Displaying images on mobile devices: capabilities, issues, and solutions

Transcription

1 WIRELESS COMMUNICATIONS AND MOBILE COMPUTING Wirel. Commun. Mob. Comput. 2002; 2: (DOI: /wcm.82) Displaying images on mobile devices: capabilities, issues, and solutions Jiebo Luo*, Amit Singhal, Gustav Braun and Robert T. Gray Electronic Imaging Products, R&D, Eastman Kodak Company, Rochester, NY, U.S.A. Nicolas Touchard and Olivier Seignol Electronic Imaging Products, R&D, Eastman Kodak Company, Paris, France Summary Wireless imaging is enabling visual communication anytime anywhere to become a reality. Apart from wireless communication issues, a key technical challenge is how to achieve the best-perceived image quality given the limited screen size and display bit-depth of the mobile devices. In this paper, we give an overview of the current capabilities of various mobile devices, highlight some of the technical issues, and present potential solutions. In addition, we present a review of some of the software products on the market and look ahead to the trend toward more capable devices. Copyright 2002 John Wiley & Sons, Ltd. KEY WORDS mobile devices image display image rendering 1. Introduction Communicating and sharing are principal ways of fostering feelings of belonging between family and friends. We share special moments with each other to stay connected when we are apart. Pictures and videos play a major role in how we share, express, and remember our lives. Digital imaging, wireless, and broadband communications are evolving technologies that enable new, richer ways for families and friends to stay connected. Friends and family members can share their vacation pictures, babies first steps, graduation ceremony, and wedding moments all on the screen of a mobile device ( share the life on the go ). Images and videos are also effective ways of communicating commercial information. Wireless imaging is poised to capture a big market in the business world. Potential applications for wireless imaging include news agency photographers sending pictures from the world s hot spots out to hundreds of US newspapers; people reading news footage dramatically enhanced by the insertion of images and videos related to the events; insurance adjusters filing images of a burned house or a wrecked car from the field; construction company engineers sending pictures of a project back to the home office; or even marketing spies sending back images of rivals products on display at a trade show just to name a few. Ł Correspondence to: Jiebo Luo, Eastman Kodak Company, 1700 Dewey Avenue, Rochester, NY 14650, U.S.A. jiebo.luo@kodak.com Copyright 2002 John Wiley & Sons, Ltd.

2 586 J. LUO ET AL. For all of the above to become a reality, many hurdles need to be crossed in wireless imaging technology [1,2]. Major technical challenges include the following: Connection bandwidth Error correction Multiple standards Image management Coverage area Power supply Rendering/display Ease of use. In particular, conflicting requirements for wireless imaging imposes further challenges on some of these technologies in terms of displayed image quality. Most mobile devices with image display capabilities have small-sized display screens, a result of form factor limitations associated with the physical devices. The smaller screen size makes it imperative that the displays have the highest color bit-depth possible to achieve good image-rendering results. However, the computing technology behind most of the currently popular cellular phones and personal digital assistants (PDAs) does not permit a color bit-depth higher than 8 bits in total (instead of per color channel), without increasing the bulk and price of the devices. A second conflict arises because of the limited bandwidth associated with the wireless transmission of data. The constraints on the bandwidth necessitate small image file sizes, which can be harder to achieve without significant levels of compression, especially at higher image resolutions and color bit-depth. To help illustrate the strong conflicting factors between display size, bit-depth, transmission bandwidth, and the quality of the displayed images in wireless imaging, consider the case in which a transmission channel bandwidth limitation (e.g. 3 kb) constrains the images to be encoded between 0.12 to 0.94 bits/pixel for display sizes of (320 ð 240) and (160 ð 160) pixels, respectively. While these bit rates seem to be comparable to those from a 10 : 1 lossy Joint Photographic Experts Group (JPEG) compression of an 1800 ð 1200 pixel 24-bit digital camera image to 648 kb (i.e. 0.1 bits/pixel), the challenge for the small-sized image display is more problematic. Because the total bit-depth of the display device is often only 8 bits/pixel and most displays use fixed system palettes, the images need to be rendered on a server according to the display palette and losslessly transmitted to the client (assuming image rendering is done in the server). The requirement for lossless encoding of the rendered images is evidenced by the fact that requantization of these images can result in very large