Color Management: Principles and solutions

Color Management: Principles and solutions Jon Yngve Hardeberg DeviceGuys / Conexant Systems, Inc., Redmond, WA, USA [email protected], [email protected], www.deviceguys.com/jonh/

Abstract

mainstream of computing. But where is the problem in all this? For example, one might say: “I know that my scanner provides me with a description of each color as a unique combination of red, green, and blue (RGB) and so does my monitor, and even my inkjet printer accepts RGB images!” The problem is that even if these devices all ’speak’ RGB, the way they describe colors (”scanner-RGB”, ”monitor-RGB”, and ”printer-RGB”) are substantially different, even for peripherals of the same type. An obvious example of this is that an image printed on glossy paper by a sublimation printer is considerably more colorful than the same image printed on plain paper by an old ink-jet printer. In another example, when displaying scanned images directly on a computer monitor without color management, the color differences between the original and the monitor images can be very large, e.g. with a mean Eab > , which is way beyond the limits of acceptable color differences [3]. Digital color imaging systems process electronic information from various sources; images may come from the internet, a remote sensing device, or a local scanner, etc. After processing, a document is usually compressed and transmitted to several places via a computer network for viewing, editing or printing. To achieve color consistency throughout such an extremely widely distributed system, it is necessary to understand and control the way in which the different devices involved in the entire color imaging chain treat colors. This can be achieved by calibrating the color image acquisition and reproduction devices so that the device-dependent color representations of the scanner, the monitor, the printer, and other color imaging devices, can be linked to a device-independent color space. The exchange of images can then be done in this color space, which should conform to international standards. In my research [1], I have investigated several of the problems mentioned above. In this article, I will hopefully be able to give you a small taste of the subject of color management. I start by presenting rapidly the principles of color management, emphasizing the physics of image acquisition and reproduction, and the state-of-the-art of color management software. Then I present my original solutions to the problems of the colorimetric characterization of scanners and printers providing efficient and colorimetrically accurate means of conversion between a device-independent color space such as the CIELAB space [4, 5], and the device-dependent color spaces of a scanner and a printer.

The term color management designs an ensemble of algorithms that provides a framework in which color information can be processed consistently throughout a digital imaging system. This is most commonly achieved through the use of special software packages, known as Color Management System (CMS) software. The purpose of this article is to present some of the scientific principles of color management, to evangelize its importance whenever a computer is used for the acquisition, visualization, or reproduction of color images, and to present some of the original color management algorithms and solutions I have developed during my Ph.D. studies [1].

1 Introduction The use of color in imaging continues to grow at an ever increasing pace. Every day, most people in the industrialized parts of the world are users of color images coming from different imaging devices. For example color photographs, magazines, and television at home, computers with color displays, and color printers in the office. With the increased use of color images, people’s quality requirements have increased considerably. Just a few years ago, a computer graphics system capable of producing 256 different colors was more than enough for most users, while today, most computers that are sold have true color capabilities, being able to produce 16.7 million1 colors. When working with images on a computer, people expect the colors to be ”right”. Furthermore, several professions have particular needs for high-quality color images. Artists are very concerned about colors in their works, and so are the art historians and curators studying their works. The printing, graphic arts, and photography industry have been concerned about color imaging for a long time. Most of the color imaging standards and equipment used today have their roots in these industries. But the past twenty years have seen the field of digital color imaging emerging from specialized scientific applications into the 1 Note

that this number represents only the number of different colors that can be specified to the monitor (28 28 28 = 16777216); the actual number of distinguishable resulting colors is much lower, approximatively in the order of 1 million [2].

