Automatic building of granular fuzzy color spaces

In this paper, several approaches to learn granular fuzzy color spaces from images are proposed. Granular fuzzy colors were proposed as an appropriate way to model color categories, even when a color category has several color shades. This approach uses crisp color prototypes to build granular fuzzy color spaces, being used a standard set like ISCC-NBS ones. However, it has been demonstrated that this set lacks the versatility required to accurately capture the specific hues present in a given image. In this paper, it is proposed to learn the prototypes on the basis of the image content, and build granular fuzzy color spaces. We demonstrate the advantages of our approach through specific real-world experiments, grounded in human perception.