[argyllcms] Re: eye-one pro Rev A/Rev D IR calibration?

  • From: Gerhard Fuernkranz <nospam456@xxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Sat, 26 Jun 2010 17:53:26 +0200

Am 26.06.2010 00:22, schrieb Klaus Karcher:
> (I dare say the radiation maximum of the lamp is still above the
> relevant wavelengths at the end of the reading, but I have no idea
> which compensation mechanism might affect the result).

Hello Klaus,

the IR output of the tungsten lamp is certainly rather high, but I'm not
sure whether there is possibly an IR filter in the play? Do you know?
And the sensor sensitivity might be already down in this region, too. In
order to assess whether the instrument can still deliver reasonable
values in this region, one rather needs to take a look at the raw values
obtained from the sensor.

If the SNR of the sensor readings happens to be too low in this region,
so that the correction would just result in "wild numbers", then
shrinking the numbers (e.g. in a Wiener Filter like way) may well make
sense and may even be the only way to obtain plausibly looking readings
at all.

At very low SNR levels, even reporting a constant reflance of 50% for
any value obtained from the sensor can still result in a lower average
error than "wild" numbers like 350% or -150% that might result from a
division by small, noisy numbers (note, when reporting 50% reflectance
then the error can never exceed +/-50). I'm not sure though, whether
Graeme does some kind of shrinkage in his driver, which may lead to the
observed bias.

In your measurements, not only the inter-instrument bias increases
significantly at the IR end, but so does the standard deviation (which
is 7.5 (!) at the longest wavelength, while the median stdev over all
wavelengths is just 0.3, that's a factor of more than 20). So I'd guess
that the SNR of one (or both) of the instruments is rather low in this
region, and I'm not sure whether it will be possible to make low-error
measurements at all in this region. But this needs further
investigation, of course.


Other related posts: