tween the image and the scanner sensors and making an additional scan. .... Desktop scanners are already very popular and are a natural choice for printer ...
How to Turn your Scanner into a Colorimeter Joyce Farrell, Doron Sherman and Brian Wandell* Hewbtt Packard Labs., Palo Alto, California *Stanford Univ., Stanford, California Abstract We describe a method for converting conventional threechannel flatbed scanners into colorimeters. The method is based on the simple idea of placing a color filter between the image and the scanner sensors and making an additional scan.
Introduction Neugebauer'sl original suggestion that color printing systems should be based on device-independent representations (e.g., tristimulus coordinates such as CIE XYZ) is widely accepted at present. A consequence of this system architecture is that there is frequently a need to convert device-dependent data (such as RGB values from a scanner) into device-independent data (XYZ coordinates). One of the simplest methods for converting device-independent XYZ values into printer-dependent CMY values is to use 3D look-up tables (LUTs). The time-consuming part of creating such 3D LUTs is the process of measuring the XYZ values for every possible combination of printer CMY values. Even if one limits the 3D LUT to a subset of every possible CMY combination, the number of combinations one must measure to obtain satisfactory performance can be quite large. The measurement is exacerbated by the fact that each CMY combination must be measured sequentially by a colorimeter that is positioned over each color patch. In principle, one might speed up this measurement process by using conventional scanners. The problem, of course, is that the scanner itself is not colorimetric. Not only are the differences between the predicted and actual XYZ values unacceptably large, but errors that are introduced by the transformation process depend on the media, such as the paper and ink, used to produce the image. One way to improve on the transformation linearity is to use more sensors to measure each element2.3.Typical scanners employ arrays of charge coupled devices (CCD) as sensors. These CCD arrays are expensive. Using more CCD sensors to reduce the error in the transformation process is a very expensive approach. In this paper we describe a method for linearly transforming device-dependent scanner RGB values to device-independent XYZ values without incurring significant cost.
A Method for Converting Scanners into Colorimeters Our method for converting a conventional flatbed scanner into a scanning colorimeter is based on the simple
idea of introducing a color filter between the image and the scanner sensors and making an additional scan. There are several ways to accomplish this. First, one can position a filter in the optical path, between the sensors and the focused light. Second, one can place a color transparency between printed samples and the front glass plate of the scanner in between successive scans. In both examples, one would take successive scans of the printed samples: one scan with and one scan without the filter in place. This effectively doubles the number of color channels. For example, a three-channel color scanner has six color outuuts. Because ink reflectances a r e usually s m o o t h l y varying function^^^^,^^^ this number of color samples is frequently enough to obtain a good estimate of the ink surface reflectances and thus their XYZ coordinates under any illuminant8. In this way, by appropriate selection of colored filters and a multi-pass mode, one can use many modern scanners as colorimeters. To demonstrate the method, we scanned 24 different colored patches taken from a Macbeth ColorChecker with a conventional three-channel scanner. once with and once without a colored transparency placed between the patches and the front glass plate of the scanner. For each of the 24 patches, we obtained six color channel responses to form a 24 by 6 matrix, R. Each of the 24 patches also has correponding measured XYZ values. This forms a 24 by 3 matrix, X.t We then solve for the 6x3 coefficients, M, that minimize the quantity X - (R * M). Table One compares the colorimetric accuracy of the Agfa Horizon Scanner (expressed in terms of the AE,, difference9 between the measured XYZ values, X, and the predicted XYZ values, R*M, ) with and without the introduction of an Edmund Scientific 855 Azure Blue transparency placed between the Macbeth ColorChecker target and the front glass plate of the Afga Horizon Scanner. The data in the rightmost column of Table One are based on the 6x3 prediction of XYZ values for llluminant C. We evaluated 21 different transparencies that were purchased from Edmund Scientific. For each transparency, we found the best 6x3 linear prediction of XYZ values f o r the 2 4 colored patches in the Macbeth ColorChecker. Obviously the optimal choice of transparency depends on the scanner. Table One shows that the solution we selected reduced both the average colorimetric error and the maximum colorimetric error. Moreover, the colorimetric error associated with the Red surface reduces the colorimetric error from an unacceptable AEab difference of 14.86 to an acceptable colorimetric error of 2.98.
IS&T's Tetztl~Internrctiotzal Congress on Advances in Non-Impact Printing Technologies (1994)-579
Our results generalize to other scanners as well. With the appropriate transparency or filter, any inexpensive flatbed scanner can be transformed into a scanning colorimeter and thereby decrease the amount of time required to calibrate hardcopy output. Table One: Colorimetric Error
(ma,,)
functions that define the space of possible surface reflectances and illuminants8. Such scanners can be used as spectrophotometers and, consequently, as colorimeters for any standard illuminant. In this paper, we describe a simple method for increasing the number of sensors in a conventional flatbed scanner by introducing a color filter between the image and the ccd sensing device and making an additional scan. In this manner, we are able to achieve six channel output from a three-sensor scanner and reduce the colorimetric errors of the scanner.
