Background subtraction

Previous Topic Next Topic
 
classic Classic list List threaded Threaded
12 messages Options
Reply | Threaded
Open this post in threaded view
|

Background subtraction

Rubiu, Elena (ext)
Hi everybody,

I am trying to analyze the fluorescence intensity of some pictures.
I usually have three different pictures of the same object, that not only differ in fluorescence intensity but also in background noise (the less the intensity, the more the noise). To deal with the noise I'm using the "subtract background" function (rolling ball algorithm), then I'm selecting the area of interest with the threshold.

My idea was to use the same parameters for all three pics (same rolling ball radius, same threshold limit) but the problem is that if I subtract the background of a relatively non-noisy pic then the threshold selection is working in a different way than with a noisy picture where I subtracted the background.

Is there a way to "quantify" noise in a picture, so that I can decide if I should subtract the background or not?
Or should I just treat every picture in the same way, not caring if the non-noisy pic's data is completely affected by the subtraction?

I hope I could explain myself clearly and that you will be willing to help me.

Thank you for your time!

Kind regards,
Elena Rubiu


--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Background subtraction

karo03
Hi Elena Rubiu,

I think "subtract background" is not the tool do reduce noise. It is the choice to flatten uneven illumination or background. You should use it only if the background, after some smoothing, shows deviations AND these deviations are also influencing the object. In case of fluorescence you should better control your microscope to have an even exitation light!

For intensity estimation I think it would be sufficient to generate one mask by thresholding and to apply it to all three pictures of the same object, provided there is no shift, and measure with this mask the intensities.

If you are fixed to "subtract background" estimate the noise using one or more roi's in the background for measurement, really not recommended.

Regards
Karsten

Am 17.09.2014 um 10:01 schrieb Rubiu, Elena (ext) <[hidden email]>:

> Hi everybody,
>
> I am trying to analyze the fluorescence intensity of some pictures.
> I usually have three different pictures of the same object, that not only differ in fluorescence intensity but also in background noise (the less the intensity, the more the noise). To deal with the noise I'm using the "subtract background" function (rolling ball algorithm), then I'm selecting the area of interest with the threshold.
>
> My idea was to use the same parameters for all three pics (same rolling ball radius, same threshold limit) but the problem is that if I subtract the background of a relatively non-noisy pic then the threshold selection is working in a different way than with a noisy picture where I subtracted the background.
>
> Is there a way to "quantify" noise in a picture, so that I can decide if I should subtract the background or not?
> Or should I just treat every picture in the same way, not caring if the non-noisy pic's data is completely affected by the subtraction?
>
> I hope I could explain myself clearly and that you will be willing to help me.
>
> Thank you for your time!
>
> Kind regards,
> Elena Rubiu
>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

AW: Background subtraction

Rubiu, Elena (ext)
Hi Karsten,

thank you for your answer. I guess then I should just apply the threshold - but what do you mean with "provided there is no shift"?

Kind regards
Elena

-----Ursprüngliche Nachricht-----
Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Karsten
Gesendet: Mittwoch, 17. September 2014 10:38
An: [hidden email]
Betreff: Re: Background subtraction

Hi Elena Rubiu,

I think "subtract background" is not the tool do reduce noise. It is the choice to flatten uneven illumination or background. You should use it only if the background, after some smoothing, shows deviations AND these deviations are also influencing the object. In case of fluorescence you should better control your microscope to have an even exitation light!

For intensity estimation I think it would be sufficient to generate one mask by thresholding and to apply it to all three pictures of the same object, provided there is no shift, and measure with this mask the intensities.

If you are fixed to "subtract background" estimate the noise using one or more roi's in the background for measurement, really not recommended.

