SEM background correction - uneven Threshold

[Alvin Orbaek]

hi everyone,
I'm looking for help with correcting random and uneven backgrounds  
from Scanning Electron Microscope (SEM) images of single walled carbon  
nanotubes (SWNTs).

I have been acquiring data that have some anomaly or artifact on the  
substrate so I can come back to this place and see changes after I  
have completed subsequent growths of SWNTs under different conditions.  
However, these artifacts that I use as markers are also presenting  
problems when trying to Threshold the images.

The artifacts tend make the background subtraction rather difficult, i  
think this is physically due to a large difference in secondary  
electrons around from these locations. The difference is not  
noticeable when looking at the images, but become evident when trying  
to Threshold. The baseline emanates from the artifacts instead of from  
all across the substrate, so i do not get an even correction.

I have tried several techniques that i have found on this list serve,  
but still I am not able to normalize the background, and successfully  
Threshold the image.

Eventually the goal is to render the images and count all the SWNTs on  
the surface, which I can do manually. But on some images there are so  
many SWNTs that it has taken me 6 hours for just one image to count  
manually. I did this using the ROI manager, by marking all the SWNTs  
with lines, then applying a multi measure, and outputting the data.  
But this is extremely tedious, needless to say.

I would be really happy if someone could help me with flattening out  
the images so I can then apply a similar technique like particle  
counting to the images. Or if anyone has a better suggestion for  
counting particles with high aspect ratios, like these SWNTs, then I  
would also be glad to hear your suggestions.

I have placed an example image here:
http://www.aokengineer.com/swnt-example.html
Note: this is now a JPG image, the data i work from is a TIFF image. I  
had to convert it to JPG to load it to the website.


The eventual goal is to count all the SWNTs and then compare the  
images after continued growth, which I think i can do relatively easy.  
But so far the bottleneck in my work is getting the images to  
Threshold adequately.


thanks in advance for your help.

alvin


--
Alvin W. Orbaek, Ph.D. Student. [hidden email]
Professor Andrew R. Barron group

Rice University, Chemistry Department, MS-60
6100 Main Street, Houston, TX 77005
Phone: [phone], Fax: [phone]
Ext: x3456, Cell: 713-851-2653
--

[Michael Schmid]

Hi Alvin,

this looks difficult.

The white particle in the image (defect of substrate?) can't be fully  
removed by filtering because of its sharp edges. The best I could get  
was with the Fast Filters plugin (type:eliminate maxima, radius=3,  
preprocessing median, subtract filtered with offset e.g.=20)
   http://imagejdocu.tudor.lu/doku.php?
id=plugin:filter:fast_filters:start

Also, the images are too noisy for thresholding (this is partly due  
to the jpg artifacts, however; the original data should be better).  
So you should do milder background subtraction that is less sensitive  
to noise (Fast filters: larger radius, type:background from minima)
Unfortunately, I am not aware of any plugin that enhances linear  
features - one could write such a plugin based on the principle of  
the Hough transform, but it would be a hell of work...
The usual edge-preserving blur filters are rather poor for thin  
linear features.

Best wishes,

Michael
________________________________________________________________

On 22 Feb 2010, at 21:15, Alvin Orbaek wrote:

> hi everyone,
> I'm looking for help with correcting random and uneven backgrounds  
> from Scanning Electron Microscope (SEM) images of single walled  
> carbon nanotubes (SWNTs).
>
> I have been acquiring data that have some anomaly or artifact on  
> the substrate so I can come back to this place and see changes  
> after I have completed subsequent growths of SWNTs under different  
> conditions. However, these artifacts that I use as markers are also  
> presenting problems when trying to Threshold the images.
>
> The artifacts tend make the background subtraction rather  
> difficult, i think this is physically due to a large difference in  
> secondary electrons around from these locations. The difference is  
> not noticeable when looking at the images, but become evident when  
> trying to Threshold. The baseline emanates from the artifacts  
> instead of from all across the substrate, so i do not get an even  
> correction.
>
> I have tried several techniques that i have found on this list  
> serve, but still I am not able to normalize the background, and  
> successfully Threshold the image.
>
> Eventually the goal is to render the images and count all the SWNTs  
> on the surface, which I can do manually. But on some images there  
> are so many SWNTs that it has taken me 6 hours for just one image  
> to count manually. I did this using the ROI manager, by marking all  
> the SWNTs with lines, then applying a multi measure, and outputting  
> the data. But this is extremely tedious, needless to say.
>
> I would be really happy if someone could help me with flattening  
> out the images so I can then apply a similar technique like  
> particle counting to the images. Or if anyone has a better  
> suggestion for counting particles with high aspect ratios, like  
> these SWNTs, then I would also be glad to hear your suggestions.
>
> I have placed an example image here:
> http://www.aokengineer.com/swnt-example.html
> Note: this is now a JPG image, the data i work from is a TIFF  
> image. I had to convert it to JPG to load it to the website.
>
>
> The eventual goal is to count all the SWNTs and then compare the  
> images after continued growth, which I think i can do relatively  
> easy. But so far the bottleneck in my work is getting the images to  
> Threshold adequately.
>
>
> thanks in advance for your help.
>
> alvin
>
>
> --
> Alvin W. Orbaek, Ph.D. Student. [hidden email]
> Professor Andrew R. Barron group
>
> Rice University, Chemistry Department, MS-60
> 6100 Main Street, Houston, TX 77005
> Phone: [phone], Fax: [phone]
> Ext: x3456, Cell: 713-851-2653
> --

[Alvin Orbaek]

Hi Michael,
  thanks for your help, I should have mentioned that I also used to  
just eliminate the artifact by cutting it from the image, but still  
when I try threshold it makes no difference. The background forms in a  
circle around where the artifact is in the image, so I still do not  
create an even background.

