Michael Rudd: Lightness
computation by a neural filling-in mechanism
Abstract
Lightness perception is both an ancient
and currently very active research area within visual perception. Over the years,
many different types of explanations have been devised to account for lightness
phenomena, including theories based on neural, ecological, gestalt-organizational,
and cognitive principles. All of these perspectives are alive and well today.
One important aspect of lightness appearance is its dependence on spatial context.
For example, a gray paper viewed against a white background appears darker than
the same paper viewed against a black background. This illusion, known in the
trade as "simultaneous contrast," has been studied scientifically since at least
the mid-19th Century, but it is not yet fully understood. I will survey some
key findings and theories of contrast phenomena, then present my own quantitative
theory of lightness computation based on a neural network mechanism. The model
assumes the existence of visual maps in the brain in which lightness and darkness
signals are separately represented, prior to being combined to create an achromatic
color appearance signal. Within each map, color induction (influencing) signals
spread from neurons that encode the luminance ratios at edges within the image,
to fill-in regions between the edges, thus "painting" in the areas between the
edges in the image with color. I will demonstrate that the model gives a good
quantitative account of data from lightness matching experiments from my own
lab, and I will show that the model can also account for phenomena that have
been cited as evidence for alternative theories of lightness based on cognitive,
ecological, and gestalt principles..