A question about ImageJ's threshold values

[Avital Steinberg]

Hi,
When I choose any auto thresholding method in ImageJ, I get a histogram. I
can change the threshold values using a slide bar and I see percent values.
Are these percent cumulative frequencies for the histogram, or something
else?

I would like to write a macro which, when given a percent value from this
slide bar, would give me a threshold value. I tried doing this with the
following macro:

http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html

but the values it gave were different from what I expected.

Also, there must be something that I don't understand, because the
percentile auto thresholding method shows a value of 70.17% percent on the
slide bar. I thought it would give me 50%.

I'm confused and would appreciate your help.

Thanks,

Avital

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Michael Schmid]

Hi Avital,

yes, the small number below the histogram in the Threshold panel is the percentage of pixels within the histogram bounds selected by the threshold.
Here is a macro that sets the threshold such that roughly 80% of the pixels are included:


percentage = 80;
nBins = 256;
resetMinAndMax();
getHistogram(values, counts, nBins);
// find culmulative sum
nPixels = 0;
for (i = 0; i<counts.length; i++)
  nPixels += counts[i];
nBelowThreshold = nPixels * percentage / 100;
sum = 0;
for (i = 0; i<counts.length; i++) {
  sum = sum + counts[i];
  if (sum >= nBelowThreshold) {
    setThreshold(values[0], values[i]);
    print(values[0]+"-"+values[i]+": "+sum/nPixels*100+"%");
    i = 99999999;//break
  }
}

For 16-bit and floating-point images, there will be a small difference between the exact number of pixels actually selected and the number printed by the macro. That's because the macro uses only an 8-bit histogram (256 values). When using an 8-bit histogram, for 16-bit and 32-bit images it would be usually better to include one more histogram bin.

Michael
________________________________________________________________
Michael Schmid                    email: [hidden email]
Institut für Angewandte Physik, Technische Universität Wien
Wiedner Hauptstr. 8-10/E134, A 1040 Wien, Austria
Tel. [phone] or -13453, Fax [phone]
________________________________________________________________

On Jun 16, 2015, at 14:36, Avital Steinberg wrote:

> Hi,
> When I choose any auto thresholding method in ImageJ, I get a histogram. I
> can change the threshold values using a slide bar and I see percent values.
> Are these percent cumulative frequencies for the histogram, or something
> else?
>
> I would like to write a macro which, when given a percent value from this
> slide bar, would give me a threshold value. I tried doing this with the
> following macro:
>
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
>
> but the values it gave were different from what I expected.
>
> Also, there must be something that I don't understand, because the
> percentile auto thresholding method shows a value of 70.17% percent on the
> slide bar. I thought it would give me 50%.
>
> I'm confused and would appreciate your help.
>
> Thanks,
>
> Avital
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Avital Steinberg]

Thanks - I get the same answer when I use the macro you wrote and the macro
I found in this link:

http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html

But when I choose adjust, threshold and manually adjust the sliding bar, I
get a different result.

Why would this be?

Thanks,

Avital


On Tue, Jun 16, 2015 at 4:10 PM, Michael Schmid <[hidden email]>
wrote:

> Hi Avital,
>
> yes, the small number below the histogram in the Threshold panel is the
> percentage of pixels within the histogram bounds selected by the threshold.
> Here is a macro that sets the threshold such that roughly 80% of the
> pixels are included:
>
>
> percentage = 80;
> nBins = 256;
> resetMinAndMax();
> getHistogram(values, counts, nBins);
> // find culmulative sum
> nPixels = 0;
> for (i = 0; i<counts.length; i++)
>   nPixels += counts[i];
> nBelowThreshold = nPixels * percentage / 100;
> sum = 0;
> for (i = 0; i<counts.length; i++) {
>   sum = sum + counts[i];
>   if (sum >= nBelowThreshold) {
>     setThreshold(values[0], values[i]);
>     print(values[0]+"-"+values[i]+": "+sum/nPixels*100+"%");
>     i = 99999999;//break
>   }
> }
>
> For 16-bit and floating-point images, there will be a small difference
> between the exact number of pixels actually selected and the number printed
> by the macro. That's because the macro uses only an 8-bit histogram (256
> values). When using an 8-bit histogram, for 16-bit and 32-bit images it
> would be usually better to include one more histogram bin.
>
> Michael
> ________________________________________________________________
> Michael Schmid                    email: [hidden email]
> Institut für Angewandte Physik, Technische Universität Wien
> Wiedner Hauptstr. 8-10/E134, A 1040 Wien, Austria
> Tel. [phone] or -13453, Fax [phone]
> ________________________________________________________________
>
> On Jun 16, 2015, at 14:36, Avital Steinberg wrote:
>
> > Hi,
> > When I choose any auto thresholding method in ImageJ, I get a histogram.
> I
> > can change the threshold values using a slide bar and I see percent
> values.
> > Are these percent cumulative frequencies for the histogram, or something
> > else?
> >
> > I would like to write a macro which, when given a percent value from this
> > slide bar, would give me a threshold value. I tried doing this with the
> > following macro:
> >
> >
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
> >
> > but the values it gave were different from what I expected.
> >
> > Also, there must be something that I don't understand, because the
> > percentile auto thresholding method shows a value of 70.17% percent on
> the
> > slide bar. I thought it would give me 50%.
> >
> > I'm confused and would appreciate your help.
> >
> > Thanks,
> >
> > Avital
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Avital Steinberg]

Thanks, Michael - I tweaked your macro a little and now it works. I had to
subtract the background pixels.

I appreciate your help - it really made things clear,

Avital

On Tue, Jun 16, 2015 at 4:10 PM, Michael Schmid <[hidden email]>
wrote:

> Hi Avital,
>
> yes, the small number below the histogram in the Threshold panel is the
> percentage of pixels within the histogram bounds selected by the threshold.
> Here is a macro that sets the threshold such that roughly 80% of the
> pixels are included:
>
>
> percentage = 80;
> nBins = 256;
> resetMinAndMax();
> getHistogram(values, counts, nBins);
> // find culmulative sum
> nPixels = 0;
> for (i = 0; i<counts.length; i++)
>   nPixels += counts[i];
> nBelowThreshold = nPixels * percentage / 100;
> sum = 0;
> for (i = 0; i<counts.length; i++) {
>   sum = sum + counts[i];
>   if (sum >= nBelowThreshold) {
>     setThreshold(values[0], values[i]);
>     print(values[0]+"-"+values[i]+": "+sum/nPixels*100+"%");
>     i = 99999999;//break
>   }
> }
>
> For 16-bit and floating-point images, there will be a small difference
> between the exact number of pixels actually selected and the number printed
> by the macro. That's because the macro uses only an 8-bit histogram (256
> values). When using an 8-bit histogram, for 16-bit and 32-bit images it
> would be usually better to include one more histogram bin.
>
> Michael
> ________________________________________________________________
> Michael Schmid                    email: [hidden email]
> Institut für Angewandte Physik, Technische Universität Wien
> Wiedner Hauptstr. 8-10/E134, A 1040 Wien, Austria
> Tel. [phone] or -13453, Fax [phone]
> ________________________________________________________________
>
> On Jun 16, 2015, at 14:36, Avital Steinberg wrote:
>
> > Hi,
> > When I choose any auto thresholding method in ImageJ, I get a histogram.
> I
> > can change the threshold values using a slide bar and I see percent
> values.
> > Are these percent cumulative frequencies for the histogram, or something
> > else?
> >
> > I would like to write a macro which, when given a percent value from this
> > slide bar, would give me a threshold value. I tried doing this with the
> > following macro:
> >
> >
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
> >
> > but the values it gave were different from what I expected.
> >
> > Also, there must be something that I don't understand, because the
> > percentile auto thresholding method shows a value of 70.17% percent on
> the
> > slide bar. I thought it would give me 50%.
> >
> > I'm confused and would appreciate your help.
> >
> > Thanks,
> >
> > Avital
> >
> > --
> > ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Michael Schmid]

Hi Avital and everyone,

there was a bug in the percentile display of the Threshold panel; the highest pixel value (255 for uncalibrated 8-bit images) was not included in the statistics if it was not inside the thresholded range.
Wayne has fixed this in the latest daily build, 1.49v4.