Regards
Karsten

Am 17.09.2014 um 10:01 schrieb Rubiu, Elena (ext) <[hidden email]>:

> Hi everybody,
>
> I am trying to analyze the fluorescence intensity of some pictures.
> I usually have three different pictures of the same object, that not only differ in fluorescence intensity but also in background noise (the less the intensity, the more the noise). To deal with the noise I'm using the "subtract background" function (rolling ball algorithm), then I'm selecting the area of interest with the threshold.
>
> My idea was to use the same parameters for all three pics (same rolling ball radius, same threshold limit) but the problem is that if I subtract the background of a relatively non-noisy pic then the threshold selection is working in a different way than with a noisy picture where I subtracted the background.
>
> Is there a way to "quantify" noise in a picture, so that I can decide if I should subtract the background or not?
> Or should I just treat every picture in the same way, not caring if the non-noisy pic's data is completely affected by the subtraction?
>
> I hope I could explain myself clearly and that you will be willing to help me.
>
> Thank you for your time!
>
> Kind regards,
> Elena Rubiu
>
>
> --
> 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
Reply | Threaded
Open this post in threaded view
|

Polygon surrounding multiple point selections

Richard Mort-2
Hello list,

Before I go through the effort of trying to solve this problem I thought
I'd see if somene else has already done the same. I have a spatial
pattern composed of lots of xy positions (see attached plot). I can find
the bounding box for these positions using the code pasted at the end.
How can I find a polygon that joins all the points at the edge of the
pattern?

Best
R

//Find bounding box of multiple xy positions


run("Analyze Particles...", "display");

//some variables
max_x = 0;
max_y = 0;
min_x = 0;
min_y = 0;

//Max value in "XM" column
for (a=0; a<nResults(); a++) {
     if (getResult("XM",a)>max_x)
     {
      max_x = getResult("XM",a);
         }
         else{};
}

//Min value in "XM" column (note: requires max value)
min_x=max_x;
for (b=0; b<nResults(); b++) {
     if (getResult("XM",b)<min_x)
     {
      min_x = getResult("XM",b);
         }
         else{};
}


//Max value in "YM" column
for (c=0; c<nResults(); c++) {
     if (getResult("YM",c)>max_y)
     {
      max_y = getResult("YM",c);
         }
         else{};
}

//Min value in "YM" column (note: requires max value)
min_y=max_y;
for (d=0; d<nResults(); d++) {
     if (getResult("YM",d)<min_y)
     {
      min_y = getResult("YM",d);
         }
         else{};
}

print(min_x);
print(max_x);
print(min_y);
print(max_y);

makeRectangle(min_x, min_y, max_x-min_x, max_y-min_y);

//end

--
Dr Richard Mort
MRC Human Genetics Unit
MRC IGMM
University of Edinburgh
Western General Hospital
Crewe Road
Edinburgh.
EH4 2XU, UK

Tel: +44 (0)131 332 2471
Fax: +44 (0)131 467 8456


The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.


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

Pattern.tif (352K) Download Attachment
Reply | Threaded
Open this post in threaded view
|

Re: Background subtraction

karo03
In reply to this post by Rubiu, Elena (ext)
Ja, select the threshold with the best image of the three and apply it to all three, provided the three images don't show any local displacements (shifts) of the object.

Karsten

Am 17.09.2014 um 11:11 schrieb Rubiu, Elena (ext) <[hidden email]>:

> Hi Karsten,
>
> thank you for your answer. I guess then I should just apply the threshold - but what do you mean with "provided there is no shift"?
>
> Kind regards
> Elena
>
> -----Ursprüngliche Nachricht-----
> Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Karsten
> Gesendet: Mittwoch, 17. September 2014 10:38
> An: [hidden email]
> Betreff: Re: Background subtraction
>
> Hi Elena Rubiu,
>
> I think "subtract background" is not the tool do reduce noise. It is the choice to flatten uneven illumination or background. You should use it only if the background, after some smoothing, shows deviations AND these deviations are also influencing the object. In case of fluorescence you should better control your microscope to have an even exitation light!
>
> For intensity estimation I think it would be sufficient to generate one mask by thresholding and to apply it to all three pictures of the same object, provided there is no shift, and measure with this mask the intensities.
>
> If you are fixed to "subtract background" estimate the noise using one or more roi's in the background for measurement, really not recommended.
>
> Regards
> Karsten
>
> Am 17.09.2014 um 10:01 schrieb Rubiu, Elena (ext) <[hidden email]>:
>
>> Hi everybody,
>>
>> I am trying to analyze the fluorescence intensity of some pictures.
>> I usually have three different pictures of the same object, that not only differ in fluorescence intensity but also in background noise (the less the intensity, the more the noise). To deal with the noise I'm using the "subtract background" function (rolling ball algorithm), then I'm selecting the area of interest with the threshold.
>>
>> My idea was to use the same parameters for all three pics (same rolling ball radius, same threshold limit) but the problem is that if I subtract the background of a relatively non-noisy pic then the threshold selection is working in a different way than with a noisy picture where I subtracted the background.
>>
>> Is there a way to "quantify" noise in a picture, so that I can decide if I should subtract the background or not?
>> Or should I just treat every picture in the same way, not caring if the non-noisy pic's data is completely affected by the subtraction?
>>
>> I hope I could explain myself clearly and that you will be willing to help me.
>>
>> Thank you for your time!
>>
>> Kind regards,
>> Elena Rubiu
>>
>>
>> --
>> 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

