Object Recognition
From the example of speech perception, we can derive a general approach that
researchers bring to the bottom-up study of recognition: they try to determine
the building blocks that perceptual systems use to recognize whole percepts.
For language, your speech perception processes combine environmental infor-
mation about series of sounds to recognize individual words. What are the
units from which you construct your representations of objects in the world?
How, for example, do you decide that a gray, oddly shaped, medium-size,
furry thing is actually a cat? Presumably, you have a memory representation of
a cat. The identification process consists in matching the information in the
percept to your memory representation of the cat. But how are these matches
accomplished? One possibility is that the memory representations of various
objects consist of components and information about the way these compo-
nents are attached to each other (Marr & Nishihara, 1978).Irving Biederman
(1985, 1987) has proposed that all objects can be assembled from a set ofgeo-
metrical ions,orgeons. Geons are not a large or arbitrary set of shapes. Bieder-
manarguedthatasetof36geonscanbedefinedbyfollowingtherulethat
each three-dimensional geon creates a unique pattern of stimulation on the
two-dimensional retina. This uniqueness rule would allow you to work back-
ward from a pattern of sensory stimulation to a strong guess at what the envi-
ronmental object was like. Figure 7.37 gives examples of the way in which
objects can be assembled from this collection of standard parts.
Figure 7.36
Phonemic restoration.
178 Philip G. Zimbardo and Richard J. Gerrig