[argyllcms] Re: 1st post (improved estimation of blackpoint using a digital camera)
- From: Knut Inge <knutinh@xxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Thu, 9 Dec 2010 10:05:21 +0100
I am assuming that my camera has unknown spectral properties, but that it is
linear over a wide range of input levels. My display has unknown spectral
properties and is non-linear.
If I made a display ramp from 0-255 in e.g. the "red" channel, and analyzed
the camera sensor response in its "red" channel. Even though absolute
brightness would be unknown, is it not sensible to believe that relative
brightness will be accurate? So if my camera tells me that rgb-code [1,0,0]
produce a number of photons that is 1/500 that of rgb-code [255,0,0], this
can be used to build a per-channel non-linearity/gamma scale of high
precision of the range 0-255? If the spyder 3 whitelevel can be depended
upon, the two sets of measurements could (?) be combined to provide one
absolute scale.
I am assuming that however complex lcd displays are, they do not appear as
one 3x3-matrix at low input levels and a very different 3x3-matrix at higher
levels (level-dependent gamut "twisting"). Rather, I would think that they
can be modelled as 3 sets of some spectral responses that is processed by
one linear 3x3-matrix and per-channel non-linearity.
Spectral characterization and whitepoint and such, I would leave to the
calibrator, but you mentioned that accurate black-point measurements was
difficult using cheap calibrators.
I think that grabbing camera raw files using dcraw and processing them using
MATLAB is no big obstacle. The real question is if this makes physical sense
or if it is only wishful thinking, wasting everybodys time.
-k
On Thu, Dec 9, 2010 at 8:44 AM, Graeme Gill <graeme@xxxxxxxxxxxxx> wrote:
>
> In principle you can. It sounds like a big, expensive project through.
> First you would have to characterise the cameras spectral sensitivities -
> that takes a monochrometer to do well, or a lot of fancy software
> and a ColorCheckerDC/SG to do approximately. Then you have to spectrally
> characterize the display (needing a spectrometer), and compute a camera
> RGB to CIE XYZ calibration matrix. Then you have to write software to
> control the camera and process the resulting RAW file. Then you have
> to interface it to the calibration software.
>
> In all, it is probably cheaper and faster to buy a better instrument,
> such as the spectrometer you would need to spectrally characterise the
> display.
>
> [People have tried some hacky ways of using a camera as a colorimeter,
> and even used it on displays - Argyll will let you do that - but
> the results are never very good due to the spectral issues, and
> it won't work with Argyll's display calibration code, because the
> calibration code is interactive - unlike profiling it creates test
> values on the fly.]
>
> Graeme Gill.
>
>
Other related posts:
