help_standardizing images

[Ryan Garrick]

Dear All,

I have a question about whether I can perform image standardization in
ImageJ. I have a series of digital photographs (of an insect), and want to
score some morphological traits. As a first step, I want to use the
graycard I have included in each photo to standardize all images, relative
to a single 'reference' image (i.e. just one of the photos, the choice of
which one is to be used as the reference is arbitrary, because I am only
interested in relative measures of phenotypic variation, not absolute
values).

The spirit of what I want to do is something like this:
(1) select the graycard in the reference image, then select the graycard
in the target image;
(2) use ImageJ to calculate what adjustments need to be made to make the
ROI in the target image match the ROI in the reference image;
(3) apply the calculated adjustment to the entire target image;
(4) repeat the procedure for all remaining target images

I hope to end up with all of the images having been standardized, so that
measurements of phenotypic characters that make us of setting thresholds
to objectively delineate edges are not impacted by slight differences in
illumination of the original digital photographs.

Is this possible with ImageJ, and if not, can you direct me towards
software that will perform this sort of task?

Thanks for your help.

cheers,
Ryan

email: [hidden email]

[Frederick Ross]

Certainly it's possible.  Since you only have a few images, doing it
by hand won't be that onerous.  I assume your card's histogram is just
going to be scaled and offset, not distorted.  If it is, then you have
to do something more sophisticated (search on 'histogram
equalization'...it's not hard, it's just not as simple).

First, open all your images.  In each image, draw an ROI on each
reference card, and take the histogram of it (Analyze->Histogram).
Find the upper and lower limits (or any two distinctive features) of
the histogram for each image and note them down (note that the
interface tells you exactly where your mouse is, so just point to the
ends).

Then choose the reference, and figure out what offset and scaling
factor you need to apply to make the lower and upper ends of the
histogram line up.  Then go through and scale the image
(Process->Maths->Multiply) and add the offset (Process->Maths->Add).
The images will all be standardized.

I would use the narrowest histogram as the reference image.  When you
expand the greyscale range, you end up with gaps in the histogram
which can be a pain if you're trying to compare greyscales ("these two
are in bins separated by one, but the neighboring bins are both
zero...is it an artifact?").  You lose precision by contracting, but
it doesn't put in holes and makes interpretation very simple.  Though
if your measurements are entirely insensitive to having holes in the
histogram, then this obviously doesn't apply.

Hope this helps.

On 4/23/07, Ryan Garrick <[hidden email]> wrote:

> Dear All,
>
> I have a question about whether I can perform image standardization in
> ImageJ. I have a series of digital photographs (of an insect), and want to
> score some morphological traits. As a first step, I want to use the
> graycard I have included in each photo to standardize all images, relative
> to a single 'reference' image (i.e. just one of the photos, the choice of
> which one is to be used as the reference is arbitrary, because I am only
> interested in relative measures of phenotypic variation, not absolute
> values).
>
> The spirit of what I want to do is something like this:
> (1) select the graycard in the reference image, then select the graycard
> in the target image;
> (2) use ImageJ to calculate what adjustments need to be made to make the
> ROI in the target image match the ROI in the reference image;
> (3) apply the calculated adjustment to the entire target image;
> (4) repeat the procedure for all remaining target images
>
> I hope to end up with all of the images having been standardized, so that
> measurements of phenotypic characters that make us of setting thresholds
> to objectively delineate edges are not impacted by slight differences in
> illumination of the original digital photographs.
>
> Is this possible with ImageJ, and if not, can you direct me towards
> software that will perform this sort of task?
>
> Thanks for your help.
>
> cheers,
> Ryan
>
> email: [hidden email]
>

--
Frederick Ross
Graduate Fellow, (|Siggia> + |McKinney>)/sqrt(2) Lab
The Rockefeller University
Je ne suis pas Fred Cross!

