Colour is an important cue in many applications in machine vision and image processing. Nevertheless, colour greatly depends upon illumination changes. Colour constancy goal is to keep colour images stable. This paperrsquos contribution to colour constancy lies in estimating both the set and the likelihood of feasible colour mappings. Then, the most likely mapping is selected and the image is rendered as it would be seen under a canonical illuminant. This approach is helpful in tasks where light can be neither controlled nor easily measured since it only makes use of image data, avoiding a common drawback in other colour constancy algorithms. Finally, we check its performance using several sets of images of objects under quite different illuminants and the results are compared to those obtained if the true illuminant colour were known.

This work was supported by the project 'Active vision systems based in automatic learning for industrial applications'.

Iberoamerican Congress on Pattern Recognition (CIARP), 2003, Havana (Cuba)

9 páginas,3 figuras, 1 tabla

Peer Reviewed