Seeing colors in real scenes

摘要

Color perception in real conditions is determined by the spectral and spatial properties of objects and illumination. These properties are best evaluated by spectral imaging, a technique that records the reflecting spectral profile for each point of the scene. Using this technique on a set of natural scenes it was found that the color gamut expressed in the CIELAB color space is much smaller than the theoretical limits defined for the object colors. Moreover, the colors more frequent are those around the white point and their frequency of occurrence can be well described by a power law. Spatial variations of the spectral composition of the illumination across natural scenes were also quantified by placing small reflecting spheres in different locations of the scenes. The extent of these variations across scenes was found to be large and of the same order of magnitude as the variations of daylight along the day. These findings show that colors in nature are considerable constrained and that constancy mechanisms must be efficient over a wide range of stimuli variations to compensate for large natural variations of illumination.