Determination of optimal general edge detectors by global minimization of a cost function

摘要

We present a numerical optimization technique based on genetic algorithms or simulated annealing functional search engines, and use it to optimize linear and nonlinear versions of Canny's performance measure for general edge operators. No prior knowledge of the point-spread-function is assumed. We demonstrate that the results obtained are superior both for an ideal and a real edge to those arrived at by previous researchers.