A Subjective Method for Image Segmentation Evaluation

Image segmentation is an important processing step in many image understanding algorithms and practical vision systems. Various image segmentation algorithms have been proposed and most of them claim their superiority over others. But in fact, no general acceptance has been gained of the goodness of these algorithms. In this paper, we present a subjective method to assess the quality of image segmentation algorithms. Our method involves the collection of a set of images belonging to different categories, optimizing the input parameters for each algorithm, conducting visual evaluation experiments and analyzing the final results. We outline the framework through an evaluation of four state-of-the-art image segmentation algorithms—mean-shift segmentation, JSEG, efficient graph based segmentation and statistical region merging, and give a detailed comparison of their different aspects.