Using reflectance models for color scanner calibration.

We examine the use of linear spectral reflectance models for calibrating a color scanner to generate device-independent CIE XYZ values from scanner vectors. Polynomial regression approaches to color scanner calibration use parameterized functions to approximate the calibration mapping over a set of training colors. These approaches can perform poorly if the parameterized functions do not accurately model the structure of the desired calibration mapping. Several studies have shown that linear reflectance models accurately characterize a wide range of materials. By viewing color scanner calibration as reflectance estimation, we can incorporate linear reflectance models into the calibration process. We show that in most cases linear models do not constrain the calibration problem sufficiently to allow exact recovery of X, Y, Z from a scanner vector obtained with three filters. By examining a series of methods that exploit information about reflectance functions, however, we show that reflectance information can be used to improve the accuracy of calibration over that of standard methods applied to the same set of inputs.