Does the geometry of perceptual colour space reflect the colours in our environment?

Abstract

There are universal patterns in colour naming (Berlin & Kay, 1991; Cook et al., 2005) and the colour categories in a language influence colour perception during perceptual uncertainty (Bae et al., 2015). This suggests that that colour categorisation is not structured at random, but rather the structure is constrained to serve an adaptive purpose. Previous research has focused on identifying constraints on language and the human visual system to explain the origins of universal colour categories, but there was little evidence demonstrating ecological constraints on colour categories.

This thesis proposes that the distribution of colours across objects in the environment place an ecological constraint on colour perception and categorisation. If there are systematic patterns in colouration across behaviourally relevant objects, then colour categories may be formed to guide perceptual processes towards visual stimuli which are the most important. To explore this, two tasks were used: a colour estimation task to measure the geometry of perceptual colour space, and an object recognition task to measure the distribution of colours across objects.

Humans exhibited categorical biases in colour perception across hue and saturation towards universal colour categories, corresponding to red, blue, green, pink, orange and grey. These categorical biases were successfully estimated from the statistical regularities of colours across objects for two independent image datasets: colour perception was biased towards category foci which were most informative about objects, and away from category boundaries which were least informative about objects. These findings support a universal mechanism through which sensory systems adapt to environmental statistics, allowing for optimal representation of sensory features under uncertainty for both perception and language.