Unlike the human visual system, image-capture sensors lack the ability to adapt to the ambient exposure conditions. Accordingly, post-capture adjustments are required to compensate for variations in the exposure level (brightness) and capture illuminant chromaticity (white balance). These adjustments are typically made by scaling linear RGB sensor signals or by applying additive adjustments to logarithmic encodings of these signals. Even if such adjustments result in perfect balance for a given reference color in the scene (such as a reference neutral gray card), some degree of color error will typically be introduced into other scene colors. Such errors can be particularly problematic if they result in significant hue errors or if the color name is changed. The magnitude and direction of color errors introduced by balance adjustments will depend on the color space in which the balance adjustments are applied. For additive RGB color spaces, the fundamental attribute affecting these color errors is the chromaticities of the color space primaries. This paper develops a method for characterizing such color errors, and evaluates the performance of a number of commonly used color spaces. A method to determine the optimal color space primaries for which the application of balance adjustments produces minimal color errors is also presented.
Geoffrey Woolfe, Kevin Spaulding, Edward Giorgianni, "Optimal Color Spaces for Balancing Digital Color Images" in Proc. IS&T 11th Color and Imaging Conf., 2003, pp 66 - 70, https://doi.org/10.2352/CIC.2003.11.1.art00012