It is shown that color constancy can be achieved only under a restricted set of circumstances. An algorithm is discussed that utilizes only structure intrinsic to the problem of color constancy. This algorithm is robust and flexible, and can easily be learned by observing photoreceptor responses. In particular this approach generalizes naturally to the case when there is spatial variation in the illuminant content, and has been shown to work on real color images. This formulation is also shown to be capable of reproducing qualitative aspects of E. H. Land's two color experiments (1986).
|Original language||English (US)|
|Title of host publication||Second Int Conf on Comput Vision|
|Publisher||Publ by IEEE|
|Number of pages||10|
|State||Published - 1988|
ASJC Scopus subject areas