = (X*Y)*(No.)+ (No.)*(X1*Y1) (3-5) It is an elaborate computation, and it is extremely hard to be realized on hardware.
Therefore, an alternative method, “Fourier transformation process,” is utilized to get backlight distribution.
A Fourier Transformation with low-pass filter is used to simplify the backlight distribution generation as shown in Fig. 3-10. At first, the locally controlled backlight signal is extended into the image size, and then the signal in frequency domain is gotten by making a Fourier Transformation to the extended locally controlled backlight signal. Next, a low-pass filter covers the locally controlled backlight signal in frequency domain to block the information of high frequency which is the edge part in spatial domain. In the step, a Gaussian low-pass filter is chosen because it can prevent from the ringing phenomenon in the image as shown in Fig. 3-11[30]. Finally, a blur image can be gotten by making an inverse Fourier Transformation to the locally controlled backlight signal passed filter in frequency domain. The computation
complexity is equal to Eq. 3-6, and the X and Y stand for image size.
Computation complexity =4*[(X*Y)+(X*Y)+ (X*Y)] (3-6) In order to prove the computation simplification, a comparison of computation complexity is made between the convolution process and the Fourier Transformation process. The image and light spread function size are set as 1366*768, and the number of backlight regions is divided into 16*12. By inserting the parameters into Eq. (3) and Eq. (4), Fourier transformation process has only 3% computation compared to convolution process. Therefore, generating the backlight distribution by Fourier Transformation process can simplify the computation effectively, and it can enhance the feasibility of the Stencil-FSC method on the hardware.
Fig. 3-9 Convolution process. (a) Light spread function,(b)the locally controlled backlight signal, and (c) the convolution backlight.
Fig. 3-10 Fourier transformation process. (a) The extended locally controlled backlight, (b) the backlight signal in frequency domain, (c) the Gaussian low-pass filter, and (d) The blurred image.
FT Low-pass filter (FT)-1
Locally controlled B/L (Spatial domain)
Locally controlled B/L (Frequency domain)
(a) (b) (c) (d)
12*16
=
Light spread function B/L distribution
X Y
X1
(a) (b) (c)
Y X
B\L distribution
Fig. 3-11 Ringing phenomenon. (a) Target image, (b) blurred image with ringing, and (c) blurred image without ringing.
(From Digital Image Process, Gonzalez and Woods, p.180-185)
3.3 Summary
Base on the concept of “stencil” which is a technique of painting, a Stencil-FSC method was proposed to suppress the CBU on the FSC-LCD. The method was based on locally controlled backlight system, and it displays a multi-color image in the first multi-color field, so it could reduce the color and luminance of red, green, and blue fields to suppress CBU. However, the original algorithm had some issues, so the backlight colorful method was proposed to get more colorful multi-color field, and the Fourier transformation process was utilized to simplify the computation complexity.
By the optimized algorithm, the Stencil-FSC can not only suppress CBU effectively, but also enhance the feasibility on the hardware. Moreover, Stencil-FSC will be verified by simulation and demonstration in the next two chapters.
(a) (b) (c)
Chapter 4 Experimental Demonstration
Stencil-FSC method was proposed to suppress CBU on the FSC-LCD.
Furthermore, two optimize algorithms, the backlight colorful method and the Fourier Transformation process, have been utilized to enhance CBU suppression and reduce the computation complexity. In order to verify the Stencil-FSC method, a demonstration on the real FSC-LCD will be completed.
4.1Approximation of Backlight Intensity Distribution
In the demonstration, the Fourier Transformation (FT) process was utilized instead of the convolution method to generate light distribution of locally controlled backlight and simplify the computation complexity as mentioned in 3.2.2. Therefore, in order to demonstrate correctly, it is important to make the backlight intensity distribution generated form Fourier Transformation process the same as that form convolution process.
