Rationalizing edge detectors

It is contended that two stages can be isolated in the formal derivation of an operator; the obtaining of masks meaningful with respect to the phenomenon being sought, and the deriving of the procedural part.

Abstract

It is contended that two stages can be isolated in the formal derivation of an operator; the obtaining of masks meaningful with respect to the phenomenon being sought, and the deriving of the procedural part. The first of these stages is achieved either by surface fitting or by use of an orthogonal basis of two-dimensional functions; the second, by adopting the method of convolution. Within this framework, we comment on the rationale underlying some of the currently used operators.