On the quantitative performance evaluation of image analysis algorithms

The paper contains a brief review of main approaches to the comparative performance evaluation of image analysis algorithms. Some empirical methods used for the comparative evaluation of edge detectors and image segmentation algorithms are considered and quantitative criteria employed in these methods are studied. Problems associated with the use of these criteria are described. Finally, using the edge detector evaluation as an example, we propose an empirical method, called EDEM, which is implemented using our proprietary software system PICASSO.