Shape Representation in Parallel Systems

There has been a recent revival of interest in parallel systems in which computation is performed by excitatory and inhibitory interactions within a network of relatively simple, neuronlike units [1 2 3 4]. At the early stages of visual processing, individual units can represent hypotheses about how small local fragments of the visual input should be interpreted, and interactions between units can encode knowledge about the constraints between local interpretations. Higher up in the visual system, the representational issues are more complex. This paper considers the difficulties involved in representing shapes in parallel systems, and suggests ways of overcoming them. In doing so, it provides a mechanism for shape perception and visual attention which allows a novel interpretation of the Gestalt slogan that the whole is more than the sum of its parts.