Performance evaluation of image segmentation. Application to parameters fitting

This paper deals with the performance evaluation of image segmentation. The goal of this work is to show some techniques that enable the comparison of different segmentation results. We first present a visualization method that facilitates the visual evaluation of a segmentation result. Next, we focus on unsupervised evaluation criteria that do not take into account any a priori knowledge to quantify the quality of a segmentation result. Finally, we use these evaluation criteria to determine the best fitted parameters of a segmentation method for a given image and a desired level of precision.