Scene-Adaptive RGB-to-RGBW Conversion Using Retinex Theory-Based Color Preservation

Transcription

1 684 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 Scene-Adaptive RGB-to-RGBW Conversion Using Retinex Theory-Based Color Preservation Kyung Joon Kwon, Member, IEEE, and Young Hwan Kim, Member, IEEE Abstract This paper proposes a retinex theory-based approach to RGB-to-RGBW conversion that preserves the human color perception within a pre-determined level of color distortion for RGBW displays. The proposed method primarily consists of two procedures. In the first, it searches for the maximum intensity level that induces no color distortion for a given image by extracting the white spectra from the common components of the RGB primary colors and adjusting all the pixels gains uniformly. In the second, the proposed method applies an additional gain to each pixel based on its chromaticness and controls the color distortions arising from the individual gains using the color perception estimated by retinex theory and a feedback mechanism. Experimental results showed that the proposed method was more effective than conventional methods in terms of intensity increment and color preservation. For Kodak test images, the proposed method increased the average intensity by times with a color-distortion level of compared to reference RGB displays, whereas the conventional methods increased the average intensity by times with color-distortion levels of The surplus intensity yielded by the proposed method can be used to dynamically reduce the power consumption of a liquid crystal display (LCD) backlight or to provide brighter images on LCDs. Index Terms Color distortion, human visual system (HVS), low power display, retinex theory, RGBW display. I. INTRODUCTION T HE colors of most flat-panel displays, such as liquid crystal displays (LCDs), electrophoretic displays and active-matrix organic light-emitting diodes, are implemented using color filters [1]. Display structures that use color filters can easily display colors, but a loss of transmittance is inevitable because color filters, by their nature, transmit only the fraction of incoming light at a specified wavelength. To compensate for this transmittance loss in color-filtered displays, the RGBW display, which adds a white sub-pixel to the RGB display as shown in Fig. 1, has become increasingly popular [2] [5]. Because a white sub-pixel passes the entire visible spectrum, the loss of transmittance can be minimized for that white sub-pixel area. Let the aperture ratios of each display be and, and the transmittance values of the primary color and white filters be and, respectively. Then, the maximum brightness of each display is computed as Manuscript received April 19, 2012; revised July 23, 2012; accepted August 08, Date of publication October 04, 2012; date of current version November 20, This work was supported by IT Consilience Creative Program of MKE and NIPA (C ), BK21, and LG Display. The authors are with the Division of Electronic and Electrical Engineering, Pohang University of Science and Technology, Pohang, Korea. ( primalfear@postech.ac.kr; youngk@postech.ac.kr). Color versions of one or more of the figures are available online at ieeexplore.ieee.org. Digital Object Identifier /JDT (1) Fig. 1. Color-filter structures of patterned displays. (a) RGB vertical-stripe structure. (b) RGBW checkerboard structure. where and denote the maximum brightness of the RGBW and RGB displays, and is the constant determined by other common factors related to the display s brightness, e.g., the backlight intensities and transmittance of the polarizer in LCDs. The first term in (1) is the brightness from the RGB primary sub-pixels of the RGBW display, and the second term is the brightness from a white sub-pixel. By dividing (1) by (2), the relationship of the maximum brightness for the RGB and RGBW displays can be determined as follows: where denotes the brightness gain of the RGBW display relative to the RGB display at the maximum brightness level. In LCD TVs, the typical values of the parameters of (3) are,,and [6]. Then, is computed to be approximately 1.6; that is, the RGBW display is 1.6 times brighter than the RGB display under the maximum brightness condition, i.e., a pure white display. In contrast, when displaying a primary color, because the second term of (1) becomes zero, the ratio of the brightness is 0.75; that is, the RGBW display is 25% darker than the RGB display. To use RGBW displays, an RGB-to-RGBW signal conversion process is needed because display systems usually assume that the display devices have an RGB structure and adopt an RGB signal compatible with the devices. In addition, because the RGBW display has unconventional color space characteristics, which are summarized as brighter white and darker RGB primary colors than those of the RGB display, the image quality of the RGBW display depends strongly on the RGB-to-RGBW conversion method. To achieve better visual quality in the RGBW display, several RGB-to-RGBW conversion methods have been proposed [7] [10]. They were developed on the basis of two prerequisites: first, the ratio of the light intensities of primary colors for each pixel before and (2) (3) X/$ IEEE

