Purchase this article with an account.
Vijay Singh, Johannes Burge, David Brainard; Equivalent noise characterization of human lightness constancy. Journal of Vision 2020;20(11):610. doi: https://doi.org/10.1167/jov.20.11.610.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
An important goal for vision is to provide stable perceptual representations of task-relevant scene properties (e.g. target object size, shape, reflectance) despite variation in task-irrelevant scene properties (e.g. illumination, reflectance of other nearby objects). To study such stability, we measured how variation in a task-irrelevant scene property affects threshold for discriminating changes in a task-relevant property. Four subjects viewed computer-rendered images of a 1-degree sphere, within a 2-degree scene containing naturalistic background objects. The sphere’s reflectance was spectrally flat but varied in albedo. On each trial, two images of the scene were presented in sequence and subjects indicated which 0.25s interval contained the sphere with higher albedo. Across intervals, the reflectances of the background objects were randomized by sampling from a probabilistic model of naturally occurring surface reflectances. This reflectance distribution was varied systematically by applying a scalar to its covariance matrix. Discrimination thresholds were measured as a function of the scalar. When plotted as a function of log covariance scalar, log squared thresholds were initially constant, and then rose approximately linearly with a slope of 0.20 +/- 0.03. The equivalent noise, the log covariance scalar value at which threshold elevation began, was -2.08 +/- 0.21. We compared the data to predictions of a recently published computational model of lightness constancy. Model thresholds were aligned with human thresholds at covariance scalar equal to zero by adding noise to the computational observer estimates. The model predicted human equivalent noise to reasonable approximation (model value, -2.44), but model thresholds increased more rapidly than those of the subjects (model slope, 0.38). Our experiment characterizes the intrusion that background variability has on perceived object lightness. Our computational model accounts reasonably for the equivalent noise, but is challenged by the low slope of threshold rise shown by human subjects.
This PDF is available to Subscribers Only