Power Minimization in a Backlit TFT-LCD Display by Concurrent Brightness and Contrast Scaling

Transcription

1 Power Minimization in a Backlit TFT-LCD Display y Concurrent Brightness and Contrast Scaling Wei-Chung Cheng, Yu Hou, Massoud Pedram Department of EE-Systems, University of Southern California Los Angeles, CA 989, USA {wccheng, yhou, massoud@sahand.usc.edu Astract This paper presents a Concurrent Brightness and Contrast Scaling (CBCS) technique for a cold cathode fluorescent lamp (CCFL) acklit TFT-LCD display. The proposed technique aims at conserving power y reducing the acklight illumination while retaining the image fidelity through preservation of the image contrast. First, we eplain how CCFL works and show how to model the non-linearity etween its acklight illumination and power consumption. Net, we propose the contrast distortion metric to quantify the image quality loss after acklight scaling. Finally, we formulate and optimally solve the CBCS optimization prolem with the ojective of minimizing the fidelity and power metrics. Eperimental results show that an average of 3.7X power saving can e achieved with only % of contrast distortion. Introduction Previous studies on attery-powered electronics point out that the cold cathode fluorescent lamp (CCFL) acklight of an LCD display dominates the energy consumption of the whole system []. In the SmartBadge system, for instance, the display consumes 28.6%, 28.6%, and 5% of the total power in the active, idle, and standy modes, respectively []. To reduce the power consumed y the acklight, researchers [2][3] have proposed the concept of Backlight Scaling. The acklight scaling technique dynamically dims the acklight to conserve its power consumption while increasing the transmissivity of the LCD panel to compensate for the image fidelity loss due to reduced acklight. Image fidelity is defined as the resemlance etween the original and acklight-scaled image. If the acklight-scaled image is identical to the original image in terms of the rightness of each piel (these approaches are called rightness-invariant acklight scaling), then there is no fidelity loss after acklight scaling. However, when a more aggressive acklight scaling policy is used to gain greater power savings, the rightness invariance is no longer attainale and the induced distortion degrades the image fidelity. Image fidelity can e measured y the rightness variance after acklight scaling. However, the This research was supported in part y DARPA PAC/C program under contract DAAB7-2-C-P32 and y NSF under grant no rightness metric is too strict for efficacious acklight scaling policies. Since dimming the acklight directly limits the dynamic range of the image rightness, a rightness-invariant acklight scaling policy is usually too conservative to deliver great energy savings. In this paper, we propose using the image contrast as a metric to measure the image fidelity after acklight scaling.. Terminology The following photometric quantities are illustrated around a acklit TFT-LCD display in Figure. Luminous flu (lumen) is the emission rate of light energy corrected for the standardized spectral response of human vision. Luminous intensity (candela) is defined as one lumen of luminous flu per steradiam (sr -- unit of solid ane). Luminous intensity can e used to characterize the optical power emitted from a spot light source, such as a light ul. Illuminance (lu) is defined as one lumen of luminous flu per area (cd/m 2 ). Illuminance can e used to characterize the luminous power emitted from a surface. Most light meters (e.g., for photographic purpose) measure the illuminance quantity. The luminous flu may not travel in parallel after passing the surface, so that the light intensity decreases as the travel distance increases. Luminance (nit) is defined as lumen per area per steradiam (lm/m 2 /sr). Luminance is used to rate the maimum rightness of CRT or LCD monitors [4]. Luminous Intensity candala(cd)= lumen(lm)/sr Luminous Flu lumen(lm) Illuminance lu=lumen(lm)/m 2 Reflector CCFL TFT-LCD Panel Luminance nit=lumen(lm)/m 2 /sr Light meter Figure : Illustration of CCFL acklight and photometric terms. In this paper, we use acklight factor to epress the percentage of the acklight illumination, and transmissivity to epress the translucence of the TFT-LCD. The acklight factor and TFT-LCD transmissivity determine the perceived luminance from the TFT-LCD display..2 Backlit TFT-LCD display The major components of a acklit (or transmissive) TFT- LCD display susystem include the video controller, frame uffer, video interface, TFT-LCD panel, and acklight. The 53-59/4 $2. (c) 24 IEEE