2 KWON AND KIM: SCENE-ADAPTIVE RGB-TO-RGBW CONVERSIO 685 Fig. 2. Color space of RGB and RGBW displays with normalized intensity. (a) Color space of RGB displays,. (b) Color space of RGBW displays. (c) Projected color space of RGB (green broken line) and RGBW (blue solid line) displays in the RG domain. after conversion must be the same in order to maintain the hue and saturation; second, a gain is applied to each pixel separately according to its chromaticness in order to use the asymmetric color space of the RGBW display. Ultimately, the hue and saturation for each pixel are preserved, and the intensity is altered after conversion. However, the human visual system (HVS) does not perceive the color of an arbitrary single point but determines its color by normalizing the color signal from that point with the color signals of the entire scene surrounding it [11] [13]. Thus, the nonuniform change in the intensities of the point and its surroundings make one perceive the point s color differently, although the same hue and saturation are retained after the conversion. Therefore, the prerequisites used by the conventional methods are not appropriate for preserving the perceived color of an original RGB image. This paper proposes a novel RGB-to-RGBW conversion method that can preserve the perceived color within a pre-determined color-distortion level and that can provide brighter displays which are the major purpose of RGBW displays. First, the proposed method extracts the white data for each pixel with a spectrum exchange manner [5], [18] and applies the uniform gain to all pixels as largely as possible without inducing color distortion. Then, it controls the trade-off between an additional increase in intensity and an increase in color distortion using retinex theory. This paper is organized as follows. In Section II, we review the color space of the RGBW display and retinex theory to explain existing RGB-to-RGBW conversion methods and the reason for the associated color distortion. The details of the proposed approach are described in Section III. The experimental results regarding the intensity increment and color distortion for the proposed and benchmark methods are presented in Section IV. Finally, conclusions are drawn in Section V. II. BACKGROUND A. Color Space of the RGBW Display The color-space diagrams [19] of RGB and RGBW displays are shown in Fig. 2, and the intensity scales are normalized by the maximum intensity of the RGB display. The shape of the color space of the RGB display is a cube [Fig. 2(a)], and the shape of the color space of the RGBW display is a trace of a cube moving along the diagonal axis of the three-dimensional color space [Fig. 2(b) and (c)]. RGB sub-pixels in the RGBW display form a cube scaled to approximately 75% of the size of the RGB display, and adding a white sub-pixel including all the primary colors causes the trace along the diagonal axis. Fig. 2(d) shows the projected color space of the RGB and RGBW displays with respect to the RG domain [8], [20]. This analysis of projected domain makes it easier to understand the properties of the color space of the RGBW display. In this domain, the blue data are assumed to be intermediate values between red and green data and do not affect the analysis. The points near the diagonal axis indicate achromatic colors, and those near the red or green axis indicate chromatic colors. The color space of the RGBW display is characterized by this asymmetric structure, i.e., a wide achromatic color space and a narrow chromatic color space. Thus, most images displayed by RGBW displays have brighter achromatic colors and darker chromatic colors than RGB display images. B. Retinex Theory Land [12] suggested that the perceived color of a unit area is determined by three-dimensional coordinates [13], which are calculated as the relative reflectance affected by the randomly distributed unit areas for the red, green, and blue wavebands. The relative reflectance means the interpreted surface color of an object influenced by colors of surroundings.,which denotes the relative reflectance of a unit area affected by another unit area, is computed as follows: where denotes the intensity of the arbitrary unit area on the path from to and denotes one of the three principal wavebands. A threshold operation using on the ratios is included to remove the effects of nonuniform illumination over the scene: variations gradual enough to be below the threshold (4)

3 686 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 are neglected [13]. If the number of unit areas is,theaverage relative reflectance of is Then, for each waveband,, we can obtain the three relative reflectance values of the unit area, which are called designators. In retinex theory, these three values correspond to one color point in the 3D coordinates; if two color points are close together, their colors are perceived as being similar although the colors of the two points are actually different. Therefore, retinex theory can explain perceived colors and the effects of their surroundings more effectively than the color of an object itself. C. Existing RGB-to-RGBW Conversion Methods Conventional RGB-to-RGBW conversion methods consist of three steps [7] [10]. First, the white data are extracted pixel by pixel as a function of the minimum and maximum intensities among,,and. Second, the individual gains are calculated as to increase the intensities of the display. Finally, the individual gains are applied to the input data, and then the target RGBW data are generated by subtracting the white data as follows: where denotes the input light intensity of primary colors, and denote the output light intensities of primary and white colors after conversion, and and indicate the minimum and maximum values of in each pixel, respectively. All notations represent,,and. These conventional methods can be expressed as four different algorithms according to the manner in which the white component is extracted and scaled, as described in (7) [7] [10]. All of these white functions are satisfied by two conditions: when the input data are purely chromatic, i.e., ; when the input data are purely achromatic, i.e., (5) (6) 4 (7) From the second equation in (6), the overall intensity of a specified primary color,, is expressed as where denotes the pixel gain, which is determined by the minimum and maximum intensity values of each pixel. From (7) and (8), is the largest in achromatic input data and the (8) smallest in chromatic input data. This reflects the asymmetric color space of the RGBW display well. To evaluate the