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Background subtraction

Csúcs  Gábor
In reply to this post by Rubiu, Elena (ext)
Dear Elena,

I fully agree with the previous comments of Karsten: you should use
background substraction not to reduce the noise but correct for uneven
illumination. However, if you want to do quantification (measure intensity)
I would not use the rolling ball algorithm. There are more appropriate
methods (like recording a fluorescent calibration image) that could be
used and with those you could also avoid your current problems. If you
look for papers you may search for "flat-field correction", "vignetting
correction" or "shading correction".

Greetings     Gabor



>>On 9/17/14 10:01 AM, "Rubiu, Elena (ext)" <[hidden email]>
>>wrote:
>>
>>>Hi everybody,
>>>
>>>I am trying to analyze the fluorescence intensity of some pictures.
>>>I usually have three different pictures of the same object, that not
>>>only
>>>differ in fluorescence intensity but also in background noise (the less
>>>the intensity, the more the noise). To deal with the noise I'm using the
>>>"subtract background" function (rolling ball algorithm), then I'm
>>>selecting the area of interest with the threshold.
>>>
>>>My idea was to use the same parameters for all three pics (same rolling
>>>ball radius, same threshold limit) but the problem is that if I subtract
>>>the background of a relatively non-noisy pic then the threshold
>>>selection
>>>is working in a different way than with a noisy picture where I
>>>subtracted the background.
>>>
>>>Is there a way to "quantify" noise in a picture, so that I can decide if
>>>I should subtract the background or not?
>>>Or should I just treat every picture in the same way, not caring if the
>>>non-noisy pic's data is completely affected by the subtraction?
>>>
>>>I hope I could explain myself clearly and that you will be willing to
>>>help me.
>>>
>>>Thank you for your time!
>>>
>>>Kind regards,
>>>Elena Rubiu
>>>
>>>
>>>--
>>>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

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Polygon surrounding multiple point selections

Michael Schmid
In reply to this post by Richard Mort-2
Hi Richard,

to get the bounding polygon of a selection, use Edit>Selection>Convex Hull.
If you don't have a point selection but an image with single points to start with, use 'Find Maxima' with Output Type=Point Selection.

In a macro, you can retrieve the vertices of the polygon selection with
  getSelectionCoordinates(xpoints, ypoints)

Michael
________________________________________________________________
On Sep 17, 2014, at 12:08, Richard Mort wrote:

> Hello list,
>
> Before I go through the effort of trying to solve this problem I thought I'd see if somene else has already done the same. I have a spatial pattern composed of lots of xy positions (see attached plot). I can find the bounding box for these positions using the code pasted at the end. How can I find a polygon that joins all the points at the edge of the pattern?
>
> Best
> R
>
> //Find bounding box of multiple xy positions
>
>
> run("Analyze Particles...", "display");
>
> //some variables
> max_x = 0;
> max_y = 0;
> min_x = 0;
> min_y = 0;
>
> //Max value in "XM" column
> for (a=0; a<nResults(); a++) {
>    if (getResult("XM",a)>max_x)
>    {
>     max_x = getResult("XM",a);
>        }
>        else{};
> }
>
> //Min value in "XM" column (note: requires max value)
> min_x=max_x;
> for (b=0; b<nResults(); b++) {
>    if (getResult("XM",b)<min_x)
>    {
>     min_x = getResult("XM",b);
>        }
>        else{};
> }
>
>
> //Max value in "YM" column
> for (c=0; c<nResults(); c++) {
>    if (getResult("YM",c)>max_y)
>    {
>     max_y = getResult("YM",c);
>        }
>        else{};
> }
>
> //Min value in "YM" column (note: requires max value)
> min_y=max_y;
> for (d=0; d<nResults(); d++) {
>    if (getResult("YM",d)<min_y)
>    {
>     min_y = getResult("YM",d);
>        }
>        else{};
> }
>
> print(min_x);
> print(max_x);
> print(min_y);
> print(max_y);
>
> makeRectangle(min_x, min_y, max_x-min_x, max_y-min_y);
>
> //end
>
> --
> Dr Richard Mort
> MRC Human Genetics Unit
> MRC IGMM
> University of Edinburgh
> Western General Hospital
> Crewe Road
> Edinburgh.
> EH4 2XU, UK
>
> Tel: +44 (0)131 332 2471
> Fax: +44 (0)131 467 8456
>
>
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
> <Pattern.tif>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Polygon surrounding multiple point selections

