Imaging systems are multivariate, involving response functions or even sets of response functions—the red, green, and blue characteristic curves of photographic systems, for example—that relate image properties to scene properties. These functions must be simultaneously optimized to produce the best possible system. While the preferred methods for empirically optimizing the characteristics of a product or process are those of designed experimentation and response surface methodology, there is no widely accepted method that enables the application of these techniques to multivariate problems and therefore to imaging products and processes. This situation is changed with the advent of the desktop computer. We will describe a conceptually simple, though computationally intensive, method that enables application of designed experimentation and response surface methodology to multivariate systems and imaging systems. The method discussed will produce a more robust manufacturing process as well as a better product.
Allan Ames, Neil Mattucci, Douglas Hawkins, "Empirical Optimization of Imaging Processes by Use of Designed Experiments and Quality Loss Functions" in Journal of Imaging Science and Technology, 1999, pp 166 - 169, https://doi.org/10.2352/J.ImagingSci.Technol.1999.43.2.art00012