A Computational View of Perception

A computational approach to visual perception is outlined. The investigation centres on two problems. One concerns the shape, and the other the edges, of dot figures. In determining edges it is not possible to delimit arbitrarily the region on which each point computes. So in this context a global computation is defined. This leads to a novel way of investigating clustering. When the edges of a figure have been sequenced to yield the boundary it is then possible to determine its shape. The study of shape has been confined to convex figures. In these an attempt is made to extract global units which correspond to corner-like regions. Finally, it is suggested that a theory which evolves in the manner described has a more unitary aspect than other approaches.