changes in colors after RGB-to-RGBW conversion using the conventional methods, we adopt the HSI color space model [21] where,,and are the light intensities of the primary colors and,,and denote the hue, saturation, and intensity of each pixel, respectively. Let,,and denote the results of (9) when the input intensity is substituted for,,and in (9), and let,,and denote the results of (9) when the overall intensities after conversion,,aresubstitutedinthesameway.then,the two HSI calculations are related as follows:,,and. These results show that in conventional conversion methods, the hue and saturation for each pixel are preserved, and only the intensity for each pixel increases by. However, varies with the chromaticness of the input data, and this makes the data of the same waveband be stretched with different degrees. As a result, the viewers interpretation about the input colors, which is called the relative reflectance in the retinex theory, is altered by the different intensity changes corresponding to the input colors. For example, the red data for the red color are less stretched than the red data for the white color because of the different values for each type of data, and this process changes the relative reflectance values of the red data for the red and white colors. Thus, the conventional methods produce the perceived color differences compared to RGB displays from the retinex point of view. III. PROPOSED METHOD The main purpose of the proposed method is to reduce color distortions and to increase intensities as much as possible after RGB-to-RGBW conversion. As the gain of the pixels increases at a uniform rate, color preservation and the increase in intensity can be achieved simultaneously to the point where data clipping of the pixels starts to appear. Beyond this point, the intensities of all pixels cannot be increased at a uniform rate. Because RGBW displays have a wide space along the direction of their achromatic color axis, achromatic pixels have relatively larger margins that can be stretched without data clipping than chromatic pixels. Thus, achromatic pixels need to have additional gains to maximize the intensities of the RGBW display. However, these non-uniform gains between chromatic and achromatic pixels begin to preclude color preservation and force a trade-off between color preservation and intensity increment. The proposed method is mainly composed of two stages, color-preserving conversion (CPC) and retinex-optimized control(roc),asshowninfig.3.inthecpcstage,rgbsignals are converted into RGBW signals and uniform gain is applied scene-adaptively considering the data clipping of the pixels. In (9)

4 KWON AND KIM: SCENE-ADAPTIVE RGB-TO-RGBW CONVERSIO 687 Fig. 3. Block diagram of the proposed method. (a) Color-preserving conversion stage. (b) Retinex-optimized control stage.d (c) Data modulation stage. the ROC stage, additional gains are calculated pixel by pixel considering the chromaticness of each pixel, and the trade-off between color preservation and intensity increase is quantified and adjusted by retinex theory. A. Color-Preserving Conversion Stage The major roles of CPC are to convert RGB signals into RGBW signals and then to calculate the maximum gain that generates clipped pixels within a pre-defined level. In CPC, a color distortion does not occur except for clipped pixels, because the RGBW signals are acquired in a spectrum-exchange manner, and the maximum gain is applied uniformly to all RGBW intensity values. The first step of the CPC is de-gamma processing. As the input data increase linearly, the light intensities increase non-linearly because the data are processed with gamma values in the display system. Thus, these input data must be converted into linearly scaled intensities before the application of linear operations such as addition, subtraction, and multiplication, which are used in RGB-to-RGBW conversion. If input bit widths are assumed to be 8 bits, the linearly scaled intensities are described as follows: (10) where is the linearly scaled light intensity, represents the input data, and is the gamma value used in the display system. All notations represent,,and. This power-law gamma is just for the simple description of the linearization process. For real implementation, the piecewise-defined functions described in ITU Rec. 709 [2], [22] can be used. In the second step, these RGB light intensities are weighted to have the same scale as the RGB light intensities coming from the white sub-pixel. Let the ratio of the maximum light intensities at a red wavelength coming from a red sub-pixel and white subpixel be defined as and let and be defined in the same way. Then, the weighted light intensities are calculated as - (11) and can be determined by the properties of a white color resinorbythemanufacturingprocessofthewhitecolorfilter [14]. In the third step, the white light intensity is extracted from common components of the weighted RGB intensities because the white light intensity includes all of the light intensities with red, green and blue wavelengths. The amount of common components extracted from RGB intensities is controlled to obtain the maximum frame gain without clippings in the next step; if all of the common components are extracted into the white intensity, the white intensity becomes too large to obtain a large frame gain. Conversely, if very few of the common components are extracted into the white intensity, the remaining RGB intensities also become too large. Thus, to minimize the maximum of the converted intensities, the following rules are applied: - (12) where and are the minimum and maximum values of -, - and - for each pixel. After extracting the white intensity, the remaining RGB intensities are converted again to the original scale using. - - (13) where - denotes the light intensity of the primary colors with the original scale to meet their physical transmittance properties.

