A measure of quality for evaluating methods of segmentation and edge detection

摘要

A new measure of quality is proposed for evaluating the performance of available methods of image segmentation and edge detection. The technique is intended for the evaluation of low error results and features an objective assessment of discrepancy with respect to the theoretical edge, in tandem with subjective visual evaluation using both the neighbourhood and error-interaction criteria. The proposed mathematical model is extremely simple, even from the perspective of computational execution. A training of the measure has been put in practice, which uses visual evaluation of a set of error patterns by a team of observers. Encouraging results were obtained for a selection of test images, especially in relation to other recently proposed and/or currently employed quality measures.