Gabriel Landini
In reply to this post by Richard Mort-2
Hi Richard,
In addition to Michael's answer, if the points via the maxima are not enough
(suppose that they you are using regions instead of points), the plugin
Convex_Hull_Plus:
http://www.mecourse.com/landinig/software/software.html
gives you the convex hull of the set of regions as a ROI.

Cheers

Gabriel

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

Re: Polygon surrounding multiple point selections

Paul van Schayck
In reply to this post by Richard Mort-2
Hey Richard,

Are you looking for the convex hull command?

http://imagejdocu.tudor.lu/doku.php?id=gui:edit:selection#convex_hull

Kind regards,

Paul

On Wed, Sep 17, 2014 at 12:08 PM, Richard Mort <[hidden email]>
wrote:

> Hello list,
>
> Before I go through the effort of trying to solve this problem I thought
> I'd see if somene else has already done the same. I have a spatial pattern
> composed of lots of xy positions (see attached plot). I can find the
> bounding box for these positions using the code pasted at the end. How can
> I find a polygon that joins all the points at the edge of the pattern?
>
> Best
> R
>
> //Find bounding box of multiple xy positions
>
>
> run("Analyze Particles...", "display");
>
> //some variables
> max_x = 0;
> max_y = 0;
> min_x = 0;
> min_y = 0;
>
> //Max value in "XM" column
> for (a=0; a<nResults(); a++) {
>     if (getResult("XM",a)>max_x)
>     {
>      max_x = getResult("XM",a);
>         }
>         else{};
> }
>
> //Min value in "XM" column (note: requires max value)
> min_x=max_x;
> for (b=0; b<nResults(); b++) {
>     if (getResult("XM",b)<min_x)
>     {
>      min_x = getResult("XM",b);
>         }
>         else{};
> }
>
>
> //Max value in "YM" column
> for (c=0; c<nResults(); c++) {
>     if (getResult("YM",c)>max_y)
>     {
>      max_y = getResult("YM",c);
>         }
>         else{};
> }
>
> //Min value in "YM" column (note: requires max value)
> min_y=max_y;
> for (d=0; d<nResults(); d++) {
>     if (getResult("YM",d)<min_y)
>     {
>      min_y = getResult("YM",d);
>         }
>         else{};
> }
>
> print(min_x);
> print(max_x);
> print(min_y);
> print(max_y);
>
> makeRectangle(min_x, min_y, max_x-min_x, max_y-min_y);
>
> //end
>
> --
> Dr Richard Mort
> MRC Human Genetics Unit
> MRC IGMM
> University of Edinburgh
> Western General Hospital
> Crewe Road
> Edinburgh.
> EH4 2XU, UK
>
> Tel: +44 (0)131 332 2471
> Fax: +44 (0)131 467 8456
>
>
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.
>
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

AW: Background subtraction

Rubiu, Elena (ext)
In reply to this post by Csúcs Gábor
Hi Gabor,
thank you for your answer.

I had a look for those methods but it seems that it would take me more time than what I should actually dedicate to this.

Unfortunately I started using Imagej very little ago, and I'm not very confident with the right terminology. I'm not really able to tell if my images are "noisy" or they just present an uneven illumination.

What I would like to obtain in the end is that using the same threshold for all images, I get only my "interesting area" highlighted and not also some parts of the background.

I read that maybe filtering the image and then subtracting it from the original one could help with the final thresholding?

Thank you for your help!
 

