Other Free Encyclopedias » Online Encyclopedia » Encyclopedia - Featured Articles » Contributed Topics from P-T

Spectral Imaging

color reproduction systems rosen

Rochester Institute of Technology

Color reproduction has always relied on both metamerism and on the fact that an observer will confuse pairs of spectral power distributions as being the same color under particular viewing conditions. Metamerism in humans forgives the fact that colorants used to create a copy will typically radiate photonic energies in ratios different from those that come from the color-forming materials of an original. The engineering conveniences of basing color reproduction systems on metamerism are clear, but many applications demand reproductions that maintain their relationship under various viewing conditions or when viewed by various observers. In recent years, spectral reproduction has been explored to address the disadvantages of metameric imaging and as a source of previously unattainable features.

Traditionally, color capture devices have been engineered to emulate the spectral integration found within the human visual system. Thus, typical color capture devices have three channels each of spectrally wide sensitivity, much like the human eye. For image output, a computer monitor is similar to the typical image-capture device in that it relies on three spectrally wide channels. Standard printers have four channels, where a non-spectrally selective black is added to a set of three spectrally wide subtractive primaries. Among the purposes for including black is increasing color gamut for dark colors and improving stability in the neutral tones.

Building cameras that “see” color similarly to humans is a very difficult task. It turns out that for many reasons, the spectral sensitivities of typical cameras are different than those of the standard observer. In Figure 60, the spectral sensitivities of an area array camera popular with high-end imaging studios are optimally transformed to match the human color-matching functions as closely as possible. Note how different the camera curves are from the human functions. Thus, even under perfect conditions, 3-channel cameras will have different metamerism properties than humans.

Spectral color reproduction addresses many of the drawbacks of metameric reproduction. Hill has compared the evolution of three-band systems with the revolution of spectral imaging, concluding, “the solution to the fundamental problems of tristimulus reproduction systems is offered by multi-spectral technology” [Hill, 2002a]. He found that the following weak points of metameric systems would be addressed by spectral-based systems:

  • Differences between device and observer metamerism
  • Sensitivities of device profiles to changes in paper and colorants
  • Illumination dependence
  • Gamut mapping effects
  • Deviations of color-matching functions among humans

The advantages of spectral color reproduction are all a consequence of the additional information captured about a source scene. Thus, storing the spectral signal is a far more precise way to archive the original information. Also, image processing can take advantage of the richer data to perform analysis or manipulation routines not otherwise achievable. Material identification, object segmentation, background removal, and facial recognition are all enhanced. Arbitrary output rendering characteristics may be fully exploited.

Spectral imaging captures sufficient information to estimate the spectral radiance, transmittance, or reflectance from an original. Relying on reflectance matching improves the ability of a system to be robust to changes in illuminant and observer. Radiance matching allows for emulation of a reproduction under a change in viewing environment. The reflectance-and radiance-matching tasks referred to above are only the most obvious spectral-based extensions of a color reproduction infrastructure. New capabilities will also present themselves. Some examples include an increased ability to analyze original scenes, improving the recognition and reproduction of certain memory colors such as skin tones, grass, and sky; higher levels of within-gamut reproduction quality through the use of multi-channel input devices that exhibit hardly any instrumental metamerism, something currently unavoidable and the scourge of today’s color reproduction world; the introduction of fluorescence as a positive force in color reproduction, unmasked through the acquisition of multiple spectral images taken under different light sources; and improvement in making spot-color and speciality ink choices.

Modern investigation of this branch of color reproduction spans over 30 years. On the input side, as early as 1975, Kotera and co-workers at Matsushita started working with the use of multiple interference filters for spectral scanning [Kotera, 1975, 1976], [Hayami, 1976]. A number of worldwide patents were issued on that work starting in 1978 [Kan, 1978], [Hayami, 1979]. Ohta and co-workers at FujiFilm® demonstrated the use of a spectral camera in 1981 for use in simulating a photographic system [Ohta, 1981a, 1981b], [Takahashi, 1981]. By the mid-1990s an explosion of interest had grown up around spectral estimation from multi-channel camera signals. Imai and Berns showed that off-the-shelf cameras could be configured for high-quality spectral imaging [Imai, 1999, 2002].

On the output side, Ohta was the first to put a stake in the ground with his 1972 simulation of a 12-colorant system for minimizing spectral reproduction error [Ohta, 1972]. Berns of the Munsell Color Science Laboratory began publishing on the topic in 1993 [Berns, 1993] and his group has established the concept of least-metameric reproduction through the research of Berns, Iino, Tzeng, Rosen, Imai, and Taplin. In 1999, Rosen demonstrated a spectral end-to-end reproduction of a complex scene [Rosen, 1999].

