Statistical evaluation of image segmentation

Image segmentation form an important preliminary step in many high level image processing and computer vision applications. Its importance necessitates the quantitative evaluation of image segmentation results. A few methods have been developed, based on the general principals. In this paper, we propose a novel segmentation evaluation method based on region cardinality ratio and variance. It addresses the limitations in the prior methods and attempts to remove them. The results of our method are superior to the prior quantitative segmentation evaluation techniques due to the explicit usage of inter-cluster relation.