2 frame uffer is a portion of memory used y software applications to deliver video data to the video controller. The video data from the application is stored in the frame uffer y the CPU. The video controller fetches the video data and generates appropriate analog (VGA) or digital (DVI) video signals to the video interface. The video interface carries the video signals etween the video controller and the TFT-LCD display. The TFT-LCD display receives the video data and generates proper shade transmissivity for each piel according to its piel value. All of the piels on the transmissive LCD panel are illuminated y the acklight from ehind. To the oserver, a displayed piel looks right if its transmissivity is high (i.e., in the 'on' state), meaning it passes the acklight. On the other hand, a displayed piel looks dark if its transmissivity is low (i.e., in the 'off' state), meaning it locks the acklight. If the transmissivity can e adjusted to more than two different levels etween the 'on' and 'off' states, then the piels can e displayed in grayscale. If the shade can e colored as red, green, or lue y using different color filters, then piels can e displayed in color y miing three su-piels in different colors at different grayscales. In other words, the perceived rightness of a piel is determined y its transmissivity and the acklight illumination. Most of current TFT-LCD displays use CCFL acklighting thanks to its unrivaled luminance density emitting the most light within the minimum form factor. The CCFL can e designed to generate aritrary color, which is critical to reproducing pure white in the acklighting applications. The technology of manufacturing CCFL is mature so that its cost has een minimized. The power consumption of the CCFL acklight, however, is consideraly high compared with that of the TFT- LCD panel. The oserved luminance of a transmissive oject L is the product of the luminance of the light source and the transmissivity of the oject t [5]. For a piel on a acklit TFT- LCD display, its transmissivity is a function of its piel value. Thus, its oserved luminance L is L = t() () The amient light is not considered here ecause it has little effect for a transmissive TFT-LCD when compared with a reflective or transflective one. Figure 2 depicts the relation in Equation () assuming that the transmissivity is a linear function of the piel value. t * = CCFL Backlight Factor TFT-LCD Transimissivity Function Luminance Function Figure 2: The luminance function of normalized piel value (right) is the product of the acklight factor and the TFT-LCD transmissivity function (center). In a non-acklight-scaled TFT-LCD display, the acklight is always fied at full power. The acklight scaling techniques in [2][3] reduce while increasing the piel value from to ' y =+ (2) =/, (3) to maintain the same L. These approaches have the following drawacks: Equation (2) cannot preserve rightness invariance according to Equation (). The contrast distortion among the unsaturated piels is not considered. The software-ased approach has high energy/performance overhead. The CCFL illumination is incorrectly modeled as a linear function of power. In this paper, we propose solutions to surmount the aovementioned drawacks. In Section 2 we characterize the CCFL illumination and power consumption. In Section 3 we propose adjusting the transmissivity function t rather than the piel value. The optimal CBCS prolem is introduced in Section. Section 5 presents the eperimental results followed y conclusions in Section 6. 