Michael
________________________________________________________________

On Jun 16, 2015, at 14:36, Avital Steinberg wrote:

> Hi,
> When I choose any auto thresholding method in ImageJ, I get a histogram. I
> can change the threshold values using a slide bar and I see percent values.
> Are these percent cumulative frequencies for the histogram, or something
> else?
>
> I would like to write a macro which, when given a percent value from this
> slide bar, would give me a threshold value. I tried doing this with the
> following macro:
>
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
>
> but the values it gave were different from what I expected.
>
> Also, there must be something that I don't understand, because the
> percentile auto thresholding method shows a value of 70.17% percent on the
> slide bar. I thought it would give me 50%.
>
> I'm confused and would appreciate your help.
>
> Thanks,
>
> Avital

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[Avital Steinberg]

Thank you Michael and Wayne!

On Wed, Jun 17, 2015 at 9:20 PM, Michael Schmid <[hidden email]>
wrote:

> Hi Avital and everyone,
>
> there was a bug in the percentile display of the Threshold panel; the
> highest pixel value (255 for uncalibrated 8-bit images) was not included in
> the statistics if it was not inside the thresholded range.
> Wayne has fixed this in the latest daily build, 1.49v4.
>
> Michael
> ________________________________________________________________
>
> On Jun 16, 2015, at 14:36, Avital Steinberg wrote:
>
> > Hi,
> > When I choose any auto thresholding method in ImageJ, I get a histogram.
> I
> > can change the threshold values using a slide bar and I see percent
> values.
> > Are these percent cumulative frequencies for the histogram, or something
> > else?
> >
> > I would like to write a macro which, when given a percent value from this
> > slide bar, would give me a threshold value. I tried doing this with the
> > following macro:
> >
> >
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
> >
> > but the values it gave were different from what I expected.
> >
> > Also, there must be something that I don't understand, because the
> > percentile auto thresholding method shows a value of 70.17% percent on
> the
> > slide bar. I thought it would give me 50%.
> >
> > I'm confused and would appreciate your help.
> >
> > Thanks,
> >
> > Avital
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html

[melkor2.0]

Hi,

I'm running through a similar problem here, I wonder how the macro was tweaked to make it run. I work with 12-bit depth images, and I am interested in thresholding and analyzing at a particular % of each image. The script seemed to work, but I noticed that the values are different as if I do it manually. I was wondering whether this might be related to the max value of the histogram (in the images with the greatest deviation the max value is around 2500 [saturated pixels are 4095], so when I manually do the 95% for that particular image the threshold value is around 1200 but the macro prints a value of 1800...

What am I missing?

origen = getDirectory("Images to process");
lista = getFileList(origen);
setBatchMode(true);
for (j=0; j<lista.length; j++) {
       showProgress(j+1, lista.length);
       open(origen+lista[j]);
       nombre = lista[j];
       run("Split Channels");
       id = getImageID();
       selectImage(id);
       run("Duplicate...", " ");
        tissueThreshPerc = 95;
                nBins = 4095;
                getHistogram(values, count, nBins);
                size = count.length;
                cumSum = getWidth() * getHeight();
                tissueValue = cumSum * tissueThreshPerc / 100;
                cumSumValues = count;
                for (i = 1; i<count.length; i++)
                        {
  cumSumValues[i] += cumSumValues[i-1];
                        }
                for (i = 1; i<cumSumValues.length; i++)
  if (cumSumValues[i-1] <= tissueValue && tissueValue <= cumSumValues[i])
                {
  setThreshold(i,4095);
  print(i);
                }
       run("Convert to Mask");
       run("Open");
       run("Close-");
       run("Create Selection");
       run("Copy");
       run("Analyze Particles...", "size=75-Infinity summarize");
       selectImage(id);
       run("Restore Selection");
       run("Measure");
  }

I just tried to change the cumSum loop by the total pixel size as suggested in the other thread:
http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html

But if I try to reset min and max right after i give the value of nBins, it doesn't work :(

Thanks for your patience and help!!

