Local and relational judgements of surface colour: constancy indices and discrimination performance.

Colour constancy is generally assumed to arise from a combination of perceptual constancy mechanisms operating to partially discount illumination changes and relational mechanisms involved in judging the colour relationships between object surfaces. Here we examined the characteristics of these mechanisms using a 'yes/no' task. Subjects judged whether a target colour patch embedded in an array of coloured patches (a) stayed the same across a simulated temporal illuminant change (local colour judgement), or (b) changed in a manner consistent with the illuminant change (relational colour judgement). The colour of the target patch remained constant in one-third of the trials, changed in accord with the illuminant shift in another third, and shifted partially with the illuminant change in the remaining third. We found that perceptual constancy was relatively weak and relational constancy strong, as assessed using a modified colour constancy index. Randomising the spatial positions of coloured patches across the illuminant change did not affect subjects' constancy indices. Application of signal detection analysis revealed some otherwise hidden effects. In the case of relational judgements, subjects adopted more conservative criteria (fewer true and false positives) with randomisation, maintaining a constant level of discrimination performance (d'). For local judgements, randomisation led to small increases in performance but no changes in criteria. We conclude that signal detection theory provides a useful tool to supplement conventional approaches to understanding colour constancy.