1

9

2 Principles of color management

[1, 6]. A characterization based on analytical models, however, seeks to define the desired transformation by minimizing the color differences on a given color target with known 2.1 Digital image acquisition colors. The method we present in Section 3.1 uses this apIn order to process images on a computer, the continuous- proach. space, analog, real-world images need to be sampled and quantized. This is typically done by a digital camera or scanner. The actual hardware assembly may vary, but schemati- 2.2 Digital image reproduction cally, we may depict the main components involved in a typ- Color may be produced in many different ways. According to ical image acquisition process as in Figure 1.2 We denote Nassau [7], as many as fifteen distinct physical mechanisms the spectral radiance of the illuminant by lR , the spectral are responsible for color in nature. Only few of these mechreflectance of the object surface imaged in a pixel by r , anisms are suitable for digital image reproduction, but there the spectral transmittance of the optical systems in front of exists nevertheless considerable diversity in available techthe detector array by o , the spectral transmittance of the nologies for displaying and printing color images. Image rek th optical color filter by k , the spectral sensitivity of the production devices can can be broadly classified in two cateCCD array by a , and the acquisition noise by k . Suppos- gories, additive and subtractive devices.4 ing a linear optoelectronic transfer function of the acquisition system, the scanner3 response ck for an image pixel is then 2.2.1 Additive color devices equal to

()

()

()

ck

=

Z max min

()

()

( ) ( ) ( ) ( ) ( ) + k :

lR r o k a d

Illumination lR(λ) o(λ)

φk(λ)

In additive color devices, colors are produced by adding dif(1) ferent colored lights, the primary colors. The most common choice of additive primary colors is red, green and blue (RGB). Visual display units, e.g. computer monitors, are typical examples of additive devices. They can be characterized almost completely in terms of a few parameters, such as the white point, the gamma curve etc. When these parameters are known, the required RGB drive signals needed to produce a given color can be calculated, see e.g. Berns et al. [8], and Chapter 14 of Kang [9].

a(λ)

2.2.2 Subtractive color devices

r(λ)

Observed object

Optical path Color filter

ck Scanner Sensor response

In subtractive color devices, the colors are produced by multiplying a white spectrum by the spectral transmission curves of the three subtractive primary colors cyan, magenta and yellow (CMY). Thus, for each of the subtractive primaries, frequency components are removed from the white spectrum. An ideal subtractive color device can be presented as in Figure 2, where we observe that the result of a multiplication of an ideal white spectrum with the three ideal rectangular bandstop filters gives a resulting color spectrum exactly equal to the one obtained in an ideal additive system. We remark that no concepts in the field of color have traditionally been more confused than that of additive and subtractive color mixture. This confusion can be traced to two prevalent misnomers: the subtractive primary cyan, which is properly a blue-green, is commonly called blue; and the subtractive primary magenta is commonly called red. In these terms, the subtractive primaries become red, yellow, and blue; and those whose experience is confined for the most part to subtractive mixtures have good cause to wonder why the physicist insists on regarding red, green, and blue as the primary

()

Figure 1: Schematic view of the image acquisition process. (Note that only one optical color filter is represented in the figure. In a color scanner, three filters would be used.) If color is to be ”managed” properly, it is necessary to perform a colorimetric characterization of the scanner, meaning to define the transformation from the scanner’s RGB output to some device-independent color space. To this end, two different approaches are typically used, applying spectral and analytical models. The goal of a spectral characterization technique will be to estimate the unknown parameters of Equation 1, and then use the spectral model to define the desired transformation. We have investigated this problem in 2 Note that an electronic version of this article, in which the figures are in color, is available at http://www.deviceguys.com/jonh/ biblio.html#norsig. 3 We use for simplicity the term scanner here, meaning typically either camera or scanner.

4 Some devices such as halftone printers combine these two principles, they are called hybrid devices.

2

1

Red

Cyan

values.