In the Fourier transformation process, a Gaussian low-pass filter was used to filter the high frequency component of locally controlled backlight image in frequency domain, and a blur backlight image can be got to be the simulated light distribution. The equation of the Gaussian low-pass filter is shown in Eq. 4-1, and the diagram is shown in Fig. 4-1
2 2( ,)/2 0
) ,
(u v e D uv D
H = − (4-1) H is the magnitude of the low-pass filter, D is a coordinate in frequency domain, and D is the cut-off frequency to determine how blurry the image is. The definition of D
is the ratio of the coordinate where the maximum magnitude deceases to 60.7% and the total width of the Gaussian profile (W). Therefore, the lager D0 denote the Gaussian profile is broader, the more high frequency component of image can pass through, and the clearer image can be generated as shown in Fig. 4-2(b). Conversely, the smaller D0 denote the Gaussian profile is narrower, less high frequency component of image can pass through, and the more blurry image can be generated, as in Fig.
4-2(c). Therefore, the blurry image with different blur level can be generated by adjusting the D0 parameter, and it can be utilized to simulate the backlight intensity distributions with different light spread function.
D0
W
H
60.7%
02 2(, )/2
) ,
(u v e D uv D
H = −
D
Fig. 4-1 Gaussian low-pass filter(a) (b) (c)
Fig. 4-2 (a) Target image. Blurry images by using FT process with (b) D0=0.01 and (c) D0=0.001
Then, an approximation of the real backlight intensity distribution on the 32” FSC-LCD supported by C-Company was made by Fourier Transformation process. Parameter, D0, is adjusted to get different blurry backlight image, and the color difference (ΔE*ab) is
used (Fig. 4-3 (a)-(c)), and the results are shown in Fig. 4-4. When D0 is close to 0.0023, the minimum ΔE*ab can be gotten, that means the blurry backlight is closest to convolution backlight, as in Fig. 4-5. Therefore, when implementing Fourier Transformation process on the 32” FSC-LCD, the D0 will be chosen to be 0.0023 to get the most realistic backlight distribution, and the compensated LC signal can be gotten.
(a) (b)
(c)
Fig. 4-3 Test images. (a) Soccer, (b) color balls, and (c) girl.
0
0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 0.0045 Do
0 0.0005 0.001 0.0015 0.002 0.0025 0.003 0.0035 0.004 0.0045 Do
Fig. 4-5 (a)Target images, (b) convolution backlights, and (c) FT blurry backlights with D0 =0.0023 of three test images.
(c)FT blurry B/L (a)Target image (b)Convolution B/L
4.2 Experimental Demonstration I-15Hz/Frame
A Conventional LCD was used to simulate a FSC-LCD with Stencil-FSC method, and the Stencil-FSC method can be verified by using the experimental demonstration.
At first, simulated field images of the Stencil-FSC method are generated by a simulation program created by Matlab software, and the simulation parameters were set to match the real 32” FSC-LCD. The number of backlight sub-regions equals 12*16, D0 in Gaussian low-pass filter equals 0.0023, and the dimming ratio was set 50% to prevent image distortion. In the following, the simulate field images were displayed sequentially on the conventional LCD with 60Hz frame rate. However, Stencil-FSC method was a color sequence with four fields so the frame rate of the simulated FSC-LCD was only 15Hz, which was so low that human will perceive flicker phenomenon. Therefore, a High-speed CCD (Charge Coupled Device) camera was utilized to capture images with 15Hz capture frequency, and the images were displayed in 60Hz frame rate. Finally, Stencil-FSC images with 60Hz frame rate (240Hz field rate) can be simulated successfully. Finally, a moving camera was used to simulate eye movement and capture the CBU phenomenon in a conventional RGB color sequence and the Stencil-FSC method on the experimental LCD. Two test images, Lily and Girl, were used to verify the Stencil-FSC method as shown in Fig.
4-6. Lily is a white image which causes serious CBU using conventional RGB color sequence, and Girl is a colorful image utilized to test the CBU and image distortion for each color. The experimental results are presented in Fig. 4-7 and Fig. 4-8. In both test images, the CBU phenomenon is effectively suppressed by utilizing Stencil-FSC method, and there is no image distortion. Therefore, the Stencil-FSC method was
successfully verified by the experimental demonstration.
