Color Constancy
how constant descriptors of surface
color are extracted from variable sensory signals
Color Induction
how changes in background properties affect the color appearance of a target stimulus
inverse problem
visual system faces the challenge of parsing the retinal light signal into object and illumination components
computational models
needs prior assumptions
relied on the fact that natural surface reflectances and daylight illuminants can be represented with a limited number of basis functions
assuming certain regularities (e.g. mean chromaticity is neutral) in the visual scene
could estimate the illuminant in a simple scene but might not apply to human color constancy
hypothesis of color constancy
adaptation to mean luminance
adaptation to local contrast
adaptation to the brightest surface in the scene
results showed a combination of several mechanisms
observers used information from several cues, such as highlights, cast shadows, and depth cues when estimating surface color
research paradigm
measuring changes in color appearance caused by a change in viewing context similar to simultaneous contrast illusion
quantify the “compensation” for the difference in surrounds in terms of color constancy, or color constancy index
hypothesis
Helmholtz
simultaneous contrast effect is produced by an error of judgement, where the target is mistakenly assumed to be illuminated with a light having the color of the surround
Due to the discounting of the illumination, the perceived color of the target shifts in a direction opposite in color space to the color of the surround
Hering
both color constancy and simultaneous contrast result from lateral inhibition between neighboring receptors at the retina
lead to assumption in later works that simultaneous contrast is due to mechanisms subserving color constancy that is broadly adopted in the field
HOWEVER, any given observable induction effect may be due to a host of different mechanisms, subserving different functional goals
Type I and Type II Constancy
challenge
we hardly experience the strong changes in perceived when an object moves across a multicolored background
Against the idea that simultaneous contrast and color constancy are due to mechanisms encoding contrast information at the borders between surfaces
Type I constancy
the mechanisms that provide color constancy across illumination changes
Type II constancy
mechanisms that provide color constancy across background changes (background independence)
computational model
Initially, color difference is coded at edges (differentiation)
The concatenation of these two operations yields the original image up to an unknown additive constant (determined by some kind of anchoring rule)
The second stage integrates color difference across space (integration)
simultaneous contrast effects (a failure of background independence) are to be understood as failures of integration
raise emphasis on
how to classify reflectance and illuminant edges
how the postulated integration works
quantify contextual effects
Asymmetric Color Matching
Two target patches embedded in different surrounds
The observer adjusts the color coordinates of one of the targets to match the appearance of the other
offer a direct link to computational models of constancy that estimate the illuminant
Cannot decompose the combined effect of two surrounds into single surround-specific effects without relying on theoretical assumptions (such as neutral surround)
Asymmetric Matching with HSD
Haploscopically Superimposed Displays (HSD)
one target-surround stimulus is presented separately to each eye such that the observer experiences the two targets to be matched as embedded in the same surround
Subjective matching problem seems to be absent compared to conventional paradigm
Achromatic Settings
the observer adjusts the chromaticity of a target embedded in a surround such that it appears achromatic
Obviate the need for presenting a second comparison target
measurements can only be made for a small subset of all possible colors
Unique Hue Settings
based on a notion of opponent color theory that the four unique hues have special properties that make them as landmarks in color space
measurements can be made not only for targets that appear achromatic in a given surround, but also for targets that appear unique red, green, blue, or yellow
interpretation and modeling requires more assumptions than achromatic settings because the set of colors that appear in a given unique hue is a two-dimensional manifold
Threshold and Scaling Measurements
measure color discrimination by threshold measurements or supra-threshold technique
threshold or scaling measurements can be used to estimate the derivative of the color code f(t,s) with respect to t, which cannot be determined based on asymmetric matching
Assumptions in Measuring and Modeling Color Appearance
The Continuity Assumption
perceived color is a continuous function of the color coordinates of the target and the surround
Functionally Neutral Surrounds
Some surrounds (e.g. complete dark) are considered “neutral” to serve as the matching target
The Compensation Assumption
the observer could compensate the net effect of the surrounds on the perceived colors of the targets by adjusting the tristimulus values of one of the targets
Assume the perceived color space is 3-D
The theoretical color space could be up to 6-D
(3 values for target and surround each)
there is often a discontinuity in color appearance when the target color coordinates reach the coordinates of the surround
crispening
color discrimination is best for target colors close to the surround color