06 ECVP
Do we “Discount” or ignore illumination?
Changes in the spectral composition of illuminants produce small, but consistent, departures from perfect color constancy. These departures from perfect constancy provide a signature of the mechanism. This experiment uses a constant set of colored papers in 27 different spectral illuminations. [All possible combinations of 3 intensities (1, 2 and 4 LEDs) for wavelengths (455, 530 and 625nm)]. Observers matched a gray, a purple and an orange paper in each illuminant to chips in a Munsell book in constant spectral illumination. If we discount illumination we need to first identify it. Departures from perfect constancy are held to be incomplete adaptation to each illuminant. The pattern in color space of these departures should follow the pattern of illuminants. If, instead, we ignore the illumination and build the color appearance from spatial comparisons (ratios), then changes of illumination affects equally both the numerator and the denominator of these spatial ratios. The overlap of cone spectral sensitivities creates chromatic crosstalk, which in turn affects spatial ratios for colorful papers, but not for grays. If we ignore illumination then the signature of departures from constancy should vary with the reflectances of the papers, not the illuminants. Observer matches confirm the second, spatial-reflectance, hypothesis.
Optical veiling glare limitations to in-camera scene radiance measurements (J. McCann & A. Rizzi)
Multiple exposure techniques are helpful in improving the digital segmentation of digital camera images. The combination of underexposed, normal and overexposed images allow much more accurate digital segmentation of the image falling on the sensor. However, the dynamic range of the image of the scene on that sensor is controlled by the veiling glare of the lens, the light reflected from the camera body, and light reflected by the sensor. The point spread function of all cameras falls off very quickly with distance. This rapid decrease is offset by the millions of pixels in an image that all contribute a small, but significant, amount of glare to all other pixels. This paper measures the radiances from test scenes to accurately document the radiances in the world. Further, it uses a uses a variety of High Dynamic Range (HDR) calibration algorithms to measure their accuracy. In the worst case, with a large-high-radiance surround, (recommended by the ISO standard for Veiling Glare measurements ISO 9358:1994), the veiling glare limit in commercial camera typically varies from 1% to 8% of the maximum radiance. That is a film-plane image dynamic range of only ~25:1. Veiling glare is an image-dependent limit to the radiances measurable in real-life scenes and must be considered in HDR imaging.
Firelight color images from Rod-Lcone interactions
Wrangham has emphasized the role of cooking in hominid evolution. He cited evidence for the controlled use of fire as early as 1.6 mya. We measured the spectral exitance from flames in wood fires. It is very similar to that from a 1700°K blackbody radiator. It has 9 times more relative emission than moonlight in the 600-700nm region. Experiments have shown that complex color images are produced by rod-Lcone at very low light levels, as long as there is sufficient long-wave light to excite Lcones. These measurements identified the radiances for each wavelength at the transition from rod vision to cone vision. Stimuli below this cone-threshold radiance activated only rods. These techniques included; dark-adaptation curves, measured action spectra, the Stiles-Crawford effect, flicker fusion photometry. Combining complex records in 500nm light (at 1/600th the radiance required for cone threshold) with complex records in 656nm light at L-cone threshold generated multi-colored images. We discuss low-light level color and the types of illumination available 1.6 mya. Moonlight spectra are nearly flat from 500 to 700nm. Using moonlight to generate multi-color images requires more light than firelight. Firelight has a spectral emission that is optimal for seeing color at very low-light levels from rod/Lcone interactions.