2 CCFL Illumination and Power Modeling A CCFL acklight unit consists of the fluorescent lamp, the driving DC-AC inverter, and the light reflector. A CCFL is a sealed ass tue with electrodes on oth ends. The tue is filled with an inert gas (argon) and mercury. The inner ass surface of the tue is coated with phosphor, which emits visile light when ecited y photons. The wavelength or color of the visile light depends on the type of the gas and phosphor. In the LCD acklighting application, a proper mi of red, green, and lue phosphors produces the desired three-and white light. Otherwise, the displayed image will e color-shifted. The CCFL converts electrical energy into visile light, which is called the gas discharge phenomenon. When a high voltage is applied to the electrodes turning on the lamp, electrical arcs are generated that ionize the gas and allow the electrical current to flow. The collision among the moving ions injects energy to the mercury atoms. The electrons of the mercury atoms receive energy and jump to a higher energy level followed y emitting ultraviolet photons when falling ack to their original energy level. The ionized gas conducts the electrical current. The impedance of the gas conductor, unlike that of the metal conductor having a linear ehavior, decreases as the current increases. Therefore, the CCFL has to e driven y an alternative current (AC) to avoid a potential eplosion. A DC-AC inverter is usually used to drive a CCFL in atterypowered applications. A DC-AC inverter is asically a switching oscillator circuit that supplies high-voltage AC current from a low-voltage attery. The nominal AC frequency of modern CCFL is in the range of 5- khz to avoid flickering. The nominal operate voltage has to e higher than 5 V RMS to keep inert gas ionized. To conserve energy in attery-powered applications, dimming control is a desired feature for DC-AC inverters. Different methods of dimming CCFL have een used, including linear current, pulse-width-modulation, and current chopping [6]. In a DC-AC inverter with dimming control, an analog or digital input signal is eposed for adjusting the CCFL illumination. Most well designed DC-AC inverters have high electrical efficiency (>8%) and linear response of output electrical power to input power. Most fluorescent lamps, however, have low optical efficiency (<2%) and non-linear response of output optical power versus input power [7].

3 2. CCFL Characteristics The CCFL illumination is a comple function of the driving current, amient temperature, warm-up time, lamp age, driving waveform, lamp dimensions, and reflector design [7]. For CBCS, only the driving current is controllale. Therefore, we model the CCFL illumination as a function of the driving current only and ignore the other parameters. The typical relationship etween the CCFL illumination and the driving power is shown in Figure 3a. The CCFL illumination increases monotonically as the driving power increases efore reaching 8% of the full driving power. Beyond 8%, the CCFL illumination starts to saturate. This saturation phenomenon is ecause the enclosed ionized gas has een fully discharged and cannot release more photons. Additionally, the increased temperature and pressure inside the tue inhiit further discharge [7][8]. This oservation suggests that the decreased optical efficiency of CCFL in the saturated region is not favored y power-aware applications. 2.2 CCFL Illumination/Power Characterization We use a stepwise function of illumination to characterize the power consumption of CCFL as a function of illumination: P C, B ( ) Lin + Lin P = s acklight (Watt). (4) P Sat + CSat, Bs The acklight factor [,] represents the normalized acklight illumination, which is dynamically controllale y the CBCS policy. The analog or digital dimming control input of the DC-AC inverter is not always linearly proportional to the output acklight illumination. Careful caliration is needed to derive the correct mapping etween the acklight factor and dimming control input q(). A precision luminance meter such as that in [9] provides accurate asolute illuminance readings. These epensive meters, however, are commonly unavailale to electronic laoratories. We find that the asolute illuminance readings are not required to calirate the CCFL in the acklight scaling applications. An accurate photographic light meter can serve the purpose so far as it is capale of sensing minor luminance variance. We use the light meter as a weight scale and adjust the acklight and TFT-LCD simultaneously while maintaining the same illuminance. We start with measuring the illuminance for the maimum CCFL acklight = when applying dimming control q(=) and minimum LCD transmissivity = [,255]. The transmissivity is oained y displaying a pure gray image, in which Red=Green=Blue= for every piel. The transmissivity is increased until the light meter can sense a variation and report a different reading. Then reduce the acklight