Abstract

Perception depends on change, as is shown by the fading of artificially stabilized retinal images, which fade gradually to disappear after some seconds. Likewise, retinal signals generally indicate change in space or time, not absolute values of local stimulation. In Land’s early retinex model, a central integrator provides estimates of lightness and color by combining retinal signals that specify factors of change. Any such central integration is evidently imperfect since stabilized images do fade. The physiological basis of this fading remains obscure. Rushton attributed it to sensitivity regulation, and he supported this by showing that a faded border remains invisible when the intensity is changed by the same factor on both sides of the border, implying reciprocal adjustments of sensitivity on the two sides. In experiments with Dirk Beer and other collaborators, we make similar observaions using unstabilized stimuli in the form of diffuse blobs (with Gaussian excitation profiles). These readily fade under steady fixation much as artificially stabilized images do. Once faded, blobs of sufficiently high intensity remain invisible when the image intensity changes by a constant factor across the non-uniform intensity profile presented by blob and background. If steady-state sensitivity is inversely proportional to input intensity, their product is constant, providing a plausible basis for fading. Yet the time constant of sensitivity regulation is too fast to account for the fading. And the afferent neural signals are either too transient or too sustained to account for it. To connect neurophysiology with perception one can either assume a central leaky integrator for the transient signals, or an AC coupling of the sustained signals. In one form of the integrator model, the afferent input is used to update a centrally generated model or hypothesis. Lack of input does not call for immediate change in the perceptual model, but the eventual fading of a stabilized contour, or its sudden disappearance (often during an eye movement), might be viewed as the abandonment of the hypothesis that the contour is there, absent any verification by transient signals associated with eye movement. Equal factors of change preserve perceptual inequality after fading, but they also preserve inequality when fading has not yet reached completion. Thus the significance of incremental or decremental retinal signals for perception depends upon the preexisting perceived lightness and color prevailing when the new stimulus arrives. Veridical perception requires that the perceptual significance of afferent signals be contingent in this way on the pre-existing perceptual baseline. The appearance of new stimuli applied during the fading of effectively stable images or afterimages supports this principle of perceptual concatenation of factors of change, and provides new support for Land’s proposals. Prior to any fading, perception is roughly veridical and remains so if the stimulus changes. But when fading alters perception, the perception of new stimulation is correspondingly modified, even though incremental sensitivity by detection criteria may be almost constant during the fading. Although fading and sensitivity regulation are distinct processes, both can be at least roughly accounted for by. independent changes for the different photoreceptor systems. Thus the appearance of stable images and afterimages during fading, and also of new stimuli that replace them, need not, on present evidence, involve color opponent mechanisms.