Estimation of fuzzy sets for computational colour categorization

Colour is an important visual cue for computer vision applications. However, until recently, the automatic assignment of names to image regions has not been widely used due to the nonexistence of a general computational model for colour categorization. In this article we present a model for colour naming based on fuzzy-set theory, in which each of the 11 basic colour terms defined by Berlin and Kay(1) is modeled as a fuzzy set with a characteristic function that assigns a membership value to the category to any colour sample. The model is based on combining two well known functions, a sigmoid and a Gaussian, to define a membership function for colour categories. It is denoted here as the sigmoid-Gaussian function and it fulfills a set of properties that make it adequate to this purpose. The characteristic functions for each colour category have been fitted to data obtained from a psychophysical experiment and the model has been tested on the Munsell colour array to show its validity. The results obtained indicate that our approach can be very useful as a first step to expand the use of colour-naming information in computer vision applications. (C) 2004 Wiley Periodicals, Inc.