Spectral reproduction was actually solved more than 100 years ago. At the turn of the twentieth century, Gabriel Lippmann perfected the production of “grainless” albumen emulsions, enabling a form of photography that recorded self-interference of light waves reflected from a mirror in contact with the photographic plate (see Lippmann history). When appropriately illuminated and placed back in contact with a mirror, the prints accurately reproduced scene spectra [Friedman, 1968]. Lippmann’s process was practiced by only a few and represents an anachronism with no influence on current attempts to bring spectra into the color-reproduction workflow.

The remote sensing field made great strides in the direction of spectral imaging. In 1914, Nikolai Morizov traveled in a balloon to obtain spectrograms of the earth’s surface during the full eclipse of the sun. E. L. Krinov, whose airborne spec-trometry experiments started in 1930, included the use of multiple filters to expose standard photographs at specific spectral regions [Krinov, 1947]. The 1960s saw the first of many launches of multi-channel detectors into space that have produced data with widespread applications in meteorology, astronomy, geology, ecology, and the military. Most of the multi-spectral and hyper-spectral remote sensing detection bands lie outside the visible region of the electromagnetic spectrum. This is in contrast to the practice of color reproduction, which obviously focuses on the visible wavelengths. Another major difference between remote sensing and color reproduction is found in the fact that color reproduction places primary weighting on maximizing the quality of scene rendering. Remote sensing is usually far more interested in deriving non-pictorial information from scene data.

In the 1970s, technology had matured to the point that scientists began to envision spectral reproduction systems, build prototypes, and work through theoretical questions [Kotera, 1975, 1976], [Ohta, 1972]. In the past few years, increased sophistication of filtering techniques, digital cameras, computers, printers, and projectors has brought to the field the potential realization of commercial use of spectral techniques [Hill, 2002b], [Rosen, 2002]. At some future point, consumer-level packages could support a spectral-imaging workflow.

Early contemporary examples of work in the field of spectral photography include that of Jussi Parkkinen and Timo Jaaskelainen of Joensuu University [Parkkinen, 1988], [Jaaskel-ainen, 1994], in which an acusto-optical tunable filter was used for capturing selected spectral bandpasses of a large variety of objects and scenes (spectral databases are found at http://cs.joensuu.fi/~spectral/databases/index.html). Bernhard Hill and his group at Aachen applied to the German government in 1987 for work on a multispectral camera for color reproduction [Hill, 2002b] and obtained a patent in 1991 on behalf of Linotype Hell for a multispectral scanner [Hill, 1991]. They have engineered spectral capture devices based on multiple interference filters. Yoichi Miyake’s lab at Chiba University demonstrated some of the earliest practical uses of visible-spectrum imaging through endoscopic research [Miyake, 1989], [Shiobara, 1996]. They have also made large contributions toward the use of spectral imaging for archiving works of art. [Haneishi, 1997]. Burns and Berns at RIT’s Munsell Color Science Laboratory made important analyses of error propagation through multi-channel capture systems [Burns, 1997, 1999] and Imai and Berns advocated the use of commercial trichromatic systems with either supplemental filtration or varying light sources to estimate scene spectra [Imai, 1998, 1999]. The Munsell Lab has also been heavily involved with the use of these systems for fine arts reproduction. [Taplin, 2005].

Spectral spaces include spectral reflectance, spectral transmittance, and spectral radiance. It is possible to calculate spectral reflectance or transmittance from spectral radiance and vice versa if certain parameters are known or assumed. Such parameters include knowledge of the spectral power distribution of the light source hitting an object and knowledge or assumptions around fluorescent properties of media being imaged. As summarized in Table 1, each device type has a spectral space most appropriate for its characterization [Rosen, 2000].

Color input systems interact with spectral radiance. For scanners, where the light source is usually well known and the scanned media is usually flat and uniformly illuminated, it is easy to derive spectral reflectance or spectral transmit-tance from spectral radiance measurements. Camera systems make the derivation of spectral reflectance much harder. First, the ambient light source is separate from the imaging system, is often unknown, and can be mixed. Also, because of non-uniformity of illumination and inter-reflections between objects in a scene, it is very difficult to know exactly the spectral power distribution of the light that falls onto a scene object. Only under the most controlled circumstances can accurate calculation of reflectance from radiance be accomplished in a camera application. It has been shown in much research concerning spectral reconstruction from cameras
that minimizing spectral RMS error is often associated with non-minimum colorimetric error (e.g., [Berns, 2005], [Rosen, 2003] and [Day, 2003]). A method for combining optimal colorimetric transformation with optimal spectral transformation based on parametric decomposition [Fairman, 1987] has been used to produce excellent performance [Berns, 2005].