(a) (b)
Fig. 4-6 The test image. (a) white image: Lily and (b) colorful image : Girl
(a) RGB color sequence (b) Stencil-FSC method Fig. 4-7 The captured CBU image of Lily with (a) conventional RGB color sequence and (b) The Stencil-FSC method
(a) RGB color sequence (b) Stencil-FSC method Fig. 4-8 The captured CBU image of Gril with (a) conventional RGB color sequence and (b) The Stencil-FSC method
4.3 Experimental Demonstration II-60Hz/frame
The second experimental demonstration verified the Stencil-FSC method on a 32” FSC-LCD supported by C-company with local dimming backlight of 12*16 sub-regions and 180Hz field rate. The schemes of the FSC-LCD are presented in Table. 4-1.Because the field rate of the FSC-LCD was only 180Hz, and the frame rate must be set higher than 60Hz to prevent flickering, the panel could only display three fields sequentially in a frame. However, the Stencil-FSC method is a color sequence with four fields, so one of the fields must be rejected. Because human eye is less sensitive to blue compared to red or green, the blue field image can be rejected with less sensitivity to color distortion. Moreover, white and yellowish images were chosen to be the test images since the color contribution from blue field images were not important when utilizing the Stencil-FSC method. Therefore, two test images, Lily and Sunflower, were used in the experimental demonstration, and the field images are shown in Fig. 4-9.
Table. 4-1 Schemes of the 32” FSC-LCD 32-inch FSC-LCD
OCB-mode LC
1366 × 768 16 × 12 Divisions BL
48 × 24 (1152) LEDs Field rate 180 Hz (3-field)
(a)
(b)
Fig. 4-9 Test images of (a) Lily and (b) Sunflower and their field images
In the experimental demonstration, the backlight signal for 12*16 sub-regions was sent to the backlight system to get locally controlled color backlight, as in Fig.
4-10. The LC signal for the multi-color field, red-field, and green-field were implemented into the LC layers. Then, the multi-color field, red-field, and green field would light in sequence at 180Hz field rate, and a colorful image could be perceived as shown in Fig. 4-11. Finally, a moving camera was utilized to capture the CBU image and verify the Stencil-FSC method. The results are presented in Fig. 4-12 and Fig. 4-13, and verified that the Stencil-FSC method can suppress CBU effectively.
Moreover, as mentioned previously, the Stencil-FSC method not only suppressed CBU, but it also performed better compared to a conventional LCD and an LED based LCD, such as ultra-low power, high contrast ratio, and high NTSC. Therefore, the display performances were measured and summarized in the Table. 4-2. By utilizing the Stencil-FSC method on the FSC-LCD, The average power consumption was reduced to 44W which is only about 24% of a conventional LED based LCD. The result is better than the objective, 33%, because the locally controlled backlight was
applied with Stencil-FSC method. Moreover, the contrast ratio was enhanced to 5900:1 in the normal image, Sunflower, which was ten times larger than conventional CCFL-LCDs. The contrast ratio can even be increased to 27000:1 in high contrast image, Lily. The NTSC can also be increased to 114% compared to 72% in conventional CCFL-LCDs. Therefore, the Stencil-FSC can suppress CBU effectively and get the attractive performances, and they have been verified by the experimental demonstration on the 32” FSC-LCD.
(a) (b)
Fig. 4-10 The color locally controlled backlights of (a) Lily and (b) Sunflower
(a) (b)
Fig. 4-11 The Stencil-FSC images of (a) Lily and (b) Sunflower
(a) (b)
Fig. 4-12 The CBU image of Lily using (a) the Stencil-FSC method and (b) the conventional RGB color sequence
(a) (b)
Fig. 4-13 The CBU image of Sunflower using (a) the Stencil-FSC method and (b) the conventional RGB color sequence
Table. 4-2 The measured performance of CCFL-LCD, LED-LCD, and FSC-LCD with the Stencil-FSC method
114 % 114 %
72 % NTSC
Power consumption of a 32” conventional LED-based TV: ~180W
CCFL-LCD FSC-LCD
(RGB_ Driving ) Stencil-FSCLCD
P (W) 105 67 44
CR 579 : 1 442:1 5,973:1
114 % 114 %
72 % NTSC
Power consumption of a 32” conventional LED-based TV: ~180W
CCFL-LCD FSC-LCD
(RGB_ Driving ) Stencil-FSCLCD
P (W) 105 67 44
CR 579 : 1 442:1 5,973:1
4.4 Summary
The approximation for the real backlight intensity distribution has been completed by Fourier Transformation process, and when the D0 parameter in Gaussian low-pass filter equals 0.0023, the backlight intensity distribution is most similar to the real one on the 180Hz field rate, locally controlled backlight, 32” FSC-LCD supported by C-company. By using the hardware approximation, two experimental demonstrations of Stencil-FSC method have been completed on the conventional LCD and the 180Hz field rate, 32” FSC-LCD. By the results, the CBU suppression of the Stencil-FSC method has been verified. Moreover, the FSC-LCD with the
Stencil-FSC method could achieve attractive performances, such as average power consumption of 44W, which is only 24% of conventional LED-LCD, 114% NTSC, a Contrast ratio of 27000:1 in high contrast image.
