Why Whites May Not Match
I have been asked why white on two displays of different types - usually a CRT and an OLED monitor - look different even though they have the same CIE XYZ when measured with a spectrometer. Here is my answer for OLEDs and CRTs: the same arguments can probably be applied to other displays. Please cut-and-paste as you see fit.
Richard Kirk 19:08, 19 January 2012 (UTC)
Comparing an OLED and a CRT...
The OLED red is a single broad peak, while the CRT red is a complex spectrum with two sharp peaks. All video monitors used to agree ten years ago because all quality monitors were CRTs with the same primaries. If you went back 30-odd years and found NTSC primaries with the orangey-red phosphor then you might have seen two CRTs with whites that did not match. If digital projectors were the same size as CRTs then we might have noticed the whites did not match five years ago, but they aren't and nobody did.
Today we have flat-panel monitors which look respectable next to a BVM CRT. We see sharp detail and good shadow tones. The obvious next step is to calibrate the monitors and see how similar they look. We match the monitor white points in XYZ using a spectrometer, but the whites look different. Different people see different shifts. Why? The only explanation is that the CIE standard observer is not accurate enough to predict what we see.
Can we improve on the 1931 standard observer? This is surprisingly hard to do. The CIE model is based on the assumption that we have three detectors in the eye sensitive to red, green, and blue, a bit like a digital camera mosaic. If two spectra give the same signal at these three detectors, they should look the same.
The human eye is not really like a digital camera. We have a central region on the retina that we use for our precision vision, and we move this around to capture detail as we need it. The rest of our peripheral vision is very fuzzy and has a lot of chromatic aberration which our brain filters out. The central region has a yellow spot of dye over the top and no blue detectors underneath. The colour image we 'see in our head' is a weird synthesis of the sharp, central, yellow filtered RG image and the surrounding unfiltered RGB image. Different people have different densities and different sizes of yellow spot, and this affects how their brains merge their central and perhipheral vision.
The original researchers back in 1931 knew about this. They did colour matching experiments where you mixed red, green, and blue lights to match a reference colour. The two colours patches were halves of a circle with a dark surround. They saw they got different results with different sizes of circle, so they settled on a 2-degree circle - about a third bigger than your thumbnail at arm's length - because this gave the most consistent results. When you look at a white patch the size of a monitor, this is a lot larger than the CIE 2-degree patch, so the colour we see is influenced much more by the peripheral vision. This means we can get the paradoxical situation when we match all the colours throughout an image at the 2-degree level, but are left with an overall sense that the white does not match at a larger scale. The Arri image is particularly good as showing this.
This is not something that can be easily corrected. We can try colour matching experiments with larger circles, but our eyes move with subjects larger than 2 degrees so it is hard to be sure which part of our eyes we are using. There is a CIE 1964 standard 10-degree observer, but it is not a trusted standard. Also, different people see different matches, so one standard will not fit all.
Can I grade on an OLED display?
The OLED is not the same as the CRT, but that does not make the OLED 'wrong' and the CRT 'right'. Ten years ago we might have said that anyone looking seriously at an image used a CRT, and so a CRT represented the best working practice. Now CRTs are rare in home and grading suites, and the choice is much less obvious.
My advice is to use an OLED monitor or a CRT when grading. Either will work, but do not use both, and certainly do not put them next to each other. The OLED primaries are at least as good as the old CRT ones. The fault that creates the visual difference is not in the CRT or the OLED but because out eyes do not fit the CIE standard observer.
D65 white has a spectral definition: it was based on the spectrum of noonday sun at Rochester. However, most of us only judge whether our display matches to D65 by measuring XYZ and invoking the CIE standard observer. If we compared any display to true D65 white, it probably would not match: true D65 has a lot more deep blues and violets than almost all display and projector whites.
Why do we not see colour errors on other displays?
This is quite a recent thing. Ten years ago, people would not have expected to get a visual match on different media.
I have not seen a digital projector look as different to a CRT when displaying matched in-gamut colours, though some digital projectors have red spectra very similar to the OLED displays. However, we have always had to tweak the white point when doing butterfly tests, which is a similar error when comparing a digital projector to film. We also see similar differences: we can get almost all the image colours to a good match, but we still have an impression that the overall white point is different, with the film version looking more yellow-green and less bright.
I believe when we have two similar media, such as two flat panel displays or two projections on a screen, we are conscious of small differences in the white point; but when we compare different media such as a CRT and projected film, or a CRT and a digital projector, then our brains register the different sizes and natures of the displays and ignore the small colour difference.
Should we use something newer than the CIE 1931 observer?
We know our eyes do not fit the assumptions behind the CIE standard observer. At FilmLight, we got fifty people to match the white of an OLED display to a fixed white on a CRT, and measured the whites with a spectrometer. There was a significant scatter to the results: what looked correct for one person looked significantly out for another. This means we cannot fix the CIE for one observer without making it worse for another. The problem is not that the 1931 standard is 80 years old. The 1931 experiments have been repeated, and you can get more modern versions of the XYZ weights. These new weights are slightly different in the deep violets and reds, but these changes make little difference to the XYZ values of sensible displays with little signal at these wavelengths. The 1931 values are sufficiently good that most people today prefer to use the standard values that everyone knows, rather than the more modern variants.
The following was posted by Jim Houston to the TIG mailinglist and deserves mention:
On Jan 29, 2012, at 5:58 PM, Rob Lingelbach wrote: > "D65 white has a spectral definition: it was based on the spectrum of noonday sun at Rochester."
Sometimes stories develop that come through the same route as Urban Legends... something we somehow would like to believe is true, but is not really.
An accurate description of where all of the "D" colors come from can be found on page 7 of Wyszecki and Stiles (1982), the bible for early color research. There is no doubt that a major center of early color research was Rochester and Kodak literally drove the field of inquiry for decades. There is also likely a spectral definition that the Kodak lab used that was measured at Rochester and used for quite some time. But the Daylight locus was calculated from 622 spectral power distributions measured at a minimum in Rochester, Ottawa, and London (and possibly some other locations).
The daylight curve was designed as an offset from the pure Physics "blackbody radiation curve". The D65 location is where the correlated color temperature of 6500 degrees hits the daylight locus. The CIE standard daylight (circa 1964) came from NASA above the atmosphere measurements, and was also calculated as a reference for 1 "air-mass" at the earth's surface as a combination of direct sunlight and integrated skylight. For reference, the direct color temperature of the sun is about 5800K. The rest of the 'blue' that takes daylight to D65 is the 'blue' skylight calculation.
Too much information, but the gist of it is the work to create a daylight standard reference was a lot more involved than the story suggests.
--Rob Lingelbach 18:38, 1 February 2012 (UTC)