color errors. Because the displays often utilize fixed color palettes, it is also necessary to utilize dithering (or diffusion) algorithms to minimize the effects of the coarse color quantization. Lossless encoding efficiency depends heavily on the entropy of the original image. For typical imaging applications, the lossless encoding efficiency ranges from between 1.5 to 3.0 bits/pixel. This is larger than the to 0.94-bits/pixel constraint for a 3- kb-limited transmission channel. A rendering process that utilizes dithering (or diffusion) increases the entropy of the image, making it more difficult to apply efficient lossless compression. As the quality of display devices moves to 16 or 24 bits per pixel, lossless compression techniques can be used more effectively in the transmission of the images. Advancements in the processing power (the CPUs) of the mobile devices will enable image-rendering operations to be transferred to the client for execution. Finally, because these images are very small, they are subject to spatial compression artifacts that can lead to blocky, reconstructed images. JPEG 2000 compression schemes could be very beneficial for the compression and transmission of small-sized images. In this paper, we focus on one issue that affects the direct perception of image quality on mobile devices image rendering and display. We begin by reviewing the current device/os capabilities supporting image display, revealing the practical issues and motivating the potential technology solutions. Classical dithering and palettization techniques are revisited to find solutions to this old problem that is faced with new challenges. Client-centric and servercentric image-rendering paths are discussed, along with the implications on image compression for bandwidth control. Other important characteristics of a display, including viewing flare and display gamma also need to be accounted through preprocessing of images to minimize visible artifacts. A review of current market-leading imaging software packages is provided. Finally, we look ahead at the directions wireless imaging is heading amidst the trend toward more capable mobile devices. 2. Device/Display/OS Capabilities Many color image output devices are not capable of displaying all the colors in an input image because they must be stored in a memory buffer with a

3 DISPLAYING IMAGES ON MOBILE DEVICES 587 reduced bit-depth. It may also be desirable to represent an image using a reduced bit-depth in order to reduce the amount of bandwidth needed for transmission or the amount of memory needed to store an image. In the early years, many computers used an 8-bit color representation to store an image that was to be displayed on a soft-copy display such as a cathode ray tube (CRT) or a liquid crystal display (LCD) screen. Such representations allow only 256 unique color values. This is significantly less than the possible color values associated with a typical 24-bit color image. This problem has attracted renewed interest with the recent boom of cell phones and PDAs. The problem is made more acute this time around by the severely limited display size in addition to limited display bit-depth. Table I presents a summary of some of the popular PDAs along with their OS, display resolution, bitdepth, and other characteristics. As can be seen, PDAs based on the Windows CE (recently renamed as PocketPC) OS platform generally have higher color bit-depths and screen resolution. In addition, the PocketPC devices tend to have 32 to 64 MB of integral storage versus 8 to 16 MB on the Palm-based devices. Most of the newest Palm and PocketPC Table I. Selected PDAs and their display characteristics. OS Bit-depth Resolution Screen Compaq IPAQ 3700 WinCE 12 bit 240 ð 320 Reflective Compaq IPAQ 3800 WinCE 16 bit 240 ð 320 Reflective HP Jornada 54X WinCE 12 bit 240 ð 320 Backlit HP Jornada 56X WinCE 16 bit 240 ð 320 Reflective Palm IIIc Palm 8 bit 160 ð 160 Backlit Palm m505/515 Palm 16 bit 160 ð 160 Reflective Visor Prism Palm 16 bit 160 ð 160 Backlit Sony Clie XXXc Palm 16 bit 320 ð 320 Reflective models also have slots for expanding the storage capacity via various memory cards. In terms of thirdparty software solutions for various imaging and other communication needs, there are ample choices for both the platforms. A detailed summary of major software solutions is presented in Section 5. Table II presents a summary of the latest cellular phones available (or soon to be released) for use in various worldwide markets. As a rule, the newest cellular phones available in the European and Asia-Pacific markets have more features [3], especially with regard to image display, than those currently available in the US. Many of the cell phones with color displays are actually integrated PDAs and cellular phones. These tend to be more expensive than cellular phones without integrated PDAs. However, color-capable cellular phones without integrated PDAs tend to have smaller screen sizes and poorer display qualities. The Samsung SGH-T100 is an exception to this rule and features one of the most brilliant color displays in the market, but it