Spectral Bandwidths of Colour Detection Mechanisms

We measured the spectral properties of human colour detection mechanisms with a noise masking paradigm, using two new methods to overcome a potential artefact of previous procedures. Estimation of the spectral bandwidth mediating detection of binary colour signals (eg orange) would be misleadingly narrow if the observer changes strategy as a function of the colour of the masking noise (eg detecting the red component of an orange signal in yellow noise and vice versa). We induce observers to use a single detection mechanism throughout an experiment by adding either biaxial noise or sectored noise to the signal. With biaxial noise, one can eliminate detectability of an orange signal by a red-sensitive mechanism, for instance, by adding a steady source of noise along a red axis. We then measure the remaining (yellow) sensitivity by adding noise along a second, variable axis. Sectored noise consists of noise samples from sectors of variable width in the colour plane, centred around the signal axis. Detection with the use of off-axis mechanisms is a less optimal strategy with such noise. A decline in the potency of such noise with increasing sector width would suggest that the spectral bandwidth is narrow and nonlinear rather than broad and more nearly linear.