Abstract
Human colour perception depends initially on the responses of the long(L-), middle(M-) and short(S-) wavelength-sensitive cones. These signals are then transformed post-receptorally into cone-opponent (L-M and S-(L+M)) and colour-opponent (red/green and blue/yellow) signals and perhaps at some later stage into categorical colour signals. Here, we investigate the transformations from the cone spectral sensitivities to the hypothetical internal representations of 8 colour categories by applying a novel technique known as "Rank-Based Spectral Estimation". Pairs of colours were ranked by 12 observers according to which appeared more representative of eight different colour categories: red-green-blue-yellow-pink-purple-brown-orange. Stimuli comprised circular patches of 32 colours presented on a CRT monitor chosen to cover as large a volume of LMS colour space as possible. In separate blocks, observers judged pairs of colours as to which appeared more like the reference colour name. Pairs for which judgement could not be made, because neither colour appeared like the reference, were recorded but not used. To derive the spectral sensitivities of the colour categories (the 8 "colour sensors") using the rank-based technique, we assumed that the relationship between cone responses and colour appearance can be described by a linear transform followed by a rank-preserving non-linearity. The estimated sensor transformations could account for over 85% of the rank orders. Sensor shapes were generally plausible; those for red and green were consistent across observers, while the yellow and blue ones showed more variability in spectral position. Other sensors, such as brown, pink and purple, showed large inter-observer variability, which might be due in part to cultural differences in colour naming. Sensors were generally restricted to limited regions of colour space. As expected from colour opponent theory, the red and green sensors formed relatively distinct regions with limited overlap as did the yellow and blue ones. Other sensors were spectrally shifted or bimodal.
Meeting abstract presented at VSS 2016