costs more than some of the PDA-based cellular phones. While many third-party software applications for image display and storage exist for PDAs (and, therefore, PDAintegrated cellular phones), there are very limited options for non-pda-based cellular phones. Some of the newest phones (such as the Ericsson T68) allow users to download an image via an infrared port to the cell phone. Usually, very strict limitations are posed on the image size and resolution (the Ericsson T68 restricts the image to a 3-K GIF file using a further reduced 8-bit Rainbow color palette). Applications that will allow cellular phone users to download images from the Web without needing PDA capabilities are under investigation. In the next section, we discuss issues related to rendering images for display on PDAs and cellular Table II. Selected cellular phones and their display characteristics. Availability Bit-depth Resolution Screen Comments Sony Ericsson T68 US 8 bit 100 ð 80 Backlit AT&T Wireless, Europe and Asia, limited US Sanyo SCP-5150 US 8 bit 120 ð 160 Backlit Sprint PCS Samsung SGH-T100 Non-US 12 bit 128 ð 160 TFT Europe, Africa, and Asia-Pacific Handspring Treo 270 US 12 bit 160 ð 160 Backlit PDA-based Samsung SPH-1300 US 8 bit 160 ð 240 Backlit PDA-based, Sprint PCS Audiovox Thera US 16 bit 240 ð 320 Reflective PDA-based, Verizon Nokia 9210i Non-US 12 bit 640 ð 200 Backlit Symbian-based PDA, Europe, Africa, and Asia-Pacific, US version 9290 expected Pogo Non-US 8 bit 320 ð 240 Reflective PDA-based, full featured Web browser (not WAP), Europe Samsung Nexio Non-US 16 bit 800 ð 480 Reflective PDA-based, Korea, largest screen. Note: WAP: wireless application protocol

4 588 J. LUO ET AL. phones. The restrictions on color bit-depth, image resolution, file size, and the quality of the physical display require us to revisit the image-rendering problem and look for novel approaches to color palettization and dithering. 3. Dithering/Palettization Revisited Color palettization refers to the process that converts an input color image with a higher bit-depth (a larger set of possible colors) to an output color image with a reduced bit-depth (smaller set of palette colors). For example, a typical 24-bit input image can have up to colors, whereas a typical 8-bit color palette has only 256 colors. To achieve the most desirable image rendering, the set of output palette colors should be based on the distribution of colors in the input image. Furthermore, it is important to preserve important colors such as human skin tones in the palettized image. Once the set of palette colors is determined, a palettized color image can be generated. Generally, the palette color for each pixel of the image is identified by an index value indicating the palette color to be used for that pixel. For example, if there are 256 palette colors used for a particular image, each pixel of the output image can be represented by an 8-bit number, that is, palette index, in the range 0 to 255. When the image is displayed, the palette index can be used to determine the corresponding color (red, green, and blue) value for each of the palette colors. Various schemes for color palettization have been proposed and used in a variety of applications. A common simple color palette ( Web Safe palette) sets six levels of quantization in each of the three channels (red, green, and blue). Other fixed color palettes may choose to set the quantization levels in a nonuniform manner. For example, 3 bits of color information (8 different levels) may be used for the red and green channels of an image and 2 bits of color information (4 different levels) may be used for the blue channel of an image. The result of using any fixed color palette is an image that has quantization errors that can produce visible contours in the image. Because a fixed color palette distributes the 256 colors throughout the entire color space, a particular image that does not contain colors in all parts of the color space may be rendered using fewer than the 256 colors possible. This can lead to unnecessary increases in the quantization errors. One method for minimizing the visibility of the quantization errors in a reduced bit-depth image is to use a multilevel halftoning algorithm to preserve the local mean of the color value [4]. Since halftoning essentially trades spatial resolution for bit-depth, it is, in general, not suitable for displays with a very small screen size, although it is a viable option for PDAs with larger screens (e.g. 8-bit 320 ð 240). An alternative approach to minimizing the visibility of the quantization errors in the reduced bit-depth image is to select the palette of colors based on the distribution of color values in the actual image. This avoids the problem of wasting color values that will never be used to represent that particular image. One such image-dependent palette selection method is vector quantization (VQ), which typically involves the selection of an initial color palette, followed by an iterative refinement scheme [5]. Another VQ method [6] starts with all of the colors of an image and groups them into clusters by merging