Comparable colour constancy for average colour and single colour percepts

Colour constancy refers to the visual system’s ability to correctly estimate surface colours under varying illumination. Most previous studies have employed stimulus surfaces with uniform chroma-ticity or limited chromatic variation. Can colour constancy operate equally well—or better—for sur-faces with significant chromatic variation? We characterised percepts of average colour across an illumination change for uniform and polychromatic surfaces.
We used realistically rendered images of 3D scenes that were divided horizontally to yellowish and bluish illumination regions. Four 3.3° irregular “blob” objects were embedded in the scene in a two-by-two configuration (separated by 2.2° vertically and 3.6° horizontally), with two objects in each illumination region. The object surfaces were either uniform or polychromatic, and background sur-face was either uniform grey or made of differently coloured squares. On each trial, three objects had the same (average) surface colour. The fourth surface colour (target) varied from yellowish to bluish from trial to trial. The stimulus was shown for up to 2000 ms, and observers (N = 20) responded which of the four objects had a different (average) surface colour. Correctly choosing the illumination region containing the target reflects the ability to discriminate colour. Given correct choice of re-gion, correctly choosing the target object reflects colour constancy relative to discriminability. Thresholds and biases for both choices were estimated from psychometric function fits to the pro-portion of correct responses.
Based on within-subject contrast analysis in a linear mixed model, discriminability was poorer for polychromatic than uniform objects (t(139) = -11.76, p < .001), but colour constancy was the same for both. Compared to a uniform grey background, a multi-coloured background made discriminabil-ity slightly poorer (t(139) = -2.34, p = .021), but had no effect on colour constancy. We conclude that colour constancy operates equally well for average and single colour estimates.