5 688 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 In the final step of CPC, the frame gain is acquired sceneadaptively. To maintain the color of the scene, a uniform gain has to be applied to all the pixels in a frame. As this uniform gain increases, the intensity of the scene also increases, but data clipping begins to occur in parts of the scene. Thus, the maximum gain that does not generate clipping artifacts must be found. If data clipping is not allowed, the maximum data of the frame determine the frame gain as follows: (14) where denotes the frame gain and denotes the maximum data of the frame. However, several approaches to obtaining a higher frame gain than that in the no-clipping case by allowing imperceptible amounts of clipping have been proposed [15], [16]. Although all of these techniques can be applied to increase the frame gain, the fixed-rate clipping method was chosen in this study because it is easy to implement and the clipping-permission methods are not the focus of this study. For fixed-rate clipping, first, the maxima of -, -, - and - for each pixel are selected, and the number of maximum data with level is counted in the bin,. After an entire frame is scanned, a histogram of the maximum data for each pixel is constructed, and the modified maximum data of the frame,,using the fixed-rate clipping method is calculated as (15) where denotes the pre-defined number of clipped pixels. Then, the modified frame gain,, is also calculated as (16) Because is smaller than, is larger than. Finally, is clipped to the pre-defined value to avoid an excessive dynamic range in the frame gain. B. Retinex-Optimized Control (ROC) Stage In the ROC, the additional gain for each pixel is calculated and applied considering its chromaticness in order to increase the light intensity. However, as the gains of the pixels are different from one another, color distortion appears. This color distortion is quantified by calculating the differences in the designators between RGBW conversion data and original RGB data using retinex theory. Then, the amount of the additional gain is controlled using the feedback regarding the differences in the designators. This process maintains the color distortion below a pre-defined level while simultaneously displaying brighter scenes. The first step of the ROC stage is to calculate the chromaticness of each pixel as follows: (17) where and are the minimum and maximum values of in each pixel, respectively. After the chromaticness of each pixel is calculated, the additional gain,, is acquired differentially for the chromaticness and the additional gain portion Fig. 4. (a) Frame gain ( ) and (b) overall gain ( ) applied to each pixel according to its chromaticness. (AGP) which is determined by the feedback values using retinex theory, as shown in Fig. 4 and (18). For a given AGP value, additional gain increases more for the lower chromatic data to reflect the asymmetric color space of the RGBW display. Then, the overall gain is computed as follows: (18) (19) In the second step, the color distortion is evaluated to control the AGP value for the next frame by a feedback mechanism. In retinex theory, the color distortions perceived by human eyes are calculated as the distances in the relative reflectance between original colors and processed colors in the 3D RGB color space. Thus, to calculate the distances, we first compute the average relative reflectance of the original and the processed images, which stands for the perceived color affected by an entire scene. If the illumination is assumed to be uniform, we can ignore the threshold operations in (4), and it is reduced to an independent form of the path from to : (20) where and denote the intensities of the initial and final unit areas along the path from to,i.e., and, respectively. From (20), we can compute the relative reflectance of a unit area affected by a unit area, although it is remote from area. Thus, the average relative reflectance of area is (21) where is the total number of unit areas. If a unit pixel is assumed to be the unit area and therefore the intensity of the pixel in the waveband isassumedtobe, then the intensity of the th pixel of an RGBW display,, can be expressed as follows: (22) where is the ratio of the primary pixel s size between the RGB and RGBW display, approximately 0.75, and is the overall gain of the th pixel of the RGBW display. From (21) and (22),

