[argyllcms] Re: using a digital camera as a colorimeter
- From: "Alastair M. Robinson" <blackfive@xxxxxxxxxxxxxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Fri, 07 Oct 2005 12:22:09 +0100
Stephan Bourgeois wrote:
I used the IT8 to profile the camera. Then took a snap of the print
target in a separate shot, without changing the lighting conditions. The
first step is necessary to relate the camera's exposure range to CIE L
values, and compensate for eventual white balance shifts.
OK - so your camera has very little variation between shots then?
I am with Linux. I am afraid the clipping is due to the scanner
calibration strip. I read posts about this on the newsgroup.
Same here. VueScan is available under Linux AFAIK - not used it myself
How detailed is the calibration strip? Just black / white or are there
colours there too? Could it be replaced (albeit with a need to
dismantle the scanner)?
Hmm, this is very clever, and would explain the cost of the instrument.
I thought LED spectral emission to be spikey, so a collection of them
would be needed to create a smooth overlap. The sensor is likely to be a
Indeed. There seem to be several different red, green, blue and orange
LEDs, which shine in sequence. The colourmouse takes approximately 4
seconds to take a sample.
The values I get when doing "profile" on the IT8 are:
Snapscan1212: peak err = 7.703943, avg err = 0.807319
DX6440: peak err = 8.134495, avg err = 1.193943
They're pretty typical for scanners. That's quite an impressive score
for the camera though...
No big difference between the two, but an order of magnitude larger than
when using the colourmouse. This is spiking my interest for the
OK - one thing to bear in mind is that you might have been using a
full-page 780-odd patch target for printer profiling.
With the colourmouse, reading huge numbers of patches is tedious, and
since for the most part I'm profiling to replace Gutenprint's own colour
adjustment with something that gives more pleasing photographic output,
and not looking for absolute precision, I'm using much smaller numbers
of patches - typically under 150.
The more patches there are, the harder it is to fit a model to them, and
so the more likely it is that the peak error will go up.
If you look back in the archives, you'll see that a few weeks ago I was
finding that a profile made using D50 XYZ measurements was giving me a
yellowy cast, and that using D65 XYZ measurements gave better results.
I've now figured out what the problem was - third-party inks!
In my R300 I've been trying out various types of ink, since the Epson
ones are so expensive. The best results so far have come from
7DayShop's own compatibles, but when I was experimenting a few weeks
ago, I still had black and cyan cartridges from a different source.
Yesterday I pulled them out, so I now have a complete set of 7DayShop
carts, and reprofiled using D50 measurements. The colour cast is gone...
All the best,
Alastair M. Robinson
Other related posts: