Color spaces for discrimination and categorization in natural scenes

摘要: Physical measurements of surfaces' color-causing properties are typically spectroradiometric, whereas color-differencing comparisons colormetric ones performed in some 3-D color space. In general, this downprojection high-dimensional spectral data into 3-dimensional space incurs a loss information, that could be more critical one than another. One ecologically valid way assessing the extent information is to determine how likely it pair surfaces which have distinctly different would colorimetrically indistinguishable. We describe virtual ideal color-difference detector uses standard metrics but has access absolute difference signals surface pair. Only when classes as "different" yet them "same" said metameric. This paradigm applied dataset hyperspectral natural images using wide variety spaces. The results show that, around thresholds approximate human performance, overal metamerism rate very low, most pixels an image will metameric with at least other pixel. Thus, downprojecting onto may compromise discriminability, unlikely affect categorization finding accord evolutionary theories regarding function vision.