| |
Abstract:
Look at a gray object indoors or out: it still looks gray,
even though the light falling upon it (and thus the light
reflecting from it) varies by a thousandfold. This "lightness
constancy" is a central feature of visual perception. A clever
experimenter can trick the visual system into seeing two identical
grays as being quite different. The lightness illusions appartently
result from the same mechanisms that produce lightness constancy.
Illusions are worth studying because (1) they allow a researcher to
expose the inner workings of the visual system and (2) they make
really cool demos.
The classical approaches to lightness perception involve
normalization by a image statistics, e.g., dividing or subtracting
out the mean luminance. Low-level operations such as lateral
inhibition can implement such normalization locally. However, we
have devised stimuli that dramatically demonstrate the inadequacy
of these traditional models. The new illusions indicate the
importance of mid-level visual processes that utilize information
about contours, junctions, and regions. We propose that the visual
system estimates "optical atmosphere" at each point in a scene, and
propagates lightness constraints within regions defined by
atmospheric boundaries. Lightness constancy can be considered as a
statistical estimation problem, where the statistics are gathered
within an adaptive window that prevents the mixture of samples from
different lighting conditions.}
|
