HDR and Color Constancy--New Psychophysical Results
8Google Tech Talks
January 22, 2009
ABSTRACT
Helmholtz was the champion of Thomas Young's trichromacy and J.C. Maxwell's colorimetry. He described in detail a bottom-up model of retinal response. Psychologists, in the second half of the nineteenth century, confronted such models of vision with "Color Constancy" experiments. Constancy presented strong evidence that quanta catch at a pixel was irrelevant to color appearance. Helmholtz's response was to introduce the concept of "discounting the illuminant". He did not describe any mechanism that could recognize illumination. He simply asserted that it was necessary. Nevertheless, the idea of using information from the rest of the image, (e.g. averages, or local averages,) to derive an illuminant estimate remains popular today. If the model succeeds in deriving the illumination, the quanta catch at each pixel can be corrected to have a calculated reflectance value for the image. These models of constancy assert that the goal of the calculation is to find the object's reflectance.
Alternative constancy mechanisms ignore the illumination. They look to building appearances from spatial comparisons in scenes. The color constancy debate continues to be split into two camps: those estimating the illuminant to discount it, and those using local relationships in the image to synthesize color appearances.
Recent experiments with Alessandro Rizzi and Carinna Parrman provide a new set of observer data designed to resolve the debate. The experiment uses a collection of 3-D wooden blocks in uniform and non-uniform illumination.
To make a long story short, when you ask observers what they see in these 3-D Mondrians, they are unable to discount complex illuminations from shadows, multiple illuminants, and colored illuminants.
These new experiments go a long way to resolve the debate, and refocus color constancy research on multi-resolution spatial image processing. Human constancy does not work well enough in complex 3-D scenes to perfectly generate appearances that correlate with reflectances. It is the high-contrast response to edges, and the low-contrast response to gradients that correlates with appearance. Human vision treats edges in illumination the same as edges in reflectance. Humans do not discount the illuminant.
Speaker: John McCann
John McCann received a B.A. degree in Biology from Harvard University in 1964. He worked in, and later managed, the Vision Research Laboratory at Polaroid from 1961 to 1996. He has studied human color vision, digital image processing, large format instant photography and the reproduction of fine art. He is a Fellow of IS He is a past President of IS and the Artists Foundation, Boston. He is currently consulting and continuing his research on color vision. He is the IS/OSA 2002 Edwin H. Land Medalist and IS 2005 Honorary Member and is a 2008 Fellow of the Optical Society of America.