[argyllcms] Re: Shadow detail problem
- From: Gerhard Fuernkranz <nospam456@xxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 01 Feb 2008 23:34:13 +0100
Dave Wagner wrote:
> My question is, should an ICC profile preserve this dynamic range of the
> monitor calibration? Or should it push
> everything below level 20 to below 4 or 5 so that it looks indistinguishable
> from black?
The result to be expected depends on the rendering intent. A
reproduction with colorimetric rendering intents are expected to
strictly preserve the colorimetry for in-gamut colors, and to clip
out-of-gamut colors (including clipping of colors which are darker than
the display's black point).
Perceptual intent on the other hand is expected to do gamut mapping, so
that "black" in the source colorspace get mapped to the display's black
point.
Some CMMs offer a pseudo-intent "relative colorimetric with black point
compensation". This intent performs a simple gamut mapping at run time,
which is limited to lightness mapping and which also maps the source
colorspace's "black" to the display's black point. This is however not a
property of the profile, but a feature of the CMM which applies the
profile. LCMS for instance supports this, but the application must
explicitly make use of it, if desired.
For typical image editing work, you will likely prefer either perceptual
intent, or relative colorimetric with BPC, and I guess you rather won't
like the clipping of dark colors which you get when you choose a pure
colorimetric reproduction, particularly if your display happens not to
have a sufficiently dark black point.
> The LUT profile on my LCD does this, the shaper profile does not. With the
> shaper profile, the shadows are still distinguishable at 4 or 5.
Are both created from the same measurement set, and both created with
Argyll? Which "quality" was used for creating the LUT profile? For
instance, for my notebook LCD, -qm is not sufficient to describe the
behaviour of the display in dark areas accurately enough, I did need -qh.
> See these for reference, and flame them or me if they are bad advice,
> or if it only applies to calibration and not to profiling:
>
> http://www.drycreekphoto.com/Learn/Calibration/monitor_black.htm
> http://www.xrite.com/product_overview.aspx?ID=756&Action=Support&SupportID=3558
>
> Chapter 9 (page 218) of "Real World Color Management" (Peach Pit
> Press, 2003 printing) gives similar advice for checking calibration.
> This is speaking of evaluating calibration before evaluating profiles
> and CMS, but it never makes it clear how CMS should change it.
>
> Speaking of exactly the process I spoke of before: create a black
> image in photoshop and then select a rectangle and adjust the black up
> one level at a time in curves, they write...
>
> "With excellent calibration systems, you may see a difference between
> level 0 and level 1. More typically, you won't see a change until
> somewhere around levels 5 to 7, or sometimes even higher. If you
> don't see any change when cycling through the first twelve levels,
> your black point is definitely set too low and you should recalibrate,
> requesting a slightly higher black point."
I would not take particular number too serious, since they depend on the
response curve which was used as calibration target. To get a rough
feeling, for instance in sRGB space, L* = 1, 2, 3, 4, 5, 6,...
corresponds to R=G=B levels of 4, 7, 11, 14, 17, 19,... (on a 0..255 RGB
scale). And if you use a power function with gamma 2.2 instead (without
the linear segment near zero), then L* = 1, 2, 3, 4, 5, 6,...
corresponds to R=G=B levels of 11 (!), 16, 19, 22, 24, 26,...
I have for instance calibrated my notebook LCD display (i.e. not a
high-end device) with an sRGB response target. And with the VCGT loaded
I can distinguish brush strokes with R=G=B=4 (in the display's native
color space) from the black background (R=G=B=0) on which they are
painted. The difference is not significant, but just noticeable. And if
I convert the sRGB image with the RGB=4 brush strokes on black
background to the colorspace of my LCD display, by applying the display
profile with perceptual intent, then I can still distinguish the brush
strokes from the background (with the pipette I can measure RGB=[4,4,5]
now). With relative colorimetric intent I get only a completely black
area of course - as expected the colors darker than the display black
point are clipped. And relative colorimetric with BPC, applied with LCMS
(tifficc -b...), results in RGB=[3,3,4] in the display's RGB color
space, which is also still just distuinguishable from the black background.
Regards,
Gerhard
Other related posts:
