Comparison and evaluation of edge detection algorithms by using edges selected by human

摘要

We propose a method for evaluating edge detection algorithms from the viewpoint of a visual task of humans through psychological tests. First we show the framework of our evaluation procedures. We make ground truths on each method for an image, by selecting candidates for edge points of each method by human beings. Parameters of a edge detector are adjusted to the ground truth by using our proposed optimizing method. These results are evaluated by matched-pair test as visual task of human. This proposed method is applied to evaluate the detection performance of algorithms proposed by Canny, Deriche, Fukui, Marr-Hildreth and Sugiyama-Abe.