Re: color calibration of microscopic images

From: David Littlewood (david_at_nospam.demon.co.uk)
Date: 09/13/04 

Date: Mon, 13 Sep 2004 18:21:06 +0100

In article <56fab26.0409130636.3b579c48@posting.google.com>, Beatrice
<pappagallo16@libero.it> writes
>David Littlewood <david@nospam.demon.co.uk> wrote in message
>news:<nZ9TYhI+nfQBFwqs@dlittlewood.demon.co.uk>...
>> >
>> >> Do you find a colour shift as you change the exposure time or ISO
>> >> setting on the camera?
>> >
>> >Besides a change in brightness? I would not know for sure. But I can
>> >tell you that I have acquired images of the grey patches of the Gretag
>> >Macbeth chart, at different exposure times, and I see a nonlinear
>> >effect (even before the histogram becomes saturated).
>>
>> If you mean non-linear changes in luminance, why would this be a
>> problem? I did not think you were trying to get an absolute value for
>> luminance. I agree, if there is non-linear effect on the chrominance
>> this would be a problem.
>
>One of the things I would like to do is to be able to compare images
>acquired at different exposure times. At the beginning I thought that
>I could, in some way, simply 'rescale' the intesity values of these
>images so as if I had acquired them at a single exposure time.
>
>So I measured the color of the Gretag Macbeth grey patches at
>different exposure times to determing the scaling factors.
>But then it turned out that:
>1 - the color of the same grey patch at different exporure times
>acquisitions does not increase linearly.
>2 - the colors of different grey patches at the same exposure time do
>not increase linearly.
>These 'crossed' non-linearities are a problem for me as they make (at
>least for me) this rescaling more complicated, and I would not know
>any more how to do it.
>
>(BTW, then I thought that color calibration would have also solved
>this problem, besides having more accurate color reproduction).
>

I guess this is a crossover problem with your sensor - film often shows
the same problem.
>
>>
>> Birefringence in the specimen would not be a problem here*; you would
>> only using the two polarisers as a light attenuator, and the sample is
>> not between them. The retardation of a birefringent sample will only
>> show up as a colour effect if it is between the two polarisers. The
>> advantage of using crossed polars as an attenuator, compared with using
>> ND filters, is that any residual colour from absorption in the
>> polarisers should be constant for sample and reference (Macbeth chart or
>> whatever) whereas adding ND filters to just one of them risks adding a
>> colour cast.
>>
>> *You might get some anomalies if your specimen was pleochroic.
>> Pleochroism is quite different from birefringence, and less common - it
>> is having a different absorption spectrum along different axes of the
>> sample crystals. However, if your specimens do show pleochroism, this
>> would be a whole new ball game - and something that showed it up could
>> even be useful.
>
>All right, but with polarizers I would cut away diffused light, which
>I think is an important component of the color. Think for example at
>the difference between matt and glossy materials.
>It is also true that by working in dark field I get rid of specular
>reflections, so same problem. But since paint in cross-sections is
>more matt than glossy, I think this would be the preferred solution.
>What do you think?
>
If you are worried that the light from the specimens is preferentially
polarised, would it help to put a quarter-wave plate between the
specimen and the attenuator? This is a standard technique for microscope
designers (and for those using cameras) where polarisation may cause
anomalous results.

>
>> OK, I understand. However, a spectrophotometer would at least allow you
>> to calibrate the system to show that your reference samples were
>> consistent and your attenuation method was not causing colour shifts.
>
>Good point.
>I can add that since I am going to make my own reference samples, I
>was thinking to measure their color by using my spectrophotometer and
>then calculate RGB coordinates from the spectra. In this case I would
>not have any problems of calibration, since for our system I already
>have a calibration procedure by comparison with the spectral lines of
>a gas discharge lamp.
>The spectrophotometer in my lab consists of an Imspector spectrograph
>connected to the microscope on one side and to a digital camera to the
>other side.
>The only thing is that I still have to see how to perform this step
>exactly. I think that I would have to take into account the
>tristimulus and the characteristics of the light source, don't know
>yet exactly how and how easily.
>

It sounds interesting. I have reached - or even passed - the limits of
any knowledge I have here though!
>_________________________________________________
>
>Aaron <nghy@comcast.net> wrote in message
>news:<0ik6k0pnqgkscg9mj27ej0gmnufk802lc6@4ax.com>...
>> Hello Beatrice,
>>
>
>1)
>> Now that I have refelcted a bit on your comments, I favor a
>> tristimulus colorimeter which will read out coordinates in the "color
>> solid" that represents human color perception. These numerical
>> readings plus your own recorded comments will provide the body of
>> knowledge that you desire.
>>
>
>2)
>> The instrumental measurements will make sense over time and will
>> perform another function for which the eye/brain is inept, i.e.,
>> remebering colors over time.
>>
>
>3)
>> Konic-Minotla makes a non-contact tristimulus coloimeter Model
>> CS-100A which seems to have many characteristics that would be useful
>> in your application. I think an instrument of this type could be
>> adapted to the output of a microscope and could be used hand held near
>> the surfaces of paintings.
>>
>> See.
>>
>> http://konicaminolta.com.hk/ph/eng/pdf/catalogue/cs_100a_e.pdf
>>
>> Good Luck
>> Aaron
>
>So, if I understand you, you prefer a tristimulus colorimeter because:
>1 - I would have directly color coordinates in a CIELab or similar
>space.
>2 - since the colorimeter has his own illumination source (if I
>understand correctly), I would not have to rely on the microscope
>light source whose characteristics are not well known.
>
>About the first point, couldn't I just convert my RGB readings to
>CIELab values (which is what I was planning to do after calibration)
>without necessarily using a colorimeter?
>About the second point, well, this is a good point.
>What is not clear to me is whether such colorimeters make
>spatially-resolved measurements. If not, then I can't use them.
>Please correct me if I am wrong.
>
>
>Beatrice 

-- 
David Littlewood