Abstract
Purpose. The goal of the analysis is to derive human color perception, as indexed by color discrimination performance throughout color space, from the fundamental properties of the retinal cone array. There have been numerous attempts to determine a uniform metric for color discrimination data such as the MacAdam ellipses, but all rely on linear or nonlinear transforms of CIE or other color spaces with multiple free parameters, without reference to retinal cone densities. Methods. Published data for multidirectional color discrimination throughout equiluminant color space are reanalyzed with a new version of the Helmholtz/Stiles line element (Euclidean vector space) theory of color space, in which color discrimination is limited by the intrinsic noise in each of the three cone pathways. This constitutes a linear vector space prediction for color thresholds after applying the nonlinear transformation of the square-root law of signal/noise ratios, based on spatial summation of the stimulus information across the photoreceptor array. Results. This approach gives rise to a one-parameter intrinsic-noise prediction for the shape and orientation of the color discrimination ellipses that is closer to the empirical results than previous analyses, and is an improvement over all the CIE approaches to uniform color space. Conclusion. This intrinsic noise analysis provides a comprehensive approach to the individual variations in human color discrimination from first principles, allowing predictions to be derived in cases of variations in cone spectral sensitivities and proportions, macular pigment densities and rod-cone interactions.