[argyllcms] Re: Colorimeter vs. Spectro-colorimeter for Display caracterization - Derived from:"Displays with internal gamut emulation"

  • From: Graeme Gill <graeme@xxxxxxxxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Tue, 31 Aug 2010 13:45:36 +1000

Matthieu DUBAIL wrote:
Here is what I thought:

Colorimeters are better for displays. Unlike printed paper, displays have
deep blacks. Spectrocolorimeters are weak on low luminance measurement cause
they have narrow spectral bandwith, unlike colorimeters.

Think about it: If you take the same amount of light and split
it up into narrow spectral bands, then yes, the measurement for
each band has a lower s/n ratio than splitting the light into
(say) 3 bands. But then those spectral measurements are weighted
and summed again using the standard observer curves, cancelling the
noise out in proportion to the number of bands being summed,
and improving the s/n ratio. But a colorimeter using passive filters
(rather than a diffraction grating or dichroic filters),
would need 3 times the sensor area because it is throwing light
away in the filters. So it is not the spectral nature of the spectrometer
that is important in the overall s/n ratio of XYZ values, it is the amount
of light captured and the efficiency of the filtering and measuring
that is important. In general it is easier (ie. cheaper) to make a colorimeter
that collects a lot of light than a spectrometer.

But have you examined the types of sensors used in most cheap display 
colorimeters ?
I have, and they are typically off the shelf light to frequency converters, with
very small sensor areas. So no advantage is being taken of this possible
avenue of advantage towards colorimeters. [The only exception is the DTP92/94]

So in practice the s/n ratio of something like the i1pro or
ColorMunki is actually better than these cheap colorimeters when
using comparable integration times.

Every  display models I heard of (form simplest to most complicated) are
colorimetry based.

If you are talking about device models, then this simply reflects the fact
that displays are typically additive devices, and therefore tri-stimulus models
are good enough. But this isn't always the case :- for instance, if you
want to model a projector reflecting from a screen, you need to use a spectral
model to account for the screen.

Spectrocolorimeters might miss spectral peaks (like those found on CRT red
primaries, fluo and led LCD backlight) depending their spectral bandwith and
overlap.

They typically don't miss reading energy from any wavelength, since the gaps 
between
the sensors is quite small, and the optics spreads narrow peaks somewhat.
The chief drawbacks of (an affordable) spectrometer is wavelength calibration
accuracy and stability, and the observer curve quantization that occurs
due the discrete nature of the wavelength measurements.

Spectrocolorimeters are better for printer.
Some printers models are based on Neugebauer primaries or other
multispectral model, computed form spectral datas.

Sure, but this has nothing directly to do with the measurement device.
Of course you need spectral measurements to feed into a spectral
model, but this doesn't imply that you need a colorimeter to
make measurements to feed into a tri-colorimetric model.
You just need tri-colorimetric values, which can (much more
flexibly) be derived from spectral values.

Graeme Gill.

Other related posts: