denoising methods to distinguish between system noise and real signal

[Haitham Shaban]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Dear list members,

I am using particle image velocimetry (PIV) methods to estimate velocity fields between image pair from confocal microscopy, which is usually subject to a mixture of noise (basically Gaussian and Poisson). As the motion estimation is very sensible to noise, are there any denoising methods which can be used to distinguish between noise and real movement?, especially when the motion is expected to be small.

I already tried with fixed samples and the noise signal is still giving velocity fields (close to the real motion). I thought in a direction like the following: calculate the temporal derivative of image 1 and 2 (in the simplest case, subtract the images), if the resulting image contains only noise, don't proceed it. However, this does not denoise images.
 Also, I don't know how to distinguish if the resulting image is noise or not?.

Thank you
Haitham

[Tim Feinstein]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Hi Haitham,

One very (VERY) rudimentary approach that I have used for this problem is apply 3D Gaussian blur to the image series in Fiji. Generally I start with small settings like 0.7/0.7/0.7 and increase them as needed; higher blur settings will smooth noise more aggressively but moving objects will become much more blurred.  Gaussian 3D filtering will significantly reduce noise while persistent and slow-moving objects will stand out much better from the background.  The trick is you need either a slow-moving target or very fast acquisition.  If your target moves its own diameter or more per frame then it will get filtered out as noise.  

I should caution that we mostly used this trick to visually identify rhythmically moving structures (beating cilia) rather than to do proper PIV analysis. For quantitative experiments your mileage may vary.  

Best,


Tim

Timothy Feinstein, Ph.D.
Research Scientist
University of Pittsburgh Department of Developmental Biology


On 5/30/17, 12:35 PM, "Confocal Microscopy List on behalf of Haitham Shaban" <[hidden email] on behalf of [hidden email]> wrote:

    *****
    To join, leave or search the confocal microscopy listserv, go to:
    https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Flists.umn.edu%2Fcgi-bin%2Fwa%3FA0%3Dconfocalmicroscopy&data=01%7C01%7Ctnf8%40PITT.EDU%7C7d7bd1896538490f326608d4a779e4ef%7C9ef9f489e0a04eeb87cc3a526112fd0d%7C1&sdata=MzMYg%2FFCAluayOpzdY3DFP5TOcMXQGV%2BxrBDIpfmrW8%3D&reserved=0
    Post images on https://na01.safelinks.protection.outlook.com/?url=http%3A%2F%2Fwww.imgur.com&data=01%7C01%7Ctnf8%40PITT.EDU%7C7d7bd1896538490f326608d4a779e4ef%7C9ef9f489e0a04eeb87cc3a526112fd0d%7C1&sdata=MY1tlwFbnH%2FpJTW1EYL%2Bn88eZUXeDBLhHOWDW9fiXPU%3D&reserved=0 and include the link in your posting.
    *****
   
    Dear list members,
   
    I am using particle image velocimetry (PIV) methods to estimate velocity fields between image pair from confocal microscopy, which is usually subject to a mixture of noise (basically Gaussian and Poisson). As the motion estimation is very sensible to noise, are there any denoising methods which can be used to distinguish between noise and real movement?, especially when the motion is expected to be small.
   
    I already tried with fixed samples and the noise signal is still giving velocity fields (close to the real motion). I thought in a direction like the following: calculate the temporal derivative of image 1 and 2 (in the simplest case, subtract the images), if the resulting image contains only noise, don't proceed it. However, this does not denoise images.
     Also, I don't know how to distinguish if the resulting image is noise or not?.
   
    Thank you
    Haitham
   

[Davide Calebiro]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

I think the best would be to use a band-pass filter in Fourier Space. You should be able to do that with ImageJ.

Best,
Davide

Prof. Davide Calebiro MD PhD DSc
Institute of Pharmacology and Toxicology & Bio-Imaging Center, University of Würzburg
Institute of Metabolism and Systems Research, University of Birmingham
Versbacher Str. 9
97078 Würzburg
Germany
Tel. [phone]
Fax. [phone]

[Haitham Shaban]

*****
To join, leave or search the confocal microscopy listserv, go to:
http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
Post images on http://www.imgur.com and include the link in your posting.
*****

Thanks Timothy and David for your suggestions,

After I tried both suggestions, the results came out that 3D Gaussian blur
method is more appropriate in my experiment.

Best
Haitham

On Wed, May 31, 2017 at 9:47 AM, Davide Calebiro <[hidden email]>
wrote:

> *****
> To join, leave or search the confocal microscopy listserv, go to:
> http://lists.umn.edu/cgi-bin/wa?A0=confocalmicroscopy
> Post images on http://www.imgur.com and include the link in your posting.
> *****
>
> I think the best would be to use a band-pass filter in Fourier Space. You
> should be able to do that with ImageJ.
>
> Best,
> Davide
>
> Prof. Davide Calebiro MD PhD DSc
> Institute of Pharmacology and Toxicology & Bio-Imaging Center, University
> of Würzburg
> Institute of Metabolism and Systems Research, University of Birmingham
> Versbacher Str. 9
> 97078 Würzburg
> Germany
> Tel. [phone]
> Fax. [phone]
>