factor y reducing the dimming control q until the meter reports the previous reading. Since the change of the TFT- LCD grayscale (transmissivity) is known, the change of the acklight is asserted to e the same. Record q as the dimming control value for the acklight factor =(255-)/256. At the same time, the power consumption of the acklight P acklight is also measured and recorded. Repeat the aove procedure for = After interpolation, we can oain q() and P acklight (). The results for a color acklit TFT-LCD [] are shown in Figure 3a. Plugging into Equation 4, the following parameters are oained: P Lin=.499, P Sat=.489, C Lin=.3, C Sat=.69, B s=.8666 (5) This power model will e incorporated in Section to solve the optimal CBCS prolem. Normalized CCFL Luminance Power Consumption (W) Figure 3: (a) Luminance/Power characterization of CCFL () Transmissivity/Power characterization of TFT-LCD panel. 3 TFT-LCD Grayscale Control and Power Modeling In a TFT-LCD display, each su-piel has an individual liquid crystal cell, a thin-film-transistor (TFT) and a capacitor. The electrical field of the capacitor controls the orientation of the liquid crystals within the cell, which indeed determines the transmissivity. The capacitor is charged and discharged y its own TFT. The gate electrode of the TFT controls the timing of charging/discharging when the piel is scanned for refreshing its content. The source electrode of the TFT controls the amount of charge that determines the transmissivity of the liquid crystal cell. The gate electrodes and source electrodes of all TFTs are driven y a set of gate drivers and source drivers, respectively. A sine gate driver drives all gate electrodes of the piels on the same row. The gate electrodes are enaled at the same time the row is scanned. A sine source driver drives all source electrodes of the piels on the same column. The source driver supplies the desired voltage level (called grayscale voltage) according to the piel value. In other words, ideally, the transmissivity t(v()) is a linear function of the grayscale voltage v(), which is a linear function of the piel value. If there are 256 grayscales, then the source driver must e ale to supply 256 different grayscale voltage levels. For the source driver to provide a wide range of grayscales, a numer of reference voltages are required. The source driver mies different reference voltages to oain the desired grayscale voltages. Typically, these different reference voltages are fied and designed as a voltage divider. For eample in [], an LCD reference driver [] is used with a -way voltage divider. Assume that the transmissivity of the TFT-LCD is linear and the resistors of the voltage divider are identical. If k identical resistors r r k are connected in series etween V k and ground, then the output voltage from r k is i Vi = V. k (6) k 3. Programmale LCD Reference Driver Our approach to CBCS is to control the mapping of v() in order to control the transmissivity function t(). We propose using a programmale LCD reference driver (PLRD) descried as follows. The PLRD is implemented y adding an etra logic to the original voltage divider epressed y Equation (6). The logic contains a numer of p-channel and n-channel switches and multipleers. The PLRD takes two input arguments and gu, Normalized Transmissivity Normalized TFT-LCD Power

4 and then connects r gu, r gu+ r k to V k and r, r r to ground. In this way, the output voltage seen from r k ecomes Vk, gu i k ' i V,, k, i gu. V = igu < (7) gu, i Clearly, the PLRD performs a linear transformation (limited y and V k ) on the original reference voltages, and therefore, provides the CBCS policy a mechanism for adjusting the TFT- LCD transmissivity function as shown in Figure 4a. The luminance function is shown in Figure 4. * (a) Figure 4: The transmissivity function (a) and luminance function () when using a programmale LCD reference driver. The similar concept of PLRD has een implemented in TFT- LCD controllers such as [2] to control contrast. The PLRD represents a class of linear transformations on the acklightscaled image. It covers oth rightness scaling (adjusting gu and simultaneously) and contrast scaling (adjusting gu-). On the other hand, non-linear transformations are not desired in acklight scaling ecause they cannot preserve the uniformity of contrast. 