Categorisation through evidence accumulation in an active vision system

Abstract. In this paper we present an artificial vision system that is trained through a genetic algorithm for categorising five different kinds of images (letters) of different sizes. The visual system is composed of a periphery (large visual field with low resolution) and a fovea (small visual field with high resolution) that can freely move over the images. As a consequence, the system solves its task by exploiting its sensory-motor interactions with its (visual) environment. The analysis of the behaviour of the evolved system indicates that categorisation is achieved by: (1) acting so to fall in behavioural attractors that are specific for each category, and (2) integrating perceptual and/or motor information over time through a process of accumulation of partially conflicting evidences. We discuss our results with respect to the general theory of categorisation and, in particular, to the possible roles that action can play in perception