Ramon

[Avital Steinberg]

Hi Ramon,
The following code is what I used when I tweaked my macro, but I think that
the tweaking is not needed anymore. I didn't take into account the pixel
with the highest grayvalue, which is 255 for an 8-bit image. I was looking
for the threshold which covers 55% of the histogram. Notice that I didn't
set the threshold here at all, and this can be done only after finding the
optimal threshold.

With tweaking, this is the code for finding the threshold when I want to
threshold according to a specific percentile:

percentVal = 55;
nBins = 256;
resetMinAndMax();
getHistogram(values, counts, nBins);
// find culmulative sum
nPixels = 0;
percentage = newArray(counts.length);
for (i = 0; i<counts.length; i++){
  nPixels += counts[i];
  nForgroundPixels = nPixels - counts[255];
  }
for (i = 0; i<255; i++){
    sum = sum + counts[i];
    percentage[i] = sum*100/nForgroundPixels;
    if (percentage[i] > percentVal) {
        idealT = i;
        i = 999999;
    }
}
print("T is: " + idealT);

I hope this helps - good luck,

Avital



On Thu, Apr 7, 2016 at 4:40 PM, melkor2.0 <[hidden email]> wrote:

> Hi,
>
> I'm running through a similar problem here, I wonder how the macro was
> tweaked to make it run. I work with 12-bit depth images, and I am
> interested
> in thresholding and analyzing at a particular % of each image. The script
> seemed to work, but I noticed that the values are different as if I do it
> manually. I was wondering whether this might be related to the max value of
> the histogram (in the images with the greatest deviation the max value is
> around 2500 [saturated pixels are 4095], so when I manually do the 95% for
> that particular image the threshold value is around 1200 but the macro
> prints a value of 1800...
>
> What am I missing?
>
> origen = getDirectory("Images to process");
> lista = getFileList(origen);
> setBatchMode(true);
> for (j=0; j<lista.length; j++) {
>        showProgress(j+1, lista.length);
>        open(origen+lista[j]);
>        nombre = lista[j];
>        run("Split Channels");
>        id = getImageID();
>        selectImage(id);
>        run("Duplicate...", " ");
>         tissueThreshPerc = 95;
>                 nBins = 4095;
>                 getHistogram(values, count, nBins);
>                 size = count.length;
>                 cumSum = getWidth() * getHeight();
>                 tissueValue = cumSum * tissueThreshPerc / 100;
>                 cumSumValues = count;
>                 for (i = 1; i<count.length; i++)
>                         {
>                         cumSumValues[i] += cumSumValues[i-1];
>                         }
>                 for (i = 1; i<cumSumValues.length; i++)
>                         if (cumSumValues[i-1] <= tissueValue &&
> tissueValue <= cumSumValues[i])
>                 {
>                         setThreshold(i,4095);
>                         print(i);
>                 }
>        run("Convert to Mask");
>        run("Open");
>        run("Close-");
>        run("Create Selection");
>        run("Copy");
>        run("Analyze Particles...", "size=75-Infinity summarize");
>        selectImage(id);
>        run("Restore Selection");
>        run("Measure");
>   }
>
> I just tried to change the cumSum loop by the total pixel size as suggested
> in the other thread:
>
> http://imagej.1557.x6.nabble.com/Threshold-as-a-percentage-of-image-histogram-td3695671.html
>
> But if I try to reset min and max right after i give the value of nBins, it
> doesn't work :(
>
> Thanks for your patience and help!!
>
> Ramon
>
>
>
> --
> View this message in context:
> http://imagej.1557.x6.nabble.com/A-question-about-ImageJ-s-threshold-values-tp5013177p5016073.html
> Sent from the ImageJ mailing list archive at Nabble.com.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html