3.2 TFT-LCD Power Characterization The TFT power can e modeled y a quadratic function of piel value [,255] [3]: P TFT ()=c +c +c 2 2 (Watt). (8) We performed the current and power measurements on []. The measurement data are shown in Figure 3. Plugging into Equation 8, the coefficients are found as: c =2.73E-3, c =2.82E-4, c 2 =2.87E-5. (9) The TFT-LCD power consumption decreases as the transmissivity increases. In other words, while maintaining the same luminance, the power consumption of the TFT-LCD decreases when dimming the acklight. In addition, the variation of TFT-LCD power consumption is very small. Therefore, we do not consider the TFT-LCD power consumption in the CBCS framework. t (a) Original () 5% contrast (c) 5% rightness (d) 5% CBCS Figure 5: Luminance functions and visual effects of adjusting rightness (), contrast (c), and oth (d) when the acklight is dimmed to 5%. = gu t=c+d () gu 4 Optimal CBCS Policy Prolem 4. Contrast Fidelity The term contrast descries the concept of the differences etween the dark and right piels. Brightness and contrast are the two most important properties of any image. In the Human Visual System [5][4], which models the perception of human vision as a three-stage processing, the rightness and contrast are perceived in the first two stages. Virtually every sine display permits the users to adjust the rightness and contrast settings. For acklit LCD displays, the rightness control changes the acklight illumination and the contrast control changes the LCD transmissivity function. Figure 5 shows how the rightness and contrast controls change the luminance function and their visual effects when the maimum rightness is limited to 5%. In Figure 5, when the acklight is reduced to 5%, the image contrast is noticealy reduced. If we compensate for the contrast loss as shown in Figure 5c, then the darker (<5%) piels preserve their original rightness while the righter (>5%) piels overshoot completely and there is no contrast present among these piels. Figure 5d shows how the concurrent rightness and contrast scaling generates a etter image y alancing the contrast loss and numer of overshot piels. The luminance function in Figure 5d or Figure 4 represents the following class of linear transformations that can e implemented y the PLRD epressed y Equation (7):, d = t( ) = c d, gu, where c +. d (), gu gu= c Here (,) and (gu,) are the points where y=c+d intersects y= and y=, respectively. The luminance function consists of three regions: the undershot region [,], the linear region [,gu], and the overshot region [gu,]. In other words, the and gu are the darkest and the rightest piel values that can e displayed without contrast distortion (overshooting or undershooting). Notice that the slope of the linear region is very close to that of the original luminance function, which is unity. The image has very few piels in the undershot and overshot regions. Its histogram is shown in Figure 6a. The kernel of CBCS is to find the dissimilarity etween the original and acklight-scaled image, which can e solely determined y eamining the luminance function (). We define the contrast fidelity function as the derivative of ():, < f ( ) = c, gu, c c. (), gu< The c is limited etween and. If c>, the contrast increases and deviates from that of the original image and the dynamic range [,gu] shrinks. The overall contrast fidelity will decrease from this point, so we do not include c> in our solution space. The contrast fidelity is defined without quantifying contrast itself, which has no universal definition [5] and cannot help solve the optimal CBCS policy prolem. However, the definition of contrast fidelity does convey the concept of the classic definitions of contrast such as Weer's or Michelson's that epress contrast as the ratio of the luminance difference to the maimum luminance [5][4][5]. If the normalized image histogram providing the proaility distriution of the occurrence of piel value in the image is given as

