[EXPERT: CONSTRUCTED EYE] Day 10 — Colour Constancy and Illuminant Ambiguity

Day 10: Colour Constancy and Illuminant Ambiguity

A digital demonstration of colour constancy. Both red squares reflect the same RGB values, but the right patch appears darker and less saturated due to simulated shadow. Still, most viewers will correctly perceive both as the same surface red. Compare your experience to Land's Mondrian experiments (Land, 1977).

Expert Objective

This lesson investigates the neural and computational basis for colour constancy: the visual system's ability to maintain stable surface colours despite changes in environmental lighting (illuminant ambiguity). An advanced understanding here underlies both perceptual artwork and digital colour design, especially when seeking or subverting realism.

Observed Effects

• Colour Constancy: Viewers perceive object colour as roughly stable despite large differences in illumination. Pioneered by Land’s experiments, this phenomenon is robust in complex scenes. For instance, a grey card appears greyer in sunlight or under tungsten, even though its cone-excitation changes massively (Land, 1977; Brainard & Wandell, 1992).
• Illuminant Ambiguity: In scenes lacking diagnostic cues, identical patch colours can be attributed to differences in lighting or surface — i.e., the famous “The Dress” viral debate (Lafer-Sousa et al., 2015).
• Failures: Constancy is poor under unnatural/carefully mixed illuminants or a reduced-cue environment; painted shadow illusions by Kitaoka (colour legend illusions) exploit these failures to produce ambiguous or reversed colour perception (Kitaoka, 2011).

An illustration of illuminant ambiguity: Are the yellow patches identically coloured, or are their apparent differences due to lighting? Standard surface-reflectance mechanisms would predict stable colour identification, but ambiguity can persist if context is ambiguous or manipulated, as exploited in illusion art.

Supported Mechanisms

• Retinex Theory (Land, 1977): The eye-brain system simultaneously assesses local and global scene relationships, enabling the discounting of illuminant fluctuations by comparing spatial ratios of cone responses. It is a multi-scale spatial operation, not a per-pixel calculation, and is broadly supported in complex scene studies (Brainard & Wandell, 1992; Foster, 2011).
• Bayesian and Statistical Models: Modern theories (e.g., Brainard et al., 2006) emphasize statistical priors. The system combines sensory evidence with an expectation that natural scene illuminants cluster within a restricted distribution (‘natural scene statistics’). This priors-based disambiguation explains variants of the Dress illusion (Lafer-Sousa et al., 2015).
• Neural Correlates: Single-neuron and fMRI studies suggest V4 is involved in stable surface-colour coding, but the operations are distributed and context-sensitive (Bannert & Bartels, 2017).

Retinex spatial ratio principle: The perceived colour of a surface (here, red patches) depends not only on direct luminance or RGB but on its ratio to the surrounding context. This enables constancy under variable illumination (Land, 1977).

Evidence and Competing Explanations

• Classical Retinex: Land’s controlled experiments show that manipulating one patch’s surround can entirely flip its perceived colour, sometimes without direct change to the patch’s pixel values (Land, 1977). Replicated across laboratories.
• Neural Population Coding: Bannert & Bartels (2017) demonstrate the surface colour representation in higher visual areas remains correlated to perceived, not retinal, colour—supporting an active, relabeling mechanism but not simple signal averaging.
• Recognition and Prior Knowledge: Brainard et al. (2006) argue that cognitive priors (from scene memory and typical illumination) bias the interpretation toward canonical daylight, especially in ambiguous scenes. This Bayesian approach better predicts failures (as with the Dress) than low-level models alone.
• Alternative (Edge Integration): Some researchers propose that constancy arises from edge integration and not global sampling, but these models require precise edge-tracing and do not fully explain constancy in cluttered natural scenes (Foster, 2011).
• Unresolved: The exact neural algorithm for discounting illuminant, and the sequence of circuit-level computations, remains undetermined. Most studies agree a hybrid of spatial, statistical, and prior-based strategies is required but the integration site and weighting system are not fully known (Foster, 2020).

Digital Experiment: Shadow Constancy

Create a digital test scene of three identical colour patches (e.g., RGB: #cc3333) in different simulated lighting scenarios. Place one in direct light, one under a neutral shadow (desaturate/gray overlayer), and one under a coloured shadow (e.g., cool or warm overlay) using controlled opacity in layers.

Protocol: View at standard calibration, measure the RGB codes to confirm identity, and observe the subjective colour. Record descriptions of whether patches appear as the same or categorically different reds.
Controlled Variables: Surface color (hold RGB constant), shadow layer opacity and tint, monitor calibration.
Limitation: Digital displays cannot emulate full spectral differences of real-world illumination; consciously knowing the RGB is identical may bias judgement. Note the value of contextual cues and compare with how painted shadow illusions manipulate such contextual cues to create greater ambiguity.

Retrieval Question

Reflect: How do scene context and statistical priors enable the visual system to maintain the perception of stable surface colour, and in what situations does this process break down? Illustrate your answer with evidence from Land’s Mondrian experiments and at least one modern digital illusion.

Sources

• Land, E. H. (1977). The Retinex Theory of Color Vision. Scientific American, 237(6), 108-129. JSTOR
• Brainard, D. H., & Wandell, B. A. (1992). Asymmetric color matching: How color appearance depends on the illuminant. Journal of the Optical Society of America A, 9(9), 1433-1448. Optica
• Foster, D. H. (2011). Color constancy. Vision Research, 51, 674-700. ScienceDirect
• Bannert, M. M., & Bartels, A. (2017). Invariance of surface color representations across illuminant changes in the human cortex. Nature Communications, 8, 1-14. Nature
• Lafer-Sousa, R., Hermann, K. L., & Conway, B. R. (2015). Striking individual differences in color perception uncovered by 'the dress' photograph. Current Biology, 25(13), R545-R546. Cell
• Kitaoka, A. (2011). Trick Eyes and Color Illusions. Professor Kitaoka’s Optical Illusions
• Brainard, D. H., & Maloney, L. T. (2006). Surface color perception and equivalent illumination models: An introduction. Journal of Vision, 6(9). JoV
• Foster, D. H. (2020). The challenges of color constancy: Contributions from the study of natural scenes. i-Perception, 11(5), 1-24. SAGE Journals