A comparative study of supervised evaluation criteria for image segmentation

This paper presents a comparative study of five supervised evaluation criteria for image segmentation. The different criteria have been tested on a selection of hundred images extracted from the c Corel database for which manual segmentation results provided by experts are available. Nine segmentation algorithms have been considered, most of which are based on threshold selection. In order to compare the behavior of the different criteria towards over- and undersegmentation, three thresholds have been taken into account, for each selected image, to simulate the various situations. Experimental results permit to reveal the advantages and limitations of the studied criteria.