RE: Humans (and cameras and scanners) do not have a color gamut (?)
RE: Humans (and cameras and scanners) do not have a color gamut (?)
- Subject: RE: Humans (and cameras and scanners) do not have a color gamut (?)
- From: Wayne Bretl via colorsync-users <email@hidden>
- Date: Wed, 8 Jan 2020 15:47:37 -0700
I think we agree on all this. If you want precise colorimetric results using
only a linear matrix for conversion, the sensor responses must be precise
linear combinations of the CIE curves. If they are not, errors may be
reduced by non-linear means (LUTs). If the responses are too narrow and far
from linear combinations of the CIE curves, the errors will be more extreme
and the result will be far from colorimetric, especially for saturated
colors, even with a LUT. When the sensor output ratio limit is reached due
to narrowness of the sensor and input spectrum, there is no further
variation to be teased out with a LUT. This was the overall point of my
paper, which apparently had not sunk in on some (many?) still-camera/
video-camera/ digital-cinema-camera sensor designers at the time.
An extreme example is the color film studied by Holm, where some wavelengths
that should have distinguishable hues excite only the green channel and
therefore convert to a single chromaticity point at the corner of a
triangular gamut. You can change what that point is by changing the
conversion, but you there is no way to distinguish the hues of those
wavelengths from each other.
-----Original Message-----
From: Iliah Borg <email@hidden>
Sent: Wednesday, January 08, 2020 3:21 PM
To: Wayne Bretl <email@hidden>
Cc: 'Andrew Rodney via colorsync-users' <email@hidden>
Subject: Re: Humans (and cameras and scanners) do not have a color gamut (?)
Dear Wayne,
I'm afraid we are going nowhere with this. Gamut is a certain subset of
colours. No colour output is available from the sensor until conversion. The
sensor output and the colour rendering are in different physical "units".
We even have different understanding what a camera observer is, what I mean
saying camera observer is quite parallel to what a colorimetric observer is.
I'm well past the idea of using sensors with narrow responses in
photographic cameras. Even for 5 / 7-capture monochrome cameras we still use
broad responses of the filters in front of the lens.
Still, we use LUTs, not matrices, to cope with some issues introduced by
matrices.
On Jan 8, 2020, at 5:01 PM, Wayne Bretl wrote:
> " That matrix defines the relation between the camera observer and
> colorimetric observer."
>
> That is not quite correct. The relationship between the camera
> observer and colorimetric observer is defined by the concatenation of
> the sensor spectral response and the matrix. On the input side of the
> sensor, it is a multidimensional problem with the number of dimensions
> determined by the number of significantly different wavelength
> response points. I like to think of it as related to sampling theory.
> If the spectra of the light source, the object reflectivity, and the
> sensor responses are all broad and smooth, almost any case can be
> matrixed to give a close approximation to a colorimetric camera. This
> also holds true pretty well if only one of the three items'
> (illumination, object, sensor) spectra is narrow or spiky. So, for
> example, you can get a pretty good color rendering index with a lamp
> spectrum that is not really much like daylight or incandescent. But
> mix that with a sensor that has narrow responses, and all bets are off
> - for example, note the greenish rendition obtained with old-fashioned
> office fluorescent lamps and color film, and the still-off rendition of
various objects when the white balance was corrected with a magenta filter.
>
> Another example: have you heard of the selected ordinate method of
> computing CIE tri-stimulus values? This was a method in which, instead
> of uniformly spaced and variously weighted spectral samples,
> non-uniformly spaced and unity-weighted samples were used. Quite a
> simplification in the days before computers, and it gave approximate
> but good enough results when applied to broad-spectrum objects under
> broad spectrum illumination. In this case, each "sensor" channel had a
> response at only 30 single wavelengths and zero response elsewhere.
> Plus, there was an abbreviated version that used only 10 discrete
wavelengths!
>
> A combination that does not work well is a sensor with narrow
> responses with an object with narrow spectrum as well. Such highly
> saturated object colors will experience an error quite different and
> greater than that of broader-spectrum objects, as show by the spectrum
> loci in Holm's paper and the gamut-interior plots in my paper. In
> those plots, a matrix was applied that produced a decent colorimetric
> correction for test chart patch spectra, but the narrower spectra show
huge errors.
