Posted by
Michael Schmid on
May 05, 2007; 10:07am
URL: http://imagej.273.s1.nabble.com/Gaussian-filter-tp3699583p3699585.html
Hi Jarek,
sorry, I don't fully understand your question.
If you truncate a Gaussian filter with sigma=35 pxl down to
kernel size 33x33 it won't have much similarities with a
Gaussian filter any more, it will almost be a uniform kernel
(similar result as ImageJ Process>Filter>Mean).
ImageJ's Gaussian Blur command currently uses a kernel radius
of 2.5 sigma, so a Gaussian filter with sigma=35 pxl will
seem to have a 176x176 kernel. Actually, it uses two
convolutions, one by a 176x1 and one by a 1x176 kernel.
ImageJ 1.38r, expected for next week or so, will have a smoother
kernel (truncating a Gaussian at 2.5 sigma still leads to a
noticable step), corresponding to roughly sigma=3.5 to 4.
So, in the current version of ImageJ, if you want a certain
value of the standard deviation sigma, you have to enter 2.5
times that value as a radius. With ImageJ 1.38r, you have to
enter the standard deviation directly, and ImageJ will calculate
an appropriate kernel size.
Michael
_______________________________________________________________
On Fri, 4 May 2007 21:33:57 +0200 Jarek Korzan <
[hidden email]> wrote:
>How to get a 1.Gaussian filter (kernel size= 30x30)
>and Gausian filter (Sigma =35 pixels, kernel size=33x 33)
>JKorzan
>