6 KWON AND KIM: SCENE-ADAPTIVE RGB-TO-RGBW CONVERSIO 689 Fig. 5. Comparison of input data transfer in various conversion methods. (a) Data transfer by conventional methods. (b) Data transfer by applying a frame gain. (c) Data transfer by the entire proposed method. Fig. 6. Kodak Lossless True Color Image Suite. (a) 001. (b) 002. (c) 003. (d) 005. (e) 006. (f) 004. (g) 007. (h) 008. (i) 011. (j) 012. (k) 013. (l) 014. (m) 015. (n) 016. (o) 020. (p) 021. (q) 009. (r) 022. (s) 023. (t) 024. (u) 017. (v) 018. (w) 019. (x) 010. the average relative reflectance of the unit area, i.e., the th pixel, is Because the gain is independent of the waveband, the average color distortion (ACD) ofthergbwdisplayis (25) (23) where is the average of the logarithmic values of the gains in a frame. Thus, the absolute difference,,ofthetwo designators of the RGB and RGBW displays at the th pixel is computed using the gains of each pixel of the RGBW display as follows: (24) ACD comes from the assumptions for the human visual characteristics in the retinex theory, and ACD value means how much different the perceived colors between the original and processed images are in terms of human perception. The ACD value is different from the other CIE metrics such as CIELAB and CIEDE2000 in that it represents the perceived color differences considering not only the color of an object, but also the color of surroundings. In the third step, the AGP for the next frame is calculated using the level of the ACD in the current frame. If the target colordistortionisassumedtobe,theagp is computed as where the frame, the (26) is the AGP value applied to the current is the AGP value that will be applied

7 690 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 Fig. 7. Intensity increment results using (30) for various RGB-to-RGBW conversion methods for Kodak Lossless True Color Image Suite. Fig. 8. Color distortion results using (25) for various RGB-to-RGBW conversion methods for Kodak Lossless True Color Image Suite. to the next frame, and is a proportional constant controlling the feedback speed. Finally, the AGP value is clipped to (Fig. 4) to prevent an excessive dynamic range in the pixel gain and the resulting value is used in (18). C. Data Modulation Stage The frame gain is computed in the CPC stage considering data clipping, and the additional gain is computed for each pixel considering its chromaticness and color distortion over the scene. Then, the overall gain is applied to - and - : - - (27) After applying the overall gains, gamma processing is performed to convert the linearly scaled intensity into gamma-processed data: (28) where and are the final output data of the RGBW display, respectively. Data transition in various conversion methods is compared in Fig. 5. The conventional methods transfer the input data based on their chromaticness [see Fig. 5(a)]. Thus, the result for the pure chromatic data ( ) is located at the same position as the input data ( ), and that of achromatic data ( ) is maximally stretched ( ). The input data with intermediate chromaticness ( ) are transferred differently according to the method used to extract white data as described in (7). The CPC stage of the proposed method transfers all of the input data at the same rate using a frame gain [Fig. 5(b)]. Then, additional gains are applied to each pixel as controlling the amount of the color distortion basedonretinextheory [Fig. 5(c)]. D. Video Sequence Consideration The proposed method can be applied to the video sequences with scene changes as well as with static scenes by using a temporal filter with the scene change detection function [23] [25]. The temporal filter should be inserted into the steps determining the frame gain and AGP value to suppress their rapid changes between successive frames. The temporal filter that we typically use for video sequences has the structure of an infinite impulse response as follows. (29) where denotes the immediate values which are determined by (16) in the case of the frame gain or by (26) in the case of the AGP value, and and denote their low-pass filtered values in the next and current frames, respectively. is a designer preference value between 0 and 1 controlling the speed of the temporal filter. If the scene change is detected, the filtered value, is reset to the immediate value, to stop filtering between uncorrelated adjacent frames. Various scene change detection algorithms [23] [25] can be adopted to our method, because the scene change detection and RGB-to-RGBW conversion are independent processes. IV. EXPERIMENTAL RESULTS The color distortion and intensity increment were evaluated using the Kodak Lossless True Color Image Suite shown in Fig. 6, and 999 images sampled from a moving picture