5 p() [,], =..255, (2) then the oal contrast fidelity of the acklight-scaled image is defined as gu FC = fc( ) p( ). (3) F c is a function of p, and gu. Finding the optimal solution that minimizes the F c is called the optimal CBCS policy prolem. The oal contrast fidelity captures the rightness distortion due to acklight scaling, also. When the acklight is dimmed, the dynamic range [,gu] is shrunk accordiny, so that more piels have contrast fidelity of zero. 4.2 Contrast fidelity Optimization Prolem To simplify the optimal CBCS policy prolem, our approach is first to find the optimal linear transmissivity function for each given acklight factor, called the contrast fidelity optimization prolem. Then we sweep the acklight factor domain to find the oally optimal solutions. Normalized CCFL Power (a) () (c) Overall Contrast Fidelity (d) (e) (f) Figure 6: (a) Histogram of the eample image () Optimal (left) and +dr (right) as functions of dynamic range dr in the y ais (c) Overall contrast fidelity F c as a function of dynamic range dr for = (upper) and =.5 (lower) (d) Optimal solutions <F c,p acklight> (e) CBCS policy (f) Brightness-invariant policy. Our goal is to find the optimal and gu that maimize the overall contrast fidelity F c. After that, the optimal coefficients c and d can e calculated from Equation (). The optimal transmissivity function t(), which should e applied to the LCD as Figure 4a, can then e determined y, < c+ d t( ) =, gu, (4), gu< and the acklight should e dimmed to concurrently. The optimal solution to the contrast fidelity optimization prolem for an aritrary histogram can e found y the following procedures. Let dr=gu- e the size of the required dynamic range [,gu] and the acklight factor e the size of the availale dynamic range [,]. For each dr, we can find the required dynamic range [,+dr] that maimizes + dr. The optimal is found y p( ) scanning =*256/k, *256/k, (k-)*256/k, where k represents the resolution of the PLRD in Equation (7). Based on the histogram shown in Figure 6a, Figure 6 shows the optimal and +dr in the ais as functions of dr in the y ais. The left and right curves are the optimal and +dr, respectively, for different dr values. This means when the acklight is dimmed to dr, using the availale dynamic range [,dr] from the acklit LCD to display the required dynamic range [,+dr] y the image will generate a acklight-scaled image that minimizes the numer of undershot or overshot piels. Now consider the contrast fidelity c in Equation (). If the availale dynamic range is larger or equal to the required dynamic range (dr ), the optimal contrast fidelity c= can e oained with d and the overall contrast fidelity F c is simply gu. Otherwise, if dr>, the highest possile contrast fidelity p ( ) is c=/dr with t= and d=. Thus, F c ecomes + dr p ( ). (5) dr Figure 6c shows F c as a function of dr for = (upper) and =.5 (lower). The F c increases as dr increases from dr= to dr=.5. For the = curve, the eample image needs no more than 7% of availale dynamic range to represent the whole histogram with the est contrast fidelity c=. For the =.5 curve, the F c decreases from dr=.5 to dr= ecause in Equation (5) the + dr increases slower than dr. The optimal F c happens at p ( ) dr=.5 and the contrast fidelity c= in the region [,+dr]. Notice that c= is not always the optimal solution when dr>. If the distriution in the histogram is not normal (e.g. has two peaks) + dr the optimal dr can e greater than, such that can e p ( ) increased. For each acklight factor, the compleity of finding the optimal F c, and gu is O(k 2 ) with k a small numer (<2). 4.3 Fidelity-Power Optimization Given the solution to the contrast fidelity optimization prolem for any acklight factor, the optimal CBCS policy prolem can e solved y sweeping the acklight factor range etween min and ma, where min and ma are user-specified minimum and maimum acklight factors, respectively. All of the optimal solutions are recorded along with their power consumptions. The inferior solutions, i.e., same fidelity ut higher power or same power ut lower fidelity, are discarded. The remaining solutions are stored for the CBCS policy to select the most suitale solution according to the user preferences. Figure 6d shows the 7 optimal solutions for =.8,.7,.2 from top to ottom. The and y coordinates of each solution indicate the oal contrast fidelity and acklight power, respectively. The two inferior solutions for =. and.9 are discarded ecause they have the same fidelity, F c =, as that of =.8. The results show that more than 5% power savings can e achieved y the CBCS policy while maintaining almost % of contrast fidelity at a acklight factor of 7%. The visual effect is shown in Figure 6e, in comparison with Figure 6f generated from the rightness-invariant policy from Equation (3). The procedures for the contrast fidelity optimized CBCS are summarized in Figure 7. 5 Eperimental Results We use a set of enchmark images from the USC SIPI Image Dataase (USID) [6]. The USID is considered the de facto enchmark suite in the signal and image processing research

