Using an illumination discrimination paradigm to investigate the role of illumination priors in colour perception

Abstract

Previous studies suggest human colour constancy is optimised for natural daylight illuminations - a \blue bias" for colour constancy - but it is unclear how such a bias is encoded in the visual system. We use an illumination discrimination task to test two hypothesised mechanisms. Both hypotheses suggest that the human visual system has a prior expectation that illuminations are more likely to vary in a bluer region of chromaticity space. One hypothesis (the nature hypothesis) suggests this has developed in the human visual system through evolution, with selection of colour mechanisms that have reduced sensitivity to global bluer changes across a scene (a species prior). The second hypothesis suggests that the prior is learnt through experience with illuminations (the nurture hypothesis - an individual prior). In Chapter 3 we expand on previous results showing a \blue bias" for colour constancy when the illumination varies from a neutral reference, to show that the \blue bias" prevails in variants of the task where the illuminations are all chromatically biased. This result supports the nature hypothesis. However, depending on the chromatic bias, di erent biases can emerge in the threshold data that are more supportive of the nurture hypothesis. In Chapter 4 we explore individual di erences in illumination discrimination ability, compare illumination discrimination ability with chromatic contrast detection ability, and develop ideal observer models for the task. The results in this Chapter are mostly in support of the nurture hypothesis. In Chapter 5 we show that illumination priors may play a role in the recent visual illusion of a dress photograph that appeared blue and black to some observers but white and gold to others. Finally, in Chapter 6, we search for evidence that observers can learn an illumination prior during a psychophysical task. We conclude that the \blue bias" is likely governed by both a learnt prior over the characteristics of daylight illuminations (the nurture hypothesis) and a generic reduction in sensitivity to bluer changes in an illumination (the nature hypothesis).