Toward a computational theory of perception

Abstract

This paper sketches a computational theory of perception. Perception is the process of acquisition of information about the environment by the senses. In this proposal, a model of the world and models of how information is projected into the senses are needed. Given these models, perception is the process of finding one configuration of the known world that corresponds well to the sensed information. The key point is how to validate a perception: a configuration of the environ- ment is validated as a perception if the expected sensory in- formation (the projection from the tested configuration into the senses) is similar to the observed one, up to the precision of the senses. Learning the models for the world and the senses is considered as different process from perception. A complete computational implementation is presented for a toy example of visual perception in a world of flat and su- perposed squares.