Second-Order Statistics of Colour Codes Modulate Transformations That Effectuate Varying Degrees of Scene Invariance and Illumination Invariance

Abstract

We argue, from an ethology-inspired perspective, that the internal concepts 'surface colours' and 'illumination colours' are part of the data format of two different representational primitives. Thus, the internal concept of 'colour' is not a unitary one but rather refers to two different types of 'data structure', each with its own proprietary types of parameters and relations. The relation of these representational structures is modulated by a class of parameterised transformations whose effects are mirrored in the idealised computational achievements of illumination invariance of colour codes, on the one hand, and scene invariance, on the other hand. Because the same characteristics of a light array reaching the eye can be physically produced in many different ways, the visual system, then, has to make an 'inference' whether a chromatic deviation of the space-averaged colour codes from the neutral point is due to a 'non-normal', ie chromatic, illumination or due to an imbalanced spectral reflectance composition. We provide evidence that the visual system uses second-order statistics of chromatic codes of a single view of a scene in order to modulate corresponding transformations. In our experiments we used centre surround configurations with inhomogeneous surrounds given by a random structure of overlapping circles, referred to as Seurat configurations. Each family of surrounds has a fixed space-average of colour codes, but differs with respect to the covariance matrix of colour codes of pixels that defines the chromatic variance along some chromatic axis and the covariance between luminance and chromatic channels. We found that dominant wavelengths of red-green equilibrium settings of the infield exhibited a stable and strong dependence on the chromatic variance of the surround. High variances resulted in a tendency towards 'scene invariance', low variances in a tendency towards 'illumination invariance' of the infield.

The Physicalistic Trap in Perception Theory

“…chromatic, illumination or due to an imbalanced spectral reflectance composition of the scene has to be achieved by a specific activation of representational primitives by signs that the sensory system provides on the basis of relevant reliable regularities of the incoming light array. We found evidence that second-order statistics of chromatic codes of the incoming light array differentially modulate, by a specific class of parametrized transformations, the relation of the two kinds of representational primitives involved (Mausfeld & Andres, 2002).…”

“…Which parameters are specified in which way and which associated classes of transformations are activated (pertaining e.g. to the idealized functional goals of ʹscene invarianceʹ or ʹillumination invarianceʹ) is then determined by specific physico-geometrical properties of the sensory input (Mausfeld & Andres, 2002;Mausfeld, 2003).…”

Recognized visual space of illumination: A new account of center‐surround simultaneous color contrast

“…13 However, we cannot ignore the very impressive findings obtained from the experiments conducted within two-dimensional environments. A number of studies to date [22][23][24][25] have also reported the ability of the visual system to estimate the illumination from two-dimensional stimuli. Also, it has been shown that the visual system takes the context into account to estimate the illumination.…”

Effects of chromatic image statistics on illumination induced color differences

“…It is clear that a model based on the gray world assumption (e.g., [1]) would fail to predict the effect of chromatic orientation, since it only relies on the mean chromaticity in the scene. There are a number of studies, however, in which the chromatic distribution of the visual scene is probed to derive information about the chromaticity of the illuminant (e.g., [2,[29][30][31][32]). But even if we suppose that this would result in the correct estimation of the exact illuminant color, it would still not predict that one illuminant leads to a higher color fidelity than another.…”