8 KWON AND KIM: SCENE-ADAPTIVE RGB-TO-RGBW CONVERSIO 691 Fig. 9. Simulation results for image 015. (a) Original image on the RGB display, and RGBW images converted using (b) benchmark1, (c) benchmark2, (d) benchmark3, (e) benchmark4, and (f) the proposed method. in IEC62087 international standard. The four conventional methods discussed in Section II were used as benchmarks. Because it is difficult to directly compare color distortion and intensity increment between RGBW and RGB displays, the intensity of the RGBW display needs to be converted to the intensity of the RGB display, with compatible spectral densities for each primary color. Thus, if the intensities of the pixel of the RGB display are assumed to be,,and,the intensities of the RGBW display can be calculated as,,and, respectively. We used 0.75 for in this experiment. The intensity increment ratio (IIR)oftheRGBWdisplayrelative to the RGB display was computed as follows: (30) The average color distortion (ACD) was also calculated using (25) for the proposed and benchmark methods. We used some pre-defined values for the following criteria: was 3; was 2, the number of clipped pixels,, was 1% of the total number of pixels in the image; the display gamma,, was 2.2; the threshold of the color distortion,, was The maximum intensities from a white sub-pixel and RGB primary sub-pixels of the RGBW display were both assumed to be 75% of those of the RGB display, and the chromaticity coordinates were assumed to be (0.680, 0.310), (0.210, 0.700), (0.147, 0.054), and (0.313, 0.329) in RGBW orders. With these assumptions,,,and were set to 1, the intensity of pure white in the RGBW display was 150% of that in the RGB display, and the intensity of the RGB primary colors in the RGBW display was 75% of that in the RGB display. Because all of the simulation results should be displayed in RGB displays such as normal LCD monitors, the intensities of all the images were scaled to the same basis by normalizing the maximum intensity of the RGBW display in order to display and compare the resulting images. The RGB intensities of the RGB display,, were scaled to ;the RGB intensities of the RGBW display,, were scaled to. The simulation results for the Kodak Lossless True Color Image Suite are shown in Figs. 7 and 8. The proposed method Fig. 10. Simulation results for image 009. (a) Original image on the RGB display, and RGBW images converted using (b) benchmark1, (c) benchmark2, (d) benchmark3, (e) benchmark4, and (f) the proposed method. generated a greater increase in the intensity increment than the benchmark methods, on average. It also maintained a lower and more constant color distortion, indicating that the proposed method can guarantee uniform image quality over successive scenes. However, the large standard deviation of IIR value is a negative effect because it means the large variation in brightness according to the input sequences. We can suppress this visual effect by using the temporal filtering described in the proposed method section and the backlight dimming in LCDs with inversely proportional operation to the intensity increment for power savings. The average results for the different conversion methods are summarized in Table I. Some examples of the original and RGBW-converted images are shown in Figs. 9 and 10 for image quality comparisons. Asshowninthefigures, simultaneous contrast effects tend to exaggerate the color errors created by benchmark methods. However, the proposed method stretched all the colors uniformly by applying a frame gain and adjusted additional gains as the color distortion was maintained under the controlled

