Characterization of segmentation methods by multidimensional metrics: application to the delimitation of structures

In order to detect coherent structures on satellite images, the authors used an extended set of segmentation algorithms. Applying this set to a group of test images, a collection of segmented images was obtained, where different methods highlighted different structures. Due to the diversity of valid solutions, it was necessary to evaluate the adequacy of the different options for specific problems. The authors propose a set of metrics that allows them to classify the methods via aggregation functions. They started working with satellite images, but, actually, they are working with general images (artificial when testing segmentation methods). Metric definitions are applicable at any field where it is required the detection of specific objects.