Good point about the noise in the image, I had never thought of this.  
But a few people have commented on this fact to me, so I think I will  
try acquire data from now on using a slower scan speed. I have also  
started to use 30kV instead of 20kV. Along with using a spot size of 6  
instead of 3.

When I get the chance I may experiment with some of the other settings  
like, working distance and the aperture size.

  Unfortunately I was following the procedure from someone that used  
to count the images by hand so for them it wasn't an issue, even  
though it took them ages to get any statistics to work with.

thanks again for the input.

alvin

On Feb 23, 2010, at 3:46 AM, Michael Schmid wrote:

> Hi Alvin,
>
> this looks difficult.
>
> The white particle in the image (defect of substrate?) can't be  
> fully removed by filtering because of its sharp edges. The best I  
> could get was with the Fast Filters plugin (type:eliminate maxima,  
> radius=3, preprocessing median, subtract filtered with offset e.g.=20)
>  http://imagejdocu.tudor.lu/doku.php?id=plugin:filter:fast_filters:start
>
> Also, the images are too noisy for thresholding (this is partly due  
> to the jpg artifacts, however; the original data should be better).  
> So you should do milder background subtraction that is less  
> sensitive to noise (Fast filters: larger radius, type:background  
> from minima)
> Unfortunately, I am not aware of any plugin that enhances linear  
> features - one could write such a plugin based on the principle of  
> the Hough transform, but it would be a hell of work...
> The usual edge-preserving blur filters are rather poor for thin  
> linear features.
>
> Best wishes,
>
> Michael
> ________________________________________________________________
>
> On 22 Feb 2010, at 21:15, Alvin Orbaek wrote:
>
>> hi everyone,
>> I'm looking for help with correcting random and uneven backgrounds  
>> from Scanning Electron Microscope (SEM) images of single walled  
>> carbon nanotubes (SWNTs).
>>
>> I have been acquiring data that have some anomaly or artifact on  
>> the substrate so I can come back to this place and see changes  
>> after I have completed subsequent growths of SWNTs under different  
>> conditions. However, these artifacts that I use as markers are also  
>> presenting problems when trying to Threshold the images.
>>
>> The artifacts tend make the background subtraction rather  
>> difficult, i think this is physically due to a large difference in  
>> secondary electrons around from these locations. The difference is  
>> not noticeable when looking at the images, but become evident when  
>> trying to Threshold. The baseline emanates from the artifacts  
>> instead of from all across the substrate, so i do not get an even  
>> correction.
>>
>> I have tried several techniques that i have found on this list  
>> serve, but still I am not able to normalize the background, and  
>> successfully Threshold the image.
>>
>> Eventually the goal is to render the images and count all the SWNTs  
>> on the surface, which I can do manually. But on some images there  
>> are so many SWNTs that it has taken me 6 hours for just one image  
>> to count manually. I did this using the ROI manager, by marking all  
>> the SWNTs with lines, then applying a multi measure, and outputting  
>> the data. But this is extremely tedious, needless to say.
>>
>> I would be really happy if someone could help me with flattening  
>> out the images so I can then apply a similar technique like  
>> particle counting to the images. Or if anyone has a better  
>> suggestion for counting particles with high aspect ratios, like  
>> these SWNTs, then I would also be glad to hear your suggestions.
>>
>> I have placed an example image here:
>> http://www.aokengineer.com/swnt-example.html
>> Note: this is now a JPG image, the data i work from is a TIFF  
>> image. I had to convert it to JPG to load it to the website.
>>
>>
>> The eventual goal is to count all the SWNTs and then compare the  
>> images after continued growth, which I think i can do relatively  
>> easy. But so far the bottleneck in my work is getting the images to  
>> Threshold adequately.
>>
>>
>> thanks in advance for your help.
>>
>> alvin
>>
>>
>> --
>> Alvin W. Orbaek, Ph.D. Student. [hidden email]
>> Professor Andrew R. Barron group
>>
>> Rice University, Chemistry Department, MS-60
>> 6100 Main Street, Houston, TX 77005
>> Phone: [phone], Fax: [phone]
>> Ext: x3456, Cell: 713-851-2653
>> --

[Bryan Bandli]

Hello Alvin,

Another thing to consider when you are acquiring the images is to
maximize the contrast between the features of interest and the
background.  If you can push the brightness of the features to the
bright end and the background to the dark end using the
brightness/contrast controls at the microscope it might help with the
image processing.  The image won't look balanced as far as the histogram
is concerned, but it will be a better image for processing.