C R λ

λ

2.3 CMS software

To obtain faithful color reproduction, a Color Management System (CMS) has two main tasks. First, colorimetric characterization of the peripherals is needed, so that the devicedependent color representations of the scanner, the printer, Yellow and the monitor can be linked to a device-independent color Blue Y space, the Profile Connection Space (PCS), see Figure 4. This B task is sometimes called device profiling. Furthermore, efficient means for processing and converting images between different representations are needed, this task is undertaken Figure 2: An ideal subtractive color reproduction system. by the Color Management Module (CMM). For further information about the architecture and usage of CMS, refer e.g. to colors. The confusion is at once resolved when it is realized [11, 12, 13, 14]. that red, green, and blue are selected as additive primaries because they provide the greatest color gamut in mixtures. For the same reason, the subtractive primaries are, respectively, red-absorbing (cyan), green-absorbing (magenta), and blueabsorbing (yellow). The principle of subtractive color mixture is used in color printers, where a white sheet is covered with layers of yellow, magenta and cyan inks or other materials. The pigments in the inks absorb certain wavelengths from the incident light, and thus constitutes a subtractive color system. To ”manage” the color of a printer consists in deciding what amounts of inks (CMY) to put on the paper to achieve the desired colors. In an ideal system, cf. Figure 2, this would be rather straightforward. However, in reality, the transmittance curves for the cyan, magenta and yellow inks are far from rectangular, see Figure 3, and this task is not trivial. White

1

1

Magenta

Resulting colour

Green

M

G

λ

λ

λ

λ

1

λ

λ

()

Ideal printer block dyes

Mitsubishi sublimation printer inks

1

1 Cyan

0 400 1

450

500

550

600

450

500

550

600

650

700

Yellow

0.5 0 400

700

Magenta

0.5 0 400 1

650

450

500 550 600 Wavelength, λ [nm]

650

0 400 1

450

500

550

600

450

500

550

600

700

700

650

The industry adoption of CMS depends strongly on standardizations [15]. The International Color Consortium (ICC) plays a very important role in this concern. The ICC [14] was established in 1993 by eight industry vendors for the purpose of creating, promoting and encouraging the standardization and evolution of an open, vendor-neutral, cross-platform CMS architecture and components. Today there is a wide acceptance of the ICC standards, even if some issues still are unsolved. Several vendors offer CMS software solutions, e.g. Microsoft, Apple, Kodak, and Agfa.5 It has been concluded in a recent study [13] that the color management solutions offered by different vendors are approximately equal, and that color

700

Yellow

0.5 0 400

650

Magenta

0.5 0 400 1

Figure 4: Different digital imaging devices connected in a Color Management System. Each device is characterized by a profile.

Cyan

0.5

Reflectance r(λ)

Reflectance r(λ)

0.5

450

500 550 600 Wavelength, λ [nm]

650

700

Figure 3: Comparison of ideal non-overlapping block inks (left) and real sublimation printer inks (right). Another aspect that has to be taken into account is that every output peripheral has a unique color range, or color gamut, that defines the range of colors it can display. A CMS should handle gamut incompatibilities by mapping out-ofgamut colors onto legal ones. This procedure is called gamut mapping [10]. We present in Section 3.2 a new printer characterization technique which provides a practical tool to transform any point of the CIELAB color space into its corresponding CMY

5 See

http://www.tsi.enst.fr/˜hardeber/work/cms. html or http://www.deviceguys.com/jonh/cms.html for a more comprehensive list of available Color Management System software, with hyperlinks to web pages.

3

independent CIELAB color space into its corresponding values in the device-dependent printer color space CMY. This process could also include a color gamut mapping technique which can be of any type. We use an approach based on computational geometry with which we construct two 3D structures which provide us with a partition of the space into two sets of non-intersecting tetrahedra, an inner structure covering the printer gamut, and a surrounding structure, the union of these two structures covering the entire definition domain of the CIELAB space. These 3D structures allow us to easily determine if a CIELAB point is inside or outside the printer color gamut, to apply a gamut mapping technique when necessary, and then to compute by non-regular tetrahedral interpolation the corresponding CMY values. We establish thus an empirical inverse printer model. Our method consists of first printing a digital color chart covering the entire color gamut of the printer to be characterized. Then we analyse colorimetrically the printed chart to obtain the CIELAB values corresponding to each sample (Figure 6). We then construct a 3D Delaunay triangulation in CMY space by taking the color sample CMY triplets as vertices. To invert the system this triangulation is transported into CIELAB space as shown in Figure 7. In order to be able to treat out-of-gamut colors, we have added a surrounding structure in such a way that, together with the inner structure, it defines a valid triangulation which includes the definition domain of the CIELAB space. Using these structures, we are able to calculate for any CIELAB color point its corresponding CMY values by tetrahedral interpolation, in particular to create a printer profile. For more information about this method, refer to [1, 20].

