Edge detection from global and local views using an ensemble of multiple edge detectors

Abstract

Edge detectors such as Canny, Sobel, Prewitt, and Laplacian consider only local grayscale transition strength or contrast. From a global point of view, considering the entire image, detection results using these operators either include too many noisy edge points when the threshold value is low, or they fail to include correct edges when the threshold value is high. A novel algorithm for edge detection that incorporates global constraints with local contrast information to address this issue is presented in this paper. We describe explorative research on edge detection using collaborative learning. An evaluation of the proposed algorithm reveals very promising results. © 2008 Springer Berlin Heidelberg.