This article is on logical aspects of uncertain, unstable perceptual information. In a review of logics of direct perception reports we stress that their semantics is often based on perfect, irretractable `points'. We argue that this way of modelling has to be replaced by a more principled one which takes the retraction of perceptual information seriously. To do so, our logic tries to stay close to the psychological models of perception developed in Marr 1982 and subsequent work. In such models perception is a multi-layered process. The different layers have filters of different gradation, which makes perception at each of them approximate. Indeed, our main task will be to formalise the layers and the ways in which they may refine each other, and to develop a logic in which description varies with such measures of refinement. We show that the logic of vision gives the right predictions for the logical behaviour of perception reports. The article ends with a study of different notions of stability within the logic. Van der Does & Van Lambalgen 1997 (LP-1997-07) gives a detailed treatment of the linguistic evidence, and of the psychological viability of the proposed model.
An updated version of this article can be obtained here: A Logic of Vision: Preliminaries. See also the information on Jaap's homepage.