Colour constancy : human mechanisms and machine algorithms
This thesis describes a quantitative experimental investigation into instantaneous colour constancy in humans. Colour constancy may be defined as the ability of the visual system to maintain a constant colour percept of a surface despite varying conditions of illumination. Instantaneous, in this context, refers to effects which happen very rapidly with the change of illumination, rather than those which may be due to long term adaptation of the photoreceptors. The results of experiments are discussed in the context of current computational models of colour constancy. Experiments on subjects with damage to the cerebral cortex are described. These highlight the different uses of chromatic signals within the cerebral cortex and provide evidence for location of the neural substrates which mediate instantaneous colour constancy. The introductory chapter describes briefly the visual system, in the first section, with particular reference to the processing of colour. The second section discusses the psychophysics of human colour vision and the third presents a summary of computational models of colour constancy described in the literature. Chapter two describes the dynamic colour matching technique developed for this investigation. This technique has the advantage of quantifying the level of constancy achieved, whilst maintaining a constant state of adaptation. The C index is defined as a measure of constancy, with 0 representing no constancy and 1 perfect constancy. Calibration procedures for the computer monitor and the necessary transformations to accurately simulate illuminant reflectance combinations are also described. Light scattered within the eye and its effect on colour constancy are discussed. Chapter three is concerned with the effects of altering the illuminant conditions on instantaneous colour constancy. The size of the illuminant shift is varied. Artificial illuminants are compared with those of the Plankian locus. The effects of overall illuminance and the luminance contrast between target and surround are investigated. Chapter four considers the spatial structure of the visual scene. Simple uniform surrounds are compared with those which have a more complex spatiochromatic structure (Mondrians). The effects of varying the test target size and shape are investigated. The decrease in constancy as a black border is placed between test target and surround is measured. Chapter five describes experiments on four subjects with damage to the cerebral cortex. Chromatic discrimination thresholds are investigated for three subjects with achromatopsia as are the contribution of both sighted and blind hemifields to constancy for a subject with hemianopia. Contrary to the predictions of many of the current computational models, using unnatural illuminants has no substantial effect on the C index, nor does the size of the illuminant shift or the luminance contrast between experimental target and surround. The complexity of the surrounding field does not effect constancy. These findings are similar to those from chromatic induction experiments reported in the literature. However, the effect of a black annulus is found to have different spatial parameters that those reported from experiments on chromatic induction, suggesting that a different mechanism may be involved. The three achromatopsics can be shown to exhibit instantaneous colour constancy. However the blind hemifield of the hemianope does not contribute. This suggests that the fusiform gyrus is not the human homologue of V4 and that the primary visual cortex is necessary for instantaneous colour constancy.