Plato's Cave: The Gestalt View
The Gestalt View of Vision
The Gestalt Objection
The Gestalt movement in psychology emerged as a reaction to the
"reductionist" view of perception which held that perception can be
decomposed into elemental components, which can then be considered
individually. This reductionist approach is still the prevailing view
in
[the conventional approach to visual processing.]
The Gestaltists produced a large number of
examples
of perceptual phenomena which cannot be explained by the reductionist
view. Nevertheless, computational models of vision continue to
reflect reductionist architectures even though these
architectures cannot address the Gestalt objection, and hold
no promise in principle of ever addressing the Gestalt
issues.
The Gestalt Laws
The Gestalt psychologists cataloged a number of
laws of perceptual grouping
which provide evidence for the nature of low-level visual processing,
as well as a clue to the nature of the visual representation.
While it is possible in theory to implement these Gestalt laws using a
reductionist architecture, in practice this can only be achieved in
highly simplified examples, and tends to lead to a
combinatorial explosion
in the required computation. The Gestalt grouping laws suggest a
totally different paradigm of computation from those proposed in the
conventional view.
The Gestalt Models
In an attempt to explain the Gestalt laws of grouping, Gestalt
theorists proposed models of perception involving analog field-like
forces whereby the final global percept emerged by a parallel
relaxation of multiple local forces in the manner of a soap bubble,
whose spherical shape is not defined by a stored template or blueprint
of the final shape, but rather it emerges naturally as a consequence
of the multiple local forces in the skin of the bubble.
While the Gestalt insight about the mechanism of perception was right
on target, Gestalt modelers were never able to define a system that
made quantifiable predictions of the apperence of the final percept
given any arbitrary input. It is for this reason that the Gestalt
view of perception has fallen by the wayside of the mainstream of
thinking in vision research.
The fundamental questions raised by the Gestalt movement have remained
unanswered for so long that they are being forgotten, or considered
irrelevent to the modern view of vision. Nevertheless the Gestalt
challenge to the reductionist view remains as valid as it ever was.
Any model of vision which does not address the Gestalt
issues is worse than a non-solution; it is a diversion from
the real questions of visual perception.
Return to argument
Return to Steve Lehar