Category knowledge can form prior constraints on scene recognition from luminance and chromatic cues

To categorize realistic stimuli such as scenes, information is available along the dimensions of luminance and chrominance. One question that arises is whether their order of integration can change depending on their category informativity. To test this, we synthesized four realistic-looking scenes by combining two different luminance patterns (flat and hilly) with two different chromatic patterns (grassy and sandy). A learning procedure induced a different, two-level, taxonomic organization. At a general level, LUMI (vs. CHRO) subjects learned to categorize the four scenes into “flat” and “hilly” (vs. “grassy” and “sandy”), using luminance (vs. chromatic) cues. At the specific level, LUMI and CHRO subjects all learned to categorize the same stimuli as either “field” (the combination of flat and grassy), “desert” (flat and sandy), “mountain” (hilly and grassy) or “dune” (hilly and sandy). The conjunctive nature of the stimuli warrants that, in both groups, the input can only be recognized as, e.g. “field,” when its flat luminance and its grassy chrominance are perceived and integrated. We can use this property to determine the existence of a knowledge-driven weighting of luminance and color. Suppose the “field” is briefly (15, 45, 75, 105, 135, 165 and 195 ms) presented on the screen, followed by a mask. Three low-level miscategorizations are possible: “dune” implies a misperception of both the flat luminance and the green chrominance; “mountain” (vs “desert”) implies a misperception of only the luminance (vs. chrominance). We found opposite miscategorizations (i.e. respond more often “desert” than “mountain” in LUMI, and “mountain” than “desert” in CHRO, when field is presented), revealing a differential sensitivity to luminance and chrominance.