6 CBCS(p[..255],k) { cdf[]=p[]; for (i=; i<256; i++) cdf[i]+=p[i]; for (=min; <=ma; +=(/k)) { P=Packlight(); for (dr=; dr<=255; dr+=(256/k)) { Rma=-; for (g=; g<=255-dr; g+=(256/k)) { R=cdf[g+dr]-cdf[g]; if (R>Rma) { =g; Rma=R; if (>=dr) Fc=R; else Fc=(/dr)*R; gu=+dr; Sol = <Fc,P,,,gu>; Search solution dataase for <Fc,*,*,*> and <*,P,*,*>; if (Sol is not inferior) Insert Sol into solution dataase; applications. Since the decision of the acklight factor is ased on each frame individually, the acklight factor may change significantly across consecutive frames ecause the histogram varies significantly. The huge change in the acklight factor will introduce inter-frame rightness distortion to the oserver. When the CBCS technique is to e applied to video applications such as an MPEG2 decoder, the change of the acklight factor should e limited such that the change is too sule to e sensed y human eyes. Tale 2: Original images (upper) vs. acklight-scaled images (lower) Figure 7: CBCS Optimization Flow. field [5]. The results reported here are from 8 color images from volume 3. All of them are 256 y 256 piels. The color depth is 24 its, i.e., 8 its per color-channel in the range of to 255. Tale : Optimal CBCS solutions to the USID enchmark images Image Backlight factor Contrast fidelity c Brightness shift d Overall fidelity CCFL Power (mw) # F c Tale and 2 show the optimal CBCS policies for the enchmark images. We use.9 as the oal contrast fidelity threshold to find the minimum acklight factor and its optimal linear transformation. The results show an average of 3.7X power savings within % of contrast distortion. 6 Conclusions and Future Work We have presented the CBCS technique for a CCFL acklit TFT-LCD display. The proposed technique aims at conserving power y reducing the acklight illumination while retaining the image fidelity through preservation of the image contrast. We have eplained how CCFL works and showed how to model the non-linearity etween its acklight illumination and power consumption. We have proposed the contrast distortion metric to quantify the image quality loss after acklight scaling. We have formulated and optimally solved the CBCS optimization prolem with the ojective of minimizing the fidelity and power metrics. Eperimental results show that an average of 3.7X power savings can e achieved with % of contrast distortion. The CBCS technique we propose in this paper is only for still images. Future studies, however, should consider applying it to video References [] T. Simunic et al., Event-driven power management, IEEE Tran. Computer-Aided Design of Integrated Circuits and Systems, vol. 2, pp , July 2. [2] I. Choi, H. Shim, and N. Chang, Low-power color TFT LCD display for hand-held emedded systems, Proc. of Symp. on Low Power Electronics and Design, Aug. 22, pp [3] F. Gatti, A. Acquaviva, L. Benini, B Ricco, Low power control techniques for TFT LCD displays, Proc. Intl. Conf. Compilers, Architecture, and Synthesis for Emedded Systems, Octoer 22, pp [4] R. L. Myers, Display Interfaces: Fundamentals and Standards, Chichester, Enand: Wiley, 22. [5] W. K. Pratt, Digital Image Processing, Wiley Interscience, 99. [6] Maim, MAX6 Digitally Controlled CCFL Backlight Power Supply. [7] J. Williams, A fourth generation of LCD acklight technology, Linear Technology Application Note 65, Nov [8] Stanley Electric Co., Ltd., [CFL] cold cathode fluorescent lamps, 23. [9] Minolta, Minolta Precision Luminance Meter LS-. [] LG Philips, LP64V Liquid Crystal Display. [] Analog Devices, AD85 -Channel, Mued Input LCD Reference Drivers. [2] Hitachi, HD dot Graphics LCD Controller/Driver with Bit-operation Functions, 23. [3] H. Aoki, Dynamic characterization of a-si TFT-LCD piels, HP Las 996 Technical Reports (HPL-96-9), Feruary 2, 996. [4] S. Daly, The visile differences predictor: an algorithm for the assessment of image fidelity, Digital Images and Human Vision, pp , Camridge: MIT Press, 993. [5] E. Peli, Contrast in comple images, J. Opt. Soc. Amer. A, vol., no., pp , Oct. 99. [6] A. G. Weer, The USC-SIPI image dataase version 5, USC-SIPI Report #35, Oct. 997.