>
> -----Original Message-----
> From: Iliah Borg <email@hidden>
> Sent: Wednesday, January 08, 2020 2:36 PM
> To: Wayne Bretl <email@hidden>
> Cc: 'Andrew Rodney via colorsync-users'
> <email@hidden>
> Subject: Re: Humans (and cameras and scanners) do not have a color
> gamut (?)
>
>> See Jack Holm's paper ... where he derives the spectral locus of a
>> combination of a sensor's outputs and a subsequent matrix
>
> I've read that paper when it was first published on the web ;) It
> doesn't define sensor gamut, or gamut of sensor outputs.
>
> To quote,
>
> "This paper reports on the observed characteristics of the capture
> color analysis gamuts resulting from a number of capture
> devices/media, and scene analysis color matrices."
>
> The result depends on the matrix. Strongly depends, as the paper
> itself demonstrates. That matrix defines the relation between the
> camera observer and colorimetric observer.
>
> On Jan 8, 2020, at 4:21 PM, Wayne Bretl via colorsync-users wrote:
>
>> Arrgh - addressing problems - went only to iliah and not the whole
>> list as intended
>>
>> -----Original Message-----
>> From: Wayne Bretl <email@hidden>
>> Sent: Wednesday, January 08, 2020 2:13 PM
>> To: 'Iliah Borg' <email@hidden>
>> Subject: RE: Humans (and cameras and scanners) do not have a color
>> gamut (?)
>>
>> See Jack Holm's paper, which I posted above, where he derives the
>> spectral locus of a combination of a sensor's outputs and a
>> subsequent
> matrix:
>> http://www.color.org/documents/CaptureColorAnalysisGamuts.pdf
>>
>> Although he does not show the sensor outputs themselves, they have
>> reached their extreme possible ratios when the stimulus is a single
>> wavelength, and therefore a pseudo chromaticity chart could be
>> plotted using the ratios of one channel to the sum of the three on
>> one axis and a ratio of a second channel to the sum on the orthogonal
axis.
>> This chart would have a defined gamut of sensor values (inside the
>> spectrum locus) but would not represent colors per se until processed
> through a matrix.
>>
>> Holm shows the resulting chromaticities for multiple sensors and
>> several reasonable matrices. He includes cases for motion picture
>> negative film that clearly show a triangular gamut limit (spectral
locus).
>>
>> -----Original Message-----
>> From: Iliah Borg <email@hidden>
>> Sent: Wednesday, January 08, 2020 1:59 PM
>> To: Wayne Bretl <email@hidden>
>> Cc: Andrew Rodney <email@hidden>; Andrew Rodney via
>> colorsync-users <email@hidden>
>> Subject: Re: Humans (and cameras and scanners) do not have a color
>> gamut (?)
>>
>>
>> On Jan 8, 2020, at 3:52 PM, Wayne Bretl via colorsync-users wrote:
>>> .
>>> The discussion has moved beyond the true but uninformative statement
>>> that cameras (sensors to be precise) do not have a color gamut , to
>>> considering how the gamut of sensor outputs
>>
>> I would very much like a definition here. What is the gamut of sensor
>> outputs?
>>
>> And of course, how to measure it?
>>
>> --
>> Best regards,
>> Iliah Borg
>> LibRaw, LLC
>> www.libraw.org
>> www.rawdigger.com
>> www.fastrawviewer.com
>>
>>
>>
>>
>> _______________________________________________
>> Do not post admin requests to the list. They will be ignored.
>> colorsync-users mailing list (email@hidden)
>> Help/Unsubscribe/Update your Subscription:
>> 4
>> 0gmail.com
>>
>> This email sent to email@hidden
>
> --
> Best regards,
> Iliah Borg
> LibRaw, LLC
> www.libraw.org
> www.rawdigger.com
> www.fastrawviewer.com
>
>
>
>
--
Best regards,
Iliah Borg
LibRaw, LLC
www.libraw.org
www.rawdigger.com
www.fastrawviewer.com
_______________________________________________
Do not post admin requests to the list. They will be ignored.
colorsync-users mailing list (email@hidden)
Help/Unsubscribe/Update your Subscription:
This email sent to email@hidden