Abstract
It has been suggested that objects are more likely to be warmer in colour, redder and more saturated than the background. Here, we investigate the colour statistics of objects, and the brain regions that are responsive to these statistics. First, we analysed the Natural Scenes Dataset (NSD), a 7T dataset in which 8 participants viewed up to 10,000 natural scenes. Our analysis of the chromaticities of the 80 segmented object classes and backgrounds confirmed that object pixels were warmer, redder, more saturated and darker than background pixels. The probability that pixels were from objects rather than backgrounds (the 'Object Colour Probability', OCP) was calculated for 240 hue bins. The mean OCP of images correlated with NSD BOLD responses mostly in the ventral visual pathway. Other image statistics (e.g., number of food pixels) better explained the responses of correlated voxels. A second fMRI study, in which colours were shown as a single patch on a grey background, was analysed to study whether ventral visual pathway is responsive to OCP in the absence of other scene statistics. To constrain our analyses to functionally relevant areas, we used independent functional localizers to identify colour- and object-selective areas and combined these with NSD defined OCP responsive areas. The OCP of the colour patches significantly correlated with BOLD in colour-selective but not object-selective visual regions. Implications for the role of colour in object vision are discussed.
 Funding: The work was funded by ERC grant 772193 COLOURMIND to AF and ERC grant 949242 COLOURCODE to JB.