Skip to Main Content
The explosion of image content is closely connected with segmentations efficiency. However, there is no agreement as to what a good segmentation is due to hard data and applications dependence. To reduce the gap between low-level features and high-level semantic, collections of image partitions produced by different segmentation algorithms are often considered. We propose, theoretically ground and experimentally explore a new metric on segmented images or on arbitrary partitions of finite sets in general.