Illuminant estimation in surface color perception
Color appearance is one of the central issues in color vision. Starting with Helmholtz, research concerning color appearance has been revived in recent years with new algorithms. The core of those algorithms is to provide a way of computing the illuminant out of the quantum catch at the photoreceptors, to arrive at surface reflectance. The visual system is assumed to take an estimate of the illuminant from the scene. The question sought here in this thesis was, ‘where in the scene are the illuminant sources, or cues?’ Chapter 1 summarizes previous research. Chapter 2 addresses a key methodological problem: correctly rendering colors on display devices in modern experiments. Chapter 3 describes experiments testing two candidate illuminant cues using a novel ‘perturbation’ method. The rendered images included spherical objects and flat surfaces, illuminated from upper left. The tested cues were specularities and uniform background. The conclusion of Chapter 3 is that the visual system takes an illuminant estimate from specularities, but not from the background, and that the visual system treats daylight illumination differently from a reddish reference illuminant. This phenomenon was further explored in Chapter 4, to see whether illuminant estimates change when the contents of scenes change. The scene was changed by varying the number of objects and moving the location of the test patch. I also tested whether illuminant estimation is statistically ‘robust’. Changes to the scene did not alter the conclusions of Chapter 3. Some individual differences were found in examining robust estimation. Chapter 5 summarizes the conclusions of this thesis.