Good luck,
Bryan Bandli

Alvin Orbaek wrote:

> Hi Michael,
>  thanks for your help, I should have mentioned that I also used to
> just eliminate the artifact by cutting it from the image, but still
> when I try threshold it makes no difference. The background forms in a
> circle around where the artifact is in the image, so I still do not
> create an even background.
>
> Good point about the noise in the image, I had never thought of this.
> But a few people have commented on this fact to me, so I think I will
> try acquire data from now on using a slower scan speed. I have also
> started to use 30kV instead of 20kV. Along with using a spot size of 6
> instead of 3.
>
> When I get the chance I may experiment with some of the other settings
> like, working distance and the aperture size.
>
>  Unfortunately I was following the procedure from someone that used to
> count the images by hand so for them it wasn't an issue, even though
> it took them ages to get any statistics to work with.
>
> thanks again for the input.
>
> alvin
>
> On Feb 23, 2010, at 3:46 AM, Michael Schmid wrote:
>
>> Hi Alvin,
>>
>> this looks difficult.
>>
>> The white particle in the image (defect of substrate?) can't be fully
>> removed by filtering because of its sharp edges. The best I could get
>> was with the Fast Filters plugin (type:eliminate maxima, radius=3,
>> preprocessing median, subtract filtered with offset e.g.=20)
>>  http://imagejdocu.tudor.lu/doku.php?id=plugin:filter:fast_filters:start
>>
>> Also, the images are too noisy for thresholding (this is partly due
>> to the jpg artifacts, however; the original data should be better).
>> So you should do milder background subtraction that is less sensitive
>> to noise (Fast filters: larger radius, type:background from minima)
>> Unfortunately, I am not aware of any plugin that enhances linear
>> features - one could write such a plugin based on the principle of
>> the Hough transform, but it would be a hell of work...
>> The usual edge-preserving blur filters are rather poor for thin
>> linear features.
>>
>> Best wishes,
>>
>> Michael
>> ________________________________________________________________
>>
>> On 22 Feb 2010, at 21:15, Alvin Orbaek wrote:
>>
>>> hi everyone,
>>> I'm looking for help with correcting random and uneven backgrounds
>>> from Scanning Electron Microscope (SEM) images of single walled
>>> carbon nanotubes (SWNTs).
>>>
>>> I have been acquiring data that have some anomaly or artifact on the
>>> substrate so I can come back to this place and see changes after I
>>> have completed subsequent growths of SWNTs under different
>>> conditions. However, these artifacts that I use as markers are also
>>> presenting problems when trying to Threshold the images.
>>>
>>> The artifacts tend make the background subtraction rather difficult,
>>> i think this is physically due to a large difference in secondary
>>> electrons around from these locations. The difference is not
>>> noticeable when looking at the images, but become evident when
>>> trying to Threshold. The baseline emanates from the artifacts
>>> instead of from all across the substrate, so i do not get an even
>>> correction.
>>>
>>> I have tried several techniques that i have found on this list
>>> serve, but still I am not able to normalize the background, and
>>> successfully Threshold the image.
>>>
>>> Eventually the goal is to render the images and count all the SWNTs
>>> on the surface, which I can do manually. But on some images there
>>> are so many SWNTs that it has taken me 6 hours for just one image to
>>> count manually. I did this using the ROI manager, by marking all the
>>> SWNTs with lines, then applying a multi measure, and outputting the
>>> data. But this is extremely tedious, needless to say.
>>>
>>> I would be really happy if someone could help me with flattening out
>>> the images so I can then apply a similar technique like particle
>>> counting to the images. Or if anyone has a better suggestion for
>>> counting particles with high aspect ratios, like these SWNTs, then I
>>> would also be glad to hear your suggestions.
>>>
>>> I have placed an example image here:
>>> http://www.aokengineer.com/swnt-example.html
>>> Note: this is now a JPG image, the data i work from is a TIFF image.
>>> I had to convert it to JPG to load it to the website.
>>>
>>>
>>> The eventual goal is to count all the SWNTs and then compare the
>>> images after continued growth, which I think i can do relatively
>>> easy. But so far the bottleneck in my work is getting the images to
>>> Threshold adequately.
>>>
>>>
>>> thanks in advance for your help.
>>>
>>> alvin
>>>
>>>
>>> --
>>> Alvin W. Orbaek, Ph.D. Student. [hidden email]
>>> Professor Andrew R. Barron group
>>>
>>> Rice University, Chemistry Department, MS-60
>>> 6100 Main Street, Houston, TX 77005
>>> Phone: [phone], Fax: [phone]
>>> Ext: x3456, Cell: 713-851-2653
>>> --
>

--
~~~~~~~~~~~~~~~~~~~~
Bryan Bandli
SEM Laboratory Manager
University of Minnesota Duluth
Life Science 93

Mail:
UMD Geological Sciences
229 Heller Hall
1114 Kirby Dr.
Duluth, MN  55812-3036

Phone: [phone]
Fax:   [phone]

www.d.umn.edu/SEM/