Chromatic Aberration: A New Tool for Colour Constancy

Maloney proposed solving the colour constancy problem by using a fi&e-dimensional linear model. Zlis method depends on the number of sensor classes being larger than the number of basis functions needed to model the surface spectral reflectance. Since the surface spectral reflectances of most natural objects require at least three basis functions for accurate modeling. xcording to Maloney's paper we must have four sensor classes in order to achieve colour constancy. However, most experiments suggest that the human visual system is tri-chromatic. That is. there are only three classes of sensors in the human visual system. As a result, we must look for other spectral information to help us achieve colour constancy. We claim that chrom&c aberration can help here. , . f'rom the chromatic aberration which occurs a t the edge between two coloured regions under unknown illumination. we derive the difference o f the spectral power distributions of the lights reflected from these regions. Using finite-dimensional linear models of illumination and surface spectral reflectances as our basic assumption, we obtain a set of equations for the coefficients that describe the illumination and the surface spectral reflectance. In combination with the equations obtailred from the sensors inside each coloured reglon. we determine the surface spectral reflectances. b hence colours. of the regions and the spectral power distribution of the illumination. ~f the number of sensor classes used is not smaller than that of the dimensions of I he f inite-dimensional linear model for surface spectral reflectance. Since u s i q degree three in the linear model approximates most of surfac; spectral reflectances very well, we only need to use three classes of sensors and chromatic aberration to recover the surface spectral reflectances. Without using chromatic aberration, Maloney had to use a t least four sensors. ~ h i s th; information provided by chromatic aberration is very valuable.