Research Overview

A Primer on Probabilistic Approaches to Visual Perception

Figure 1 View »

Figure 2 View »

The Central Problem

Figure 1

« Close

Figure 1 / The fundamental factors that determine the luminance of any stimulus (or component thereof) are illumination, reflectance, and transmittance. Because behavior in response to the stimulus will be successful only if the relative contributions of each of these factors are in some sense known, seeing lightness or brightness according to the physical intensities (luminances) in the stimulus as such would be a poor strategy of vision.

Figure 2

« Close

Figure 2 / The inherent ambiguity of any three dimensional object projected onto a plane. As indicated in this diagram, the same retinal projection can be generated by objects of different sizes at different distances from the observer, and in different orientations with respect to the observer.

The Answer Indicated By Perceptual Evidence

The apparent answer to this problem has come primarily from studies of what people actually see in response to visual stimuli. The central tenet of the theory of vision that has emerged from such studies over the last few years is that the dilemma of inherent stimulus ambiguity is solved by having the pattern of light on the retina trigger reflex patterns of neural activity that have been shaped entirely by the past consequences of visually guided behavior. In this conception of vision, perceived images accord with the cumulative probability of what the same or similar stimuli have signified in the past history of the species and the individual. This operational determination of what we see by the probability distributions of stimulus sources explains why visual percepts do not co-vary systematically with the characteristics of the light stimulus (or with the physical properties of the objects that generated the stimulus, as indicated by the various demonstrations presented on the website). Much to the advantage of the observer, percepts co-vary with the efficacy of past actions in response to visual stimuli, and thus only coincidentally with the measured properties of the stimulus or the underlying objects. This strategy ensures that visually guided responses will usually deal successfully with the objects and conditions that have given rise to retinal stimuli whose sources are, as Berkeley pointed out, unknowable in any direct way. The counterintuitive conclusion that follows from this evidence is that what we see is a statistical consequence of an accumulation of past experience rather than a veridical representation of the retinal stimulus or the objects that confront the observer in the present (the term 'past' again referring to both phylogenetic and ontogenetic experience).

For those who are interested in thinking about this in more formal terms, a primer is available that compares the present approach (called empirical ranking theory) with Bayesian decision theory.