In this paper we develop a method for verifying edges for visual inspection
purposes. This method characterizes and examines a candidate edge against
a given condition which can be designed and modified according to the requi
red specifications of an application. A criterion is established using the
edge strength, edge length and the distributions of grey level values of th
e pixels located in an edge neighbourhood. In consideration of computationa
l speed and the accuracy of the locations of the edges, we first convolve t
he image with a Gaussian impulse response with a single small smoothing sca
le factor sigma to generate candidate edges. The candidate edges are verifi
ed against the criterion and an edge is then retained or discarded on the b
asis of its local characteristics. (C) 1999 Elsevier Science B.V. All right
s reserved.