Varieties of perception 37
so understood, is divided into stages (summarised in Marr 2010, 36–38): from the
representation of the two-dimensional retinal image, a “primal sketch” is generated
as a symbolic description of properties of the input image in terms of “intensity
changes” (edges, boundaries, virtual lines, etc.). On these grounds, surfaces, textures and their orientation are described and apparent motion is constructed. Only
from there, a three-dimensional representation, now understood in a non-technical
sense, is generated that describes shapes and spatial arrangements and that is centred on objects. Only at this last stage, conscious perception is accomplished. In this
process, the physical characteristics of a “scene” are “recovered”, that is inferred,
from the image (Marr 2010, 330f ). Those physical characteristics work as objective
constraints upon perception – which remains otherwise underdetermined by what is
present in the perceiving organism’s surroundings.
Perception remains underdetermined in Marr’s theory inasmuch as the retinal
image from which the perceptual process commences is conceived of as a mere
pattern that does not convey information, and certainly no informational content of
the kind defined by Dretske, about the world. The pattern is made up of the distribution of intensity values and their transformations across the retinal image. These
intensity changes are caused by the geometry and the reflectance of the visible
surfaces, the illumination of the scene and the viewpoint (Marr 2010, 41). The distribution and changes of intensity values thus caused are all the perceptual system
has for processing. Detection of intensity changes is accomplished by algorithmic
operators that transform the input so as to produce values for mathematical functions with peaks, troughs and zero crossings as their output (Marr 2010, 54–73).
Quite self-evidently on this account, the same pattern and the same distribution of values for a set of mathematical functions may be caused by a variety of
different things, so that the organism has to make inferences as to how the world
stands from the analysis of these patterns in further stages of visual processing.
This is the problem of the ambiguity of the retinal image or of inverse projection,
as classically stated in George Berkeley’s Essay Towards a New Theory of Vision
(1709). Berkeley’s initial observation was that distances cannot be directly perceived, and that space as such cannot be seen, and that similar conditions apply
to other perceptual qualities, such as the magnitude of objects. Identical retinal
images can be caused by various objects, under various conditions, in various
constellations. Where the association of experiences was the main disambiguating
factor in Berkeley, inference of spatial properties from the retinal image serves
that purpose in Marr (as in the “geometrical” theories of visual perception by
Descartes and Malebranche to which Berkeley’s was intended as an alternative).
If, however, such inference is insufficient for disambiguation, and thus if the frog
has no means of telling apart the intensity changes for insects and lead pellets
moving across his visual field, there is not much that Marr’s theory could do about
this unfortunately ambiguous state of affairs. Issues of this kind would have to be
relayed to ecological or other biological theories that, very much independently
from a theory of perception, account for how frog’s environments are shaped and
how changes in environmental conditions will affect the frog’s survival and reproductive success.
so understood, is divided into stages (summarised in Marr 2010, 36–38): from the
representation of the two-dimensional retinal image, a “primal sketch” is generated
as a symbolic description of properties of the input image in terms of “intensity
changes” (edges, boundaries, virtual lines, etc.). On these grounds, surfaces, textures and their orientation are described and apparent motion is constructed. Only
from there, a three-dimensional representation, now understood in a non-technical
sense, is generated that describes shapes and spatial arrangements and that is centred on objects. Only at this last stage, conscious perception is accomplished. In this
process, the physical characteristics of a “scene” are “recovered”, that is inferred,
from the image (Marr 2010, 330f ). Those physical characteristics work as objective
constraints upon perception – which remains otherwise underdetermined by what is
present in the perceiving organism’s surroundings.
Perception remains underdetermined in Marr’s theory inasmuch as the retinal
image from which the perceptual process commences is conceived of as a mere
pattern that does not convey information, and certainly no informational content of
the kind defined by Dretske, about the world. The pattern is made up of the distribution of intensity values and their transformations across the retinal image. These
intensity changes are caused by the geometry and the reflectance of the visible
surfaces, the illumination of the scene and the viewpoint (Marr 2010, 41). The distribution and changes of intensity values thus caused are all the perceptual system
has for processing. Detection of intensity changes is accomplished by algorithmic
operators that transform the input so as to produce values for mathematical functions with peaks, troughs and zero crossings as their output (Marr 2010, 54–73).
Quite self-evidently on this account, the same pattern and the same distribution of values for a set of mathematical functions may be caused by a variety of
different things, so that the organism has to make inferences as to how the world
stands from the analysis of these patterns in further stages of visual processing.
This is the problem of the ambiguity of the retinal image or of inverse projection,
as classically stated in George Berkeley’s Essay Towards a New Theory of Vision
(1709). Berkeley’s initial observation was that distances cannot be directly perceived, and that space as such cannot be seen, and that similar conditions apply
to other perceptual qualities, such as the magnitude of objects. Identical retinal
images can be caused by various objects, under various conditions, in various
constellations. Where the association of experiences was the main disambiguating
factor in Berkeley, inference of spatial properties from the retinal image serves
that purpose in Marr (as in the “geometrical” theories of visual perception by
Descartes and Malebranche to which Berkeley’s was intended as an alternative).
If, however, such inference is insufficient for disambiguation, and thus if the frog
has no means of telling apart the intensity changes for insects and lead pellets
moving across his visual field, there is not much that Marr’s theory could do about
this unfortunately ambiguous state of affairs. Issues of this kind would have to be
relayed to ecological or other biological theories that, very much independently
from a theory of perception, account for how frog’s environments are shaped and
how changes in environmental conditions will affect the frog’s survival and reproductive success.