University of Southampton
The appearance of surface colours - Do we need spectral information to understand human colour constancy?
Abstract
dc:description.abstractColour plays a key role in object recognition, memory retrieval, and interaction with the environment. For colours to serve as perceptual attributes of objects, the relations between perceived colours of surfaces must remain stable despite changes in illumination—commonly referred to as (relational) colour constancy. Understanding colour constancy is challenging because surface reflectance and illuminant spectra mix before reaching the retina, and the fine-grained spectral information is lost in sensory processing. This thesis focuses on whether understanding human colour constancy requires the fine-grained spectral information given the physical, sensory, and perceptual constraints related to the natural environments. Chapter 2 assessed whether sensory reflectance matrices could approximate how surface reflectances modulate the sensory colour signals across illuminants. The findings showed that the approximation is perceptually accurate under naturalistic broadband illuminants, unlike artificial narrowband illuminants, suggesting that the fine-grained spectral information is not necessary to modulate sensory colour signals in natural environments. Given the rarity of metamers in nature, Chapter 3 evaluated whether people have formed expectations about naturally occurring colour shifts under different illuminations. The results showed that human observers have general expectations regarding the direction of colour changes; these expectations lack precision, though. Furthermore, it was found that the stability of cone-excitation ratios affected expectations, and viewers would perceive colour changes as natural when these ratios stayed constant, independent of the source of the changes—naturalistic spectra or synthetic ones. This implies that prior knowledge of surface colour shifts combined with the stability of cone-excitation ratios—rather than the high-resolution spectral data—contributes to colour recognition in natural environments. Chapter 4 focuses on whether reflectance-based cone-excitation ratios (rCERs) might compress the fine-grained spectral information in hyperspectral images while preserving visual indistinguishability when shown on a screen. The findings indicated that, although many colours seemed to be unchanged, at least one colour per scene clearly changed so that observers could tell the compressed from the original form. Despite this limitation, the rCER method can be useful for studies on colour constancy since it guarantees the whole stability of cone-excitation ratios when designing experimental stimuli. Overall, while the spectral information loss at the sensory level causes challenges to the understanding of human colour constancy, the findings of this thesis suggest that non-spectral data available at the sensory level are perceptually sufficient to model surface colours as long as spectra are broadband.
Degree
thesis:*- Name dc:type.qualificationname
- Ph.D.
- Level dc:type.qualificationlevel
- doctoral
- Grantor dc:publisher.institution
- University of Southampton
- Year dc:date.issued
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Karimipour, Hamed
- Advisors dc:contributor.advisor
-
- Witzel, Christoph
- Graf, Erich
- ORCID person:identifier.orcid
Rights
- Language dc:language
- English, eng