nearestneighbor pairs of clusters at a time until the number of clusters equals the number of desired palette colors. A third class of VQ methods uses splitting techniques to divide the color space into smaller subregions and selects a representative palette color from each subregion [7]. In general, splitting techniques are computationally more efficient than either the iterative or merging techniques and can provide a structure to the color space that enables efficient pixel mapping at the output stage. While VQ mechanisms can yield high-quality images, they are computationally intensive. To alleviate this problem, a sequential scalar quantization (SSQ) method was proposed by Allebach et al. in Reference [7]. This method sequentially partitions a histogram representing the distribution of the input colors into a number of subregions or color space cells, such that each partitioned color cell is associated with a color in the output color palette. This splitting scheme makes it more computationally efficient than VQ schemes and is used as the underlying color palettization method in our imagerendering scheme. The image-dependent palettization (IDP) methods described above have the significant advantage in that they assign the palette colors based on the distribution of input colors. However, there may still be large quantization errors in important image regions. For example, consider an image containing a human face that only occupies a small image region. The number of pixels that represent skin tone colors may be relatively small and, therefore, the likelihood that adequate palette colors are assigned to skin tone

5 DISPLAYING IMAGES ON MOBILE DEVICES 589 (a) Fig. 1. Color palettization: (a) Web Safe and (b) Web Safe and error diffusion. (b) colors is low. As a result, when the image is represented by the set of chosen palette colors, there may be objectionable colors or contours in the face. Because this image region may be very important to an observer, these artifacts may be much more objectionable than they would have been if they had occurred elsewhere. Existing techniques do not provide any mechanism for minimizing the quantization artifacts in these important regions unless they are large enough to comprise a significant portion of the color distribution. We have developed a scheme that selectively reduces contouring in important regions by supplementing the distribution of the input colors with a set of selected important colors [8]. In particular, skin tone color supplementation is achieved by appending skin tone patches generated from statistical sampling of the skin color probability density function to the input image. A major advantage of this scheme is that explicit skin detection, which can be error prone and takes additional computation, is avoided. In addition, this scheme can be used with any color palettization algorithms. Subjective evaluation has shown that this scheme provides the best rendering and display quality on reduced bit-depth mobile devices and leads to a number of key observations. First, SSQ is effective (and extremely fast); the custom 240-color palette by SSQ (16 Windows system colors are preserved) handsomely beats a fixed Web Safe palette (6 quantization levels in each channel), as can be seen by comparing Figures 1(a) and (b) to 2(b). Second, the introduction of supplementary skin colors (Figure 2c) leads to significantly better rendition of human skin areas without, in general, adversely affecting other areas (Figure 2d). The latter is mostly because each supplemented skin color patch is small enough to not claim a color in the palette unless similar colors are also present in the input image. The size of the individual patches scales with the size of the input image, while the total number of patches remains constant. A second question related to image rendering pertains to the workflow processes employed by the system. The image-rendering operations can be performed on a server, and the resulting image compressed and transmitted to the mobile device over a wireless connection, or the original image may be transmitted to the device, with rendering performed on the client. In the next section, we discuss some options for arranging the workflow of a small-sized image-rendering system. 4. Image-rendering Paths In general, two rendering paths can be conceptualized for mobile imaging systems: server-side rendering and client-side rendering. Each has its benefits and limitations. In server-side rendering (a typical client server relationship), the server engine performs the entire image preprocessing and palettization operations and transmits the palettized image to the client. The client s responsibility is to decode the image file and display the results. No further image processing is required on the client. A typical workflow for a server-side rendering is given in Figure 3(a). This architecture has the advantage of reducing the computational burden placed on the client. A typical processing path might include a resizing and preprocessing stage prior to palettization. The preprocessing stage may include sharpness, tone, and color compensation suitable for the destination display device. For many systems, this process is the only viable architecture because the devices do not have the capabilities for running third-party software or do not have the

