DoG filter for denoising

[fabrice senger-2]

Hi there,

I'm using a DoG filter for the purpose of image denoising. Has some one a
good technical paper about the use of this filter for denoising (not edge
detection). I found some Wikipedia infos but I would appreciate something
more specific on denoising.

As far I now this filter behaves linearly so it is possible to still do
intensity measurements. In my case i'm looking at the evolution of my
signal over time.

Thank you for your support,

Fabrice.

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

[Herbie]

Good day Fabrice,

to me it is not perfectly clear what you intend to do.

DoG means difference of Gaussians and this implies that this linear
filter will--in simple terms--increase contrasts. In fact it acts as an
isotropic and band-limited LaPlace-filter. The latter produces the
second derivative of an image and, as a spatially discrete operator,
can be formulated a like:

0  1  0
1 -4  1
0  1  0

If you apply this operator to an image (convolution) you will see that
it extracts edges. The same holds for isotropic versions such as

1  1  1
1 -8  1
1  1  1

and finally for band-limited versions such as the DoG-operators.

The disadvantage of the LaPlace-operator is, that it not only extracts
edges but enhances noise as well. Band-limitation then will reduce the
enhancement of high frequency noise. However the main purpose is to
extract/enhance edges.

DoG-filtering is _not_ for noise reduction, if you are not interested in
edge-extraction/enhancement.

You may experiment with test patterns to get an impression of these type
of differentiating operations.

HTH

Herbie

::::::::::::::::::::::::::::::::::::::::::::
Am 02.02.17 um 13:03 schrieb Fabrice Senger:

> Hi there,
>
> I'm using a DoG filter for the purpose of image denoising. Has some one a
> good technical paper about the use of this filter for denoising (not edge
> detection). I found some Wikipedia infos but I would appreciate something
> more specific on denoising.
>
> As far I now this filter behaves linearly so it is possible to still do
> intensity measurements. In my case i'm looking at the evolution of my
> signal over time.
>
> Thank you for your support,
>
> Fabrice.
>
> --
> ImageJ mailing list: http://imagej.nih.gov/ij/list.html
>

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

[Jeremy Adler-2]

Unless you believe that the noise varies over time why try to remove it when making an intensity measurement, presumably from a ROI that contains many pixels. The noise averages out.
Any filtering requires that you can identify a specific type of noise - say occasional high intensity pixels and aim to remove that type of noise.
A Gaussian filter smooths the image, potentially making  more pleasing to the eye, but does not alter the sum of the intensities in a ROI. DoG involves subtracting images smoothed over different radii to highlight edges - it will remove any low frequency components and affect the intensity summed or averaged over a ROI.

Try the easy way - without filtering.

-----Original Message-----
From: ImageJ Interest Group [mailto:[hidden email]] On Behalf Of Fabrice Senger
Sent: den 2 februari 2017 13:03
To: [hidden email]
Subject: DoG filter for denoising

Hi there,

I'm using a DoG filter for the purpose of image denoising. Has some one a good technical paper about the use of this filter for denoising (not edge detection). I found some Wikipedia infos but I would appreciate something more specific on denoising.

As far I now this filter behaves linearly so it is possible to still do intensity measurements. In my case i'm looking at the evolution of my signal over time.

Thank you for your support,

Fabrice.

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

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