One of the oldest puzzles in vision research involves the mechanisms by which the context of a stimulus can influence the perception of its relative brightness. For example, the two surfaces of the object shown here are physically identical (of equal luminance), but appear to be different, because of opposing light and dark luminance gradients along a step boundary. These different luminance gradients make the upper square appear darker than the lower square. Vision researchers have long thought that this phenomenon, known as the 'Cornsweet effect', could be explained by the receptive field properties of retinal ganglion cells.

Dale Purves and colleagues have re-examined this hypothesis, and in the 1 October issue of The Journal of Neuroscience, they propose that perceptions of luminance are not determined by retinal physiology, but by empirical associations made in the circuitry of the visual cortex. Investigation of different factors that enhance or reduce the Cornsweet effect suggest that the brain accumulates information from past visual stimuli and applies it to the stimulus confronting the observer at any given moment. Thus, the various pieces of information included in this image, such as the luminance gradient, perspective, orientation, texture and background, all indicate to the viewer, based on past visual experience, that the two areas are likely to differ in brightness. Accordingly, this is the perception experienced.

The authors of the paper believe that this empirical strategy of visual processing has been selected for throughout evolution. "Since all visual stimuli are ambiguous, the most efficient way for an animal to respond is to generate an association based on what the stimulus has usually turned out to be, and then act upon that," explained Purves. "The animal that couldn't do this effectively probably got eaten by a saber-toothed tiger."

Purves and colleagues are also investigating this empirical theory in terms of other visual qualities that affect luminance perception, such as color (reported in the November issue of Nature Neuroscience).