Visual Perception and Representation of Shape

Shape is a crucial source of information for identifying and inferring the functional properties of an object. Classic work dating back to the Gestalt psychologists has shown that the information we encode about an object’s shape is different from the sum of all visual elements present during sensation. For example, we often see clouds shaped like horses or other objects even though there is very little similarity between the physical features of a horse and any cloud. We study the information processing systems by which sensory elements are transformed into shape representations.

How does the visual system encode information present in sensation into more abstract, symbolic descriptions of shape? Abstract shape representations should be more informationally compact than the physical contours of objects while still preserving contour information that is essential to an object’s shape. One hypothesis we are investigating is that the visual system segments contours into regions of similar curvature and represents a shape as a relatively constrained set of smoothly connected constant curvature segments.

Arrangements of unconnected dots sometimes give rise to shape percepts despite having no explicit orientation or curvature information. We study the geometric relationships between dots that result in a shape percept. These simplified dot displays can also be informative about what visual information is used in encoding the shape of an object. For example, changing the position of individual dots along a virtual shape contour reveals very little sensitivity in the visual system to local feature changes.

Visual Perception of Contours
Sample shape and constant curvature representation.

Spatiotemporal Boundary Formation


  • Philip J. Kellman Philip J. Kellman
  • Everett Mettler Everett Mettler
  • Suzy Carrigan Suzy Carrigan
  • Nicholas Baker Nicholas Baker


  • Gennady Erlikhman Gennady Erlikhman
  • face Jennifer Mnookin (UCLA)
  • face Itiel Dror (UCL)
  • face Hongjing Lu (UCLA)
  • face Patrick Garrigan (SJU)
  • face Brian Keane (Rutgers)
  • face Tandra Ghose (TU Kaiserslautern)

Selected Publications