AimsImageSmoothing

AIMS library and commands usage

Moderators: denghien, riviere

Post Reply
guillaume
Posts: 44
Joined: Wed Mar 15, 2006 10:29 am
Location: Institut des Neurosciences de la Timone
Contact:

AimsImageSmoothing

Post by guillaume »

Hello everybody,

when i apply a gaussian smoothing to a volume, the resulting image shows multiple voxels at around +or- 10^ -40.
( comand : ~/SHFJ_pack-stable-linux-3.0.2/bin/AimsImageSmoothing -i myVol.ima -o myVolSmooth.ima -t 2 -s 0)


I tried to convert my volume from FLOAT to DOUBLE (this looks like a cast error), but AimsImageSmoothing doesn't take DOUBLE.
The problem already occurs with ohter version i've installed (more recent is cartopack-Fedora-4-i686-3.1.3-2008_10_01)

Don't you know what goes wrong?

thanks
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: AimsImageSmoothing

Post by Jean-Francois Mangin »

I do not thing anything goes wrong. As far as I am aware, this command is based
on the recursive implementation of R. Deriche, which is a very efficient way
to approximate Gaussian smoothing. Hence getting this +/- epsilon in some places
is possible if you smooth a binary image. If it is a pb for you, you can perform
some thresholding afterwards?
guillaume
Posts: 44
Joined: Wed Mar 15, 2006 10:29 am
Location: Institut des Neurosciences de la Timone
Contact:

Re: AimsImageSmoothing

Post by guillaume »

Hi Jeff,
The thresholding solves this issue, but it is strange to obtain negative values by smoothing a non negative mask...
Matthieu said to me that it comes from numerical approximations, as you confirm.

Thank you!
User avatar
Jean-Francois Mangin
Posts: 337
Joined: Mon Mar 01, 2004 10:24 am
Location: Neurospin, CEA, France
Contact:

Re: AimsImageSmoothing

Post by Jean-Francois Mangin »

Not so strange, if you resample an image with a cubic spline you get worse...
Post Reply