Device-independent color imaging demands a reliable color-appearance model.
We present a method for faithfully approximating color-appearance models b
y means of feed-forward neural networks trained with the error back-propaga
tion algorithm. In particular we present experimental evidence that in seve
ral "standard" viewing conditions recommended for testing color-appearance
models, the same network architecture is capable of learning quite satisfac
torily the transformations performed by different color-appearance models.
(C) 1999 John Wiley & Sons, Inc.