9 692 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 Fig. 11. Intensity increment results using (30) for various RGB-to-RGBW conversion methods for randomly sampled IEC62087 images. Fig. 12. Color distortion results using (25) for various RGB-to-RGBW conversion methods for randomly sampled IEC62087 images. TABLE I INTENSITY INCREMENT (TIMES) AND COLOR DISTORTION RESULTS OF VARIOUS CONVERSION METHODS FOR KODAK IMAGES TABLE II INTENSITY INCREMENT (TIMES) AND COLORDISTORTION RESULTS OFVARIOUS CONVERSION METHODS FOR RANDOMLY SAMPLED IEC62087 IMAGES level. Thus, the resulting images processed by the proposed method looked brighter and less distorted. Secondly, we simulated the effectiveness of the proposed method for randomly sampled pictures because the intensity increment efficiency of the RGBW display depends on the scenes displayed. We used the sampled images from IEC62087 international standard moving pictures, which are used for measuring the power consumption of TVs. As shown in Figs. 11 and 12, the simulation results for the randomly sampled moving pictures were very similar to those for the Kodak still images. The average intensity increment of the proposed method was slightly higher than that for the still-image case (Table II) because the moving pictures were composed of less-saturated scenes, which can have a relatively larger frame gain. Finally, we performed the psychological testing for the RGBW images converted by the proposed and benchmark methods to evaluate the perceived color consistency with the original RGB images. After normalizing the RGBW-converted images with the IIR values for each image and each method, we compared them with the original RGB images. We used 5-point scale measurement and nine people participated in the subjective evaluation test. As showninfig.13andtableiii, these subjective evaluation results were very consistent with the results of the color distortion measurement using ACD described in Fig. 8 and Table I. In this experiment, the limit of color distortion ( ) were set to under 0.04, and the proposed method acquired 4.61 mean opinion score (5: imperceptible; 4: perceptible, but not annoying) while most of conventional

10 KWON AND KIM: SCENE-ADAPTIVE RGB-TO-RGBW CONVERSIO 693 Fig. 13. Subjective evaluation results about the perceived color consistency with the original RGB images for various RGB-to-RGBW conversion methods for Kodak Lossless True Color Image Suite. TABLE III SUBJECTIVE TEST RESULTS WITH 5-POINT OPINION SCORES OF VARIOUS CONVERSION METHODS FOR KODAK IMAGES methods acquired around 3 (slightly annoying). Thus, this psychophysical evaluation result can be one example that helps us understand the relations between ACD values and real psychophysics. V. CONCLUSION RGBW displays have the benefit of better transmittance characteristics than RGB displays, but their asymmetric color space makes RGB-to-RGBW conversion complicated. This paper proposed a novel RGB-to-RGBW conversion method that uses the asymmetric color space of RGBW displays effectively. It also presented a technique that uses retinex theory to quantify the amount of color distortion that appears in the form of the simultaneous contrast. The proposed method defined the perceived color distortion of RGBW displays from the HVS point of view, and controlled this color distortion under the imperceptible level as maximizing the intensities for a given image. In experimental results, the proposed method allowed for a higher increase in intensity than the benchmark methods while maintaining a lower and more constant level of color distortion. The intensity increment of the proposed method can be used to reduce the power consumption of LCD backlight units, employing dynamic backlight scaling techniques [17], or to provide brighter images on LCDs. REFERENCES [1] S. Yang, J. Heikenfeld, E. Kreit, M. Hagedon, K. Dean, K. Zhou, S. Smith, and J. Rudolph, Electrofluidic displays: Fundamental platforms and unique performance attributes, J. SID, vol.19,no.9, pp , Sep [2] M. E. Miller and M. J. Murdoch, RGB to RGBW conversion with current limiting for OLED displays, J. SID, vol. 17, no. 3, pp , Mar [3] E.H.A.Langendijk,O.Belik,F.Budzelaar,andF.Vossen, Dynamic wide-color-gamut RGBW display, in SID Symp. Dig. Tech. Papers, 2007, vol. 38, pp , 1. [4] M. Ito, M. Kon, C. Miyazaki, N. Ikeda, M. Ishizaki, Y. Ugajin, and N. Sekine, Front drive display structure for color electronic paper using fully transparent amorphous oxide TFT array, IEICE Trans. Electron., vol. E90 C, no. 11, pp , Nov [5] Y. Xiong, L. Wang, W. Xu, J. Zou, H. Wu, Y. Xu, J. Peng, J. Wang, Y. Cao, and G. Yu, Performance analysis of PLED based flat panel display with RGBW sub-pixel layout, Org. Electron., vol. 10, no. 5, pp , Aug [6] R. Lu, Q. Hong, Z. Ge, and S. Wu, Color shift reduction of multidomain IPS-LCD using RGB-LED backlight, Opt. Express, vol. 14, no. 13, pp , [7] C.LaiandC.Tsai, Amodified stripe-rgbw TFT-LCD with imageprocessing engine for mobile phone displays, IEEE Trans. Consum. Electron., vol. 53, no. 4, pp , Nov [8] L. Wang, Y. Tu, and