

Observers measured the Color Constancy
of 3 papers
matched in 27 different spectral illuminants
3 Papers 27 Illuminants
This experiment used 27 different spectral illuminants.
• Mixtures of 625, 530, 455 nm narrowband LEDs
• 3 colored papers on a white background
• Observers matched to Munsell Book
• Plot matches in L*a*b* Color Space
RESULTS
Discount Illumination Edge Ratios
Predictions Predictions
Compare Color Constnacy mechanisms:
• Chromatic Adaptation - Discount Illumination
• Predicts cloud of 27 different matches (left)
• Build scene from Reflectance Edge Ratios (right)
• Predicts line of matches for chromatic papers
• Predicts spot of matches for achromatic papers
Matches support Edge Ratio mechanism:
Predicted Edges Ratios fit Matches
Obsever Matches in L*a*b* space of 3 papers in 27 spectral illuminants
Video of predicted Matches in L*a*b* based on
edge ratios of cone responses
Predictions respond to Crosstalk between L, M, S cone quanta catches
Orange and violet predictions form linear shapes,
and grays form a small spot
Achromatic papers have constant preictions
Do Humans “Discount the Illuminant”?
J. J. McCann, Proc. SPIE Vol. 5666, 9-16, 2005.
Mechanism of Color Constancy
J. J. McCann, Proc. IS&T Color Imaging Conference, Scottsdale, Arizona, 12, 29-36, 2004 2004c CIC12.pdf
The shape of 3-D plots of observed color matches is the signatue of Color Constancy's mechanism
• 27 illuminants forms clouds of matches in L*a*b* space
• Chromatic Adaption predictions maintain the chromatl shifts caused by illumination (I)
• All three paper reflectances (R) behave the same - All papers form a 3-D grid
• Shifted by changes in illuminations
• Edge Ratio predictions use ratios of spectral integrated cone responses [R * I]
• Achromatic reflectance predictions are a single spot ( Cone Crosstalks cancel )
• Chromatic reflectance predictions forms lines in L*a*b*
• L*a*b* Signature of Color Constancy Matches predicted by Cone Crosstalk
• Edge Ratio predictions fit Observer matches

Video of predicted Matches in L*a*b* based on CIECAM incomplete adaptation
Predictions respond to shifts in illumination
Orange, violet, gray predictions form identical 3-D cloud shapes
All paper's respond solely to shifts in illumination
Observe matched orange and violet papers to form linear shapes,
and grays form a small spot
Cone Crosstalk, not the spectra of illumination,
models the small departures from percect Color Constancy
with changes in illumination spectra