Chapter 5
Optimization of Color Breakup Suppression
In previous chapter, a demonstration of 32” FSC-LCD with Stencil-FSC method was presented. However, the hardware parameters used in the demonstration were limited, thus the CBU suppression was not the best performance. In order to suppress CBU more effectively, the optimization of the hardware parameters were calculated, and these optimized results can be recommended to the FSC-LCD producers in the future.
The optimization works were done by nine test images with different detail content and color complexity. Three major parameters in Stencil-FSC method:
Number of backlight divisions, light spread function of backlight, and the dimming ratio in backlight colorful process, was analyzed to achieve best performance.
5.1 Classification of Test Images
In order to make the optimizations to be reliable and general results, the test images must be chosen carefully so that all kinds of images can be included in the test images. Therefore, images are chosen and classified by two defined indexes, detail complexity and color complexity.
5.1.1 Detail complexity
Detail complexity is used to evaluate how complex the detail is, and it can be defined easily by finding the edge component of the image. How to find the edge component in the image is a popular and well developed work in digital image
process[30] and utilizing image gradient is a common and effective method. Image gradient is defined by Eq. 5-1, and it is defined in the direction achieving maximum gradient. R, G, and B mean the gray level signal of red, green, and blue in an image; x and y indicate the horizontal and vertical direction, respectively; θstands for the direction causing the maximum gradient.
(5-1)
Generally speaking, because the edge component occur at the position causing a larger gradient, the edge can be found easily by set a threshold of the gradient as shown in Fig. 5-1. Finally, a summation of the edge parts is made, and the value is used to define the detail complexity. Larger value means the image has more detail complexity; contrarily, the smaller value indicates the image has less detail complexity. Examples of two images, lily and color balls, are presented in the Fig. 5-2.
The summation value of the simpler image, Lily, is 9437, and the value of the more complex image, Color balls, is 50648. Therefore, the detail complexity can be evaluated effectively by the summation value of edge component.
(a) (b)
Fig. 5-1 (a) Target image, and (b) Edge image gotten by calculating image gradient
2
ΣEdge=9437 ΣEdge= 50648 (a) (b)
Fig. 5-2 (a) Image, Lily, with less detail complexity and the edge summation=9437 (b) Image, Color balls, with more detail complexity and the edge
summation=50648
5.1.2 Color complexity
The other index, color complexity, is utilized to analyze the color abundance, and the concept of entropy is used to evaluate it. The index, entropy, is usually used in image compression to define the uncertainty of source[24], and the equation is Eq. 5-2.
P (i) means the probability of appearance of the i component, and entropy can stand for the dispersive level of each component.
=
∑
i
i P i
P
Entropy ()*log( ( )) (5-2) The concept is applied to determine the color complexity. At first, a color space, CIELAB, is divided into 20*20 regions such as Fig. 5-3(a). Then, the probability of color appearance in each color region is calculated as shown in Fig. 5-3(b). Finally, the entropy can be calculated by Eq. 5-2. The lager entropy means the image has more color complexity, and smaller entropy means the image has less color complexity contrarily. The example images, Lily and Color balls, are used again to verify the index as shown in Fig. 5-4. The image, Lily, is a single color image so the entropy is only 0.88. On the other hand, the entropy of the multi-color image, Color ball, is 3.41.
Thus, the entropy index can evaluate the color complexity correctly and will be
Thus, the entropy index can evaluate the color complexity correctly and will be