A Formal Framework for the Objective Evaluation of Edge Detectors

Edge detection is one of the most-studied problems in the field of computervision. However,there is no standard method of objectively and accurately evaluating the performance of an edge detector. This work proposes such a framework. The framework is based on using real images, hand-specified ground truth, and a receiver operating characteristic (ROC)curve style analysis of true positive and false positive edgepixels. The method is illustrated by developing a comparison of six edgedetectors.

• Type: Conference
• Year: 1998
• Where: FLAIRS
• Authors: Sean Dougherty, Kevin W. Bowyer
• Keywords: Artificial Intelligence, False Positive Edgepixels, FLAIRS 1998, Hand-specified Ground Truth, Receiver Operating Characteristic