From the article Color Analyzers

Perhaps the most basic approach to color analysis is a commercial off-the-shelf color CCD camera. This device does not contain dispersive elements, but with the optimum software, it can compare a sample under test against a standard to determine whether the two match. Besides cost, the advantage to this approach versus a fiber optic spectrometer is that the detector produces a two-dimensional image, whereas a conventional spectrometer analyzes only a point on the object under test. For applications such as color matching or for detecting the transition from one object to another based on color, the imagers can provide an effective, economical solution.

What’s more, advances in dichroic filter technologies can enhance imaging at red, green and blue wavelengths, essentially tuning the spectral characteristics of the image sensor to reflect or transmit over specific bands to produce the optimum response. New processes also make it easier to produce multilayer coatings, so that a single CCD array becomes a multispectral imager: Instead of measuring color at three wavelengths, one can measure up to 11 wavelengths.

Typical color imagers do have their limitations, however. CCD-based color analyzers, and even filter-based colorimeters that use silicon photodiodes, measure only a limited range of wavelengths. Physiologically speaking, human color perception is centered on three color stimuli (approximately red, green and blue), but each response spans a broad spectrum of wavelengths. This has been modeled by the CIE as the (x,y,z) color space, which is illustrated by the (x,y) chromaticity diagram for color matching.

In generating color images from the monochromatic sensitivity of a CCD detector, manufacturers overlay the device with red, green and blue filters. These define a triangle on the (x,y) chromaticity diagram but cannot encompass the entire range. In other words, the response of the color CCD imager is not a true representation of all the colors the human eye sees.

What we think of as color is essentially a broadband response between 380 and 780 nm, but analysis by a color imager must take that spectral data and collapse it down to a three-point representation. Certain regions of the color space will never be encompassed by even the best CCD imagers. Transforming the response from a broadband spectrum to the three-point representation requires sacrificing a significant amount of information.

As a result, imager-based analyzers are strongly dependent on lighting conditions because a material’s appearance tends to vary depending on the illumination. With the right lighting, two materials can appear to be identical in color even if the reflected spectral power distributions differ – an effect called metamerism. If the lighting changes, however, the colors can look significantly different. Controlled lighting conditions thus become important for consistent results. A spectrometer, in contrast, captures the light reflected, transmitted or emitted by a sample and uses a dispersing element to split it into discrete wavelengths. Because the instrument captures the complete spectral power distribution rather than merely measuring power at three specific wavelengths, the resulting color measurement is more robust.