Abstract
Perceived relative achromatic reflectance of a surface is referred to as lightness, and it depends not only on the luminance but also on the context within which the surface is viewed. For example modest changes in two-dimensional configuration or three-dimensional geometry may lead to profound variations in lightness even when the luminance remains unchanged. Perceptual grouping is another factor that affects the lightness of a surface. Here we study possible mechanisms underlying perceptual grouping that affect lightness perception. First, using a novel stimulus we behaviorally investigate how lightness depends on spatially distant image features, including luminance and contrast. Next we propose a computational model to estimate the lightness at every pixel in the image. The first term in the model takes luminance as the zeroth order approximation to lightness. The next term preserves luminance discontinuities, i.e. contrast edges. The third term facilitates homogeneity and encourages uniform solutions. The last term facilitates grouping of pixels that have similar lightness values even when they are spatially separated. We tested the model using an expectation maximization (EM) algorithm and found that it successfully predicts the human data. This result demonstrates that a few simple intuitive rules can be implemented in a computational framework to predict human lightness perception and to estimate surface reflectance.
Acknowledgment: Author HB is supported by a research grant from Turkish Scientific and Technological Association (TUBITAK 1001-108K398)