Abstract
In this paper, we present an original method to segment color images using a classification of the image histogram in the 3D color space. As color modes in natural images usually do not fit a well-known statistical model, we propose a classifier that rely on mathematical morphology and, more particularly, on the watershed algorithm. We show on various images that the expected color modes are correctly identified and, in order to obtain coherent region, we extend the method to make the segmentation contextual.