Assumes we have a static image that has been pre-processed:

• objects have been identified
  (we use unique identifier, but any set of "distinguishing characterstics" would be ok)

• objects have been segmented
  (set of pixels that comprise the object is available ... gives object pointset/region)

Approach draws on work by

• [Chang et al.] on 2D strings

• [Allen] on interval relationships

• [Egenhofer] point-set topological relationships

Basic idea: define relationships between objects in terms of

• topological relationship   (defined by pointsets (boundary,interior))

• interval relationships of object projections on x and y axes