New performance criteria of edge detection algorithms

A method of evaluating edge detection algorithms is presented, in which a contour picture is segmented into lines. These lines are classified and used for feature extraction. The procedure can be applied either in connection with a reference contour picture for comparing binary pictures or by evaluating the output of edge detectors applied to test pictures with unknown edges. This evaluating method combines two possibilities: One is to determine thresholds for given properties and the other is the comparison of the performance of various edge detection algorithms.