[Michael Doube-2]

Hi Frederick,

This explanation has been very helpful; I do the same when standardising
grey levels in backscattered electron micrographs.

A macro to do it goes like this:

macro "Convert Kontron IMG File [b]" {

Dialog.create("Enter Values");
Dialog.addNumber("Br-PMMA", br);
Dialog.addNumber("I-PMMA", io);
Dialog.addNumber("Width", width);
Dialog.addNumber("Height", height);
Dialog.show();
br = Dialog.getNumber();
io = Dialog.getNumber();
width = Dialog.getNumber();
height = Dialog.getNumber();
setBatchMode(true);
run("Raw...", "image=8-bit width="+width+" height="+height+" offset=128
number=1 gap=0");
name = getTitle();
name = split(name, ".");
rename(name[0]+".tif");
run("Subtract...", "value="+br);
run("Multiply...", "value="+255/(io-br));
path = ""+getDirectory("image")+getTitle+"";
saveAs("Tiff", path);
close();
}

Mike

Frederick Ross wrote:

> Certainly it's possible.  Since you only have a few images, doing it
> by hand won't be that onerous.  I assume your card's histogram is just
> going to be scaled and offset, not distorted.  If it is, then you have
> to do something more sophisticated (search on 'histogram
> equalization'...it's not hard, it's just not as simple).
>
> First, open all your images.  In each image, draw an ROI on each
> reference card, and take the histogram of it (Analyze->Histogram).
> Find the upper and lower limits (or any two distinctive features) of
> the histogram for each image and note them down (note that the
> interface tells you exactly where your mouse is, so just point to the
> ends).
>
> Then choose the reference, and figure out what offset and scaling
> factor you need to apply to make the lower and upper ends of the
> histogram line up.  Then go through and scale the image
> (Process->Maths->Multiply) and add the offset (Process->Maths->Add).
> The images will all be standardized.
>
> I would use the narrowest histogram as the reference image.  When you
> expand the greyscale range, you end up with gaps in the histogram
> which can be a pain if you're trying to compare greyscales ("these two
> are in bins separated by one, but the neighboring bins are both
> zero...is it an artifact?").  You lose precision by contracting, but
> it doesn't put in holes and makes interpretation very simple.  Though
> if your measurements are entirely insensitive to having holes in the
> histogram, then this obviously doesn't apply.
>
> Hope this helps.
>
> On 4/23/07, Ryan Garrick <[hidden email]> wrote:
>> Dear All,
>>
>> I have a question about whether I can perform image standardization in
>> ImageJ. I have a series of digital photographs (of an insect), and
>> want to
>> score some morphological traits. As a first step, I want to use the
>> graycard I have included in each photo to standardize all images,
>> relative
>> to a single 'reference' image (i.e. just one of the photos, the
>> choice of
>> which one is to be used as the reference is arbitrary, because I am only
>> interested in relative measures of phenotypic variation, not absolute
>> values).
>>
>> The spirit of what I want to do is something like this:
>> (1) select the graycard in the reference image, then select the graycard
>> in the target image;
>> (2) use ImageJ to calculate what adjustments need to be made to make the
>> ROI in the target image match the ROI in the reference image;
>> (3) apply the calculated adjustment to the entire target image;
>> (4) repeat the procedure for all remaining target images
>>
>> I hope to end up with all of the images having been standardized, so
>> that
>> measurements of phenotypic characters that make us of setting thresholds
>> to objectively delineate edges are not impacted by slight differences in
>> illumination of the original digital photographs.
>>
>> Is this possible with ImageJ, and if not, can you direct me towards
>> software that will perform this sort of task?
>>
>> Thanks for your help.
>>
>> cheers,
>> Ryan
>>
>> email: [hidden email]
>>
>
>

--
Michael Doube  BPhil BVSc MRCVS
PhD Student
Dental Institute
Queen Mary, University of London
New Rd
London  E1 1BB
United Kingdom

Phone [phone] ext 2681

[Gabriel Landini]

 Frederick Ross wrote:
> by hand won't be that onerous.  I assume your card's histogram is just
> going to be scaled and offset, not distorted.  If it is, then you have
> to do something more sophisticated (search on 'histogram
> equalization'...it's not hard, it's just not as simple).