6 590 J. LUO ET AL. (a) (b) (c) Fig. 2. Color palettization results: (a) 24 bit, (b) SSQ, (c) composite image with optimal skin patches, and (d) SSQ and optimal skin patches. (d) computational capabilities to perform many rendering calculations (processing speed and memory are distinct limiting factors in applying client-side rendering for VQ-based palettization schemes). However, one of the major limitations of a serverside rendering path is that the palettized images need to be transmitted using a lossless compression scheme. This rules out typical lossy compression schemes such as JPEG. File formats such as PNG and GIF offer modest lossless compression schemes for palettized images but are generally limited in their flexibility to create compressed images of fixed file sizes. For some systems, having file sizes below some hard limit is required because of memory limitations on the devices or tolerable waiting period in wireless transmission. In contrast, client-side rendering architectures place all or a portion of the computational burden onto the client (see Figure 3b). In this example, only the palettization process has been moved to the client. However, in general, the preprocessing algorithms, such as resizing and custom tone scale adjustments, could be performed in the client as well. One advantage of this architecture is that the server has less responsibility to keep track of the display/operating system characteristics of a large set of clients. This makes the client server relationship simpler in that less metadata needs to be communicated to the server to ensure that the image is rendered according to the proper screen size, display palette, and tonal response of the client. Clearly, client-side rendering requires that the devices have the ability to run complex image-processing applications, both in terms of computing power and memory. Therefore, most of the current third-party software applications, discussed in the next section, use a server-side rendering path for displaying images on mobile devices. 5. Imaging Software in the Market There are a large number of third-party software packages available for image display on Windows CEand Palm OS-based handheld devices. These include: PhotoSuite Mobile Edition, FireViewer Suite, Album- To-Go, SplashPhoto, PocketPhoto, ACDSee Mobile, ImagerX, ipaint, JPGview, Mobil-Photo, PhotoAlbum, PhotoUtil, PictPocket, TealPaint, Snapshot, Tiny Viewer, Presto! Mr. Photo, ClipSync Pro, ImagerX, Image Viewer, BrowseIt!, and YiShow Explorer. In addition, Windows CE-based PDAs can also use Web browsers to display the images.

7 DISPLAYING IMAGES ON MOBILE DEVICES 591 (a) Original JPEG file Server Decompress Preprocess Palettize PNG, GIF, BMP, etc. Client Transmit Decompress Display (b) Server Original JPEG file Decompress Preprocess JPEG compress Client Transmit Decompress Palettize Display Fig. 3. (a) Typical server-side rendering process. (b) Typical client-side rendering process. Most of these packages consist of a desktop component that is used to create a PDA-displayable version of an image and a handheld component that can be used to display (and sometimes organize) the images. Because of the limited processing power of handheld devices (especially the Palm OS-based devices), most of the image-processing and imagerendering operations are performed in the desktop component of the package. The operations can include zoom/crop, rotate, contrast/brightness adjustment, and color bit-depth reduction. Table III shows a comparison of the features of some of the Palm-based software packages for image display. In an effort to understand the current issues with image rendering and display on PDAs, we performed a subjective study that evaluated five popular imaging software packages. The scope of the study was limited to performing a qualitative comparison of the quality of the reduced bit-depth image produced by each package for display on the Palm device. To perform the qualitative comparison, we installed the software packages on 8-bit color Palm-based devices (Palm IIIc and Handspring Visor Prism) and then downloaded 11 images with varied content to each device. Note that the Visor Prism supports up to 16-bit color (as does the Palm m505 and some other new handheld devices). However, we restricted the bit-depth of the Prism to 8 bits for the purpose of this study. A set of five judges (mix of image scientists and consumers) ranked the software packages in terms of image quality (on a per image basis) of the reduced bit-depth image by viewing images from the five packages simultaneously on multiple PDAs. Other factors, such as ease of use, cost, and platform support, were not evaluated. The results showed that PhotoSuite ME,

