Abstract

Bilinear models of low dimensionality can accurately model interactions of naturally occurring illuminant and reflectance spectra. There is, therefore, a basis for an observer to recover surface characteristics independently of the illuminant. Much previous work has assumed knowledge that certain colors corresponded to surfaces sharing a common illuminant. Such information would not be available, however, to a low level process analyzing the possibility that an χ-junction in the image arose from a colored shadow crossing a reflection boundary. Bilinear models that use different basis functions for illuminants and reflectances are necessarily asymmetrical: the predicted appearance of a red surface under yellow light is generally different than that of a yellow surface under red light. Because of this, two analyses must be performed at each χ-junction, corresponding to each of the limbs of the χ being considered the reflectance boundary while the other is the illumination boundary. Symmetric models may be constructed by using a single set of basis functions to model both illumination and reflectance spectra. This reduction is advantageous in the description of color transparency with more than two surfaces, since instead of surfaces and illuminants interacting to form a color, everything may simply be represented as a color.

© 1992 Optical Society of America