James M. Kraft

Bayesian model of human color constancy.

Vision is difficult because images are ambiguous about the structure of the world. For object color, the ambiguity arises because the same object reflects a different spectrum to the eye under different illuminations. Human vision typically does a good job of resolving this ambiguity-an ability known as color constancy. The past 20 years have seen an explosion of work on color constancy, with advances in both experimental methods and computational algorithms. Here, we connect these two lines of research by developing a quantitative model of human color constancy. The model includes an explicit link between psychophysical data and illuminant estimates obtained via a Bayesian algorithm. The model is fit to the data through a parameterization of the prior distribution of illuminant spectral properties. The fit to the data is good, and the derived prior provides a succinct description of human performance.

Surface-Illuminant Ambiguity and Color Constancy: Effects of Scene Complexity and Depth Cues

Two experiments were conducted to study how scene complexity and cues to depth affect human color constancy. Specifically, two levels of scene complexity were compared. The low-complexity scene contained two walls with the same surface reflectance and a test patch which provided no information about the illuminant. In addition to the surfaces visible in the low-complexity scene, the high-complexity scene contained two rectangular solid objects and 24 paper samples with diverse surface reflectances. Observers viewed illuminated objects in an experimental chamber and adjusted the test patch until it appeared achromatic. Achromatic settings made under two different illuminants were used to compute an index that quantified the degree of constancy. Two experiments were conducted: one in which observers viewed the stimuli directly, and one in which they viewed the scenes through an optical system that reduced cues to depth. In each experiment, constancy was assessed for two conditions. In the valid-cue condition, many cues provided valid information about the illuminant change. In the invalid-cue condition, some image cues provided invalid information. Four broad conclusions are drawn from the data: (a) constancy is generally better in the valid-cue condition than in the invalid-cue condition; (b) for the stimulus configuration used, increasing image complexity has little effect in the valid-cue condition but leads to increased constancy in the invalid-cue condition; (c) for the stimulus configuration used, reducing cues to depth has little effect for either constancy condition; and (d) there is moderate individual variation in the degree of constancy exhibited, particularly in the degree to which the complexity manipulation affects performance.

Effects of known variations in photopigments on L/M cone ratios estimated from luminous efficiency functions

The extent to which known variations in photopigment lambda max and optical density may affect cone ratios estimated from the spectral luminous efficiency function (LEF) was examined. LEFs were generated using L- and M-cone fundamentals, one of which had been shifted in lambda max (+/- 1, 2, 4 or 6 nm) or varied in peak optical density (increased or decreased by 10, 25 or 50%). A curve-fitting program was then used to estimate the L/M cone ratios for the generated LEFs assuming standard L- and M-cone fundamentals. These modeling exercises indicate that L/M cone ratios estimated from LEFs are highly correlated with long-wave sensitivity and with known variations in L-cone lambda max. Variations in M-cone lambda max and photopigment optical density for both cone types are also correlated with L/M cone ratios, but have much less impact on the estimated ratios.