When Some Clipping is Allowed

The below chart shows the colors of a CRT at maximum luminance (Y=1) with the criteria: For every RGB triple clipping of larger value than 255 is allowed on one of the color channels. So the two remaining channels are colorimetricly accurate.

Y=1.0, one ch max clip.

Because the mapping for the above image was done at Y=1 only the whitepoint is reproduced correctly, the colored areas have one channel (either red, green or blue) clipped to 255 but there is no clipping to level 0 on any channel, for colors that are grayed clipping occurs on two channels either to 255 or to 0.

Things get more interesting when clipping is also allowed for negative channel values. The below chart shows the colors of a CRT at maximum luminance (Y=1) with the criteria: For every RGB triple clipping of larger than 255 channel value is allowed for one of the color channels and clipping of negative value is allowed for one of the color channels. So, one of the color channels in every RGB triple is still colorimetricly correct, the two other may be clipped.

Y=1.0, one ch negative clip, one ch max clip.

Again the mapping for the above image was done at Y=1 so only the whitepoint is reproduced correctly. Interestingly now when also one negative channel value is allowed to be clipped we see that a CRT can somewhat faithfully represent a larger gamut than it natively has, there is quite a lot of hue differences in the out-of-Trinitron area.

Because of this phenomena (clipping of only one or two color channels does not remove all the color/image information) it is possible e.g. to show a very accurate representation of the Kodak Q-60 Color Input Target on a native CRT monitor even if the Q-60 target has far larger gamut than the CRT has.

Finally a look at Y=0.5, with the same criteria as above: For every RGB triple clipping of larger than 255 channel value is allowed for one of the color channels and clipping of negative value is allowed for one of the color channels. So, one color channel in every RGB triple still is colorimetricly correct the two other may be clipped.

Y=0.5, one ch negative clip, one ch max clip.

Compare this against the 128/255 from the first on-top-of-each-other maps were no clipping was allowed. There is huge increase in coverage and out-of-Trinitron areas do show plenty of hue variations.

Conclusions and Discussion

The main finding surely is that a simple gamut map does not tell much. The gamut of a trichromatic device will always shrink to a point (to the whitepoint) when the luminance reaches maximum. The full triangle gamut of a device is available only up to the luminous efficacy of the blue channel, above that the gamut starts to be clipped at the blue corner, then when the the luminous efficacy of the red channel is reached also red red corner of the gamut starts to be clipped and finally when the luminous efficacy of the green channel is reached also the green corner of the gamut starts to be clipped.

All the color/image information is not lost when only one or two channels are clipped. In fact in order to show a faithful reproduction of a large gamut scene/original on the CRT monitor some channel clipping is necessary, slight gamut clipping allows to render a large gamut scene/original more realistically.

These maps most probably also explains the so called purple fringe problem that encumber many consumer grade digicams. When the whitepoint of the image is detected slightly incorrectly and/or colorimetricly inaccurate conversions are used, then an out-of gamut blue can sometimes be mapped as purple. In the green and red corners even a large error that is induced by the color conversions do not create a visually large error, green stays as green, red may only turn slightly towards orange. In the blue corner the color can turn slightly towards cyan (this is visually a small error) but a rather small error can turn the blue to magenta and this is often very visible.