ABSTRACT
In an attempt to tackle this problem, we started from recent work on object recognition in perceptual psychology as well as in computer vision. For example, Biederman (1987) has suggested that humans recognize objects fast and efficiently because almost all objects can be defined as a particular configuration of a small number of components that can be identified in parallel. In addition, the characteristics determining the identity of the components can be recovered from non-accidental properties in the image of the object such as parallelism, symmetry and convergence. Computer vision researchers have specified how these properties can be detected (Lowe, 1987) and how an object can be segmented in its basic parts (Hoffman and Richards, 1984). Furthermore, alternative theories stressing the role of parts in object recognition have been formulated (e.g. Pentland, 1987).
