The distribution of discernible colours in natural scenes

Abstract

Although the total number of discernible colours in natural scenes has been estimated from physical data, the distribution of colours has received little attention. To address this issue, an analysis was made of hyperspectral images taken from a database of 50 urban and rural scenes. The colour volume of the database was computed by expressing each pixel of each image in CIELAB space. To derive the distribution of colours, the volume was segmented into just distinguishable sub-volumes, approximated by unit cubes, with each non-empty sub-volume assumed to represent one discernible colour and its pixel population the total number of times it occurs over the 50 scenes. It was found that the fraction of discernible colours expressed as a function of occurrence could be described by a negative power-law and that most discernible colours appeared only rarely. Thus, 70% of the total number of discernible colours constituted only about 3% of the pixels sampled, implying that most of the natural chromatic distribution fell within a small colour gamut