Local Image Operators and Iconic Structure

We study the image structure revealed by a coherent set of local image operators (e.g., edge finders, ... ). An observation specifies an equivalence class of images of infinite cardinality. A given set of image operators thus induces a parcellation of all images into such equivalence classes. Taking tolerances into account leads to a further growth of the equivalence classes. We analyze the qualitatively different image structures that can be classified by the set of image operators, that is the discriminatory power of the image operators.