Gamut mapping and the printing of digital color images

Color Gamut Mapping and the Printing of Digital Color Images MAUREEN C. STONE Xerox Palo Alto Research Center WILLIAM B. COWAN National Research Council of Canada and JOHN C. BEATTY University of Waterloo Principles and techniques useful for calibrated color reproduction are defined. These results are derived from a project to take digital images designed on a variety of different color monitors and accurately reproduce them in a journal using digital offset printing. Most of the images printed were reproduced without access to the image as viewed in its original form; the color specification was derived entirely from calorimetric specification. The techniques described here are not specific to offset printing and can be applied equally well to other digital color devices. The reproduction system described is calibrated using CIE tristimulus values. An image is represented as a set of three-dimensional points, and the color output device as a three-dimensional solid surrounding the set of all reproducible colors for that device, called its gamut. The shapes of the monitor and the printer gamuts are very different, so it is necessary to transform the image points to fit into the destination gamut, a process we call gamut mopping. This paper describes the principles that control gamut mapping. Included also are some details on monitor and printer calibration, and a brief description of how digital halftone screens for offset printing are prepared. Categories and Subject Descriptors: 1.3.4 [Computer Graphics]: Graphics Utilities; 1.4.3 [Image Processing]: Enhancement General Terms: Algorithms, Experimentation Additional Keywords and Phrases: Color, color correction, color reproduction, color printing INTRODUCTION There are many different computer-controlled devices for making color images, and people eagerly use them all. Unfortunately, images prepared for one device are often disappointing when viewed on a different one. Why this is so and what can be done to improve the situation are the crux of cross-rendering, the difficult Authors’ addresses: M. C. Stone, Computer Science Laboratory, Xerox Palo Alto Research Center, 3333 Coyote Hill Road, Palo Alto, CA 94304; W. B. Cowan, National Research Council of Canada, Ottawa, Ontario, Canada; current address: J. C. Beatty, Computer Science, University of Waterloo, Waterloo, Ontario, N2L 3Gl Canada. Permission to copy without fee all or part of this material is granted provided that the copies are not made or distributed for direct commercial advantage, the ACM copyright notice and the title of the publication and its date appear, and notice is given that copying is by permission of the Association for Computing Machinery. To copy otherwise, or to republish, requires a fee and/or specific permission. 0 1988 ACM 0730-0301/88/1000-0249 $01.50 ACM Transactions on Graphics, Vol. 7, No. 4, October 1988, Pages 249-292. 250 l M. C. Stone et al. problem of making two examples of an image appear similar when produced on output devices that are very different. Successful cross-rendering depends on two radically different types of expertise: the color-rendering properties of output devices, and the psychophysics of color appearance. Neither is a solved problem, so cross-rendering is very (difficult. Yet, in many applications, it is solvable, since image reproduction is routinely done in the graphic arts business. The advent of affordable digital color printing devices is bringing problems familiar to the graphic arts community to the attention of the computer graphics community. A typical graphics environment has significant experience and equipment available to ma.nipulate colored images. What is lacking is the knowl- edge of which manipulations are necessary to get visually pleasing results. Successful color reproduction in the graphic arts, however, is a combination of experience, folklore, taste, and quantitative controls. Making this knowledge explicit will enable automatic or semiautomatic production of images under computer control. Furthe.rmore, graphic arts practice will benefit from a con- trolled scientific treatment of its craft. The results that follow put some of the graphic arts “know-how” onto a scientific basis, demonstrating some of the principles and algorithms that underlie cross-rendering. We claim that these advances bring the compu.ter generation of high-quality images across a variety of output media significantly closer to reality. The project that acted as a forcing function for our work on the cross-rendering problem was a commitment made by the Xerox Palo Alto Research Center to prepare, in printable form, the images illustrating the special issue of Color Research and Application that constitutes the proceedings for the 1986 AIC (Association Internationale de la Couleur) Interim Meeting on Color in Computer Generated Displays. Images supplied in digital form by authors of the conference papers were used to produce screened color separations for offset printing. A short paper presented at the conference [19] gave a narrative account of the project but supplied little technical detail, since it was written while the work was in progress. Because this was a real project with a real deadline, we were unable to pursue every scientific opportunity and occasionally had to be satisfied with algorithms that were merely adequate rather than elegant. Nevertheless, we believe that the principles discovered during this project and presented here significantly advance the :&ate of the art of digital color reproduction, and that leads we were unable to follow owing to time pressures point the way to a great deal of productive research. Examples of images from Color Research and Appli- cation are used as illustrations throughout this paper. The results reported here deal with the following aspects of the cross-rendering problem: Given images defined by monitor coordinates, what transformations should be performed on them before they are rendered on a variety of color printers, with color offset being the definitive output medium? The expertise necessary to perform these transformations requires a mastery of monitor cali- bration, printer calibration, screen technology, and color transformations. The first three are gained by introducing techniques of radiometry and graphic arts into the digital domain. The fourth amalgamates the psychophysics of color appearance and the folk ,wisdom of the printer into interactive techniques for color transformation. Our