next up previous
Next: The factor models Up: Methods Previous: The color matching

The spectral distributions

Vrhel et al [6] measured the spectral distributions of a diverse sample of natural and man--made objects. They measured reflectance spectra of 64 Munsell chips, 120 Du Pont paint chips and 170 objects from the environment using a high precision spectrometer and illuminance with a color--temperature of kelvin to avoid fluorescence effects. A calibrated white reflectance standard (X-RITE reference tile) was used to compensate for the illuminant spectrum when calculating the object reflectance spectrum.

For each object, the reflectance intensity was measured within the 31 different 10nm wavelength bands located between 395nm and 705nm. These measurements comprise 31 different scores representing the reflectance spectrum for each individual object. The factor analyses were performed on a covariance matrix of these scores. Only the 15 odd numbered 10nm bands were used for the maximum likelihood factor analysis since the correlation between successive bands was so high that maximum likelihood estimation procedures had trouble converging.

The natural objects included 31 measurements of human hair and skin, and measurements of 78 different samples of rocks, leaves, bark, wood, flowers, fruit, vegetables and grains. The man--made objects included a wide variety of clothing and personal objects. While this sample of environmental reflectance spectra doesn't claim to be a representative sample of the natural spectral environment, it is relatively broad based and complex. For purposes of these analyses, the ``natural'' objects and the man--made objects were each analyzed separately as well as combined together. Since partitioning the objects into natural versus man--made categories did not affect the results, only the analyses from the combined data set will be reported here.

Steven M. Boker
Tue Aug 15 16:09:36 EDT 1995