UNSUPERVISED FILTERING OF COLOR SPECTRA

We describe a class of unsupervised systems that extract features from databases of reflectance spectra that sample color space in a way tha t reflects the properties of human color perception. The systems find the internal weight coefficients by optimizing an energy function. We describe several energy functions based on second- and fourth-order st atistical moments of the computed output values. We also investigate t he effects of imposing boundary conditions on the filter coefficients and the performance of the resulting systems for the databases with th e reflectance spectra. The experiments show that the weight matrix for one of the systems is very similar to the eigenvector system, whereas the second type of system tries to rotate the eigenvector system in s uch a way that the resulting filters partition the spectrum into diffe rent bands. We also show how the system can be forced to use weight ve ctors with positive coefficients. Systems consisting of positive weigh t vectors are then approximated with Gaussian quadrature methods. In t he experimental part of the paper we investigate the properties of thr ee databases consisting of reflectance spectra. We compare the statist ical structure of the different databases and investigate how these sy stems can be used to explore the structure of the space of reflectance spectra. (C) 1996 Optical Society of America