I think that you are describing more histogram stretching rather than
equalisation.

Shouldn't one sample all the greys in between too and then fit a function that
is applied to the image?

Assuming that the sensor is linear may not be a very valid assumption.

Cheers.

G.

[Ryan Garrick]

Thanks for your help with this everyone.
 
Frederick, it looks like the procedure you suggested has worked quite
well. I haven't yet played around with equalization (or stretching, as
mentioned by Gabriel) that much, but might need to because although the
graycard in my reference image has a histogram of gray values with an
approx. normal distribution, the graycards in some of  target images give
a histogram with a slightly longer upper tail (c.f. a normal
distribution). If I use the min. and max. gray values alone to determine
scaling and offset, this tail is likely to have some impact on
standardization. Perhaps it would be reasonable to use the min. and modal
gray value of the histograms, the rationale being that it is more
important end up with comparable frequency distributions of gray values
rather than matching range limits?

cheers,
Ryan

[Gabriel Landini]

On Wednesday 25 April 2007 02:59:09 Ryan Garrick wrote:
> I haven't yet played around with equalization (or stretching, as
> mentioned by Gabriel) that much, but might need to because although the
> graycard in my reference image has a histogram of gray values with an
> approx. normal distribution, the graycards in some of  target images give
> a histogram with a slightly longer upper tail (c.f. a normal
> distribution).

Is your test card made of a number of squares?
If so, then wouldn't you expect to get a histogram with equally high (but
narrow) peaks for the greyscale values of each tablet samples?

Can you post an image of your imaged test card?

Cheers,

G.

[Frederick Ross]

As Gabriel said, please post a picture of your card.

On 4/24/07, Ryan Garrick <[hidden email]> wrote:

> Frederick, it looks like the procedure you suggested has worked quite
> well. I haven't yet played around with equalization (or stretching, as
> mentioned by Gabriel) that much, but might need to because although the
> graycard in my reference image has a histogram of gray values with an
> approx. normal distribution, the graycards in some of  target images give
> a histogram with a slightly longer upper tail (c.f. a normal
> distribution). If I use the min. and max. gray values alone to determine
> scaling and offset, this tail is likely to have some impact on
> standardization. Perhaps it would be reasonable to use the min. and modal
> gray value of the histograms, the rationale being that it is more
> important end up with comparable frequency distributions of gray values
> rather than matching range limits?

If you get distortions of the original Gaussian, you need to NOT use
the procedure I recommended, as it will introduce systematic errors
into your numbers.  It means that your sensor is not linear (as
Gabriel had mentioned before).  The best thing is to keep your sensor
in a range where it's linear, but I realize that's not always
possible.

This being so, life is rather harder.  Essentially, you've got to fit
a function to the histogram for each of two images, then make the
coordinate transformation that takes one to another.  For this reason
you'll want fairly manageable functions.  You might try G(x^\alpha)
where G is the Gaussian, and alpha is some power close to 1 (so you're
fitting three parameters: the mean, standard deviation, and alpha).
If your sensor isn't too far from linear, you may be able to get
satisfactory fits this way, and the transformation is fairly
straightforwards.

I personally think the easiest way is to call Octave from a plugin
(JAPOS at http://jopas.sourceforge.net/ lets you do this reasonably
well), and then run some curve fitting algorithm someone's already
written there.

Otherwise, normally for a Gaussian you are better off measuring the
mean and standard deviation.  These will be more robust than min and
max.

--
Frederick Ross
Graduate Fellow, (|Siggia> + |McKinney>)/sqrt(2) Lab
The Rockefeller University
Je ne suis pas Fred Cross!