mastery of the fourth problem is not, however, com- plete; from time to time we had to take arbitrary leaps over chasms of ignorance. ACM Transactions on Graphics, Vol. 7, No. 4, October 1988. Color Gamut Mapping and the Printing of Digital Color Images 251 Consequently, it is important to describe significant areas of this problem where more research can profitably be undertaken. An essential part of cross-rendering is to quantify precisely both the images to be rendered and the characteristics of the source and destination devices. For this purpose, techniques and terminology standardized by the Commission Inter- nationale de 1’Eclairage (CIE) serve well. Each color device requires a calibration that maps each value of the device coordinates to the corresponding CIE tristim- ulus values. The calibration makes it possible to convert device coordinates to tristimulus values and back again. Monitor calibration is, by now, routine [2, 31. Characterizing the offset printing process, on the other hand, is novel. In offset printing, a set of color separations is a set of films containing halftone patterns, one film for each color. These separations are used to produce the printing plates, as well as a color proof, which is a cheaper and simpler way of viewing the separations than setting up an offset press. Standard practice is to adjust the separations until the proof is satisfactory and trust to the skill of the printer to duplicate the appearance of the color proof on the printed page. In line with this practice and as it was impractical to do otherwise, the output we measured to calibrate the printing process was the proof. Once the journal was printed, however, we were also able to measure how well the printer duplicated the proof. A naive view of color reproduction might consider the problem to be solved once device colors are specified in a common form: Simply reproduce the input tristimulus values on the output device. However, unless the input and output devices are very similar (e.g., two color monitors), this approach does not produce satisfactory results [S]. Even on similar devices there will be a problem with colors that exist in the set of possible colors for one device (its gamut), but not in the other. To create a satisfactory reproduction, the tristimulus values must be modified significantly to accommodate the difference between the reproduction properties of the printer, the monitor, and the associated viewing environments. We call this process gamut mapping. Gamut mapping is similar to the adjustments a graphic arts professional makes when preparing color separations from original artwork. It is not yet possible to do this step algorithmically; we developed computer tools to produce the trans- formations interactively, using human aesthetic judgment as the criterion, just as in the graphic arts. We also developed a tool that creates a fixed mapping between a monitor and a printer so that a user may interactively select colors that lie inside both gamuts. The entire process is summarized in Figure 1, and the rest of this paper expands upon the boxes in this diagram. Our system for printing color separa- tions, “Color Printing,” is not particularly novel. -Nevertheless, since color print- ing is increasingly important in the computer graphics community and it is necessary to understand this process to understand this paper, we include a brief description in Section 2. More detail on our specific implementation is provided in a technical report [20]. Sections 3-5 discuss the specific algorithms used to perform the processes in the half of the diagram labeled “Color Correction.” Section 3 describes how to calibrate input and output devices, and discusses the color gamuts of these devices. Section 4 describes the device-to-tristimulus transformation and its inverse, the blocks labeled “Input Conversion” and ACM Transactions on Graphics, Vol. 7, No. 4, October 1988. 252 - M. C. Stone et al. Color Correction Color Printing Fig. 1. A block diagram showing the process used to convert images designed on a monitor to digital halftone screens for offset printing. Monitor PrImarIes Printer PrImarIes Input Converslon XYZ Gamut Mapping XYZ output Conversion Prmter PrImarIes to Offset Printer “Output Conversion” in Fibmre 1. Section 5 describes the tools and techniques used to perform image-to-device “Gamut Mapping.” Most of the images used in this experiment were defined by authors who had no specific knowledge of the printing process used for the journal. One set of simply colored illustrations, however, was required for one of our own papers [18]. Section 6 describes an experiment using a calibrated tool that took advan- tage of our knowledge about the different devices to define colors that would reproduce well on both a monitor and a printed page. Section 7 presents some numerical results describing how effectively the printing process was controlled and the effect of gamut mapping on the tristimulus values. The paper concludes with a discussion of the wider significance of this work as a partial solution to the cross-rendering problem. 2. COLOR PRINTING Color offset printing reproduces color by combining three primary inks: cyan, magenta, and yellow. To improve contrast, a fourth ink, black, is often added. Different colors are produced using a technique called halftoning, which repro- duces the gradations in a picture using patterns of dots of different sizes. Originally, halftone patterns were produced by photographing an image through a fine screen, called a halftone screen. Modern systems produce the halftone patterns digitally by scanning the original artwork. The standard interface to, a commercial offset printing service is a set of three or four halftoned films, one for each color separation, plus a proof made from those films. There have been some efforts to provide guidelines to make it easier for the printer to match the proof. One is a set of recommendations for standard printing inks, proofing colors, and proofing densities, “Recommended Specifica- tions Web Offset Publicat.ions (SWOP) [l, 121, which, although not strictly ACM Transactions on Graphics, Vol. 7, No. 4, October 1988. Color Gamut Mapping and the Printing of Digital Color Images 253 EED l 0 ::ifl 50”/ 00 o 0 0 ED l * y;; 25% (a) (b) Fig. 2. (a) Halftone patterns for 25%, 50% and 85% dot areas. (b) Digitally produced halftone patterns for 25%, 50% and 85% dot areas using Holladay’s algorithm. Original image by Chuck Haines, Xerox Corporation. controlled, provides some mechanisms for ensuring that a duplication of the proof is practical. The printer’s skill and judgment, however, are still the most significant factors in the success of the reproduction. We made the color separations for this project on an experimental, high- resolution, laser film printer [17] and developed the film in-house. The color proofs were then commercially produced from the separations using a DuPont process called Cromalin,’ which, when properly controlled, is as stable and reliable a process as is commercially available. We took great care to produce suitable high-contrast negatives, from which a proofing house made SWOP-standard proofs for the printer. It was very important for the success of this experiment that we meet the standard expectations of the printing industry, since we had no special relationship with the printer. This section describes the basic steps we performed to make our printing process work effectively. 