Neural Processing of Overlapping Shapes

Visual information from physically distinct sources often becomes overlaid or finely interspersed in the very process of image formation. For example, one may think of shadows overlaid on surface patterns, or of the multitude of tree branches and leaves that occur in images of a forest. Analyzing such images leads naturally to multi-valued fields of local features. This paper proposes a general model structure for recovering shape characteristics from such data. It uses a “blurred relation” representation to group and segment the data in a way that agrees well with psychophysical and neurophysiological evidence. Some core examples of the behaviour of the model are worked out analytically.