-----Ursprüngliche Nachricht-----
Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Csúcs Gábor
Gesendet: Mittwoch, 17. September 2014 12:20
An: [hidden email]
Betreff: Re: Background subtraction

Dear Elena,

I fully agree with the previous comments of Karsten: you should use background substraction not to reduce the noise but correct for uneven illumination. However, if you want to do quantification (measure intensity) I would not use the rolling ball algorithm. There are more appropriate methods (like recording a fluorescent calibration image) that could be used and with those you could also avoid your current problems. If you look for papers you may search for "flat-field correction", "vignetting correction" or "shading correction".

Greetings     Gabor



>>On 9/17/14 10:01 AM, "Rubiu, Elena (ext)"
>><[hidden email]>
>>wrote:
>>
>>>Hi everybody,
>>>
>>>I am trying to analyze the fluorescence intensity of some pictures.
>>>I usually have three different pictures of the same object, that not
>>>only differ in fluorescence intensity but also in background noise
>>>(the less the intensity, the more the noise). To deal with the noise
>>>I'm using the "subtract background" function (rolling ball
>>>algorithm), then I'm selecting the area of interest with the
>>>threshold.
>>>
>>>My idea was to use the same parameters for all three pics (same
>>>rolling ball radius, same threshold limit) but the problem is that if
>>>I subtract the background of a relatively non-noisy pic then the
>>>threshold selection is working in a different way than with a noisy
>>>picture where I subtracted the background.
>>>
>>>Is there a way to "quantify" noise in a picture, so that I can decide
>>>if I should subtract the background or not?
>>>Or should I just treat every picture in the same way, not caring if
>>>the non-noisy pic's data is completely affected by the subtraction?
>>>
>>>I hope I could explain myself clearly and that you will be willing to
>>>help me.
>>>
>>>Thank you for your time!
>>>
>>>Kind regards,
>>>Elena Rubiu
>>>
>>>
>>>--
>>>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

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

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

AW: Background subtraction

Rubiu, Elena (ext)
In reply to this post by Csúcs Gábor
Hi Karsten, Gabor and everybody,

I kept reading in various guides and I ended up with this method.

I have 3 pics:
The first one has usually a background with pixels with a gray value of 0 or 1.
The other two have a background with higher level pixels, around 10.
 
So if I apply the same threshold to all the pics, in the first one only the fluorescence is highlighted, while in the other 2 also some background is highlighted.
 
Therefore I shouldn't process all three pics cause, while in the last two it alters mainly the background, in the first one it alters my actual data.
 
So as a maybe valid process, I found this:
 
Duplicate my pic; in one copy apply a median filter.
 
Then, I subtract the filtered image from my original one, thus obtaining very similar fluorescence data from the original, but with a background that ranges from 1 to 3, I would say.
 
Now, if I apply the same threshold to all the three pics, only the fluorescence is highlighted.
 
What do you think? Here is an example (page 72, figure 9.5)  http://blogs.qub.ac.uk/ccbg/files/2013/06/Analyzing_fluorescence_microscopy_images.pdf

I hope this time it makes some sense and that you'll be still willing to help me.

Kind regards,
Elena


-----Ursprüngliche Nachricht-----
Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Csúcs Gábor
Gesendet: Mittwoch, 17. September 2014 12:20
An: [hidden email]
Betreff: Re: Background subtraction

Dear Elena,

I fully agree with the previous comments of Karsten: you should use background substraction not to reduce the noise but correct for uneven illumination. However, if you want to do quantification (measure intensity) I would not use the rolling ball algorithm. There are more appropriate methods (like recording a fluorescent calibration image) that could be used and with those you could also avoid your current problems. If you look for papers you may search for "flat-field correction", "vignetting correction" or "shading correction".

