The independent and causal contributions of chromatic and achromatic features to fixation selection in natural scenes

Colour and luminance provide partially redundant information about the content of natural scenes. To assess the independent contribution of colour and luminance features to fixation selection in scenes, we applied a novel method combining a scene-patch analysis with generalized linear mixed models (Nuthmann & Einhäuser, 2015). Participants freely viewed 108 colour-calibrated natural scenes. Statistical models to predict fixated locations included seven features: luminance, luminance-contrast and edge-density as achromatic features, chroma (saturation), local variation of chroma, and the density of cone-contrast edges as their chromatic counterparts, and distance to image centre as quantification of central bias. To test causality of feature effects, stimuli were displayed in three variants: unmodified, with colour features removed (greyscale), or with luminance features removed (isoluminant). For unmodified stimuli, all tested features except chroma made independent contributions to fixation selection, with the effects of edge densities (chromatic and achromatic) being the most robust across spatial scales. When colour was removed, the effects of achromatic features increased only slightly. In contrast, when luminance was removed, the effects of chromatic features substantially increased. In conclusion, chromatic and achromatic features independently contribute to fixation selection; when both are available, achromatic features overrule some – but not all – of the chromatic effects.