References t. The
Conclusions The primary job of a color management system is to provide consistent color communication across a large collection of different color devices. Printer profiles serve the color management framework by providing a standardized description of a color output device. For example, the information within a printer profile contains the transformation from device-independent color values, such as CIE XYZ, to device-dependent values, such as the amount of cyan, magenta, yellow or black to place on a specific media. Printer profiles must be customized for a given combination of printer, ink-set, paper and halftoning method. A color management system cannot always anticipate the particular combination that a person would like to use. Often times, the user, and not the manufacturer, will be required to generate a printer profile. Today's spectrophotometers and colorimeters cannot meet our needs for an affordable, easy to use and fast method for generating printer profiles. Desktop scanners are already very popular and are a natural choice for printer profile characterization. Without some modification, however, today's desktop scanners cannot be used as colorimeters because I ) their sensors are not within a linear transformation of the XYZ m a t c h i n g f u n c t i o n s c a l i b r a t e d f o r any s t a n d a r d illuminantlO.", and 2) source material surface reflectance functions require more than three dimensions in a linear vector pace^,^,^.^. Scanners with four or more sensors can be used to estimate the coefficients of four or more spectral basis 580-IS&T's
X Y Z values, X, were measured using a Minolta Chroma Meter CR-321 and undoubtedly contain some colorimetric error as well. This is because the Minolta colorirneter, like most industrial colorimeters, illuminates surfaces with a Xenon flash lamp. To estimate the XYZ values for surfaces under standard illuminants, such as Illurninant C or D65, colorimeters effectively divide the color signal measured by the colorirneter by the spectral power distribution (SPD) of the Xenon flash lamp and multiply the result with the SPD of the desired illuminant. The result of these operations is then multiplied by the XYZ color matching functions to obtain the estimated XYZ values. If the surface does not fluoresce, this procedure would produce the same result as multiplying the measured spectral radiance factor (i.e. spectral reflectance) of the surface with t h e S P D of the desired illuminant, and then multiplying the result with the XYZ color matching functions. If the surface fluoresces, however, the two methods are not equivalent. Accurate measurements of XYZ values of fluorescent materials must be based on the color signal of the materials as they are being illuminated by the standard illuminant.
1. H. J. Neugebauer. Color correction selector, US Patent #2,790,844, 1957. 2. M . J. Vrhel and 1-1. J. Trussell. Filter considerations in color correction. Proceedings of the SPIE Symposium on Electonic Imaging, San Jose, CA, February 1992. 3. B. A. Wandell and J. E. Farrell. Water into wine: Converting scanner RGB to tristimulus XYZ, Proceedings of tile SPIE, Volume 1909, pp. 92-101, 1992. 4. J. Cohen. Dependency of the spectral reflectance curves of the Munsell color chips, Psychorlomic Science, 1:369370, 1964. 5. J. P. S. Parkkinen, J. Hallikainen, and T. Jaaskelainen. Characteristic spectra of Munsell colors, Journal of tile Optical Society of America A, 11:318-322, 1989. 6. L. T. Maloney, Evaluation of linear models of surface spectral reflectance with small numbers of parameters, Jorlrnal of the Optical Society ofAmerica A , 3: 1673- 1683, 1986. 7. B. A. Wandell and D. t I . Brainard. Towards cross-media color reproduction. Proceedings of OSA Applied Vision Topical Meeting, Optical Society, San Francisco, CA, 1980. 8. J. E. Farrell, D. Sherman and B . A. Wandell, Illuminant Independent Color Calibration, Proceeding of the 2nd Color Itnagin,? Conference, IS&T and SID, November, 1994.
Tenth Itlten~crtionulCongress on Arlvances in Norl-ltnpoct Printitlg Tecllnolo~ies(1994)
9. G. Wyszecki and W. S. Stiles. Color Scietlce, John Wiley and Sons, New York, second edition, 1982. 10. P. C. Hung. Color rendition three-dimensional interpolation, In Proceedings oJtlze SPIE, Volume 900, pages 1 1 I 115. 1988.
1 I . P. Iioelling, J. Slinehous, and M. Mallz. Color character ization of a scanner. Proeedirlgs of ISGiT's Severltlz Inter national Congress on Advarzces in Nor?-Impact Printing Technologies, pages 433-440, 1991.
