PERCEPTUAL EQUATIONS - IMPLICATIONS FOR COMPUTER VISION

Vision can be understood as a problem of solving systems of equations. Several approaches to this problem have evolved within psychology. Re search on computer vision has gravitated towards one, which assumes th e task of vision is to invert the distal-proximal stimulus transformat ion. A second, the psychophysical approach, attempts to solve individu al equations separately, which leads to a search for heuristic methods of solving ill-posed equations. There are well known objections to bo th of these. In contrast a new approach, descended from the transactio nalism, accepts the need to deal with systems of equations which, cons idered separately, define invariance relations. Classical objections t o the transactional approach can be met constructively. Perceptual var iables stand in invariant relations to sensory variables, not to attri butes of the physical stimulus: recognising that sensory variables are a key intermediate stage leads to a clarification of the separate par adigms which are relevant to vision research. Equation-solving is not just a metaphor: if a pseudoscope is used to present inconsistent perc eptual data, vision produces effects very like a computer's response t o inconsistent systems of equations. The transactionalist approach pro vides a framework for solving long-standing problems such as neutral c olor constancy.