Re: color calibration of microscopic images
From: David Littlewood (david_at_nospam.demon.co.uk)
Date: 09/01/04
- Next message: Kevin Cunningham: "Re: microscopes for the faculty"
- Previous message: David Littlewood: "Re: color calibration of microscopic images"
- In reply to: Beatrice: "color calibration of microscopic images"
- Next in thread: Aaron: "Re: color calibration of microscopic images"
- Messages sorted by: [ date ] [ thread ]
Date: Wed, 1 Sep 2004 23:27:55 +0100
In article <56fab26.0409010352.d94e109@posting.google.com>, Beatrice
<pappagallo16@libero.it> writes
>Hi all,
>
>I would like to calibrate the color of images acquired with a digital
>camera attached to a microscope. I have collected a lot of information
>about color calibration and I set up a procedure, but I am having some
>problems.
>Since I do not have much practical experience, at some point it
>becomes difficult for me to establish the quality of what I obtain.
>Maybe someone with more experience could help me in this sense?
>
>
>Here is a description of what I am doing.
>
>1) Samples
>The kind of samples I am working with are thin sections made from tiny
>chips of paint taken from paintings (typical size is of the order of
>100um, with particles from 10um down to submicron size). For better
>handling, the sections (or cross-sections as we call them) are
>embedded in a resin block and polished until the paint is exposed.
These do seem too small to use with a reflectance densitometer,
certainly any common one. IIRC, the measuring spot on a typical Macbeth
instrument is about 1mm across.
>
>2) Aim
>What I would like to do is to compare the appearance of cross-section
>in light-microscopic images. One reason for doing this is for example
>the following. After re-polishing, the appearance of the
>cross-section, although still recognizable to be the same, changes
>because several microns of particles are polished away. But how much
>is this new appearance similar/different from the older one? Of course
>I might also wish to compare images of different cross-sections.
And presumably look for different coloured layers to detect overpainting
or glazes.
>
>3) System and acquisition conditions
>For acquiring images I am using a LEICA DMRX microscope with internal
>tungsten-halogen lamp (3200K at the appropriate voltage) illumination
>source, with attached a 24-bit NIKON DXM1200 color digital camera.
I assume you are using a voltage-stabilised power supply, otherwise
supply voltage fluctuations may cause significant colour temperature
variations.
>
>I work in reflection mode (the whole resin block is too thick for
>transmission), dark field at 20x magnification and I use neutral grey
>filters (for a while I used also the CB12 blue filter to get closer to
>daylight conditions, but I had the feeling that this is not necessary,
>especially if I color calibrate my images). I also use a shielding box
>in order to eliminate external light.
>
>Before acquiring images I set the white point by using (can't remember
>exactly now) filter paper with about micron pore size.
I would have thought filter paper may be a bit variable. Can you use a
Macbeth calibration white standard or something?
>Then I acquire the flat field (this is actually causing me other type
>of problems, but I will ignore it for the moment) and images of the
>patches (one by one) of a Gretag Macbeth ColorChecker chart (24
>patches). Since under microscope the color patches do not look
>homogeneous but I see their structure, I calculate and refer to the
>average (RGB) color.
>
>4) Color calibration
>So, now the reasons why I want to make a color calibration are:
>
>a. To have accurate color representation (reasonably accurate, does
>not have to be perfect).
>
>b. Most importantly, to be able to compare colors of different images.
>Color calibration would on one hand compensate for variations in the
>illumination conditions (e.g. aging of the light source), and would
>also allow me comparing images acquired at different exposure
>times/settings.
>
>
>Now, these are my problems:
>
>1) The type of color standard target.
>As far as I know there are no color calibration targets on the market
>for usage under the microscope. Perhaps someone has knowledge of any
>such target?
>The problem with the chart I am currently using is that under the
>microscope I see the structure of the patches.
You could try defocussing the microscope to blur the image. This should
not change the colour balance.
> Also, at the exposure times that I use for acquiring images of my
>samples, many of the color patches of the Gretag chart are
>overexposed. If I would use an exposure time such that the color
>patches are not overexposed, then my samples are underexposed.
Do you find a colour shift as you change the exposure time or ISO
setting on the camera? If not, problem solved. If so, do neutral density
filters change the colour? If not, problem solved. Failing these, try
using a pair of polarisers in a mount in which one can be rotated
relative to the other to give varied extinction - Zeiss used to make one
of these. You would of course still need to check that there was no
colour shift as you rotated them but the advantage would be that any
shift due to absorption should be constant at any setting.
>
>2) Calibration with my readings
>I acquired images of the Gretag chart patches individually and I
>calculated the average (RGB) color for each patch. By comparing these
>colors with the real color values I see that the colors that I measure
>are much more vivid than the real ones. What I am wondering is whether
>these color differences might be due more to the fact that I see the
>structure of the patches rather than to the instrumental setup
>characteristics.
I don't know what your patches are, in detail - but if they have been
covered in layers of faded varnish, are a couple of hundred years old,
need cleaning and were no very bright to begin with this may not be
surprising. The Macbeth chart has pretty high chroma levels.
If I were doing this I would want to compare with some modern paint
samples as a control.
>
>Still, I wanted to try how the calibration works under these
>conditions.
>I have a Matlab routine that can perform a sequence of a linear and a
>non-linear (third-order polynome) color calibration. The routine seems
>to perform quite well, by looking at the acquired chart after color
>correction. Corrected colors are much more similar to the real ones,
>maybe a bit grayish; it improves by excluding from the calibration the
>patches of lighter colors, that tend to saturate.
>
>But when I apply the obtained calibration to the sample images, the
>resulting image has a very dull appearance. By looking at the
>histograms of the images before and after calibration, it's clear that
>there is a loss of contrast, as the histograms become narrower.
>
>I might somewhat expect this result, as the calibration is obtained by
>comparing an image of the calibration chart with very vivid and bright
>colors (that is what I acquire) with one with darker colors (the real
>ones). My problem is that I have no idea of how a correctly
>color-calibrated image should look like, therefore how much my results
>are close (or far) from the correct result. Unfortunately nothing of
>what I have read do any mention of the problems I have found.
>I am wondering whether the source of the problem resides in the
>calibration routine or in the type of color standards that I am using.
>
>Can anyone help me having a better understanding? Should I maybe try a
>different type of color standard (what about the IT8 charts on
>photographic paper?)?
>
Sorry, can't help with the software or the calibration standards, not
really my area. I do suggest though that this is the kind of thing that
forensics labs all around the world do routinely, as I imagine do art
conservation departments in major galleries.
BTW, I still think a spectrophotometer would give a much more rigorous
definition of spectral properties, if you could find one which operated
at small enough dimensions. I'm sure someone must have made a microscope
spectrophotometer at some point (and if so, I bet it's Zeiss).
Sorry this could not answer many of your questions, but I hope it might
stimulate some ideas anyway.
Regards,
David
-- David Littlewood
- Next message: Kevin Cunningham: "Re: microscopes for the faculty"
- Previous message: David Littlewood: "Re: color calibration of microscopic images"
- In reply to: Beatrice: "color calibration of microscopic images"
- Next in thread: Aaron: "Re: color calibration of microscopic images"
- Messages sorted by: [ date ] [ thread ]
Relevant Pages
|
|
