I noticed that with Gaussian blur, the Radius setting in GIMP means something different than in Photoshop CS3.

As a test, I made a black square on a white background and used Gaussian blur on it with Photoshop CS3 and the current development version of GIMP with radius 6.0 pixels. Photoshop blurs the image much more heavily than GIMP; to get the same effect with GIMP, I had to set the radius to approximately 19.0 pixels.

Is Gaussian blur not a standard algorithm that has a well-defined meaning for the radius? See for example http://en.wikipedia.org/wiki/Gaussian_blur

If it is, then which one is doing it wrong, CS3 or GIMP?

Hello David,

Both Photoshop CS3 and GIMP are set to pixels.

Jesper

2007/11/20, David Gowers :

Hi Jespar,
On Nov 20, 2007 9:23 PM, Jesper de Jong wrote:

I noticed that with Gaussian blur, the Radius setting in GIMP means something different than in Photoshop CS3.

As a test, I made a black square on a white background and used Gaussian blur on it with Photoshop CS3 and the current development version of

GIMP

with radius 6.0 pixels. Photoshop blurs the image much more heavily than GIMP; to get the same effect with GIMP, I had to set the radius to approximately 19.0 pixels.

Have you checked that the radius is in pixels in BOTH cases, and not in inches or points?

Is Gaussian blur not a standard algorithm that has a well-defined

meaning

for the radius? See for example

http://en.wikipedia.org/wiki/Gaussian_blur

If it is, then which one is doing it wrong, CS3 or GIMP?

I understand that the amount of blurring is determined by the standard deviation. So the conclusion is that there is no standard way to compute the standard deviation from the radius parameter that you specify for gaussian blur, and Photoshop and GIMP do this with different formulas - so it's not the case that either Photoshop or GIMP is wrong.

I looked in the sources and found the formula that GIMP uses:

plug-ins/common/gauss.c:932: vert = fabs (vert) + 1.0; plug-ins/common/gauss.c:933: std_dev = sqrt (-(vert * vert) / (2 * log (1.0/ 255.0)));

(and the same for horizontal about 120 lines below).

Thanks, Jesper

2007/11/20, Øyvind Kolås :

On Nov 20, 2007 12:05 PM, Jesper de Jong wrote:

Is Gaussian blur not a standard algorithm that has a well-defined

meaning

for the radius? See for example

http://en.wikipedia.org/wiki/Gaussian_blur

If it is, then which one is doing it wrong, CS3 or GIMP?

The amount of blurring achieved by a gaussian blur is specified by the standard deviation of the blurring kernel not the size of the blurring kernel. This probably means that GIMP and Photoshop use different ways of computing the standard deviation used from the user specified radius.

Using the size of the kernel (which probably is ~2.5x the standard deviation) might be leading to behavior similar to what GIMP is doing.

/Øyvind K. --
«The future is already here. It's just not very evenly distributed» -- William Gibson http://pippin.gimp.org/ http://ffii.org/