management now has passed the breakthrough phase and can be considered a valid and useful tool in image reproduction. However, there is still a long way to go, both when it comes to software development (integration of CMS in operating systems, user-friendliness, simplicity, : : : ) and research in color science (better color consistency, gamut mapping, color appearance models, : : : ). Color imaging is a very active research domain [16]. In the next section we present briefly some of our original solutions for the colorimetric characterization of scanners and printers, providing profiles for these devices.

3 Device characterization 3.1 Colorimetric scanner characterization The proposed algorithm for the colorimetric characterization of the scanner is the following (see Figure 5): A standard color target (IT8.7/2) is scanned, the scanner output values of each color patch is compared to the theoretical CIELAB values. From this data set we establish a model of the scanner response using a polynomial regression technique.

L*

Spectrophotometric Data

a* b*

Characterisation Process

Colour Chart R

Scanner

G B

Scanner profile Polynomial Coefficients

B

G

CMS

3D Polynomial Transformations

Y M C

Numerical CMY chart

R RGB-to-L*a*b* 3D look-up tables

R

Figure 5: Scanner characterization.

Printer

Calibrated Scanner

G B

L*

Non-linear 3D Transformation

a* b*

L*

Spectrophotometer

Spectra

Colorimetric Calculations

Calibration Process & Color Gamut Mapping

a*

We apply a 3rd order polynomial model in the 3 parameters R, G, and B, giving 60 coefficients. We optimize these coeffiPrinter profile cients by minimizing the mean square error in CIELAB space Non-regular tetrahedral Printed Color Chart of the 288 patches of the IT8.7/2 color target. This colorimetinterpolation ric characterization then allow us to perform the transformation between the scanner RGB space and CIELAB space. In particular, a scanner profile is calculated. The originality of Figure 6: Printer characterization. our approach relies mostly on the fact that we optimize directly in CIELAB space and not in CIEXYZ space as done traditionally. This has the effect that the errors that we minimize correspond to visual color differences. 4 Conclusion For more information about this method and its applicaI hope that this article has provided an introduction and some tions, refer to [1, 3, 17, 18, 19]. useful insight into the issue of color management: why it is so important, and how it may be done. 3.2 Colorimetric printer characterization For further reading on this subject, Sharma and Trussell We propose a characterization technique which provides [16] have written an excellent introduction to digital color a practical tool to transform any point of the device- imaging, and my Ph.D. dissertation [1] is of course highly L*

b*

a*

L*a*b*-to-CMY 3D look-up tables

4

b*

White

[10] Ján Morovic. To Develop a Universal Gamut Mapping Algorithm. PhD thesis, Colour & Imaging Institute, University of Derby, October 1998.

Y

90

100

80

L*

M [%]

70

50

60 50 M

[11] Lindsay W. MacDonald. Device independent colour reproduction. In Eurodisplay, pages B–3/1–B–3/36, August 1993.

R

40

0 100

30

100 50 C [%]

50 0 0

Y [%]

B 20 10 −50

Black 0

50

[12] Gerald Murch. Color management on the desktop. In Proceedings of IS&T and SID’s Color Imaging Conference: Transforms and Transportability of Color, pages 95–99, Scottsdale, Arizona, November 1993.