8 592 J. LUO ET AL. Table III. Comparison of some Palm-based software packages. PhotoSuite ME FireViewer Album-To-Go SplashPhoto PocketPhoto Maximum bit-depth 16 bit color 16 bit color 16 bit color 16 bit color 8 bit color Image size 3 sizes (S, M, L) Up to ð Up to 1280 ð ð ð 160 Input image format all popular formats all popular formats GIF, JPG, BMP GIF, JPG, BMP GIF, JPG, BMP Output image format proprietary proprietary proprietary bitmap bitmap Image compression built-in (fast) none, standard, high built-in (slow) none none Brightness/contrast Yes No Yes Yes Yes Color control Yes No No No No Crop/zoom/rotate Yes No Yes Yes Yes Album support Desktop Handheld Desktop Handheld Handheld Annotate images Yes Yes Yes No No Thumbnail view Yes No Yes Yes No Other formats Video Video, Web pages None None None Price Free $ FireViewer, and SplashPhoto produced fewer visible contours in the images (through dithering or lowering the overall contrast of the image). In general, dithering seemed to fare better, given the restricted bit-depth of the images. Lowering the overall contrast in the scene made for less visible contours in the image and was judged preferable by the observers over a sharp and contrasty image with visible contouring effects. A second study judged the image quality when using fixed color palette schemes (such as the Web Safe palette) versus an image-dependent 256-color palette. It was confirmed that the imagedependent palette produced a better 8-bit rendered image than any of the fixed color palette schemes. However, none of the software solutions currently in the marketplace seem to make use of a custom, image-dependent palette for rendering the images. 6. Future Directions Image rendering and display on mobile handheld devices is rapidly coming to maturity. The imagerendering solutions we have described here can be put into widespread practice using today s technology. One of the problems, however, is the different display characteristics of the various handheld and other mobile devices. Companies such as LightSurf are proposing architectures whereby the server receiving a request for an image from a mobile client also receives the description of the client [9]. The image can be custom tailored for best rendering on that particular client so that optimized visual content is delivered. A second solution to this problem is to shift some of the image-processing and image-rendering operations to the client rather than to the server. This is increasingly possible as the computational power and storage capacity of the mobile devices increases. Web clipping applications by Palm can also be used to enable such a solution, which includes client-side applications that run on Palm OS devices, proxy servers for handling translation between Web clipping application format to HTML, and content servers [10]. A second arena of possibilities is opened by developments in the inherent capabilities of the handheld devices and their enabling technologies. State-of-theart PDAs have increasingly higher color bit-depth, a minimum of 12 bits and usually 16 bits. This allows for the image to be rendered using 4000 to colors instead of 256 colors in the case of 8-bit PDAs. The screen resolutions are also increasing, with the Sony Clies offering a 320 ð 320 high-resolution display. Screen technology is shifting from low-contrast backlit displays to bright and colorful reflective displays that provide high-fidelity color reproduction. The cellular phone industry is seeing an even bigger jump in technology. The newest cell phones have integrated PDAs and are being offered at approximately the same price as the newest PDAs [3]. While most cellular phones have very small screens (e.g. 80 ð 100), one of the newest models boasts an 800 ð 480 pixel screen. Cell phones also have the added advantage of built-in wireless communications for effective transmission of images. In the meantime, support for various imaging and communication operations is being built into the operating systems. The operating systems for handheld devices are evolving toward supporting increasingly higher color bit-depths, larger displays, and better wireless connectivity, ultimately leading to 24-bit, reasonably large displays for handheld devices. For example, the upcoming Palm OS 5.0 will support up to 320 ð bit high-density display [11]. In

9 DISPLAYING IMAGES ON MOBILE DEVICES 593 addition, the increasing bandwidth between mobile devices and the server enables images of large size to be quickly transmitted to a client. Recent advances in display technologies are increasing the viability and quality of mobile imaging system displays. One specific technology is the use of organic light-emitting diode (OLED) displays pioneered by Kodak. These displays have several advantages to transmissive and reflective LCDs, including lower power consumption, higher screen luminance, less susceptibility to display flare, less restrictive viewing angles, and compact display design [12]. OLED displays come in both active and passive matrix forms. The passive matrix displays are ideal for low-cost, alphanumeric displays, while the active matrix models offer high-resolution capabilities for video and graphics applications. The organic molecules used in these displays can be tailored to produce various levels of color saturation, operating life, brightness, and transparency [12]. 