An Abstraction Model of Visual Perception

Lorenza Saitta

Università del Piemonte Orientale

Dipartimento di Scienze e Tecnologie Avanzate

Corso Borsalino 54, 15100 Alessandria

saitta@unipmn.it, saitta@di.unito.it

Abstract

The talk presents a computational model of abstraction, based on perceptual principles, well suited to describe and handle visual and spatial concepts and data. The model spans two dimensions: one concerns the type of representation and processing applied to visual input, starting from the signal and ending with an abstract theory; the onter one concerns the selection of a suitable level of detail for representing the spatial concepts when a specific goal has to be reached. In order to concretely implement the transformation between different levels, a number of abstraction operators are defined. The semantics of the operators, as well as conditions for applicability, will be analyzed.

The model will be shown to encompass and to give a precise basis to previous attempts, in vision, to describe changes of representation, as well as to visual perceptual findings from cognitive sciences. A practical application to cartographic generalization in GIS will also be described.