Peer groups, neighbor groups, and edge detection

Abstract

Standard edge detectors such as Canny edge detector use a local gradient operator to find the ridges in digital images. In this study we show that peer groups and neighbor groups, a new measure introduced in this paper, can be used to improve the performance of gradient operators. Some of the pixels near the threshold are weaker edge pixels of the objects in the image and others are unimportant edges created by unevenness of the background. We find that peer groups and neighbor groups of these marginal pixels can be used to make a better decision regarding them than using a single rigid threshold value.