Random walkers on morphological trees: A segmentation paradigm

The problem of image segmentation is often considered in the framework graphs. In this context, two main paradigms exist: first, vertices a non-directed graph represent pixels (leading e.g. to watershed, random walker or cut approaches); second, directed connected regions, leading so-called morphological trees (e.g. component-trees shapes). Various approaches have been proposed for carrying out from images modeled by such trees, computing cuts these selecting relevant nodes descriptive attributes. article, we propose new way trees. Our approach dedicated take advantage tree an image, enriched multiple attributes each node, using maximally stable extremal regions and defining optimal final segmentation. Experiments, carried on multimodal medical emphasize potential relevance approach.