Effect of background color on object lightness perception

Our visual system provides a stable percept of object intrinsic properties, such as surface reflectance, despite variation in object extrinsic properties of the visual scene. We developed a method to relate object extrinsic variability in visual scenes to perceived values of object intrinsic properties and use it to study the dependence of object lightness on background color. We measured lightness discrimination thresholds as a function of background color variance. Human observers viewed computer-rendered images of a 1-degree sphere within a 2-degree scene containing naturalistic background objects and reported the image in which the sphere was lighter. The sphere was spectrally flat (gray) but varied in its albedo. The color of background objects was varied by sampling from a statistical model of natural surface reflectance measurements. Amount of variability was controlled by multiplying the covariance matrix by a scalar. Two types of backgrounds were studied: one where reflectance spectra could take arbitrary shape allowing changes in lightness, hue, and saturation of the background objects, and another where the reflectance spectra were spectrally flat (gray). The results show that for both backgrounds, threshold squares were first constant and then increased linearly as a function of the covariance scalar (p-value for the hypothesis, the means for different scalars are equal, was <1.12*10^(-7)). At a given value of the covariance scalar, there was no significant difference between the thresholds for the two backgrounds (p-value 0.83, 0.46, and 0.30 respectively at covariance scalar 0.03, 0.3, and 1). Our method relates observers’ intrinsic noise to extrinsic variation in visual scenes and predicts that observers’ internal noise at lightness discrimination is ~0.6 times the extrinsic color variation in natural scenes. We also show that lightness discrimination thresholds depend on the amount rather than the type (color or gray) of spectral variation.