7. Conclusion We have discussed one issue that affects the perception of image quality on mobile devices image rendering and display. Limitations of the current device/os capabilities motivated the proposed technology solutions. Dithering and palettization were revisited to find solutions to the image-rendering problem. Client-centric and server-centric imagerendering paths were discussed, along with the implications on image compression for bandwidth control. A review of popular imaging software was presented. Finally, we provided an outlook of wireless imaging, which is evolving toward more capable mobile devices. References 1. Wireless Imaging Infrastructure and Players, Special report by Future Imaging, July Wireless Imaging Market Opportunities, Special report by InfoTrends, September Carnoy D. Six cellphones you can t have (yet). Popular Science, July Gentile RS, Walowit E, Allebach JP. Quantization and multilevel halftoning of color images for near original image quality. Journal of the Optical Society of America 1990; A7: Gentile RS, Allebach JP, Walowit E. Quantization of color images based on uniform color spaces. Journal of Imaging Technology 1990; 16: Balasubramanian R, Allebach JP. A new approach to palette selection for color images. Journal of Imaging Technology 1991; 17: Allebach JP, Bouman CA, Balasubramanian T. Sequential Product Code Quantization of Digital Color Image. U.S. Patent No.5,544,284 August Luo J, Spaulding K, Yu Q. A Novel Color Palettization Scheme for Preserving Important Colors. SPIE Electronic Imaging 2003; in press Cropper AD, Cok RS, Feldman RD. Organic LED system and applications. SPIE 2000; 4105: Authors Biographies Jiebo Luo received his Ph.D. degree in Electrical Engineering from the University of Rochester in He subsequently became a Senior Research Scientist and is currently a Senior Principal Research Scientist in the Eastman Kodak Research Laboratories. His research interests include image processing, pattern recognition, and computer vision. He has authored over 60 technical papers and holds 15 issued/allowed US patents. Dr. Luo is a member of the Organizing Committee of the 2002 IEEE International Conference on Image Processing, an At-Large Member of the Kodak Research Scientific Council, and a Senior Member of the IEEE. Amit Singhal received his M.S. (1996) and PhD (2001) in Computer Science from the University of Rochester in Rochester, NY. He received his B.S. (1995) in Computer Science from West Texas A&M University in Canyon, TX. He is a principal research scientist with the Imaging Science and Technology Lab at Eastman Kodak Company. His current research interests include image understanding, image and scene classification, pattern recognition, and knowledge engineering. Amit has authored over 2 dozen journal and conference papers in the areas of image understanding and data fusion. He is a member of SPIE and ISIF. Gustav Braun is a Principal Scientist in the Research Laboratories at the Eastman Kodak Company. He received his Ph.D. in Imaging Science from the Center for Imaging Science at the Rochester Institute of Technology in His current research interests include digital color reproduction, color gamut mapping, digital printing, and vision modeling. He is active in two CIE standards groups Division 8 technical committees for Communication of Colour Information (TC805) and Gamut Mapping (TC803).

10 594 J. LUO ET AL. Robert T. Gray received his Ph.D. in Optical Sciences from the University of Arizona in He subsequently joined the Commercial and Government Systems Division of Eastman Kodak and is currently a Research Fellow and group leader in the Imaging Science and Technology Laboratory at Eastman Kodak Company. His current research interests include image understanding, pattern recognition, and image enhancement. Dr. Gray is a member of the Kodak Research Scientific Council and a member of IEEE. Nicolas Touchard graduated from the Ecole Supérieure d Optique, Orsay, France in He first served in a company designing customized Laser Anemometer. In 1987 he joined Kodak R&D Labs in France and designed a number of computer controlled optical systems that assess photographic system performances. Since 1992, he has been heading various R&D teams across Kodak Europe, involved in traditional and digital imaging technologies. He is currently the Lab Head of the Kodak Electronic Imaging R&D Lab in Paris, France, with a strong focus on network and mobile imaging technologies. Olivier Seignol graduated from the Institut de Chimie et Physique de Lyon, France with a speciality in Image Analysis. He first worked in a small company writing software for Electronic Document Management. In 1997, he joined the Kodak Electronic Imaging R&D Lab in Paris, France, as a Software Engineer, writing software for digital cameras and scanners. Since 1999, he is involved in designing Internet imaging services as a Software Architect. His main interests are around digital imaging, with a strong focus on network and mobile imaging technologies.