2.1 Halftone Patterns Halftone patterns are defined by the spacing of the dots measured in dots per inch, called screen frequency, and the percentage area covered by ink in the resulting patterns, called dot area. The dot spacing defines the sharpness of the resulting image, with 133-150-dot-per-inch screens typical for magazine-quality offset printing. The percentage area defines the lightness/darkness of the result- ing area. Figure 2a shows halftone patterns for 25%, 50% and 85% dot areas. Figure 2a shows the round dots that would be produced using a traditional mechanical halftone screen, but other shapes can be used either to produce textures for artistic reasons or to accommodate a scanning output device such as a film plotter. When a halftone pattern is generated on a raster printer, a pattern like the one shown in Figure 2b is often used [6]. In color printing, the four separations are printed one over the other. To minimize interference between the different colors, the halftone patterns on each ’ Cromalin is a registered trademark of the DuPont Corporation. ACM Transactions on Graphics, Vol. 7, No. 4, October 1988. M. C. Stone et al. separation are oriented along lines of different angles. Mechanically, this differ- ence is produced by rotating the halftone screen when photographing the image. Digitally, the effect of this rotation must be simulated (Holladay provides a means of doing this). Typical screen angles are 105, 75, 90, and 45, for cyan, magenta, yellow, and black, respectively, though some recommendations ex- change the values for magenta and black. Figure 3 shows a magnified version of a portion of a screened color image. When printed on film, a halftone pattern should contain only opaque and clear areas. Gray areas in the pattern will make the pattern overly sensitive to exposure time when making the printing plate. The dot area of a properly screened film can be accurately measured using a densitometer, which is crucial for maintaining quality control of the film printing process. Film fog, which affects the transpar- ency of the background film, inadequately opaque dark areas, and insufficiently sharp edges will all invalidate this measurement and jeopardize the reproduction process. It is essential to use high-contrast film and carefully control the devel- opment process to achieve reliable results. 2.2 Tone Reproduction The basic measure of a reproduction technology is its tone reproduction curve, abbreviated TRC. This function defines the mapping from the input gray or tone values to the output values. In traditional printing these tone values are measured as density, a logarithmic function of reflectance. The tone reproduction curve relates density values in the original image to those in the reproduction, and an ideal reproduction maps the original values to identical values on the print. Where the original image is in digital form, however, the concept of a TRC must be redefined, because there is no set of density measurements for the original image. The optimal mapping from monitor intensity to print density has not yet ACM Transactions on Graphics, Vol. 7, No. 4, October 1988. Color Gamut Mapping and the Printing of Digital Color Images 255 been defined, so for this experiment we chose to use the default mapping for our existing printing system, which maps intensity to dot area. In this paper TRC is used to describe the function that controls the mapping between the requested tone values and those actually produced by the mechanics of the printing process. The spots produced by any mechanical device are not the idealized squares or disks shown in Figures 1-3, so the bit patterns sent to the printer must be modified to accommodate the tone reproduction characteristics of the device. This compensation can be provided by a table that is the inverse of the tone reproduction curve for the device. Significant time was spent con- trolling the film printing process so that the dot area requested was actually produced on the film. Even so, this process was not as reliable as is commercially achievable owing to the experimental nature of our printing system. 2.3 Gray Balance One of the most basic criteria for a good color reproduction process is that it be able to reproduce the neutral colors in a picture accurately. The three guns on a color monitor are usually balanced so that equal amounts of the primary colors produce a neutral gray color. In printing, it is not usually possible to use equal amounts of the primary colors to produce a gray. The process of adjusting the mix of the three primaries to produce a neutral color is called gray balancing. It is a straightforward task to find an approximately neutral progression of grays from a set of color patches containing nearly equal amounts of the primary colors. These data are used to produce a table of values that compensates each of the primaries to produce a neutral gray scale. For example, a color whose ideal value is a 50% gray might actually be produced as [C: 50%, M: 47%, Y: 47%]. Given that we adjust the mix of cyan, magenta, and yellow to produce gray colors, what about colors that are nearly gray? If we do not compensate those also, there will be a discontinuity in our color space. Most colors contain some amount of gray, called the gray component, which is simply min[C, M, Y]. For example, the color defined as [C: lOO%, M: 75%, Y: 80%] has a gray component of 75%. The obvious solution is to treat the gray component exactly as we would a gray color, and this is often done. For our process, however, we found we got better results by