Abstract
Colour signals from a given retinal location are not stable over time, since eyes and objects move and lighting conditions are dynamic. It is behaviourally useful to separate a change in illumination from a change in reflectance. For an isolated matte surface, these changes are indistinguishable. However, for a 3D glossy surface, in which light varies across the surface as a mixture of diffuse and specular components, illuminant and reflectance changes are in principle separable. A change in reflectance causes darker (less specular) locations to change in chromaticity more than the lighter (more specular) locations, whereas a change in illuminant shows no such relationship. We presented hyperspectrally raytraced movies showing surfaces undergoing illuminant or reflectance changes and asked observers to classify the change. We ensured that the magnitude of the chromatic change of the diffuse component could not be used as a cue. Performance was at chance for matte surfaces and increased with specularity. Accurate classification was possible even at low specularities (~0.2). With multiple uncorrelated changes (spatial odd-one-out) performance increased more gradually, reaching 100% only with full specularity. Previous work shows the importance of specularity in colour constancy: highly specular reflections provide a direct estimate of the illuminant, improving constancy (Yang & Maloney, 2001) and, with multiple surfaces, chromaticity variation across each surface falls on lines in colour space that intersect at the illuminant chromaticity (D'Zmura & Lennie, 1986; Lee, 1986). We show that observers can extract structured chromaticity relationships that are not available at any given instant.
Meeting abstract presented at OSA Fall Vision 2012