100

b*

Figure 7: The triangulated CMY color gamut cube and its corresponding geometrically deformed CIELAB color [13] Kurt Schläpfer, Walter Steiger, and Joanna Grönberg. Features gamut. of color management systems. UGRA Report 113/1, Association for the Promotion of Research in the Graphic Arts Industry, 1998.

recommended :-) Useful textbooks are for example those by Hunt [21], Giorgianni and Madden [22], and Kang [9], and if [14] ICC Profile Format Specification. The International Color you really want to know it all, refer to the bible by Wyszecki Consortium, November 1997. Version 3.4, See http:// and Stiles [5]. For literature in norwegian, refer to Valberg www.color.org/. [23]. [15] Michael Stokes. Industry adoption of color management systems. In Proceedings of the 8th Congress of the International Colour Association, AIC Color 97, volume I, pages 126–131, Kyoto, Japan, 1997.

References [1] Jon Yngve Hardeberg. Acquisition and reproduction of colour images: colorimetric and multispectral ap´ proaches. Ph.D dissertation, Ecole Nationale Supérieure des Télécommunications, Paris, France, 1999.

[16] Gaurav Sharma and H. Joel Trussell. Digital color imaging. IEEE Transactions on Image Processing, 6(7):901–932, July 1997.

[2] M. R. Pointer and G. G. Attridge. The number of discernible colours. Color Research and Application, 23:52–54, 1998. See also page 337 of the same volume.

[17] Jon Yngve Hardeberg. Transformations and colour consistency for the colour facsimile. Diploma thesis, The Norwegian Institute of Technology (NTH), Trondheim, Norway, April 1995.

[3] Jon Yngve Hardeberg, Francis Schmitt, Ingeborg Tastl, Hans Brettel, and Jean-Pierre Crettez. Color management for color facsimile. In Proceedings of IS&T and SID’s 4th Color Imaging Conference: Color Science, Systems and Applications, pages 108–113, Scottsdale, Arizona, November 1996. Also in R. Buckley, ed., Recent Progress in Color Management and Communications, IS&T, pages 243-247, 1998.

[18] Jon Yngve Hardeberg and Jean-Pierre Crettez. Computer aided colorimetric analysis of fine art paintings. In Oslo International Colour Conference, Colour between Art and Science, Oslo, Norway, October 1998. [19] Jon Yngve Hardeberg. Desktop scanning to sRGB. To appear in IS&T and SPIE’s 12th annual Symposium on Electronic Imaging, Color Imaging: Device-Independent Color; Color Hardcopy and Graphic Arts V, San Jose, California, January 2000.

[4] Colorimetry, volume 15.2 of CIE Publications. Central Bureau of the CIE, Vienna, Austria, 2 edition, 1986. [5] Günter Wyszecki and W. S. Stiles. Color Science: Concepts and Methods, Quantitative Data and Formulae. John Wiley & Sons, New York, 2 edition, 1982. [6] Jon Yngve Hardeberg, Hans Brettel, and Francis Schmitt. Spectral characterisation of electronic cameras. In Electronic Imaging: Processing, Printing, and Publishing in Color, volume 3409 of SPIE Proceedings, pages 100–109, May 1998.

[20] Jon Yngve Hardeberg and Francis Schmitt. Color printer characterization using a computational geometry approach. In Proceedings of IS&T and SID’s 5th Color Imaging Conference: Color Science, Systems and Applications, pages 96–99, Scottsdale, Arizona, November 1997. Also in R. Buckley, ed., Recent Progress in Color Management and Communications, IS&T, pages 88-91, 1998.

[7] Kurt Nassau. The Physics and Chemistry of Color. The Fifteen Causes of Color. John Wiley & sons, 1983.

[21] R. W. G. Hunt. The Reproduction of Colour. Fountain Press, Kingston-upon-Thames, UK, 5 edition, 1995.

[8] Roy S. Berns, Ricardo J. Motta, and Mark E. Gorzynski. CRT colorimetry. part I: Theory and practice. Color Research and Application, 18(5):299–314, October 1993.

[22] Edward J. Giorgianni and Thomas E. Madden. Digital color management: encoding solutions. Addison-Wesley, 1997. [23] Arne Valberg. Lys, Syn, Farge. Tapir Forlag, Trondheim, Norway, 1998.

[9] Henry R. Kang. Color Technology for Electronic Imaging Devices. SPIE Optical Engineering Press, 1997.

5