Color output devices fall in two categories: those that produce light and those devices that reflect or transmit light. For self-luminant displays, the only practical spectral characterization space is spectral radiance. For printers, measuring the spectral reflectance or transmittance of the media is convenient. Three categories of approaches have been undertaken to characterize printers for spectral output: purely physical [Mourad, 2001]; a combination of a physical model and empiricism [Tzeng, 1999]; and brute force empirical [Rosen, 2004]. Recently, the Cellular-Yule-Nielsen-Spectral-Neugebauer model (CYNSN) [Wyble, 2000] has shown good results [Chen, 2004]. “Cellular” means that the parameters to the Yule-Nielsen Spectral Neugebauer model are fit within small regions of colorant space, ensuring high accuracy [Viggiano, 1985].

Derhak and Rosen have described a low-dimensional transform of spectral space for use in spectral color management [Derhak, 2006], [Rosen, 2003a]. This space is called LabPQR and is in a class of spaces that fall within the category of Spectral Colorimetry, because of the fact that it has explicit colorimetric portions and explicit spectral reconstruction portions. This space makes it possible to visualize spectral gamuts and to begin thinking about the spectral gamut mapping problem [Rosen, 2006]. The first three dimensions of LabPQR are CIELAB values under a particular set of viewing conditions. A simple 3 × 31 matrix is used to convert between its CIEXYZ and a distinct spectrum—one from the infinitely large set of metamers for that CIELAB value. The PQR dimensions describe the metameric black [Cohen, 2001] difference between the Lab-derived metamer and the actual spectrum. Thus, the first three dimensions describe colorimetry and the full set of dimensions describe a spectral reconstruction. PQR is derived from a Principal Components Analysis (PCA) of the metameric black space. The PQR dimensions are related to a small number of PCA-derived vectors. Thus, there is a level of inaccuracy introduced by reducing spectral descriptions to this limited number of dimensions.

When attempting to produce a specific response from a color output device, one important consideration is determining whether a request is beyond the device’s rendering capability. The request is considered out-of-gamut when it is outside the palette of available responses. An out-of-gamut request must still generate some response from the color output device. A set of rules is needed to guide this process of spectral gamut mapping.

An approach to spectral gamut mapping conceptually has two stages. First and foremost, colorimetric error is driven as low as possible. Spectral fidelity is addressed once the color-imetric aspect of a process is complete. Figure 61 is a flow diagram for the two-stage spectral gamut mapping algorithm.

Today, there are specialized areas where capturing and reproducing the spectra of an original scene or document is considered an important goal. These include reproduction and archiving of artwork, proofing, medical imaging, and catalog sales. There are many other potential opportunities for the use of spectral information in a color reproduction workflow. Some are being actively investigated. One of the most compelling needs of the color reproduction community is to improve camera capture. Instrumental metamerism ensures that even the best cameras currently on the market do not have the same color sensitivities as humans. Spectral capture techniques will improve this situation.

Once spectral data is available, processing it for output is an obvious choice. Among the advantages of creating a spectral reproduction of an original is that the reproduction will match the original for all devices, whether cameras or humans. Another feature of a spectral copy is that as illumination changes, the match between an original and the reproduction persists. New approaches to processing spectral data are being developed. The use of LabPQR within spectral color management is showing promise. The colorimetric aspects of the space make it a very useful hybrid, ensuring that both colorimetry and spectrophotometry can be taken into consideration when spectral gamut mapping takes place.

Photography, the art and science of drawing with light, has been continually updated and improved. Almost 60 years ago, the director of the Kodak Research Laboratories, CE Kenneth Mees, observed that the original methods for photography had at that point become extinct [Mees, 1942]. Yet, what Mees considered then to be modern is found by contemporary digital photography standards to be antiquated and primitive. Each generation has striven to improve the general baseline of imaging ability. Among the possible new technologies being
developed, spectral reproduction techniques bring important new capabilities and solve many of the current problems.

Spectral Model [next] [back] Specific Problems of Silver Dye-Bleach Photography - Unsatisfactory Color Reproduction, Silver Sludge Formation in the Developer

User Comments

Your email address will be altered so spam harvesting bots can't read it easily.
Hide my email completely instead?

Cancel or