Jeremy Pronk wrote:
http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.151.6187
A problem with this paper is that they claim to have solved the problem of a change in scene illumination. I can't see how they have done that, since it's inherent in tri-stimulus colorimetry, when the camera does not meet the Luther condition. The only way of solving this is to spectrally characterize the cameras sensors, and this is very hard to do without using a monochrometer based measurement technique, and they don't refer to any of the more difficult techniques for avoiding that. If they are merely referring to the fact that by default an instrument will read D50 based XYZ values from a reflective target, then this is easily tackled by measuring the spectral reflectance, the scene illuminance spectrum, and combining them to compute the XYZ values for the scene illumination (Argyll will do this if you supply a spectral input chart reference for instance.) For their technique to work for different scene illuminants, they would have to use the scene illuminant as the backlight for their test target, and re-characterize each camera for that illuminant, which is exactly the same situation as that for a reflective target. Graeme Gill.