Abstract
"The colour-changing dress" is an internet phenomenon in which the colours of a dress are reported to be different by different people. In previous works (e.g. Winkler et al. 2015, Lafer-Sousa et al. 2015, Brainard and Hurlbert 2015, Gegenfurtner et al. 2015) this striking ambiguity was interpreted within the colour constancy framework, where the colour of a surface is perceived after estimating and discounting the colour of the illuminant (e.g, Foster 2011, Boyaci et al., 2006). Here we explicitly test the hypothesis that people who reported different colours of the dress also perceived different illuminations. We measured this by introducing a probe (Koenderink et al. 2007), a sphere, in two positions within the scene and letting the observers adjust its appearance. Twenty-four observers adjusted the colour, the intensity and the direction of the illumination of the probe rendered in front and in the background of the dress in the original scene. Observers' task was to adjust the probe to appear as a white sphere embedded in the scene. We found that people who reported the dress to be white tended to produce bluer adjustments than people who reported it as blue. A computer classification approach on the two chromatic dimensions of DKL colour space (LM & S) allowed us to classify the two groups of people with 67% accuracy, and t-tests show a significant difference in the colour adjustments along the LM dimension (p< 0.05) and at the edge of statistical significance on the S axis (p=0.067). There were no differences in the adjustments of the background probe. These results are compatible with the idea that the ambiguity in the perceived colour of the dress can be explained by the different assumptions that people made about the illumination of the scene in which the dress is embedded.
Meeting abstract presented at VSS 2016