L. Chen, Trade-off between luminance and color in RGBW displays for mobile-phone usage, in SID Symp. Dig. Tech. Papers, 2007, vol. 38, pp , 1. [9] J.G.R.MourikandJ.H.C.J.Stessen, Methodofdrivingdisplays comprising a conversion from the RGB color space to the RGBW color space, U.S. Patent2008/ A1,, [10] S. Lee, C. Kim, Y. Seo, and C. Hong, Color conversion from RGB to RGB+white while preserving hue and saturation, in Proc. IS&T/SID 10th Color Imag. Conf., 2002, pp [11] E. H. Land and J. J. McCann, Lightness and retinex theory, J. Opt. Soc. Amer., vol. 61, no. 1, pp. 1 11, Jan [12] E. H. Land, Recent advances in retinex theory, Vision Res., vol.26, no. 1, pp. 7 21, [13] E. H. Land, Recent advances in retinex theory and some implications for cortical computations: Color vision and the natural image, in Proc. Nat. Acad. Sci. USA, Aug. 1983, vol. 80, pp [14] S. Y. Noh, J. P. Kim, S. R. Park, J. Y. Yang, M. S. Yang, I. B. Kang, and I. J. Chung, Advanced RGBW panel with high planarization overcoat material, in SID Symp. Dig. Tech. Papers, 2009, vol. 40, pp , 1. [15] N. Chang, I. Choi, and H. Shim, DLS: Dynamic backlight luminance scaling of liquid crystal display, IEEE Trans. Very Large Scale Integr. (VLSI) Syst., vol. 12, no. 8, pp , Aug [16] S. Kang and Y. H. Kim, Image integrity-based gray-level error control for low power liquid crystal displays, IEEE Trans. Consum. Electron., vol. 55, no. 4, pp , Nov [17] M. A. Kao, P. Hsieh, and H. Lin, Dynamic backlight control method for RGBW LCD, in SID Symp. Dig. Tech. Papers, 2011, vol. 42, pp , 1. [18] N. Shlayan, R. Venkat, P. Ginobbi, and A. K. Singh, Energy efficient RGBW pixel configuration for light-emitting displays, J. Display Technol., vol. 5, no. 11, pp , Nov [19] L. Chen, Y. Tu, L. Wang, and F. Li, Perceptual evaluation of sub-pixel rendering in a four-primary display system, in 2nd Int. Congr.Image and Signal Process., Oct. 2009, pp [20] Y. Kwak, J. Park, D. Park, and J. Park, Generating vivid colors on red-green-blue-white electronic-paper display, Appl. Opt., vol. 47, no. 25, pp , [21] H. Qin, S. Wang, H. Lu, and X. Chen, Human-inspired order-based block feature in the HSI color space for image retrieval, in IEEE Int. Conf. on Robot. Biomimetics, Dec. 2009, pp

11 694 JOURNAL OF DISPLAY TECHNOLOGY, VOL. 8, NO. 12, DECEMBER 2012 [22] S. Juric and V. Klepac, Gamma and gamma correction in television production, in Int. Symp. ELMAR, 2009, pp [23] C. Huang and B. Liao, A robust scene-change detection method for video segmentation, IEEE Trans. Circuits Syst. Video Technol., vol. 11, no. 12, pp , Dec [24] I. K. Sethi and N. Patel, A statistical approach to scene change detection, in SPIE, 1995, vol. 2420, pp [25] B. L. Yeo and B. Liu, Rapid scene analysis on compressed video, IEEE Trans. Circuits Syst. Video Technol., vol. 5, no., pp , month? Kyung Joon Kwon (M XX) received the B.S. degree in electrical engineering from Seoul National University, Seoul, Korea, in He has been working for LG Display, Seoul, Rep. of Korea, since 2003, and he is currently working toward the Ph.D. degree in electronic and electrical engineering at Pohang University of Science and Technology, Gyeongbuk, Korea. His research interests include image-analysis methodology, real-time image processing, and HVS-based image-quality assessment. Young Hwan Kim (S 86 M 89) received the B.E. degree in electronics from Kyungpook National University, Daegu, Korea, in 1977, and the M.S. and Ph.D. degrees in electrical engineering from the University of California, Berkeley, in 1985 and 1988, respectively. He is currently a professor of electronic and electrical engineering at Pohang University of Science and Technology, Gyeongbuk, Korea. His research interests include the design of plasma display panel systems and liquid crystal display systems, multimedia circuit design, MPSoC and GPGPU system design for multimedia applications, statistical analysis and design technology for deep-submicron semiconductor devices, and power noise analysis. Dr. Kim has served as editor of the Journal of the Institute of Electronics Engineers of Korea and as General Chair and a committee member of various Korea domestic and international technical conferences. He is currently serving as the TPC of IEEE International Symposium of Circuits and Systems (ISCAS) 2012.

12 本文献由 学霸图书馆 - 文献云下载 收集自网络, 仅供学习交流使用 学霸图书馆 ( 是一个 整合众多图书馆数据库资源, 提供一站式文献检索和下载服务 的 24 小时在线不限 IP 图书馆 图书馆致力于便利 促进学习与科研, 提供最强文献下载服务 图书馆导航 : 图书馆首页文献云下载图书馆入口外文数据库大全疑难文献辅助工具