An unsupervised color image segmentation based on morphological 2D clustering and fusion

A segmentation method for color images is presented in this work. A morphological unsupervised 2D multiband histogram clustering provides an initial coarse segmntation of the image. Region information is then used and a novel technique is introduced to simplify the Region Adjacency Graph by merging candidate regions until the stabilization of a segmentation criterion. Merged regions are refined by a color watershed. The whole method requires the definition of only one single parameter which acts as a kind of multiscale parameter.