Greetings     Gabor



>>On 9/17/14 10:01 AM, "Rubiu, Elena (ext)"
>><[hidden email]>
>>wrote:
>>
>>>Hi everybody,
>>>
>>>I am trying to analyze the fluorescence intensity of some pictures.
>>>I usually have three different pictures of the same object, that not
>>>only differ in fluorescence intensity but also in background noise
>>>(the less the intensity, the more the noise). To deal with the noise
>>>I'm using the "subtract background" function (rolling ball
>>>algorithm), then I'm selecting the area of interest with the
>>>threshold.
>>>
>>>My idea was to use the same parameters for all three pics (same
>>>rolling ball radius, same threshold limit) but the problem is that if
>>>I subtract the background of a relatively non-noisy pic then the
>>>threshold selection is working in a different way than with a noisy
>>>picture where I subtracted the background.
>>>
>>>Is there a way to "quantify" noise in a picture, so that I can decide
>>>if I should subtract the background or not?
>>>Or should I just treat every picture in the same way, not caring if
>>>the non-noisy pic's data is completely affected by the subtraction?
>>>
>>>I hope I could explain myself clearly and that you will be willing to
>>>help me.
>>>
>>>Thank you for your time!
>>>
>>>Kind regards,
>>>Elena Rubiu
>>>
>>>
>>>--
>>>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

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

--
ImageJ mailing list: http://imagej.nih.gov/ij/list.html
Reply | Threaded
Open this post in threaded view
|

AW: Background subtraction

Rubiu, Elena (ext)
PS: thank you again!!

-----Ursprüngliche Nachricht-----
Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Rubiu, Elena (ext)
Gesendet: Donnerstag, 18. September 2014 14:45
An: [hidden email]
Betreff: AW: Background subtraction

Hi Karsten, Gabor and everybody,

I kept reading in various guides and I ended up with this method.

I have 3 pics:
The first one has usually a background with pixels with a gray value of 0 or 1.
The other two have a background with higher level pixels, around 10.
 
So if I apply the same threshold to all the pics, in the first one only the fluorescence is highlighted, while in the other 2 also some background is highlighted.
 
Therefore I shouldn't process all three pics cause, while in the last two it alters mainly the background, in the first one it alters my actual data.
 
So as a maybe valid process, I found this:
 
Duplicate my pic; in one copy apply a median filter.
 
Then, I subtract the filtered image from my original one, thus obtaining very similar fluorescence data from the original, but with a background that ranges from 1 to 3, I would say.
 
Now, if I apply the same threshold to all the three pics, only the fluorescence is highlighted.
 
What do you think? Here is an example (page 72, figure 9.5)  http://blogs.qub.ac.uk/ccbg/files/2013/06/Analyzing_fluorescence_microscopy_images.pdf

I hope this time it makes some sense and that you'll be still willing to help me.

Kind regards,
Elena


-----Ursprüngliche Nachricht-----
Von: ImageJ Interest Group [mailto:[hidden email]] Im Auftrag von Csúcs Gábor
Gesendet: Mittwoch, 17. September 2014 12:20
An: [hidden email]
Betreff: Re: Background subtraction

Dear Elena,

I fully agree with the previous comments of Karsten: you should use background substraction not to reduce the noise but correct for uneven illumination. However, if you want to do quantification (measure intensity) I would not use the rolling ball algorithm. There are more appropriate methods (like recording a fluorescent calibration image) that could be used and with those you could also avoid your current problems. If you look for papers you may search for "flat-field correction", "vignetting correction" or "shading correction".

Greetings     Gabor



>>On 9/17/14 10:01 AM, "Rubiu, Elena (ext)"
>><[hidden email]>
>>wrote:
>>
>>>Hi everybody,
>>>
>>>I am trying to analyze the fluorescence intensity of some pictures.
>>>I usually have three different pictures of the same object, that not
>>>only differ in fluorescence intensity but also in background noise
>>>(the less the intensity, the more the noise). To deal with the noise
>>>I'm using the "subtract background" function (rolling ball
>>>algorithm), then I'm selecting the area of interest with the
>>>threshold.
>>>
>>>My idea was to use the same parameters for all three pics (same
>>>rolling ball radius, same threshold limit) but the problem is that if
>>>I subtract the background of a relatively non-noisy pic then the
>>>threshold selection is working in a different way than with a noisy
>>>picture where I subtracted the background.
>>>
>>>Is there a way to "quantify" noise in a picture, so that I can decide
>>>if I should subtract the background or not?
>>>Or should I just treat every picture in the same way, not caring if
>>>the non-noisy pic's data is completely affected by the subtraction?
>>>
>>>I hope I could explain myself clearly and that you will be willing to
>>>help me.
>>>
>>>Thank you for your time!
>>>
>>>Kind regards,
>>>Elena Rubiu
>>>
>>>
>>>--
>>>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

--
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